Datasets:

bzantium

/

LongPerplexity

like 0

from being a scientific curiosity only a few years ago , energy harvesting ( eh ) is well on its way to becoming a game - changing technology in the field of self - sustainable , autonomous wireless networked systems . a major factor that has contributed to the growth of the eh market is the evolution of ultra - low power electronics , which can run on the minuscule amounts of energy supplied by typical solar , vibration or thermal energy harvesters @xcite . a number of companies are already offering system solutions consisting exclusively of eh sensor nodes @xcite . stimulated by these advances , eh - powered wireless communications have attracted growing interest in recent years @xcite . different from traditional communication systems , eh from environmental sources shifts the paradigm on resource allocation from reducing energy consumption to the most efficient utilization of opportunistic energy . energy availability constraints are imposed such that the energy accumulatively consumed up to any time can not exceed what has been accumulatively harvested so far . considering these new type of constraints , optimal transmission policies were characterized for point - to - point channels in @xcite , for broadcast channels in @xcite , for multi - access channels in @xcite , for interference channels in @xcite , for two - hop relay channels in @xcite , for transmitters with non - ideal circuit - power consumption @xcite , and for systems with battery imperfections in @xcite . existing works @xcite on eh communications mostly addressed offline optimizations , where the eh profiles were assumed to be known _ a priori_. in practical scenarios , complete predictability of eh profiles is clearly an over - simplified assumption . relying on past realizations of eh processes and certain statistics of their future evolutions , @xcite developed some heuristic online algorithms , which , however , lack strong analytical performance guarantees . by modeling the eh and/or data processes as markov processes , online optimizations were cast as markov decision problems ( mdp ) and numerically solved with dynamic programming tools in @xcite . however , the well - known `` curse - of - dimensionality '' with such solutions precludes their application for all but the simplest practical networks . leveraging stochastic optimization tools , a few low - complexity online schemes were developed in @xcite . considering the energy - aware routing with energy replenishment at nodes , @xcite proposed an algorithm that achieves a logarithmic competitive ratio and is asymptotically optimal as the network size grows . a simple asymptotically optimal joint energy allocation and routing scheme was developed for rechargeable sensor networks with static and non - interfering ( i.e. , orthogonal ) links in @xcite . relying on the lyapunov optimization techniques , @xcite developed and analyzed the utility - optimal resource scheduling schemes for general eh - powered multi - hop wireless networks , while @xcite addressed the effect of finite energy and data storage capacities on such resource allocation tasks . the schemes in @xcite assumed ideal energy storage devices ( i.e. , batteries ) in use . under this assumption , the energy - queue sizes at the batteries can play the role of `` stochastic '' lagrange multipliers to develop a dual - subgradient based solver to the intended problems . however , the imperfections with practical batteries could disable this approach . in this paper , we consider a practical battery model accounting for finite battery capacity , energy ( dis-)charging loss , and energy dissipation over time . by integrating and generalizing the lyapunov optimization techniques in @xcite , we re - establish a systematic framework to develop and analyze the stochastic online control schemes for eh wireless networks with such imperfect batteries . specifically , we propose a data - backpressure based scheduling and degenerated energy - queue based power allocation scheme that can maximize the utility of data rates for eh multi - hop wireless networks , without requiring any statistical knowledge of the stochastic channel , data - traffic , and eh processes . different from @xcite where an eh admission mechanism is performed to ensure finite energy queues , we apply the sample path analysis in @xcite to derive the conditions that the proposed scheme is feasible for any given finite battery capacities without eh admission , which can help fully exploit the available harvested energy . in addition , we rigorously establish the performance guarantees of the proposed scheme in form of sub - optimality bounds in the presence of practical battery imperfections . numerical results demonstrate that the proposed scheme can significantly outperform the existing alternatives . the rest of the paper is organized as follows . the system models are described in section ii . the proposed dynamic resource management scheme is developed and analyzed in section iii . numerical results are provided in section iv , followed by concluding remarks in section v. as with @xcite , we consider a general eh multi - hop wireless network that operates in slotted time . for convenience , the slot duration is normalized to unity ; thus , the terms `` energy '' and `` power '' can be sometimes used interchangeably . the network is represented by a graph @xmath0 , where @xmath1 denotes the set of network nodes , and @xmath2 , \ ; n , m \in { \cal n}\}$ ] collects the directed links between nodes . for each node @xmath3 , define two sets of neighbor nodes @xmath4 \in { \cal l}\}$ ] , and @xmath5 \in { \cal l}\}$ ] . further define @xmath6 as the maximum in- and out - degree for nodes in the network . the network delivers packets for data flows ( called commodity in @xcite ) indexed by their destination nodes @xmath7 . per time slot @xmath8 , a data admission is implemented by the network to decide the number @xmath9 of packets for flow @xmath7 that can be newly admitted at node @xmath10 . we assume that @xmath11 with some finite @xmath12 at all time . every node in the network is assumed to be capable of harvesting energy from the environmental ( e.g. , renewable ) sources to power its transmissions . the amount of harvested energy is clearly random and intermittent over time . let @xmath13 denote the amount of harvested energy by node @xmath10 at time slot @xmath8 , and let @xmath14 $ ] be called the _ energy state _ at @xmath8 . we assume that @xmath15 takes values in some finite set and there exists @xmath16 such that @xmath17 per slot @xmath8 , let @xmath18 denote time - varying , random _ channel state _ , which in general can be a @xmath19-by-@xmath19 matrix where the @xmath20 component denotes the channel condition between nodes @xmath10 and @xmath21 . we assume that @xmath18 take values in some finite set for all time . given @xmath18 , the network allocates a power vector @xmath22}(t ) , \ ; \forall [ n , m ] \in { \cal l}]$ ] for data transmissions over links , where @xmath23}(t)$ ] denotes the power allocated to node @xmath10 for link @xmath24 $ ] at time @xmath8 . we assume that each node has a peak power constraint such that @xmath25}(t ) \leq p_{\max } , \quad \forall n\ ] ] with a constant @xmath26 at all time . given the channel state @xmath18 and the power allocation vector @xmath27 , the transmission rate over the link @xmath24 $ ] is dictated by a rate - power function @xmath28}(t ) = \mu_{[n , m ] } ( \boldsymbol{s}(t ) , \boldsymbol{p}(t)).\ ] ] for any bounded power allocation @xmath27 , we also assume that there exists a finite constant @xmath29 such that @xmath30}(t ) \leq \mu_{\max } , \quad \forall [ n , m ] \in { \cal l}\ ] ] for all time under any channel state @xmath18 . now let @xmath31}^c(t)$ ] denote the rate allocated to the data flow @xmath7 over link @xmath24 $ ] at time @xmath8 . it is clear that we have : @xmath32}^c(t ) \leq \mu_{[n , m]}(t ) , \quad \forall [ n , m].\ ] ] every node in the network has an energy storage device , i.e. , battery , to save the harvested energy . consider a practical battery with : i ) a finite capacity , ii ) ( dis-)charging loss , and iii ) energy degeneration . let @xmath41 denote the battery capacity , @xmath42 $ ] the ( dis-)charging efficiency ( e.g. , @xmath43 means that only 90% of the charged or discharged energy is useful while other is conversion loss ) , and @xmath44 $ ] the storage efficiency ( e.g. , @xmath45 means that 10% of the stored energy will be `` leaked '' over a slot , even in the absence of discharging ) . we can model the battery using an energy queue . let @xmath46 denote the energy queue size , which indicates the amount of the energy left in the battery of node @xmath10 at time @xmath8 ; and let @xmath47 $ ] . as the data transmissions are powered by the harvested energy stored in the batteries , the power allocation vector @xmath27 must satisfy the following `` energy availability '' constraint : @xmath48}(t ) \leq \xi \eta e_n(t ) , \quad \forall n\ ] ] where the product @xmath49 captures the discharging loss and energy degeneration . due to the energy availability constraint ( [ eq.ea ] ) , the energy queue @xmath46 evolves according to the dynamic equation : @xmath50}(t)}{\xi } + \xi e_n(t ) , \label{eq.e}\\ & 0 \leq e_n(t ) \leq e_{\max } \label{eq.emax}\end{aligned}\ ] ] with @xmath51 , @xmath52 . define the time - average rate for data flow @xmath7 that is admitted into node @xmath10 , as @xmath53 where the expectation are taken over all sources of randomness . each flow @xmath7 is associated with a utility function @xmath54 , which is assumed to be increasing , continuously differentiable , and strictly concave . let @xmath55 denote the maximum first derivative of @xmath56 , and define @xmath57 which is assumed to be finite . note that the energy state @xmath15 and the channel state @xmath18 are random processes in our model . the eh wireless network to be controlled is thus a stochastic system . the goal is to design an online resource management scheme that chooses the data admission amounts @xmath58 $ ] , the power allocations @xmath59}(t ) , \forall [ n , m]]$ ] , as well as the routing and scheduling decisions @xmath60}^c(t),\forall [ n , m],\forall c]$ ] per slot @xmath8 , so as to maximize the aggregate utility of time - average data rates subject to ( s. t. ) network operation constraints . upon defining @xmath61 , we wish to solve @xmath62 note that here the constraints ( [ eq.rn])([eq.emax ] ) are in fact implicitly required to hold for any realization of the underlying random state @xmath63 per slot @xmath8 . the problem ( [ eq.prob ] ) is a stochastic optimization problem . the problem is challenging as the optimization variables are coupled over time due to the queue dynamics and energy availability constraints in ( [ eq.q])([eq.e ] ) . such problems usually have to be solved by dynamic programming ( dp ) , which suffers from a curse of dimensionality and requires significant knowledge of network statistics . differently , we next integrate and generalize the lyapunov optimization techniques in @xcite to develop a low - complexity online control algorithm , which can be proven to yield a feasible near - optimal solution for ( [ eq.prob ] ) under conditions , without requiring any statistical knowledge of stochastic eh and channel processes . to start , we assume that the rate - power function in ( [ eq.mu ] ) satisfies the following two properties for any given channel state @xmath64 : for two power allocation vectors @xmath65 and @xmath66 , where @xmath66 is obtained by changing any single component @xmath23}$ ] to zero , we have : * @xmath67}(\boldsymbol{s } , \boldsymbol{p } ) \leq \mu_{[n',m']}(\boldsymbol{s } , \boldsymbol{p}')$ ] , @xmath68 \neq [ n , m]$ ] ; * @xmath69}(\boldsymbol{s } , \boldsymbol{p } ) - \mu_{[n , m]}(\boldsymbol{s } , \boldsymbol{p } ' ) \leq \delta_1 p_{[n , m]}$ ] , for a finite constant @xmath70 . for two power allocation vectors @xmath65 and @xmath66 , where @xmath23}'=p_{[n , m]}+\frac{\delta p}{|{\cal n}_n^o|}$ ] , @xmath71 , and @xmath72}'=p_{[n',m]}$ ] , @xmath73 , we have : * @xmath31}(\boldsymbol{s } , \boldsymbol{p } ) \leq \mu_{[n , m]}(\boldsymbol{s } , \boldsymbol{p}')$ ] , @xmath74 ; * @xmath75}(\boldsymbol{s } , \boldsymbol{p } ) - \mu_{[n',m]}(\boldsymbol{s } , \boldsymbol{p } ' ) ] \leq \delta_2 \delta p$ ] , for a finite constant @xmath76 . property 1 was one of the keys for asymptotic optimality analysis of the esa scheme in @xcite . here , properties 1 and 2 will be the keys for our feasibility and optimality gap analysis in the sequel . these properties are actually satisfied by most rate - power functions , e.g. , when the rate function has finite directional derivatives with respect to power , and the rates do not improve with increased interference . to see it , two intriguing examples are provided below . _ example 1 ( interference channels ) _ : consider the general interference channel case . let @xmath77}$ ] denote the channel coefficient from node @xmath10 to node @xmath21 ; i.e. , @xmath78 } , \forall n , m \in { \cal n}]$ ] . the rate function is then @xmath30}= \log \left(1 + \frac{|h_{[n , m]}|^2p_{[n , m]}}{\sum_{[n',m ' ] \neq [ n , m]}(|h_{[n',m]}|^2p_{[n',m ' ] } ) + \sigma^2}\right)\ ] ] where @xmath79 is the noise variance at nodes . it can be readily derived @xmath80}|^2}{\sigma^2 } , \forall [ n , m ] , \;\forall \boldsymbol{s}\ } \\ \delta_2 = \max\{\frac{\sum_{n ' \neq n}\sum_{m ' \in { \cal n}_{n'}^o } |h_{[n , m']}|^2 } { \sigma^2 } , \;\forall \boldsymbol{s}\ } \end{cases}\ ] ] clearly , @xmath81 and @xmath82 are finite as long as all channel gains @xmath83}|^2 $ ] are bounded at all time . this is in fact also the necessary condition for @xmath29 . _ example 2 ( orthogonal channels ) _ : if the wireless links do not interfere with each other ( e.g. , when the adjacent nodes operate on orthogonal frequency bands ) , then the rate function is simply @xmath30}= \log \left(1 + \frac { |h_{[n , m]}|^2p_{[n , m ] } } { \sigma^2}\right).\ ] ] we readily have @xmath80}|^2}{\sigma^2 } , \forall [ n , m ] , \;\forall \boldsymbol{s}\ } \\ \delta_2 = 0 \end{cases}\ ] ] again , @xmath81 is finite if all channel gains @xmath83}|^2 $ ] are bounded at all time . we assume the following two conditions for the system parameters in development of the proposed algorithm : @xmath84 condition ( [ eq.a1 ] ) is a necessary condition to maintain the stability of the energy queues @xmath46 for every sample path . if @xmath85 , i.e. , the maximum energy arrival is deterministically greater than the largest energy departure possible , then there exists a sample path of energy queue @xmath46 that grows unbounded . the condition ( [ eq.a1 ] ) is thus required for establishing the sample path result in the ensuing feasibility analysis . on the other hand , condition ( [ eq.a2 ] ) dictates that the battery capacity is large enough to accommodate the largest possible charging / discharging range . this then makes the system `` controllable '' by the proposed online scheme . ) and ( [ eq.a2 ] ) , we require @xmath86 . this could be ensured by `` shaping '' or `` clipping '' the energy arrivals by some external mechanisms . ] our algorithm depends on two algorithmic parameters , namely a `` queue perturbation '' parameter @xmath87 and a weight parameter @xmath88 . derived from the feasibility requirement of the proposed algorithm ( see the proof of proposition 1 in the sequel ) , any pair @xmath89 that satisfies the following conditions can be used : @xmath90 where & v^ : = ; [ eq.gv3 ] + & ^ : = + _ 1 g _ v ; [ eq.gv1 ] + & ^ : = - _ 2 g _ v. [ eq.gv2 ] note that the interval for @xmath88 in ( [ eq.gv ] ) is well - defined under the condition ( [ eq.a2 ] ) , and the interval for @xmath87 is valid when @xmath91 . we now present the proposed algorithm : * initialization * : select a pair of @xmath89 satisfying ( [ eq.gv ] ) , and a constant @xmath92 . at every time slot @xmath8 , observe states @xmath63 , and queues @xmath93 , then determine @xmath94 $ ] , @xmath95}^*(t ) , \forall [ n , m]]$ ] , and @xmath96}^{c*}(t),\forall [ n , m],\forall c]$ ] as follows . some comments are in order . * different from the esa algorithm in @xcite , there is no eh admission mechanism in the proposed algorithm ; the available harvested energy could be then fully capitalized on for data transmission . * the perturbed energy queue - size @xmath107 is weighted by @xmath108 in the problem ( [ eq.pa ] ) to determine the optimal power allocation . these weighted are used to account for the battery degeneration and discharging loss . * solving the power allocation problem ( [ eq.pa ] ) requires centralized control and can be np - hard . for the general ad - hoc networks with interference channels , a maximal - weighted - matching based polynomial - time solver @xcite could be used to approximate the optimal scheduling within a constant factor in a distributed manner . the proposed algorithm is an online scheme , which dynamically makes _ instantaneous _ greedy control decisions for the stochastic system under consideration , without a - priori knowledge of any statistics of the underlying random processes . we next show that the proposed scheme can also yield a feasible and asymptotically near - optimal solution for the problem ( [ eq.prob ] ) of interest under conditions . note that in the proposed algorithm , energy availability constraints ( [ eq.ea ] ) and the bounded energy queue constraints ( [ eq.emax ] ) are ignored . it is then not clear whether the algorithm is feasible for the problem ( [ eq.prob ] ) . yet , we will show that by using any pair @xmath89 in ( [ eq.gv ] ) and @xmath109 , it is guaranteed that the online control policy produced by the proposed algorithm is a feasible one for ( [ eq.prob ] ) under the conditions ( [ eq.a1])([eq.a2 ] ) . to this end , we first show the following lemma . the power allocation specified by the proposed algorithm obeys : i ) @xmath110}^*(t)=0 $ ] , if @xmath111 ; and ii ) @xmath110}^*(t)=p_{\max}$ ] , if @xmath112 , @xmath52 . see appendix a. lemma 1 reveals partial characteristics of the proposed dynamic policy . specifically , when the energy queue at node @xmath10 is large enough , peak power can be afforded for its data transmissions ; i.e. , @xmath110}^*(t)=p_{\max}$ ] . on the other hand , when the energy queue at node @xmath10 is small enough , no power should be allocated ; i.e. , @xmath110}^*(t)=0 $ ] . based on the structure in lemma 1 , we can then establish the following result . under the conditions ( [ eq.a1])([eq.a2 ] ) , the proposed algorithm guarantees : i ) @xmath110}^*(t)=0 $ ] , if @xmath113 , and ii ) @xmath114 , @xmath52 , @xmath39 . see appendix b. proposition 1 establishes that node @xmath10 allocates nonzero power over any of its outgoing links only when its energy queue - size @xmath115 in the proposed scheme . hence , all the power allocation decisions are feasible , i.e. , the energy availability constraint ( [ eq.ea ] ) is indeed _ redundant _ and can be ignored in the problem ( [ eq.pa ] ) . furthermore , the energy queue - sizes @xmath46 resulted from the proposed scheme are guaranteed to be bounded within @xmath116 $ ] . these two results together imply that the proposed algorithm with proper selection of @xmath89 and @xmath117 can always yield a feasible control policy for ( [ eq.prob ] ) under the conditions ( [ eq.a1])([eq.a2 ] ) . note that proposition 1 is a _ sample path _ result ; i.e. , it holds for every time slots under _ arbitrary _ , even non - stationary , @xmath63 processes . to facilitate the optimality analysis , we assume for now that the random process for @xmath63 is i.i.d . over time slots . define @xmath118 , and @xmath119}(t)\ } \notag\end{aligned}\ ] ] since @xmath46 , @xmath110}(t)$ ] , and @xmath13 are all bounded , it follows from ( [ eq.e ] ) that @xmath120 as @xmath121 $ ] , @xmath39 , ( [ relax1 ] ) then implies @xmath122 for the queue dynamics in ( [ eq.q ] ) , we can derive similar time - average constraints @xcite : @xmath123}^c \leq \sum_{m \in { \cal n}_n^o } \bar{\mu}_{[n , m]}^c \quad \forall n , c\ ] ] where @xmath124}^c:=\lim_{t\rightarrow \infty } \frac{1}{t } \sum_{t=0}^{t-1 } \mathbb{e}\{\mu_{[n , m]}^c(t)\}.\ ] ] consider the following problem : @xmath125 note that the queue dynamic constraints ( [ eq.e ] ) and ( [ eq.q ] ) , which need to be performed per realization per slot @xmath8 , are replaced by the corresponding time - average constraints ( [ relax2 ] ) and ( [ relax3 ] ) , and the constraints ( [ eq.ea ] ) and ( [ eq.emax ] ) are ignored . it can be shown that the problem ( [ eq.relax ] ) is a relaxed version of ( [ eq.prob ] ) . specifically , any feasible solution of ( [ eq.prob ] ) , satisfying ( [ eq.q])([eq.emax ] ) , @xmath39 , also satisfies ( [ relax2 ] ) and ( [ relax3 ] ) , due to the relevant optimization variables . it then follows that @xmath126 . in ( [ eq.relax ] ) , the optimization variables are `` decoupled '' across time slots due to the removal of time - coupling constraints ( [ eq.q])([eq.emax ] ) . this problem has an easy - to - characterize stationary optimal control policy as formally stated in the next lemma . if @xmath63 is i.i.d . , there exists a stationary control policy @xmath127 that is a pure ( possibly randomized ) function of the current @xmath63 , while satisfying ( [ eq.rn])([eq.muc ] ) , and providing the following guarantees per @xmath8 : @xmath128}^{stat}(t)\}}{\xi } \leq ( 1-\eta)e_{\max } , \ ; \forall n \\ & \bar{r}_n^{c , stat } + \sum_{m \in { \cal n}_n^i } \mathbb{e}\{\mu_{[m , n]}^{c , stat}(t)\ } \leq \sum_{m \in { \cal n}_n^o}\mathbb{e}\{\mu_{[n , m]}^{c , stat}(t)\ } , \ ; \forall n , c \end{split}\ ] ] where @xmath129 , @xmath23}^{stat}(t)$ ] , and @xmath31}^{c , stat}(t)$ ] denote the data admission , power and rate allocation by policy @xmath127 , and expectations are taken over the randomization of @xmath63 and ( possibly ) @xmath127 . the proof argument is similar to the proof of ( * ? ? ? * theorem 4.5 ) ; hence , it is omitted for brevity . lemma 2 in fact holds for many non - i.i.d . scenarios as well . generalizations to other stationary processes , or even to non - stationary processes , can be found in @xcite . it is worth noting that ( [ eq.opt ] ) not only assures that the stationary control policy @xmath127 achieves the optimal utility for ( [ eq.relax ] ) , but also guarantees that the admitted data rate @xmath129 per slot @xmath8 is equal to the optimal time - average rate @xmath130 ( due to the stationarity of @xmath63 and @xmath127 ) . this will be important to establish the following result : suppose that conditions ( [ eq.a1])([eq.gv ] ) hold , and @xmath131 is i.i.d . over slots . let @xmath132 be the time - average admitted rate vector achieved by the proposed algorithm up to time @xmath133 , i.e. , @xmath134 . then @xmath135 where the constant @xmath136 with @xmath137 ^ 2 , [ -\xi e_{\max}+(1-\eta)\gamma]^2\ } , \notag \\ b_3 & = \eta(1-\eta)\max\{(e_{\max}-\gamma)^2 , \gamma^2\ } , \notag\end{aligned}\ ] ] and @xmath138 is the optimal value of ( [ eq.prob ] ) under any feasible control algorithm , even if that relies on knowing future random realizations . see appendix c. proposition 2 asserts that the proposed algorithm asymptotically yields a time - average utility with an optimality gap smaller than @xmath139 . the proposed scheme is in fact a modified version of the queue - length based stochastic optimization scheme ( e.g. , the esa in @xcite ) , where the `` perturbed '' queue lengths play the role of `` stochastic '' lagrange multipliers with a dual - subgradient solver to the problem of interest . the gap @xmath140 is inherited from the underlying stochastic subgradient method . on the other hand , the gap @xmath141 is due to the combined effect of energy - queue perturbation and battery imperfections , while the gap @xmath142 is incurred by the battery degeneration . based on propositions 1 and 2 , it is now possible to arrive at our main result . [ them : main ] suppose that conditions ( [ eq.a1])([eq.gv ] ) hold and @xmath131 is i.i.d . over slots . then the proposed algorithm yields a feasible dynamic control scheme for , which has an optimality gap @xmath139 ; i.e. , @xmath143 where @xmath134 and @xmath144 is given by ( [ eq.b ] ) . interesting comments on the minimum optimality gap with the proposed algorithm are in order . the results in theorem 1 can be generalized to the more general non - i.i.d . case where the the energy state @xmath15 and channel state @xmath18 both evolve according to some finite - state irreducible , aperiodic markov chains . note that the feasibility of the proposed algorithm is a sample path result which does not depend on the assumption on the random processes per remark 2 ; i.e. , it holds also for this non - i.i.d . case . on the other hand , the performance guarantee ( i.e. , optimality gap result ) can be obtained by applying the so - called delayed lyapunov drift technique , i.e. , the method of analyzing regenerative cycles of the markov random processes ; see ( * ? ? ? * theorem 3 ) and ( * ? ? ? * theorem 2 ) . in this section , numerical results are provided to evaluate the proposed algorithm . a simple network is considered in fig . 1 , where the nodes 1 - 4 collect data and send data to the sink node 7 through relay nodes 5 and 6 @xcite . in simulations , we assume imperfect batteries at nodes , with storage efficiency @xmath164 , and ( dis-)charging efficiency @xmath165 . table [ tab : param ] lists the values for @xmath166 ( the maximum in- and out - degree for nodes in the network ) , @xmath167 ( the maximum packets that can be newly admitted ) , @xmath168 ( the peak power ) , @xmath169 ( the maximum rate over all the links ) and @xmath170 ( the battery capacity ) . the utility function is selected as : @xmath171 . suppose that all the links are independent with each other , implying @xmath172 . the link state can be either good or bad with equal probability . one unit of power can deliver two packets when the link state is good , while it can be only used to transmit one packet upon bad link state . we also assume that the harvested energy @xmath13 is i.i.d . for each node ; @xmath13 is either @xmath173 or 0 with equal probability . as a result , we have : @xmath174 , @xmath175 and @xmath176 . to satisfy the constraint ( [ eq.a1 ] ) , we set @xmath177 , unless otherwise specified . each simulation is run for 1200 slots and each result is obtained by averaging over 10 independent runs . 2 depicts the achieved utility for the proposed algorithm with different @xmath87 , when the weight parameter @xmath178 , @xmath179 and @xmath180 . it is shown that the maximum utility is achieved when the queue perturbation " parameter @xmath87 is set to @xmath181 in this case . this is because power allocation policy ( 17 ) tends to save energy if a large @xmath87 is chosen , leading to a large data queue size @xmath182 . on the other hand , data admission policy ( [ eq.da ] ) dictates that @xmath183 , if @xmath184 , @xmath185 . this implies that large @xmath182 reduces rate utility . as a result , the utility decreases as @xmath87 grows . for this reason , we set @xmath186 unless otherwise specified in all the subsequent simulations . note that when @xmath187 and @xmath188 , we often have @xmath23}(t)=0 , \forall n , \forall t$ ] ; i.e. , no data are allowed to be transmitted over links . in this case , to maintain the queue stability , network utility becomes almost 0 . [ fig : weight ] shows the network utility of the proposed algorithm with different weight parameter @xmath88 , when @xmath179 and @xmath180 . it is observed that the network utility decreases after reaching the maximum value for a given @xmath189 . clearly , the proposed algorithm does not perform well with a small @xmath88 . on the other hand , a large @xmath88 is not a good choice in the presence of battery imperfections . to see it , we use @xmath181 in ( 14 ) and set @xmath31}(t ) = s_{[n , m]}(t ) p_{[n , m]}(t)$ ] in ( 17 ) . ( @xmath190}(t)$ ] is 2 with good link state , while it is 1 when the link state is bad . ) then power allocation problem ( 17 ) is a linear program , where the weight for @xmath23}(t)$ ] is given by @xmath101}(t)s_{[n , m]}(t)+ \eta e_n(t)-p_{\max}-\delta_1 g_{\max } v$ ] . clearly , we have @xmath23}(t)=0 $ ] , if its weight is less than zero . therefore , a large @xmath88 implies that high energy is required to render @xmath23}(t)>0 $ ] for transmission . however , it is hard to maintain high energy in the presence of energy degeneration . as a result , when @xmath187 and @xmath191 , utility is almost 0 . the results are consist with our discussion in remark 4 . the optimality gap with the proposed scheme does not monotonically decrease with @xmath88 any more in the presence of battery imperfections . determination of the best @xmath88 for the algorithm may refer to the solution of the problem formulated in remark 4-ii ) . .parameters configuration . [ cols="^,^,^,^,^",options="header " , ] .,width=336,height=192 ] .,width=336,height=192 ] the data and energy queue processes with the proposed algorithm are plotted in fig . [ fig : samplepro ] , when @xmath179 , @xmath192 and @xmath178 . clearly , the three data queue lines evolve within the region [ 0 , @xmath193 ; i.e. , the stability of the data network is maintained . in addition , all energy queues satisfy @xmath194 . this clearly corroborates the feasibility of the proposed scheme , as stated in proposition 1 . two baseline schemes are chosen to gauge the performance of the proposed algorithm . the first baseline scheme is esa in @xcite , where perfect batteries are assumed and an eh admission is performed to ensure finite energy queues . the second scheme is a heuristic greedy algorithm , where a node with the largest data queue backlog is first allowed to transmit data over its best outgoing links ( with best link quality ) . the node with second largest data queue then selects a best link to transmit , among all its outgoing links that do not conflict with the existing links . ( a link conflicts with the existing links if it shares the same transmit or receive node with any existing links . ) this continues until no nodes or no links can be selected . if a link @xmath24 $ ] is selected , the power @xmath23}(t)$ ] is set as the maximum value that satisfies both the peak power constraint ( 2 ) and the energy availability constraint ( 6 ) ; @xmath31}(t)$ ] is then determined correspondingly . finally , the greedy algorithm sets @xmath195 if @xmath196 , while @xmath197}(t)$ ] if @xmath198 . it is clear that the nodes have finite data queue backlogs ; i.e. , the system is stable . the feasibility of the greedy algorithm is also guaranteed since the energy availability constraint ( 6 ) is taken into account in power allocation policy . [ fig : results ] compares the performance of the three algorithms with different @xmath173 , when @xmath199 , @xmath192 and @xmath178 . it is shown that the proposed algorithm evidently outperforms the esa and the greedy algorithm for any given @xmath173 . since the greedy algorithm in fact schedules the links in a time division multi - access ( tdma ) manner , the resultant utility is low in general . in addition , the required energy for implementation of the greedy policy is actually also low ; hence , its performance changes only slightly for different @xmath173 . the proposed algorithm achieves higher utility than esa due to two reasons . the first reason is that esa can not make use of all available energy because of its eh admission mechanism , while the proposed algorithm harvested all available energy . the utilization of available energy ( the ratio of actually harvested energy to available energy ) for esa is depicted in fig . [ fig : uti ] . for esa , node @xmath10 stops harvesting energy when @xmath46 energy reaches @xmath200 @xcite . clearly , given a larger @xmath173 ( i.e. , renewable energy is abundant ) , esa might waste more available energy , leading to a significant performance loss . on the other hand , esa also has a performance loss for small @xmath173 case since it does not take into account the battery imperfections . for instance , when @xmath201 , the utilization of available energy for esa is 100% ; yet , the utility of the proposed algorithm is still 17.2% larger than that of esa in this case . in this paper , a dynamic resource allocation task was considered for general eh wireless networks . taking into account imperfect finite - capacity energy storage devices , a stochastic optimization was formulated to maximize the long - term utility subject to the energy availability constraints . capitalizing on lyapunov optimization technique , an online control algorithm was proposed to make data admission , power allocation and routing decisions , without requiring any statistical knowledge of channel , data - traffic , and eh processes . it was shown that the proposed algorithm can efficiently utilize the harvested energy to provide a feasible and asymptotically near - optimal control solution for general eh networks . using the perturbation weight @xmath92 in queue - backpressure , it was shown in ( * ? ? ? * theorem 2 ) that the data queue sizes satisfy : @xmath202 , @xmath98 , @xmath39 . note that this is a sample path result which holds for arbitrary @xmath15 and @xmath18 processes . * if @xmath205 , suppose @xmath110}^*(t ) > 0 $ ] ; i.e. , there exists a certain @xmath206}^*(t)>0 $ ] with @xmath207 . we can create a new power vector @xmath66 by setting only @xmath206}'=0 $ ] and all other components remain the same as in @xmath102 . let @xmath208 denote the objective function of ( [ eq.pa ] ) . it readily follows that @xmath209}(t)[\mu_{[n , m]}(\boldsymbol{p}^*(t ) ) - \mu_{[n , m]}(\boldsymbol{p } ' ) ] \notag\\ & ~~~~~ + \frac{\eta}{\xi } ( e_n(t)-\gamma ) p_{[n,\check{m}]}^*(t ) \notag \\ & \leq w_{[n,\check{m}]}(t)[\mu_{[n,\check{m}]}(\boldsymbol{p}^*(t ) ) - \mu_{[n,\check{m}]}(\boldsymbol{p } ' ) ] \notag \\ & ~~~~~ + \frac{\eta}{\xi } ( e_n(t)-\gamma ) p_{[n,\check{m}]}^*(t ) \notag \\ & < [ g_{\max}v -d_{\max}\mu_{\max}]^+ \delta_1 p_{[n,\check{m}]}^*(t ) - \delta_1 g_{\max } v p_{[n,\check{m}]}^*(t ) \notag \\ & < 0 \notag \end{aligned}\ ] ] where the first inequality is due to property 1-i ) , and the second inequality is due to property 1-ii ) , as well as the bounds @xmath101}(t ) \leq [ g_{\max}v -d_{\max}\mu_{\max}]^+$ ] and @xmath205 . this shows that @xmath102 can not have been the optimal power allocation for ( [ eq.pa ] ) , leading to a contradiction . hence , part i ) in the lemma must hold . * if @xmath210 , suppose @xmath110}^*(t)<p_{\max}$ ] . let @xmath211}^*(t)$ ] . we can create a new power vector @xmath66 by setting @xmath23}'=p_{[n , m]}^*(t)+\frac{\delta p}{|{\cal n}_n^o|}$ ] , @xmath71 , and @xmath72}'=p_{[n',m]}^*(t)$ ] , @xmath73 . it readily follows that @xmath209}(t)[\mu_{[n , m]}(\boldsymbol{p}^*(t ) ) - \mu_{[n , m]}(\boldsymbol{p } ' ) ] \notag\\ & ~~~~~ - \frac{\eta}{\xi } ( e_n(t)-\gamma ) \delta p \notag \\ & \leq \sum_{n ' \neq n } \sum_{m \in { \cal n}_{n'}^o } w_{[n',m]}(t)[\mu_{[n',m]}(\boldsymbol{p}^*(t ) ) - \mu_{[n',m]}(\boldsymbol{p } ' ) ] \notag \\ & ~~~~~ - \frac{\eta}{\xi } ( e_n(t)-\gamma ) \delta p \notag \\ & < [ g_{\max}v -d_{\max}\mu_{\max}]^+ \delta_2 \delta p - \delta_2 g_{\max } v \delta p \notag \\ & < 0 \notag \end{aligned}\ ] ] where the first inequality is due to property 2-i ) , and the second inequality is due to property 2-ii ) , as well as the bounds @xmath101}(t ) \leq [ g_{\max}v -d_{\max}\mu_{\max}]^+$ ] and @xmath210 . this leads to a contradiction . hence , part ii ) in the lemma must also hold . the proof for part ii ) proceeds by induction . first , recall that @xmath215 , @xmath52 , and suppose that it holds @xmath121 $ ] , @xmath52 , at slot @xmath8 . we will show the bounds hold for @xmath216 , @xmath52 , as well in subsequent instances . * if @xmath219 , then it follows from lemma 1 that @xmath110}^*(t)= 0 $ ] . from ( [ eq.e ] ) , we have * * @xmath220 ; * * @xmath221 , due to @xmath222 , @xmath223 , and the definition of @xmath217 in ( [ eq.gv2 ] ) . * if @xmath224 $ ] , then @xmath225}^*(t ) \leq p_{\max}$ ] . we therefore have * * @xmath226 , due to @xmath227 and the definition of @xmath181 in ( [ eq.gv1 ] ) ; * * @xmath228 , due to @xmath223 as with c1-ii ) . * if @xmath229 $ ] , it follows from lemma 1 that @xmath110}^*(t ) = p_{\max}$ ] . we have * * @xmath230 , due to @xmath227 as with c2-i ) ; * * @xmath231 , due to condition ( [ eq.a1 ] ) . consider the queues under the proposed algorithm . by squaring both sides of ( [ eq.q ] ) , and using the fact that @xmath232^+)^2 \leq x^2 $ ] , we have @xmath233 ^ 2-[q_n^c(t)]^2 \notag \\ & \leq \left[\sum_{m \in { \cal n}_n^o } \mu_{[n , m]}^{c*}(t ) \right]^2 + \left[\sum_{m \in { \cal n}_n^i } \mu_{[m , n]}^{c*}(t ) + r_n^{c*}(t ) \right]^2 \notag \\ & ~~~~ -2q_n^c(t)\left [ \sum_{m \in { \cal n}_n^o } \mu_{[n , m]}^{c*}(t ) - \sum_{m \in { \cal n}_n^i } \mu_{[m , n]}^{c*}(t ) - r_n^{c*}(t)\right ] . \notag\end{aligned}\ ] ] recall that @xmath234}^{c*}(t ) \leq d_{\max } \mu_{\max}$ ] and @xmath235}^{c*}(t ) + r_n^{c*}(t ) \leq d_{\max } \mu_{\max}+r_{\max}$ ] . upon defining @xmath236 , we readily have : @xmath237 ^ 2-[q_n^c(t)]^2 ) \notag \\ & \leq \tilde{b}_1 - q_n^c(t)\left [ \sum_{m \in { \cal n}_n^o } \mu_{[n , m]}^{c*}(t ) - \sum_{m \in { \cal n}_n^i } \mu_{[m , n]}^{c*}(t ) - r_n^{c*}(t)\right ] . \label{eq.dq}\end{aligned}\ ] ] similarly , we can also derive @xmath238 ^ 2-[e_n(t)-\gamma]^2 ) \notag \\ & \leq -\frac{1}{2}(1-\eta^2 ) [ e_n(t)-\gamma]^2 \notag \\ & ~~~~ + \frac{1}{2 } \left[\frac{\sum_{m \in { \cal n}_n^o } p_{[n , m]}^*(t)}{\xi } - \xi e_n(t ) + ( 1-\eta)\gamma\right]^2 \notag\\ & ~~~~ -\eta(e_n(t)-\gamma)\left[\frac{\sum_{m \in { \cal n}_n^o } p_{[n , m]}^*(t)}{\xi } - \xi e_n(t ) + ( 1-\eta)\gamma\right ] \notag \\ & \leq b_2 - \frac{\eta}{\xi}(e_n(t)-\gamma ) \sum_{m \in { \cal n}_n^o } p_{[n , m]}^*(t ) \notag \\ & ~~~~ + \eta ( e_n(t)-\gamma)[\xi e_n(t ) - ( 1-\eta)\gamma ] \label{eq.de}\end{aligned}\ ] ] where @xmath239 ^ 2 , [ -\xi e_{\max}+(1-\eta)\gamma]^2\}$ ] . consider the lyapunov function @xmath240 ^ 2 + \frac{1}{2 } \sum_n [ e_n(t)-\gamma]^2 $ ] . by summing ( [ eq.dq ] ) over all @xmath241 and ( [ eq.de ] ) over all @xmath10 , and by defining @xmath242 , we have @xmath243}^{c*}(t ) - \sum_{m \in { \cal n}_n^i } \mu_{[m , n]}^{c*}(t ) - r_n^{c*}(t)\right ] \notag \\ & ~~~~ - \sum_n \frac{\eta}{\xi}(e_n(t)-\gamma ) \sum_{m \in { \cal n}_n^o } p_{[n , m]}^*(t ) \notag \\ & ~~~~ + \sum_n \eta ( e_n(t)-\gamma)[\xi e_n(t ) - ( 1-\eta)\gamma]\end{aligned}\ ] ] furthermore , we have @xmath244}^{c*}(t ) \notag \\ & \leq \tilde{b } - \sum_{n , c } [ v u_n^c(r_n^{c*}(t ) ) - q_n^c(t ) r_n^{c*}(t ) ] \notag \\ & ~~~~ - \sum_n \bigl[\sum_c \sum_{m \in { \cal n}_n^o } \mu_{[n , m]}^{c*}(t ) [ q_n^c(t ) - q_m^c(t ) - \theta]\bigr . \notag \\ & ~~~~~~~~~~~~~~ \bigl . + \frac{\eta}{\xi}(e_n(t)-\gamma ) \sum_{m \in { \cal n}_n^o } p_{[n , m]}^*(t ) \bigr ] \notag \\ & ~~~~ + \sum_n \eta ( e_n(t)-\gamma)[\xi e_n(t ) - ( 1-\eta)\gamma ] \notag \\ & \leq \tilde{b } - \sum_{n , c } [ v u_n^c(r_n^{c , stat}(t ) ) - q_n^c(t ) r_n^{c , stat}(t ) ] \notag \\ & ~~~~ - \sum_n \bigl[\sum_c \sum_{m \in { \cal n}_n^o } \mu_{[n , m]}^{c , stat}(t ) [ q_n^c(t ) - q_m^c(t ) - \theta]\bigr . \notag \\ & ~~~~~~~~~~~~~~ \bigl . + \frac{\eta}{\xi}(e_n(t)-\gamma ) \sum_{m \in { \cal n}_n^o } p_{[n , m]}^{stat}(t ) \bigr ] \notag \\ & ~~~~ + \sum_n \eta ( e_n(t)-\gamma)[\xi e_n(t ) - ( 1-\eta)\gamma ] \label{eq.rhs}\end{aligned}\ ] ] where the second inequality is due to the fact that the proposed algorithm in fact minimizes the right - hand - side of ( [ eq.rhs ] ) . taking expectations for ( [ eq.rhs ] ) , we arrive at : @xmath245}^{c*}(t)\bigr\ } \notag \\ & \leq \tilde{b } - v \sum_{n , c } u_n^c(\bar{r}_n^{c , stat}(t ) ) \notag \\ & ~~ + q_n^c(t ) \bigl [ \bar{r}_n^{c , stat } + \sum_{m \in { \cal n}_n^i } \mathbb{e}\{\mu_{[m , n]}^{c , stat}(t)\ } - \sum_{m \in { \cal n}_n^o}\mathbb{e}\{\mu_{[n , m]}^{c , stat}(t)\ } \big ] \notag \\ & ~~ + \sum_n \eta ( e_n(t)-\gamma ) \bigl[\xi \bar{e}_n - \frac{\sum_{m \in { \cal n}_n^o } p_{[n , m]}^{stat}(t)}{\xi } - ( 1-\eta)\gamma \bigr ] \notag \\ & ~~ + \theta \sum_n \sum_c \sum_{m \in { \cal n}_n^o } \mathbb{e}\{\mu_{[n , m]}^{c , stat}(t)\ } \notag \\ & \leq \tilde{b } - v \tilde{u}^{opt } + n \eta(1-\eta)\max\{(e_{\max}-\gamma)^2 , \gamma^2\ } \notag \\ & ~~ + n^2 \theta d_{\max}\mu_{\max}\end{aligned}\ ] ] where the second inequality is due to ( [ eq.opt ] ) in lemma 2 , @xmath121 $ ] by proposition 1 , and @xmath246}^{c , stat}(t)\ } \leq d_{\max}\mu_{\max}$ ] . with @xmath247 and noticing @xmath248}^{c*}(t)\ } \geq 0 $ ] , we then have @xmath249 summing over all @xmath250 , we have @xmath251 which leads to @xmath252 using jensen s inequality , we see that @xmath253 the proposition follows by taking the limit @xmath254 and using the definition of @xmath132 . o. ozel , k. tutuncuoglu , y. jing , s. ulukus , and a. yener , `` transmission with energy harvesting nodes in fading wireless channels : optimal policies , '' _ ieee j. sel . areas commun . 8 , pp . 17321743 , sep . 2011 . o. ozel , y. jing , s. ulukus , `` optimal broadcast scheduling for an energy harvesting rechargeable transmitter with a finite capacity battery , '' _ ieee trans . wireless commun . 11 , no . 6 , pp . 21932203 , jun . 2012 . o. ozel , k. shahzad , and s. ulukus , `` optimal energy allocation for energy harvesting transmitters with hybrid energy storage and processing cost , '' _ ieee trans . signal process . 12 , pp . 32323245 , 2014 . x. wang , z. nan , and t. chen , `` optimal mimo broadcasting for energy harvesting transmitter with non - ideal circuit power consumption , '' _ ieee trans . wireless commun . 5 , pp . 25002512 , may 2015 . b. devillers and d. gunduz , `` a general framework for the optimization of energy harvesting communication systems with battery imperfections , '' _ j. commun 2 , pp . 130139 , apr . 2012 . r. srivastava and c. e. koksal , `` basic performance limits and tradeoffs in energy - harvesting sensor nodes with finite data and energy storage , '' _ ieee / acm trans . 4 , pp . 10491062 , 2013 . l. lin , n. shroff , and r. srikant , `` asymptotically optimal energy - aware routing for multihop wireless networks with renewable energy sources , '' _ ieee / acm trans . 5 , pp . 10211034 , 2007 . s. chen , p. sinha , n. shroff , and c. joo , `` a simple asymptocially optimal joint energy allocation and routing schemes in rechargable sensor networks , '' _ ieee / acm trans . netw . _ , vol . 4 , pp . 13251336 , aug . 2014 . z. mao , c. koksal , and n. shroff , `` near optimal power and rate control of multi - hop sensor networks with energy replenishment : basic limitations with finite energy and data storage , '' _ ieee trans . 4 , pp . 815829 , 2012 . j. qin , y. chow , j. yang , and r. rajagopal , `` online modified greedy algorithm for storage control under uncertainty , '' _ ieee trans . power syst . _ , 2015 ( accepted ) , [ online ] . available : http://arxiv.org/pdf/1405.7789v4.pdf .

it is generally agreed that low - frequency ( radio ) emission from relativistic jets in active galactic nuclei ( agn ) is synchrotron emission from nonthermal electrons . however , it is still debated whether the high - frequency ( x - ray & @xmath0-ray ) emission is due solely to compton scattering by the electrons responsible for the low - frequency synchrotron emission or due to hadronic processes resulting from protons being co - accelerated with the electrons ( e.g. , * ? ? ? if hadronic models are valid , agn would be an expected source of neutrinos , as the proton interactions responsible for the production of neutral pions which result in @xmath0-rays also produce charged pions , which result in secondary electrons , positrons , and neutrinos ( e.g. , * ? ? ? can we expect agn to be detectable neutrino sources at the level that starburst systems of similar @xmath0-ray luminosity are expected to be ? observations by the space - based @xmath0-ray telescope _ fermi _ and future observations by the hard x - ray space telescope nustar may help to further resolve this question . building on the approach of previous models ( e.g. * * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , we developed a semi - analytic model for cosmic ray interactions for the starburst galaxy m82 ( * ? ? ? * hereafter yegz ) . in developing the model , we emphasized the important role of galactic winds while limiting the number of free parameters and keeping our models observationally informed . now , we aim to determine whether the model is applicable to other starburst systems and explore benefits and limitations of such models due to the presence of agn . the starburst galaxy ngc 253 has been detected in @xmath0-rays by both space - based ( _ fermi _ ) and ground - based ( hess ) telescopes and possibly contains an active central supermassive black hole ( smbh ) @xcite . seyfert 2 galaxies ngc 1068 and ngc 4945 have also been detected with _ fermi_. though ngc 1068 ( @xmath1 mpc ; nasa / ipac extragalactic database ) is more distant than the nearby ngc 253 ( @xmath2 mpc ; * ? ? ? * ) and ngc 4945 ( @xmath3 mpc ; * ? ? ? * ) , all three have @xmath0-ray fluxes of the same order of magnitude @xcite . in this paper , we determine whether our model is applicable to a combined starburst / agn environment . we select the starburst galaxy ngc 253 with a possible low - activity agn and the high - activity seyfert galaxy ngc 1068 . applying our model to ngc 253 and ngc 1068 , we seek the best - fit cosmic ray interaction model which accurately reproduces both radio and @xmath0-ray spectra within the confines of the observed properties of the interstellar medium ( ism ) and surrounding radiation field . additionally , we make a prediction of the neutrino fluxes from both galaxies . llclc distance & 3.9 mpc & 1 & 14.4 mpc & 7 + central molecular zone ( cmz ) radius & 150 pc & 2 & 200 pc & 8 + molecular gas mass & @xmath4 @xmath5 & 3 , 4 & @xmath6 @xmath5 & 9 + average ism density & 500 - 1400 @xmath7 & & 250 @xmath7 & + ir luminosity & @xmath8 @xmath9 & 5 & @xmath10 @xmath9 & 10 + radiation field energy density & 500 - 2000 ev @xmath7 & & @xmath11 ev @xmath7 & + sn explosion rate & 0.1 yr@xmath12 & 6 & 0.07 yr@xmath12 & 11 + sn explosion energy & 10@xmath13 ergs & & 10@xmath13 ergs & + sn energy transferred to cr & 4 - 20% & & 10% & + ratio of primary protons & 50 & & 50 & + to electrons ( n@xmath14/n@xmath15 ) & & & & + slope of primary cr & 2.2/2.3 & & 2.0 & + source function & & & & + the next section describes the observed properties of ngc 253 and ngc 1068 . section 3 details how the population of energetic particles was computed . section 4 contains the results of our model for ngc 253 and ngc 1068 . in section 5 , we compare and contrast the results for ngc 253 and ngc 1068 and discuss predictions for the neutrino flux and the resulting implications . in section 6 , we present concluding remarks . ngc 253 is a giant , barred spiral galaxy with a central starburst region . tidal interactions are the cause of many starbursts in massive galaxies , and while there is no obvious companion or trigger for the starburst in ngc 253 , kinematic studies suggest ngc 253 may have been involved in a past merger @xcite . models indicate that the current high level of nuclear star formation has been ongoing for at least 20 - 30 myr @xcite . the center of ngc 253 is host to several compact objects which have been observed over a variety of wavelengths @xcite . the strongest non - thermal , compact radio source ( th2 ) has long been considered a possible agn candidate @xcite . additionally , x - ray emission from the nucleus may be powered by an obscured agn or a combination of supernova remnants and fluorescent line emission from molecular clouds @xcite . however , new observations with nustar and _ chandra _ are unable to rule out the possibility that these compact x - ray sources are ultraluminous x - ray sources ( ulxs ) and not a low - luminosity agn @xcite . one of the other major features of ngc 253 is the galactic wind emanating from the nuclear region ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? this galactic wind has been detected in both h@xmath16 and x - rays . observations of the position and size of the outflow at both wavelengths match up almost exactly @xcite . optical observations give an outflow velocity on the order of a few 100 km s@xmath12 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? additionally , @xcite suggest that , like star formation , the wind is quasi - steady . due to the obscured nature of the starburst nucleus of ngc 253 ( e.g. , * ? ? ? * ; * ? ? ? * ) , determining the properties of the nuclear region has been difficult . radio observations yield supernova rates ranging from 0.03 - 0.3 yr@xmath12 @xcite . in this work , we adopt a rate of 0.1 yr@xmath12 based on @xcite ( see table 1 for additional parameters ) . additionally , estimates of the molecular gas mass range from @xmath17 @xmath5 ( * ? ? ? * and references therein ) . we adopt a value of @xmath18 @xmath5 for the molecular gas mass for this paper , based on work by @xcite . we adopted an ionized gas mass for ngc 253 by scaling from the ratio of ionized to molecular gas in m82 , which gives a mass of @xmath19 @xmath5 . ngc 253 is well observed in the radio for both extended emission and the central starburst core . observations by @xcite and @xcite at 333 mhz and from 1 to 7 ghz show that the radio spectrum turns down only slightly in the core at low frequencies , unlike the radio spectrum for the core of m82 , which is completely absorbed at low frequencies . high frequency observations from @xcite at 18.5 and 22 ghz and from @xcite at 94 ghz require a free - free emission contribution to the flux in addition to non - thermal synchrotron emission . observed radio and @xmath0-ray spectra are presented with model spectra in section 4 . ngc 253 has been observed in @xmath0-rays at both gev and tev energies by _ fermi _ and hess . @xcite present four data points and two upper limits over 0.2 to 200 gev from analysis of 30 months of _ fermi _ data . @xcite present a combined analysis of five data points and one upper limit over 0.2 tev to 30 tev from hess data with the _ fermi _ data and fit a single power - law to both data sets , resulting in a spectral index of @xmath20 . @xcite present a more current reduction of the _ fermi _ data which we use for comparison with our models instead of the _ fermi _ data shown in @xcite . ngc 1068 is the closest , brightest seyfert galaxy . the total bolometric luminosity of the galaxy is @xmath21 @xmath9 ; half of which is accounted for by the circumnuclear disk ( cnd ) and agn @xcite . the galaxy supports a kiloparsec - scale , non - thermal radio jet @xcite and a compact radio nucleus @xcite . the complex chemical and kinematic properties of the nuclear region have been well - studied at multiple wavelengths @xcite . the circumnuclear starburst ring surrounding the 2.3 kpc stellar bar @xcite is illuminated by star formation in the past 10 - 40 myr @xcite . the stellar population inside the cnd , the inner @xmath22200 pc , developed in two bursts at 30 and 300 myr ago @xcite . @xcite give the star formation rate ( sfr ) for the inner galaxy as @xmath22 10 @xmath5 yr@xmath12 . this gives an upper bound on the supernova rate of @xmath220.07 yr@xmath12 for the nuclear region . we convert from sfr to supernova rate with by analysis similar to @xcite . both starburst regions are rich in molecular gas with masses of @xmath23 @xmath5 for the circumnuclear ring and @xmath24 @xmath5 for the cnd @xcite . co , hcn , and hco@xmath25 observations show that the gas is moderately dense ( @xmath26 @xmath7 ) with a kinetic temperature of @xmath27 k @xcite . as the nearest and brightest seyfert galaxy , ngc 1068 has been extensively studied in the radio . on the kiloparsec - scale , radio observations have mapped the star - forming spiral arms and the radio jets . there are extensive observations specifically targeting the agn on the parsec - scale @xcite . however , as even observations of the inner hundred parsecs have a dominant contribution from the agn @xcite , specifically the radio jet and counter - jet , there are essentially no observations of the central molecular zone ( cmz ) alone . radio maps show little evidence of ongoing star - formation in the central @xmath22150 pc . @xmath0-ray data for ngc 1068 are more limited than for ngc 253 . while there are four data points and two upper limits at gev energies , there is only a single upper limit for tev energies from hess @xcite . the _ fermi _ data range from 0.2 gev to 300 gev and the hess upper limit is at 0.2 tev . diffusion is thought to be of minor importance in star - forming environments with high gas densities and galactic winds . as such , following the approach in yegz , a cosmic ray spectrum can be calculated from a single quantity including the source function ( dependent on supernova rate ) and the lifetime . because we only consider radiative and advective losses , our model is not applicable to spatially extended regions , such as the inner spiral arms in ngc 1068 , where cosmic ray diffusion is important . thus , we apply our model specifically to the cmzs of galaxies . supernovae are the assumed drivers of cosmic ray acceleration , and so we adopt a power - law source function ( @xmath28 ) such that @xmath29 where @xmath30 is the supernova rate , @xmath31 is the volume of the starburst region , @xmath32 is the fraction of the supernova energy transferred to cosmic rays , and @xmath33 is 10@xmath13 ergs , the typical energy from a supernova explosion . thus , the spectrum for primary protons is given by @xmath34 where @xmath35 is the minimum cosmic ray energy , here taken to be @xmath36 gev . @xmath37 is the cosmic ray lifetime composed of an energy loss timescale ( due to radiative and collisional losses ) and an energy independent timescale : @xmath38 where @xmath39 is the scale height of the starburst region and @xmath40 is the speed of the particles in the wind of the starburst region ( see also * ? ? ? energy losses include ionization , pion production , bremsstrahlung , synchrotron emission , and the inverse compton effect . pion production from proton - proton collisions is the dominant energy loss for cosmic ray protons above the 1.3 gev threshold . as the lifetimes of both charged and neutral pions are very short , pions quickly decay into secondary electrons and positrons , neutrinos , and @xmath0-rays . the primary proton source function can be used to calculate the source function for charged and neutral pions @xcite , and the pion source function can then be used to calculate the spectrum of secondary electrons and positrons @xcite . one of the main assumptions of the model is that cosmic rays sample the mean density of the interstellar medium . the density of the interstellar medium sampled by the cosmic rays is vital to the model as it affects pion production and thus both the @xmath0-ray and radio spectra . the majority of the ism is hot , low density gas in which pockets of cold , very dense molecular gas and warm ionized gas can be found . in @xcite , we assumed a simple two phase model of cold , molecular gas and hot , ionized gas of negligible density . we found that for a wide range of cosmic ray injection conditions , magnetic field properties , and wind speeds , the cosmic rays sample the mean density of the ism to within a factor of two . the effect of a third phase of dense , warm , ionized gas on cosmic ray sampling will be considered in future work . from cosmic ray proton and primary and secondary electron / positron populations , we can calculate nonthermal radio synchrotron emission and @xmath0-ray emission by neutral pion decay , bremsstrahlung , and inverse compton ( see yegz for further details ) for comparison with observed data . primary variables include spectral index ( @xmath41 ) , cosmic ray acceleration efficiency ( @xmath32 ) , molecular gas mass ( @xmath42 ) , magnetic field strength ( @xmath43 ) , wind ( advection ) speed ( @xmath40 ) , ionized gas density ( @xmath44 ) , and interstellar radiation field energy density ( @xmath45 ) . in yegz , we found that observations of the starburst core in m82 showed a complete turn down of the radio emission at low frequencies . we attributed this to severe free - free absorption in the core and adopted a model for the starburst core with a cylindrical wall of warm , ionized gas surrounding a lower density hot medium . because some observations at low frequencies did not have enough resolution to distinguish the core from the halo in m82 , we also included a component for radio emission from the halo . observations for ngc 253 do not show the same severe free - free absorption at low frequencies as in m82 . however , there is significant free - free emission at high frequencies , as is also the case in m82 . thus , we expect that at least some fraction of the radio synchrotron emission is still absorbed by ionized gas . to account for this , we keep the thin shell geometry ( with shell width @xmath46 and radius @xmath47 ) for the nucleus , but assume only a small fraction ( @xmath48 ) of the synchrotron emission is absorbed by the ionized gas . then , the radiative intensity of the absorbed synchrotron emission is given by @xmath49 where @xmath50 is the radius of the hot , diffuse gas ( @xmath51 ) , @xmath52 is the emission coefficient for the hot gas ( @xmath53 ) , and @xmath54 is the absorption coefficient for the warm , ionized gas ( @xmath55 ) . the majority of the radio spectrum is unabsorbed synchrotron emission from the hot , diffuse gas : @xmath56 additionally , we have some portion of the radio spectrum composed of free - free emission from the ionized gas clouds : @xmath57 where @xmath58 is the emission coefficient for the warm , ionized gas ( @xmath59 ) . thus , the total emergent intensity from the starburst region is @xmath60 for an absorption fraction of @xmath61 , this reduces to @xmath62 which is the expression for the radio emission that we used for modeling m82 ( yegz ) . note that unlike in yegz , we do not include a disk / halo component for radio emission in this paper as radio observations of the disk are separable from the halo ( e.g. , * ? ? ? * ; * ? ? ? we anticipate that inverse compton scattering is a vital process for producing @xmath0-rays in ngc 253 and ngc 1068 . for inverse compton scattering , the @xmath0-ray source function is given by @xcite @xmath63 with @xcite @xmath64,\ ] ] where @xmath65 is the energy of the resulting @xmath0-ray , @xmath66 is the energy of the incident photon , @xmath0 is the energy of the electron . here , the combined cosmic ray electron / positron spectrum , @xmath67 , is in units of @xmath7 . the function @xmath68 is part of the klein - nishina cross section and is given by @xcite @xmath69 where @xmath70 for the blackbody spectrum , @xmath71 , we use an isotropic , diluted , modified blackbody spectrum ( * ? ? ? * and references therein ) @xmath72 where @xmath73 is a spatial dilution factor ( given by the normalization @xmath74 ) and @xmath75 corresponds to @xmath76 hz . as with our previous model for m82 ( yegz ) , we test a range of parameters to find the best fits to both the radio and @xmath0-ray spectra for the starburst core of ngc 253 ( see table 2 ) . we choose to vary parameters such as magnetic field strength ( @xmath77 @xmath78 g ) and radiation field energy density ( @xmath79 ev @xmath7 ) as they are not well - constrained by observations . we vary acceleration efficiency ( @xmath80 ) as it is equivalent to varying the supernova rate and test two spectral indices ( @xmath81 ) . as the observed @xmath0-ray spectrum for ngc 253 is steeper than that of m82 , we choose larger spectral indices and do not test for @xmath82 . we also test a variety of wind ( advection ) speeds ( @xmath83 km s@xmath12 ) in order to study the no - wind case and wind speeds similar to those derived from optical and x - ray observations . ll magnetic field strength ( @xmath43 ) & 50 - 350 @xmath78 g + wind ( advection ) speed ( @xmath40 ) & 0 - 1000 km s@xmath12 + ionized gas density ( @xmath44 ) & 50 - 500 @xmath7 + absorption fraction ( @xmath48 ) & 0.1 - 1.0 + in regards to the molecular gas mass , there is a wide a range of observed values and there is not yet a clear consensus on which is most accurate . varying the gas mass would have a similar ( but not identical ) effect to varying the acceleration efficiency or supernova rate . in the case of the cosmic ray electrons , varying molecular gas mass affects the losses due to ionization and bremsstrahlung but not inverse comtpon or synchrotron emission . though we adopt the value of @xmath18 @xmath5 , there is likely at least a factor of two uncertainty in this value . as such , we also test models at @xmath84 @xmath5 . while we vary the molecular gas mass , we keep the ionized gas mass constant . however , we do vary the ionized gas density ( @xmath85 @xmath7 ) , and thus the filling fraction of ionized gas , as it has a significant impact on free - free absorption and emission in the radio spectrum . to distinguish between models , we use @xmath86 tests to compare to the observed data . we can see from the contour plots of the @xmath86 values that the @xmath0-ray and radio data constrain the value for the magnetic field strength and wind speed in different ways ( see figure 1 ) . there is a degeneracy for the magnetic field strength and wind speed for the radio data such that as the magnetic field strength increases , an increase in wind speed allows for a similarly good fit . the reverse is true for the @xmath0-rays . in the radio spectrum , an increase in magnetic field strength results in an increase in synchrotron emission such that the number of electrons must be reduced ( see yegz for further explanation ) . however , in the @xmath0-ray spectrum , an increase in the magnetic field strength leads to a reduction of the inverse compton flux as fewer electrons are available as a result of the shorter energy loss lifetimes . thus , the wind speed must be reduced to allow for more electrons to be available to produce inverse compton emission . because the results are essentially non - complementary , we were not able to find a joint solution ( a best - fit model for both the @xmath0-ray and radio spectra based on a common set of parameters ) for the radio and @xmath0-ray data assuming @xmath18 @xmath5 for the molecular mass . this is highlighted in figure 2 , where the combined reduced @xmath86 result is dominated by the radio data . for the radio spectrum , optimal wind speeds are essentially the no wind case ( @xmath40 ) which is not supported by observations of ngc 253 . for the @xmath0-ray spectrum , optimal wind speeds range from @xmath87 km s@xmath12 , which is slightly higher than optical determinations of the wind speeds . additionally , while @xmath0-ray models allow for a wide range of magnetic field strengths , radio models constrain the magnetic field strength to @xmath88 @xmath78 g . these field strengths correspond to @xmath222000 to @xmath223000 ev @xmath7 . cosmic ray energy densities are nearly an order of magnitude below this , @xmath22200 to @xmath22300 ev @xmath7 for @xmath89 . while the magnetic field energy density lies at the high end of the range of the assumed radiation field energy densities , the cosmic ray energy densities are within a factor of two of the lower end of the range . taken at face value , these model results suggest that for the system the magnetic field and radiation field are in approximate equipartition while the cosmic rays are sub - equipartition . in addition to limiting the possible magnetic field strength and wind speed values , the @xmath0-ray data severely constrain the possible combinations of spectral index and acceleration efficiency . while the radio data is fit equally well by all combinations of acceleration efficiency and spectral index , the @xmath0-ray spectrum is easily overestimated because of the large radiation fields and high densities in starburst regions . as such , the majority of the combinations of acceleration efficiency and spectral index do not result in a minimized ( best - fit ) @xmath86 value for the @xmath0-ray spectrum . so , to keep the combination of pion decay and inverse compton emission in check , our model rules out acceleration efficiencies of @xmath90 and @xmath91 for both spectral indices . we also rule out the spectral index of @xmath92 for all acceleration efficiencies , as the best - fit models underestimate the @xmath0-ray spectrum at tev energies ( see figure 3 ) . thus , the only combination of acceleration efficiency and spectral index which both minimizes and fits the tev energy data is @xmath89 , @xmath93 . equivalently , a larger acceleration efficiency pared with a lower supernova rate would have the same result , as we constrain their product . the goodness of our @xmath86 fits also change slightly depending on the radiation field energy density . the best - fit @xmath0-ray spectrum occurs for an energy density of 2000 ev @xmath7 for a spectral index of @xmath89 . the resulting @xmath0-ray spectrum is dominated by pion decay emission but still has a significant contribution from inverse compton emission at tev energies ( see figure 3 ) . aside from the parameters discussed thus far , we also vary the fraction of synchrotron emission absorbed in ionized gas clouds ( see table 2 ) . observations show that at low frequencies , the radio spectrum for ngc 253 levels off , unlike the radio spectrum for m82 which turns over at low frequencies . in the case of m82 , to fit the radio spectrum properly , we required that all of the synchrotron emission from the hot , diffuse medium was absorbed in ionized gas clouds such that @xmath94 ( yegz ) . however , as the radio spectrum for ngc 253 only flattens , we varied the absorption fraction to find what absorption fraction was required to accurately fit the spectrum . we found that models with an absorption fraction of up to @xmath95 agreed with all compared radio data points ( see figure 4 ) , including models with no absorption fraction ( @xmath96 ) . the best - fit models were for an absorption fraction of @xmath97 , which was selected for use for the remainder of our models . in yegz , we found that while m82 was an electron calorimeter , it was , at best , only a partial ( @xmath2250% ) proton calorimeter . for ngc 253 , we find that it is an electron calorimeter and that it can be a proton calorimeter , depending upon the selected parameters . we based our conclusions on a comparison between energy - loss and advection timescales . in ngc 253 , we find that the average wind speeds for the best - fit @xmath0-ray models are on the order of @xmath87 km s@xmath12 . the best - fit wind speed decreases depending on the assumed magnetic field strength . comparing timescales , we find that ngc 253 is a @xmath2250% proton calorimeter . however , the best - fits for the radio spectrum were for the no wind case . in this case , ngc 253 would be a complete proton calorimeter ( cf . while the @xmath0-ray data were readily fit with our model , the radio spectrum presented much more of a challenge . as previously stated , we were not able to find a joint solution for the @xmath0-ray and radio models . while our best - fit for the @xmath0-ray spectrum had a reduced @xmath86 value of 2.0 , the best - fit for the radio spectrum had a reduced @xmath86 value of 50 . in addition to the preferred wind speed being unphysical , the best - fit model does not match up well with the @xcite data . although we did not explicitly vary molecular gas mass in our models , we did originally test a lower gas mass . models tested with a lower gas mass of @xmath98 @xmath5 resulted in significantly better radio fits ( see figures 2 and 4 ) . our best - fit model with the lower mass has a reduced @xmath86 value of 12 . this reduction in @xmath86 with the lower mass is due to higher bremsstrahlung losses for higher masses . thus , for the lower mass , more electrons are available to produce synchrotron emission at the low frequencies instead of bremsstrahlung . the discrepancy between fits for the different gas masses is , in large part , due to the limitations of the one - zone model . while there are many computational advantages to assuming a simple single zone , there are also disadvantages . in yegz , we noted that by assuming a single ionized gas density in our model , we were unable to fit both the high - frequency radio data and the low - frequency turn down . in ngc 253 , our one zone model has reached the calorimeter limit for the higher gas mass . as such , while we are producing the same amount of secondary electrons and positrons at both masses , there are fewer electrons / positrons available to produce inverse compton and synchrotron emission due to an increase in bremsstrahlung . the cmz dust temperature is the key to modeling the @xmath0-ray observations for ngc 1068 as it determines the radiation field that inverse compton emission depends on . observations by @xcite show a blackbody spectrum with temperatures in the range of @xmath99 for the inner cnd . when assuming a radiation field from dust with @xmath100 k , the photon number is significantly decreased such that our models produce negligible inverse compton @xmath0-ray emission . however , this dust temperature is attributed to the dusty torus of the agn nucleus and likely does not dominate the larger , surrounding cmz . as such , we assume that the dust temperature in the cmz is on par with the temperatures of molecular gas in the region , @xmath22100 k , and we use this to determine the radiation field spectrum . as with ngc 253 , we intended to test a variety of different sets of parameters with which to model ngc 1068 . however , we found ngc 1068 significantly harder to model than ngc 253 . the upper bound on the supernova rate produces a @xmath0-ray spectrum that is lower by a factor of only a few ( see figure 5 ) . however , a lower bound on the supernova rate results in a @xmath0-ray spectrum that is nearly two order of magnitudes lower than the observed data . because we were underestimating the @xmath0-ray emission , we selected parameters to maximize the inverse compton emission ( a magnetic field strength of @xmath101 @xmath78 g and a radiation field energy density of @xmath102 ev @xmath7 ) and the pion decay emission and bremsstrahlung ( a wind speed of @xmath103 km s@xmath12 ) . even selecting parameters to augment the @xmath0-ray emission , without invoking an extra source of cosmic rays , we were not able to produce a model which agrees with the _ fermi _ observations to better than a factor of a few . further complicating matters is the radio spectrum for ngc 1068 . while the galaxy has been extensively observed in the radio spectrum , the presence of a radio jet greatly overshadows any emission not originating from the agn or its jets . ultimately , we chose to compare our radio models against a few different radio observations . first , we plot our radio models against radio observations of the agn core ( s1 ) found in @xcite . though the observations of the agn are on the parsec - scale , they do highlight the fact that they are of a fundamentally different nature than the emission from star - forming regions . we also plot radio observations of the region labeled s2 @xcite , which could be the counter - jet as seen through the star - forming disk . additionally , we use the 18 cm radio map in @xcite to compare with our models . while the outermost contour includes the spiral arms , the second contour should give us an upper limit on the radio continuum in the cmz excluding the component from the agn and the jets . taking the brightness level ( of 74 mjy / beam ) and multiplying by the area of the cmz , we obtain an upper limit of @xmath22 13 mjy ( see red squared in figure 5 ) . while nearly all of the energy of the cosmic ray electrons goes into inverse compton emission , some small fraction of the energy is lost due to bremsstrahlung and synchrotron emission . assuming a moderate starburst magnetic field strength of 200 @xmath78 g results in a significant amount of synchrotron emission . as the upper limit ( red square ) that we plot is nearly an order of magnitude below the model , we must be significantly overestimating the synchrotron emission in the cmz . thus , while assuming the upper limit on the supernova rate results in a @xmath0-ray spectrum which comes within a factor of a few of the observed data , the radio spectrum disagrees significantly with observations ( see figure 5 ) . previously , we developed and tested a model for cosmic ray interactions in starburst nuclei ( yegz ) . in testing our model against other nuclei and combined starburst / agn environments , we have applied our updated model to two giant , barred spiral galaxies ; specifically , we have modeled both ngc 253 , with an unambiguous starburst nucleus , and ngc 1068 , the archetypal seyfert 2 galaxy . the @xmath0-ray spectrum for ngc 253 is well fit with a combination of emission from neutral pion decay , bremsstrahlung , and inverse compton , requiring both hadronic and leptonic emission mechanisms . while we find a joint solution for the @xmath0-ray and radio spectra for the lower molecular gas mass for the ngc 253 cmz , we were unable to do the same at a higher gas mass ( @xmath18 @xmath5 ) . @xcite find best - fit models for both the @xmath0-ray and radio spectra for similarly large densities ( @xmath104 @xmath7 ) . however , they do not include wind speed as a free parameter and fix their advection / diffusion timescale at 10 myr ( which is equivalent to a wind speed of @xmath22 5 km s@xmath12 for a half - thickness of 50 pc ) . this would allow for a good fit to the radio spectrum at higher masses and their lower radiation field would also allow for a similarly good fit for the @xmath0-ray spectrum . other models include those of @xcite who find joint solutions when assuming parameters closer to our lower gas mass and wind speeds between 300 and 600 km s@xmath12 . we believe that the data support our choice of gas masses in the @xmath4 @xmath5 range as well as wind speeds of @xmath105 km s@xmath12 ( see * ? ? ? * ; * ? ? ? * ) and that our models therefore contain astrophysically reasonable parameters . though located nearly four times further away , ngc 1068 has a @xmath0-ray flux comparable to that of ngc 253 . in comparing our model with observations , we find that the inner starburst region in ngc 1068 does not supply a sufficient cosmic ray population to produce the observed @xmath0-ray emission by itself . this agrees with analysis of the supernova rate , total gas mass , and @xmath0-ray luminosity by @xcite for ngc 1068 in comparison to other starburst galaxies . while a scenario which boosts the cosmic ray population in the cmz , such as a diffusion of cosmic rays from the circumnuclear ring into the cmz , is possible , it is most likely that the high - energy emission results from interactions in the jet . an agn jet origin of the @xmath0-ray emission in ngc 1068 agrees with calculations of @xcite who propose a leptonic external inverse compton ( eic ) model for the high - energy emission . while proton - proton interactions dominate as a cosmic ray energy loss mechanism in the ism , proton-@xmath0 interactions will be dominant in the jet due to the lack of molecular gas necessary for proton - proton interactions . proton - proton interactions result in both charged and neutral pions . while neutral pions decay quickly into @xmath0-rays , charged pions decay into muons and neutrinos followed by electrons and positrons and more neutrinos . thus , galaxies whose @xmath0-ray spectra are dominated by or are significantly contributed to by pion decay will also be sources of neutrinos . as proton-@xmath0 interactions still result in pion production , agn could also be a source for neutrinos as starbursts are , if @xmath0-ray emission in the jet is hadronic . while a detection of neutrinos from agn would help to determine the origin of @xmath0-ray emission , such a detection will also provide vital clues for starburst galaxies . observations of neutrinos provide direct information about the nature of the @xmath0-ray spectrum and the cosmic ray proton population . better constraining the contribution of neutral pion decay to the @xmath0-ray spectrum , also better constrains the contribution of bremsstrahlung and inverse compton emission . in the case of inverse compton , this would be another diagnostic for determining the radiation field , which can be uncertain in such environments . to make a neutrino flux prediction , we start with the source function for muon neutrinos which results from charged pion decay ( @xmath106 and @xmath107 ) depends on the source function for pions : @xmath108 where @xmath109 is the source functions for pions ( see yegz for details ) and @xmath110 @xcite . we found that the @xmath0-ray spectra for ngc 253 had a significant contribution from pion decay . while this results in a significant neutrino flux ( see figure 6 ) , the steep spectrum of ngc 253 means that m82 is a more likely candidate for eventual detection of the energetic neutrinos for which icecube has good sensitivity @xcite . though all of our models for ngc 1068 underestimated the @xmath0-ray emission , we find that the corresponding neutrino flux from the starburst region is comparable to that of ngc 253 at tev energies ( see figure 6 ) . however , if the high - energy emission in the jet is hadronic , ngc 1068 may still be a detectable source of neutrinos , particularly if the flat @xmath0-ray spectrum seen in figure 5 continues to higher energies . at energies below 1 gev , models predict a significant difference in the spectra for inverse compton and pion decay @xmath0-ray emission @xcite . thus , in the future , hard x - ray observations , such as those by nustar , should be able to distinguish between leptonic and hadronic emission mechanisms for @xmath0-ray production in such galaxies and determine whether these galaxies are likely to be detectable in neutrinos . in this paper , we explore the differences between starburst galactic nuclei and agn and their impact on cosmic ray interactions . as a typical nuclear starburst environment , ngc 253 can be fit with our model that includes inverse compton emission . we find that for an assumed gas mass of @xmath111 @xmath5 , the @xmath0-ray and radio spectra can not uniquely determine the magnetic field strength or wind speed . additionally , the @xmath0-ray spectrum also severely limits the possible combinations of spectral index ( to @xmath89 ) and acceleration efficiency ( to @xmath93 ) due to the powerful combination of effects due to inverse compton and pion decay emission . past models for ngc 253 have assumed a fixed wind ( advection ) speed which presupposes calorimetry ( e.g. * ? ? ? based on our results , a wind is a critical component for determining the degree to which a starburst system is a calorimeter . while the best - fits to the @xmath0-ray spectrum result in a @xmath2250% proton calorimeter , fits to the radio spectrum result in a complete proton calorimeter . additionally , our fits result in systems below equipartition ( in regards to the cosmic ray energy densities ) with strong magnetic fields ( see also * ? ? ? the lack of a solution which fits the @xmath0-ray and radio spectra together with a single set of parameters and the problems in fitting the radio spectrum highlight the limitations of this one - zone model . we , however , can obtain a good fit to @xmath0-rays and the radio by assuming a lower gas mass . understanding of the ism conditions is vital to applying these models to starbursts . as such , while the model works well for a late phase starburst such as m82 , it does not as easily fit an early phase starburst such as ngc 253 . as an agn environment containing a possible starburst region , ngc 1068 is not fit with our self - consistent starburst model . while we find a few models that come close to predicting the observed @xmath0-ray spectrum for ngc 1068 , all of our models underestimate the high - energy emission and we are unable to find any models of the inner starburst region that agree with the observed data . thus , we find that the cosmic ray population from the starburst region alone is not enough to accurately fit the observed @xmath0-ray emission and even those models in rough agreement with high - energy observations result in an overestimation of the radio spectrum . we conclude that the observed @xmath0-ray emission likely occurs in the agn or its jets . additionally , we find that starburst galaxies such as ngc 253 are potential sources for high energy neutrino detection . however , agn with high @xmath112 such as ngc 1068 will not be luminous neutrino sources unless their high - energy emission is of hadronic origin . this work was supported in part by nsf ast-0907837 , nsf phy-0821899 ( to the center for magnetic self - organization in laboratory and astrophysical plasmas ) , and nsf phy-0969061 ( to the icecube collaboration ) . we thank alberto bolatto for conversations concerning ngc 253 . we also thank francis halzen for his help and support . abbasi , r. , abdou , y. , abu - zayyad , t. , et al . 2011 , , 732 , 18 abramowski , a. , acero , f. , aharonian , f. , et al . 2012 , , 757 , 158 ackermann , m. , ajello , m. , allafort , a. , et al . 2012 , , 755 , 164 anchordoqui , l. a. , hooper , d. , sarkar , s. , & taylor , a. m. 2008 , aph , 29 , 1 blumenthal , g. r. , & gould , r. j. 1970 , rev . , 42 , 273 bock , j. j. , neugebauer , g. , matthews , k. , et al . 2000 , , 120 , 2904 boettcher , e. , zweibel , e. g. , yoast - hull , t. m. , & gallagher , j. s. , iii 2013 , , 779 , 12 carilli , c. l. 1996 , , 305 , 402 condon , j. j. 1992 , , 30 , 575 davidge , t. j. 2010 , , 725 , 1342 davies , r. i. , sugai , h. , & ward , m. j. 1998 , , 300 , 388 domingo - santamara , e. , & torres , d. f. 2005 , , 444 , 403 engelbracht , c. w. , rieke , m. j. , rieke , g. h. , kelly , d. m. , & achtermann , j. m. 1998 , , 505 , 639 exposito , j. , gratadour , d. , clnet , y. , & rouan , d. 2011 , , 533 , 63 fabbiano , g. , & trinchieri , g. 1984 , , 286 , 491 forbes , d. a. , polehampton , e. , stevens , i. r. , brodie , j. p. , & ward , m. j. 2000 , , 312 , 689 gallimore , j. f. , baum , s. a. , & odea , c. p. 1996a , , 464 , 198 gallimore , j. f. , baum , s. a. , odea , c. p. , & pedlar , a. 1996b , , 458 , 136 gallimore , j. f. , baum , s. a. , & odea , c. p. 2004 , , 613 , 794 hailey - dunsheath , s. , nikola , t. , stacey , g. j. , et al . 2008 , , 689 , l109 hailey - dunsheath , s. , sturm , e. , fischer , j. , et al . 2012 , , 755 , 57 halzen , f. 2006 , epjc , 46 , 669 heesen , v. , beck , r. , krause , m. , & dettmar , r .- j . 2009a , , 494 , 563 heesen , v. , krause , m. , beck , r. , & dettmar , r .- j . 2009b , , 506 , 1123 hnig , s. f. , prieto , m. a. , & beckert , t. 2008 , , 485 , 33 kamenetzky , j. , glenn , j. , maloney , p. r. , et al . 2011 , , 731 , 83 karachentsev , i. d. , grebel , e. k. , sharina , m. e. , et al . 2003 , , 404 , 93 karachentsev , i. d. , tully , r. b. , dolphin , a. , et al . 2007 , , 133 , 504 kelner , s. r. , aharonian , f. a. , & bugayov , v. v. 2006 , , 74 , 034018 kornei , k. a. , & mccrady , n. 2009 , , 697 , 1180 krips , m. , martn , s. , eckart , a. , et al . 2011 , , 736 , 37 lacki , b. c. , & beck , r. 2013 , , 430 , 3171 lacki , b. c. , thompson , t. a. , & quataert , e. 2010 , , 717 , 1 lacki , b. c. , thompson , t. a. , quataert , e. , loeb , a. , & waxman , e. 2011 , , 734 , 107 lehmer , b. d. , wik , d. r. , hornschemeier , a. e. , et al . 2013 , , 771 , 134 lenain , j .- p . , ricci , c. , trler , m. , dorner , d. , & walter , r. 2010 , , 524 , 72 lenc , e. , & tingay , s. j. 2006 , , 132 , 1333 mitsuishi , i. , yamasaki , n. y. , & takei , y. 2011 , , 742 , l31 mller - snchez , f. , gonzlez - martn , o. , fernndez - ontiveros , j. a. , acosta - pulido , j. a. , & prieto , m. a. 2010 , , 716 , 1166 muxlow , t. w. b. , pedlar , a. , holloway , a. j. , gallimore , j. f. , & antonucci , r. r. j. 1996 , , 278 , 854 paglione , t. a. d. , & abrahams , r. d. 2012 , , 755 , 106 paglione , t. a. d. , marscher , a. p. , jackson , j. m. , & bertsch , d. l. 1996 , , 460 , 295 peng , r. , zhou , s. , whiteoak , j. b. , lo , k. y. , & sutton , e. c. 1996 , , 470 , 821 persic , m. , rephaeli , y. , & arieli , y. 2008 , , 486 , 143 ricci , r. , prandoni , i. , gruppioni , c. , sault , r. j. , & de zotti , g. 2006 , , 445 , 465 rybicki , g. b. , & lightman , a. p. 1979 , radiative processes in astrophysics ( new york : wiley ) sakamoto , k. , mao , r .- q . , matsushita , s. , et al . 2011 , , 735 , 19 schinnerer , e. , eckart , a. , tacconi , l. j. , genzel , r. , & downes , d. 2000 , , 533 , 850 schlickeiser , r. 2002 , cosmic ray astrophysics ( new york : springer ) schlickeiser , r. , vainio , r. , bttcher , m. , et al . 2002 , , 393 , 69 schulz , h. , & wegner , g. 1992 , , 266 , 167 scoville , n. z. , matthews , k. , carico , d. p. , & sanders , d. b. 1988 , , 327 , l61 spinoglio , l. , pereira - santaella , m. , busquet , g. , et al . 2012 , , 758 , 108 stecker , f. w. 1979 , , 228 , 919 strickland , d. k. , heckman , t. m. , colbert , e. j. m. , hoopes , c. g. , & weaver , k. a. 2004 , , 151 , 193 strickland , d. k. , heckman , t. m. , weaver , k. a. , dahlem , m. 2000 , , 120 , 2965 storchi - bergmann , t. , riffel , r. a. , riffel , r. , et al . 2012 , , 755 , 87 telesco , c. m. , & harper , d. a. 1980 , , 235 , 392 thompson , t. a. , quataert , e. , & waxman , e. 2007 , , 654 , 219 thronson , h. a. , jr . , hereld , m. , majewski , s. , et al . 1989 , , 343 , 158 torres , d. f. 2004 , , 617 , 966 ulrich , m .- h . 1978 , , 219 , 424 ulvestad , j. s. , & antonucci , r. r. j. 1997 , , 488 , 621 van buren , d. , & greenhouse , m. a. 1994 , , 431 , 640 waller , w. h. , kleinmann , s. g. , ricker , g. r. 1988 , , 95 , 1057 wei , a. , kovcs , a. , gsten , r. , et al . 2008 , , 490 , 77 westmoquette , m. s. , smith , l. j. , & gallagher , j. s. , iii 2011 , , 414 , 3719 williams , p. k. g. , & bower , g. c. 2010 , , 710 , 1462 wilson , a. s. , & ulvestad , j. s. 1983 , , 275 , 8 yoast - hull , t. m. , everett , j. e. , gallagher , j. s. , iii , & zweibel , e. g. 2013 , , 768 , 53 ( yegz )

phylogenetic trees ( evolutionary trees ) are widely used in evolutionary biology to represent the tree - like evolution of a collection of extant species . however , due to reticulation events , such as hybridization and lateral gene transfer ( lgt ) in evolution , phylogenetic trees representing the evolutionary history of different parts of the genomes found in the same collection of the extant species may differ . in order to facilitate the comparison of these different phylogenetic trees , several metrics were proposed in the literature , such as robinson - foulds distance @xcite , nni ( nearest neighbor interchange ) distance @xcite , tbr ( tree bisection and reconnection ) distance , spr ( subtree prune and regraft ) distance @xcite , and hybridization number @xcite . the spr distance between two phylogenetic trees is the minimum number of subtree prune and regraft operations @xcite that needed to convert one tree to the other , which is equivalent to the minimum number of reticulation events to transform between the two trees . thus , the spr distance provides a lower bound on the number of such events needed to reconcile the two phylogenetic trees @xcite . and this lower bound gives an indication of the extent to which reticulation has influenced the evolutionary history of the extant species under consideration . similarly to the definition of the spr distance , the tbr distance between two phylogenetic trees is the minimum number of ` tree bisection and reconnection ' operations @xcite that needed to convert one tree to the other . although the tbr distance has no known direct biological meaning , it can be used to bound the space of phylogenetic trees . for the study of tbr distance and spr distance , a graph theoretical model , the _ maximum agreement forest _ ( maf ) of two phylogenetic trees , has been formulated . define the _ order _ of a forest to be the number of connected components in the forest . allen and steel @xcite proved that the tbr distance between two _ unrooted binary _ phylogenetic trees is equal to the order of their maf minus @xmath4 , and bordewich and semple @xcite proved that the rspr distance between two _ rooted binary _ phylogenetic trees is equal to the order of their rooted version of maf minus @xmath4 . therefore , there are extensive researches studying the maf problem , which asks for constructing an maf for the given two phylogenetic trees . traditionally , biological researchers assumed that phylogenetic trees are bifurcating @xcite , which leads to most previously work about the maf problem are restricted to binary trees . however , for many biological data sets in practice ( e.g. , @xcite ) , the constructed phylogenetic trees are not strictly bifurcating , that is , these trees contain polytomies . there are two reasons for the polytomies in phylogenetic trees . first , lack of sufficient data or inappropriate analysis of characters , which result in poor resolution of true bifurcating relationships . second , multiple , simultaneous speciation events @xcite . moreover , more recent evidences show the existence of simultaneous speciation events ( e.g. , @xcite ) . therefore , it is crucial to study the general ( i.e. , binary and non - binary ) phylogenetic trees . note that , it is not difficult to verify that the order of an maf for two rooted general trees minus 1 is equal to their spr distance , and the order of an maf for two unrooted general trees minus 1 is equal to their tbr distance . note again that we may construct multiple ( i.e. , two or more ) different phylogenetic trees for the same collection of species according to different data sets or different building methods . constructing an maf for these different trees makes more biological meaning than just for two trees . there are two reasons . first , take the maf for two rooted phylogenetic trees for instance , we have mentioned above that the spr distance between two trees provides a lower bound on the number of reticulation events needed to reconcile the two trees . but these two trees only represent the evolutionary histories of two different parts of the genomes found in the collection of species , thus , this lower bound can only give an indication of the extent to which reticulation has influenced the evolutionary histories of the two parts of the genomes found in the collection of species . if we construct a phylogenetic tree for each part of the genomes found in the collection of species and compare these different trees simultaneously , then , similarly , we can also have the same conclusion that the order of the maf for these trees provides a lower bound on the number of reticulation events needed to reconcile these trees . and this lower bound can give a more comprehensive indication of the extent to which reticulation has influenced the evolutionary history of the collection of species . second , constructing the maf for multiple trees is a critical step in studying the reticulate networks of multiple phylogenetic trees @xcite , which is a hot issue in phylogenetics . therefore , it is meaningful to study the maf problem on multiple trees . however , the maf problem on multiple trees has not been studied as extensively as that on two trees . above all , it makes perfect sense to investigate the maf problem on multiple general phylogenetic trees . in this paper , we will focus on the approximation algorithms and parameterized algorithms for the maf problem on multiple rooted general trees , and for the maf problem on multiple unrooted general trees . in the following , we firstly review the previous related work about the maf problem . in terms of computational complexity , it is known that computing the order of an maf is np - hard and max snp - hard for two unrooted binary phylogenetic trees @xcite , as well as for two rooted binary phylogenetic trees @xcite . * approximation algorithms*. for the maf problem on two rooted binary phylogenetic trees , hein et al.@xcite proposed an approximation algorithm of ratio @xmath1 . but rodrigues et al.@xcite found a subtle error in @xcite , showed that the algorithm in @xcite has ratio at least @xmath3 , and presented a new approximation algorithm which they claimed has ratio @xmath1 . bonet et al.@xcite provided a counterexample and showed that both the algorithms in @xcite and @xcite compute a @xmath5-approximation of the rspr distance between two rooted binary trees in linear time . the approximation ratio was improved to 3 by bordewich et al.@xcite , but at the expense of an increased running time of @xmath6 . a second 3-approximation algorithm presented in @xcite achieves a running time of @xmath7 . whidden et al.@xcite presented the third 3-approximation algorithm , which runs in linear - time . recently , shi et al.@xcite presented an improved approximation algorithm of ratio @xmath8 , which is the best known approximation algorithm for the maf problem on two rooted binary trees . for the maf problem on two unrooted binary phylogenetic trees , whidden et al.@xcite presented a linear - time best known approximation algorithm of ratio @xmath1 . there is also a couple of approximation algorithms for the maf problem on two general phylogenetic trees . rodrigues et al.@xcite developed an approximation algorithm of ratio @xmath9 for the maf problem on two rooted general trees , where @xmath10 is the maximum number of children a node in the input trees may have . chen et al.@xcite developed a @xmath1-approximation algorithm , which is the first constant - ratio approximation algorithm for the maf problem on two unrooted general trees . for the maf problem on multiple rooted binary phylogenetic trees , chataigner @xcite presented an @xmath11-approximation algorithm . recently , the approximation ratio was improved to @xmath1 by shi et al . @xcite . for the maf problem on multiple unrooted binary trees , shi et al . @xcite presented a @xmath3-approximation algorithm . to our best knowledge , there is currently no known approximation algorithm for the maf problem on multiple rooted ( unrooted ) general phylogenetic trees . * parameterized algorithms*. parameterized algorithms for the maf problem , parameterized by the order @xmath12 of an maf , have also been studied . a parameterized problem is _ fixed - parameter tractable _ @xcite if it is solvable in time @xmath13 , where @xmath12 is the parameter and @xmath14 is the input size . for the maf problem on two unrooted binary phylogenetic trees , allen and steel @xcite showed that is fixed - parameter tractable . hallett and mccartin @xcite developed a faster parameterized algorithm of running time @xmath15 . whidden and zeh @xcite further improved the time complexity to @xmath16 or @xmath17 . for the maf problem on two rooted binary phylogenetic trees , bordewich _ et al . _ @xcite developed a parameterized algorithm of running time @xmath18 . et al . _ @xcite improved this bound and developed an algorithm of running time @xmath19 . et al . _ @xcite presented currently the fastest algorithm of running time @xmath20 for the maf problem on two rooted binary trees . there is also a couple of parameterized algorithms for the maf problem on two general phylogenetic trees . et al . _ @xcite presented a parameterized algorithm of running time @xmath19 for the maf problem on two rooted general trees . and chen et al.@xcite developed an algorithm of running time @xmath21 for the maf problem on two unrooted general trees , which is also currently the fastest algorithm for the maf problem on two unrooted binary trees . for the maf problem on multiple rooted binary phylogenetic trees , chen et al . @xcite presented a parameterized algorithm of running time @xmath22 . shi et al . @xcite improved this bound and developed an algorithm of running time @xmath21 . for the maf problem on multiple unrooted binary phylogenetic trees , shi et al . @xcite presented the first parameterized algorithm of running time @xmath17 . to our best knowledge , there is currently no known parameterized algorithm for the maf problem on multiple rooted ( unrooted ) general phylogenetic trees . * contributions*. in the current paper , we are focused on the approximation algorithms and parameterized algorithms for the maf problem on multiple general phylogenetic trees , for both the version of rooted trees and the version of unrooted trees . our algorithms are based on careful analysis of the graph structures that takes advantage of special relations among leaves in the trees . our main contributions include two parameterized algorithms for the maximum agreement forest problem on multiple general trees : one for rooted trees that runs in time @xmath0 , and the other for unrooted trees that runs in time @xmath2 . and two approximation algorithms for the maximum agreement forest problem on multiple general trees : one for rooted trees with ratio @xmath1 , and the other for unrooted trees with ratio @xmath3 . we implement the approximation algorithm and parameterized algorithm for the maf problem on multiple rooted general phylogenetic trees , obtain programs amaf and pmaf , respectively . we test both programs on simulated data and biological data . given multiple rooted general trees , pmaf can calculate the order of an maf for these trees rapidly when the order of an maf is small . and the order of the agreement forest returned by amaf is always less than @xmath1 times the order of an maf for these trees . a tree is a _ single - vertex tree _ if it consists of a single vertex , which is the leaf of the tree . a tree is _ general _ if either it is a single - vertex tree or each of its vertices has degree either @xmath4 or greater than @xmath23 . the degree-@xmath4 vertices are _ leaves _ and the other vertices are _ non - leaves _ of the tree . there are two kinds of trees in our discussion , one is unrooted trees and the other is rooted trees . in the following , we first give the terminologies on the unrooted trees , then remark on the differences for the rooted trees . let @xmath24 be a fixed irrelevant _ label - set_. a general tree is _ unrooted _ if no root is specified in the tree in this case no ancestor - descendant relation is defined in the tree . for the label - set @xmath24 , an unrooted _ general phylogenetic @xmath24-tree _ , or simply an unrooted _ @xmath24-tree _ , is an unrooted general tree whose leaves are labeled bijectively by the label - set @xmath24 ( all non - leaves are unlabeled ) . subforest _ of an unrooted @xmath24-tree @xmath25 is a subgraph of @xmath25 . and a _ subtree _ @xmath26 of @xmath25 is a connected subgraph of @xmath25 , which contains at least one leaf in @xmath25 . denote by @xmath27 the label set that contains all labels in @xmath26 . an unrooted _ @xmath24-forest @xmath28 _ is a subforest of an unrooted @xmath24-tree @xmath25 that contains all leaves of @xmath25 such that each connected component of @xmath28 contains at least one leaf in @xmath25 . thus , an unrooted @xmath24-forest @xmath28 is a collection of subtrees of @xmath25 , moreover , the label - sets of these subtrees are disjoint and the union of the label - sets is equal to @xmath24 . define the _ order _ of the @xmath24-forest @xmath28 , denoted ord@xmath29 , to be the number of connected components in @xmath28 . a subtree @xmath26 of an unrooted @xmath24-tree may contain unlabeled vertices of degree less than @xmath1 . in this case we apply the _ forced contraction _ operation on @xmath26 , which replaces each degree-@xmath23 vertex @xmath30 and its incident edges with a single edge connecting the two neighbors of @xmath30 , and removes each unlabeled vertex that has degree smaller than @xmath23 . note that the forced contraction does not change the order of an @xmath24-forest . it has been well - known that the forced contraction operation does not affect the construction of an maf for @xmath24-trees . therefore , we will assume that the forced contraction is applied immediately whenever it is applicable . an @xmath24-forest @xmath28 is _ irreducible _ if the forced contraction can not apply to @xmath28 . thus , the @xmath24-forests in our discussion are always assumed to be irreducible . with this assumption , each unlabeled vertex in an unrooted @xmath24-forest has degree not less than @xmath1 . two @xmath24-forests @xmath31 and @xmath32 are isomorphic if there is a graph isomorphism between @xmath31 and @xmath32 in which each leaf of @xmath31 is mapped to a leaf of @xmath32 with the same label . we will simply say that an @xmath24-forest @xmath33 is a subforest of another @xmath24-forest @xmath28 if , up to the forced contraction , @xmath33 is isomorphic to a subforest of @xmath28 . a general tree is _ rooted _ if a particular leaf is designated as the root ( so it is _ both _ a root and a leaf ) , which specifies a unique ancestor - descendant relation in the tree . a rooted @xmath24-_tree _ is a rooted general tree whose leaves are labeled bijectively by the label - set @xmath24 . the root of a rooted @xmath24-tree will always be labeled by a special label @xmath34 , which is always assumed to be in the label - set @xmath24 . a subtree @xmath26 of a rooted @xmath24-tree @xmath25 is a connected subgraph of @xmath25 which contains at least one leaf in @xmath25 . in order to preserve the ancestor - descendant relation in @xmath25 , we should define the root of the subtree @xmath26 . if @xmath26 contains the leaf labeled @xmath34 , certainly , it is the root of the subtree ; otherwise , the node in @xmath26 that is in @xmath25 the least common ancestor of all the labeled leaves in @xmath26 is defined to be the root of @xmath26 . subforest _ of a rooted @xmath24-tree @xmath25 is defined to be a subgraph of @xmath25 . a rooted _ @xmath24-forest @xmath28 _ is a subforest of a rooted @xmath24-tree @xmath25 that contains a collection of subtrees whose label - sets are disjoint such that the union of the label - sets is equal to @xmath24 . thus , one of the subtrees in a rooted @xmath24-forest @xmath28 must have the leaf labeled @xmath34 as its root . we also assume that the forced contraction is applied immediately whenever it is applicable . however , if the root @xmath35 of a subtree @xmath26 is of degree @xmath23 , then the operation will _ not _ be applied on @xmath35 , in order to preserve the ancestor - descendant relation in @xmath26 . thus , all unlabeled vertices in @xmath26 that are not the root of @xmath26 have degree not less than @xmath1 . the following terminologies are used for both rooted trees and unrooted trees . an @xmath24-forest @xmath28 is an _ agreement forest _ for a collection @xmath36 of @xmath24-forests if @xmath28 is a subforest of @xmath37 , for all @xmath38 . a _ maximum agreement forest _ ( abbr . maf ) @xmath39 for @xmath40 is an agreement forest for @xmath40 with a minimum ord@xmath41 over all agreement forests for @xmath40 . the four versions of the maf problem on multiple general @xmath24-forests studied in the current paper , are formally given as follows . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ rooted parameterized maximum agreement forest ( para - rmaf ) + _ input _ : a set @xmath42 of rooted general @xmath24-forests , and a parameter @xmath12 + _ output _ : an agreement forest @xmath39 for @xmath42 with ord@xmath43 + or report that no such an agreement forest exists + unrooted parameterized maximum agreement forest ( para - umaf ) + _ input _ : a set @xmath42 of unrooted general @xmath24-forests , and a parameter @xmath12 + _ output _ : an agreement forest @xmath39 for @xmath42 with ord@xmath43 + or report that no such an agreement forest exists + rooted maximum agreement forest ( app - rmaf ) + _ input _ : a set @xmath42 of rooted general @xmath24-forests + _ output _ : a maximum agreement forest @xmath39 for @xmath42 + unrooted maximum agreement forest ( app - umaf ) + _ input _ : a set @xmath42 of unrooted general @xmath24-forests + _ output _ : a maximum agreement forest @xmath39 for @xmath42 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ every agreement forest @xmath28 for a collection @xmath36 of @xmath24-forests corresponds to a unique minimum subgraph ( contains the minimum number of edges ) of @xmath37 , denoted by @xmath44 , for all @xmath38 . thus , without any confusion , we can simply say that an edge @xmath45 is in or not in the agreement forest @xmath28 , as long as @xmath45 is in or not in the unique corresponding subgraph @xmath44 , respectively . the following concept on two @xmath24-forests will play an important role in our discussion . let @xmath31 and @xmath32 be two @xmath24-forests ( either both rooted or both unrooted ) . an agreement forest @xmath28 for @xmath31 and @xmath32 is a _ maximal agreement forest _ ( _ maximal - af _ ) for @xmath31 and @xmath32 if there is no agreement forest @xmath33 for @xmath31 and @xmath32 such that @xmath28 is a subforest of @xmath33 and @xmath46 . by definition , an maf for two @xmath24-forests @xmath31 and @xmath32 is also a maximal - af for @xmath31 and @xmath32 , but the inverse is not necessarily true . fix a label - set @xmath24 . because of the bijection between the leaves in an @xmath24-forest @xmath28 ( either rooted or unrooted ) and the labels in the label - set @xmath24 , sometimes we will use , without confusion , a label in @xmath24 to refer to the corresponding leaf in @xmath28 , or vice versa . for a subset @xmath47 of edges in an @xmath24-forest @xmath28 ( either rooted or unrooted ) , we will denote by @xmath48 the forest @xmath28 with the edges in @xmath47 removed . for any @xmath24-forest @xmath33 that is a subforest of @xmath28 , it is easy to see that there is an edge subset @xmath49 of @xmath28 that @xmath50 . for an arbitrary edge @xmath45 in an @xmath24-forest @xmath28 , removing edge @xmath45 would lead two new subtrees be constructed in @xmath51 , denoted by @xmath52 and @xmath53 , respectively . let @xmath36 be a collection of @xmath24-forests ( either all are rooted or all are unrooted ) , @xmath54 . in the following , we give a reduction rule for @xmath36 . * reduction rule 1 . * let @xmath55 be several subtrees in @xmath24-forest @xmath56 , @xmath57 , @xmath58 . if there exists an edge @xmath45 in @xmath24-forest @xmath59 , @xmath60 , @xmath61 , that @xmath62 and @xmath63 , then remove @xmath45 from @xmath59 . [ reduction rule 1 ] let @xmath64 be the collection that produced by reduction rule 1 on the collection @xmath36 , then @xmath64 and @xmath36 have the same collection of mafs . let @xmath28 be a fixed maf for @xmath36 . let @xmath65 and @xmath66 . since @xmath28 is a subforest of @xmath56 , for each subtree @xmath67 in @xmath56 , @xmath68 , we have that any label of @xmath69 can not be in the same connected component with any label of @xmath70 in @xmath28 . thus , any label of @xmath71 can not be in the same connected component with any label of @xmath72 in @xmath28 . suppose edge @xmath45 is in @xmath28 . then there would exist a path in @xmath28 that connects a label of @xmath73 and a label of @xmath74 . because @xmath75 and @xmath76 , so there would exist a path in @xmath28 that connects a label of @xmath71 and a label of @xmath72 , which contracts the fact that any label of @xmath71 can not be in the same connected component with any label of @xmath72 in @xmath28 . thus , edge @xmath45 could not be in @xmath28 . therefore , @xmath28 is still a subforest of @xmath77 , and @xmath28 is also an maf for @xmath64 . we will assume that reduction rule 1 is applied whenever it is possible . a instance ( of anyone of the four versions of maf problem ) is _ strongly reducible _ if reduction rule 1 is not applicable on it . therefore , all instances in our following discussion are always strongly reducible . before analyzing the detail parameterized algorithms for para - rmaf and para - umaf , we firstly give a few lemmas , which hold true for both para - rmaf and para - umaf . and according to these lemmas , we can present the general frame of our parameterized algorithms . the first lemma follows directly from the definition of maximal - af . [ maximal ] every agreement forest for two @xmath24-forests @xmath31 and @xmath32 is a subforest of a maximal - af @xmath33 for @xmath31 and @xmath32 , but @xmath33 may not be unique . since the maf @xmath28 for a set of @xmath24-forests @xmath36 must be an agreement forest for @xmath31 and @xmath32 , thus , by lemma [ maximal ] , there must exist a maximal - af @xmath33 for @xmath31 and @xmath32 that @xmath28 is a subforest of @xmath33 . moreover , we have the following lemma . [ maximal - maf ] let @xmath78 be a set of @xmath24-forests , and let @xmath28 be a fixed maf for it . there must exist a maximal - af @xmath33 for @xmath31 and @xmath32 that @xmath28 is also an maf for @xmath79 . let @xmath28 be a fixed maf for @xmath80 , and let @xmath33 be a maximal - af for @xmath31 and @xmath32 that @xmath28 is a subforest of @xmath33 . obviously , @xmath28 is an agreement forest for @xmath79 , thus , the order of the maf for @xmath79 is not larger than ord@xmath29 . on the other hand , every agreement forest for @xmath79 is also an agreement forest for @xmath78 , thus , the order of the maf for @xmath79 is not less than ord@xmath29 . therefore , the order of the maf for @xmath79 is ord@xmath29 , @xmath28 is an maf for @xmath79 . let @xmath81 be an instance of either para - rmaf or para - umaf . now according to lemma [ maximal - maf ] , we can give the general frame of our parameterized algorithms . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ main - algorithm + 1 . construct a collection @xmath82 of agreement forests for @xmath31 and @xmath32 that + contains all maximal - af @xmath28 for @xmath31 and @xmath32 with ord@xmath83 ; + 2 . * for * each @xmath28 in the collection @xmath84 constructed in step 1 + recursively work on the instance @xmath85 . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ for an @xmath24-subforest @xmath33 of an @xmath24-forest @xmath28 , we always have ord@xmath86 . thus , no maximal - af @xmath28 for @xmath31 and @xmath32 with ord@xmath87 can contain an maf @xmath33 for @xmath81 with ord@xmath88 . therefore , in step 1 of main - algorithm , we only need examine all maximal - afs whose order is bounded by @xmath12 . [ main - para ] the main - algorithm correctly returns an agreement forest @xmath39 with ord@xmath43 for @xmath89 if such an agreement forest exists . if there exists an agreement forest @xmath39 with ord@xmath43 for @xmath89 , by lemma [ maximal - maf ] , there must exist a maximal - af @xmath28 for @xmath31 and @xmath32 that @xmath39 is also an maf for @xmath90 , which is an instance examined in step 2 . therefore , if @xmath81 has a solution , then step 2 will return such a solution . on the other hand , if there exists an agreement forest @xmath39 for @xmath91 with ord@xmath43 , then , obviously , @xmath39 is also a solution for @xmath89 . thus , every solution for @xmath91 is also a solution for @xmath89 . that is , if @xmath81 has no solution , step 2 could not return a solution . the theorem is proved . in the following two subsections , we will discuss the detail ways of how to construct all maximal - afs for two rooted general @xmath24-forests and for two unrooted general @xmath24-forests separately . then , combining the main algorithm , we can give the detail parameterized algorithms for para - rmaf and para - umaf . two leaves of a rooted general @xmath24-forest are siblings if they have a common parent . sibling set _ is set of leaves that are siblings . maximal sibling set _ mss ) @xmath92 is a sibling set that the common parent @xmath93 of @xmath92 has degree either @xmath94 if @xmath93 has no parent or @xmath95 if @xmath93 has a parent . in this subsection , we present the way of enumerating all maximal - afs for two rooted general @xmath24-forests @xmath31 and @xmath32 . let @xmath39 be a fixed maximal - af for @xmath31 and @xmath32 . we begin with a simple lemma . [ simpleroot ] let @xmath31 and @xmath32 be two rooted general @xmath24-forests . if @xmath32 has no mss , then @xmath31 and @xmath32 have the unique maximal - af which can be constructed in linear time . if @xmath32 has no mss , then @xmath32 has at most one edge . if @xmath32 has no edge , then all connected components of @xmath32 are single - vertex trees and @xmath32 itself is the unique maximal - af for @xmath31 and @xmath32 . if @xmath32 has one edge , then all connected components of @xmath32 are single - vertex trees except one that is a single - edge tree whose root is @xmath34 with a unique child that is labeled by a label @xmath96 . if @xmath34 and @xmath96 are in the same connected component in @xmath31 , then the unique maximal - af for @xmath31 and @xmath32 is @xmath32 itself ; otherwise , the unique maximal - af for @xmath31 and @xmath32 consists of only single - vertex trees , each is labeled by an element in @xmath24 . by lemma [ simpleroot ] , in the following discussion , we will assume that @xmath32 has an mss @xmath92 . because we assumed that all instances in our discussion are strongly reducible , so none of labels in @xmath92 is a single - vertex tree in @xmath31 ; otherwise , reduction rule 1 can remove the edge incident to the label in @xmath32 which is a single - vertex tree in @xmath31 . in the following , we consider all possible cases for the labels of @xmath92 in @xmath31 . since @xmath97 , we can assume that labels @xmath98 and @xmath99 belong to @xmath92 . * case 1 . * all labels in @xmath92 consist an mss in @xmath31 . in this case , @xmath31 and @xmath32 have the same local structure in term of @xmath92 , which consists of the labels in @xmath92 and the parent of @xmath92 . thus , in the further processing of @xmath31 and @xmath32 , the local structure remains unchanged . therefore , we can treat it as an un - decomposable structure . note that @xmath39 also have the local structure . * step 1 . * group all labels in @xmath92 and their parent into an un - decomposable structure , and mark the unit with the same label in @xmath31 and @xmath32 . to implement step 1 , we simply remove all labels in @xmath92 and label the parent with @xmath100 , where @xmath100 is a combination of the labels in @xmath92 ( e.g. , assume @xmath101 , then @xmath102 ) . in the further processing of @xmath31 and @xmath32 , we can treat @xmath100 as a new leaf in the forests . this step not only changes the structures of @xmath31 and @xmath32 , but also replaces the label - set @xmath24 with a new label - set @xmath103 . if we also apply this operation on the maximal - af @xmath39 , then the new @xmath39 remains a maximal - af for the new @xmath31 and @xmath32 . * case 2 . * all labels in @xmath92 are siblings in @xmath31 . let @xmath104 be the common parent of @xmath92 in @xmath31 . and let @xmath105 be the set that contains all vertices whose parent is @xmath104 in @xmath31 , except the labels in @xmath92 . set @xmath106 could not be an empty set , otherwise , the labels in @xmath92 would consist an mss in @xmath31 , which satisfies the condition of case 1 . there are three situations for @xmath98 and @xmath99 in @xmath39 . situation 1 . @xmath98 is a single - vertex tree in @xmath39 . thus , removing the edge incident to @xmath98 in @xmath31 and @xmath32 keeps @xmath39 still a maximal - af for @xmath31 and @xmath32 . situation 2 . @xmath99 is a single - vertex tree in @xmath39 . thus , removing the edge incident to @xmath99 in @xmath31 and @xmath32 keeps @xmath39 still a maximal - af for @xmath31 and @xmath32 . situation 3 . neither @xmath98 nor @xmath99 is a single - vertex tree in @xmath39 . because @xmath98 and @xmath99 are siblings in @xmath32 , so @xmath98 and @xmath99 are siblings in @xmath39 . moreover , for this situation , we have the following lemma . [ allsiblings - rooted ] let @xmath31 and @xmath32 be two rooted general @xmath24-forests , and let @xmath92 be an mss of @xmath32 that all labels in @xmath92 are siblings in @xmath31 . for any maximal - af @xmath28 for @xmath31 and @xmath32 , if there are two labels in @xmath92 that are siblings in @xmath28 , then all labels in @xmath92 consist an mss in @xmath28 . suppose that labels @xmath98 and @xmath99 belong to @xmath92 and @xmath98 and @xmath99 are siblings in @xmath28 . at first , we show that all labels in @xmath92 are siblings in @xmath28 . there are two cases based on the cardinality of @xmath92 . case ( i ) : @xmath107 . then , @xmath108 . obviously , this case holds true . case ( ii ) : @xmath109 . assume that label @xmath110 . if @xmath98 and @xmath111 are in different connected components in @xmath28 , then because @xmath98 and @xmath111 are siblings in @xmath32 , so at least one of @xmath112 and @xmath113 can not be in @xmath28 , where @xmath112 and @xmath113 are the edges that incident to @xmath98 and @xmath111 in @xmath32 , respectively . therefore , at least one of @xmath98 and @xmath111 is a single - vertex in @xmath28 . but @xmath98 and @xmath99 are siblings in @xmath28 , so @xmath98 is not a single - vertex tree in @xmath28 , thus , @xmath111 is a single - vertex tree in @xmath28 . by attaching the single - vertex tree @xmath111 to the common parent of @xmath98 and @xmath99 in @xmath28 , we could get an agreement forest for @xmath31 and @xmath32 that consists of fewer trees , which contracts the fact that @xmath28 is a maximal - af for @xmath31 and @xmath32 . thus , @xmath98 and @xmath111 must be in the same connected component in @xmath28 . then , because @xmath98 and @xmath111 are siblings in @xmath32 , so @xmath98 and @xmath111 are also siblings in @xmath28 . therefore , all labels in @xmath92 are siblings in @xmath28 . now we show that the labels in @xmath92 consist an mss in @xmath28 . since @xmath28 is a subforest of @xmath32 , the parent of @xmath92 in @xmath32 corresponds to the the parent of @xmath92 in @xmath28 . and because the parent of @xmath92 in @xmath32 has @xmath94 children , so the parent of @xmath92 in @xmath28 has at most @xmath94 children . therefore , the labels in @xmath92 consist an mss in @xmath28 . let @xmath114 be the set that contains all edges @xmath115 $ ] , @xmath116 . by lemma [ allsiblings - rooted ] , all edges in @xmath114 could not be in @xmath39 . therefore , in situation 3 , removing the edges in @xmath114 from @xmath31 keeps @xmath39 still a maximal - af for @xmath31 and @xmath32 . summarizing above analysis , we apply the following step . one of these following three branches keeps @xmath39 a maximal - af for the new @xmath31 and @xmath32 . * step 2 . * ( branch-1 ) remove the edge incident to @xmath98 in both @xmath31 and @xmath32 ; + ( branch-2 ) remove the edge incident to @xmath99 in both @xmath31 and @xmath32 ; + ( branch-3 ) remove the edges in @xmath114 . * some labels in @xmath92 are not siblings in @xmath31 . , we assume that @xmath98 and @xmath99 are not siblings in @xmath31 . let @xmath117 be the common parent of @xmath98 and @xmath99 in @xmath32 . * subcase 3.1 . * @xmath98 and @xmath99 are not in the same connected component in @xmath31 . because @xmath98 and @xmath99 are not in the same connected component in @xmath31 , so @xmath98 and @xmath99 can not be in the same connected component in @xmath39 . thus , at least one of edges @xmath118 $ ] and @xmath119 $ ] in @xmath32 could not be in @xmath39 . therefore , at least one of @xmath98 and @xmath99 is a single - vertex tree . * step 3.1 . * ( branch-1 ) remove the edge incident to @xmath98 in both @xmath31 and @xmath32 ; + ( branch-2 ) remove the edge incident to @xmath99 in both @xmath31 and @xmath32 . one of the two branches must keep @xmath39 still a maximal - af for the new @xmath24-forests @xmath31 and @xmath32 . * subcase 3.2 . * @xmath98 and @xmath99 are in the same connected component in @xmath31 . let @xmath120 be the path in @xmath31 that connects @xmath98 and @xmath99 , in which @xmath121 is the least common ancestor of @xmath98 and @xmath99 , @xmath122 . and let @xmath123 be the edge set that contains all edges in @xmath31 that incident to @xmath124 , @xmath125 , but not on the path @xmath126 . there are also three situations for @xmath98 and @xmath99 in @xmath39 , which are the same as that for case 2 . the first two situations that either @xmath98 or @xmath99 is a single - vertex tree in @xmath39 again cause removing the edge incident to @xmath98 or @xmath99 in @xmath31 and @xmath32 . for situation 3 : neither @xmath98 nor @xmath99 is a single - vertex tree in @xmath39 , we have to analyze detailly here . in this situation , again @xmath98 and @xmath99 are siblings in @xmath39 . moreover , all edges in @xmath123 could not be in @xmath39 . note that because the subtrees in an @xmath24-forest preserve the ancestor - descendant relation , the edges incident to @xmath121 , but not on @xmath126 could not be removed in this subcase ( there may be more than one such edge ) . * step 3.2.*(branch-1 ) remove the edge incident to @xmath98 in both @xmath31 and @xmath32 ; + ( branch-2 ) remove the edge incident to @xmath99 in both @xmath31 and @xmath32 ; + ( branch-3 ) remove the edges in @xmath123 . one of these three branches keeps @xmath39 a maximal - af for the new @xmath31 and @xmath32 . for two given rooted general @xmath24-forests @xmath31 and @xmath32 , if we iteratively apply the above process , branching accordingly based on the cases , then the process will end up with a pair @xmath127 in which @xmath32 contains no mss . when this occurs , the process applies the following step : * final step . * if @xmath32 contains no mss , then construct the ( unique ) maximal - af @xmath39 for @xmath31 and @xmath32 , and convert @xmath39 into an agreement forest for the original @xmath31 and @xmath32 . when @xmath32 contains no mss , by lemma [ simpleroot ] , we can construct the unique maximal - af @xmath39 for @xmath31 and @xmath32 in linear time . the forest @xmath39 may not be a subforest of the original @xmath31 and @xmath32 because step 1 shrinks labels . for this , we should `` expand '' the shrunk labels , in a straightforward way . note that this expanding process may be applied iteratively , but in linear time . summarizing the above discussion , we conclude with the following lemma . [ rootlem ] let @xmath31 and @xmath32 be two rooted general @xmath24-forests . if we apply steps 1 - 3.2 iteratively until @xmath32 contains no mss , then for every maximal - af @xmath39 for the original @xmath31 and @xmath32 , at least one of the branches in the process produces the maximal - af @xmath39 in its final step . fix a maximal - af @xmath39 for the original @xmath31 and @xmath32 . by the above analysis , for each of the cases , at least one of the branches in the corresponding step keeps @xmath39 a maximal - af for @xmath31 and @xmath32 . moreover , when @xmath32 contains no mss , the maximal - af for @xmath31 and @xmath32 becomes unique . combining these two facts , we can conclude that at least one of the branches in the process ends up with a pair @xmath31 and @xmath32 whose maximal - af , after the final step , is @xmath39 . since @xmath39 is an arbitrary maximal - af for @xmath31 and @xmath32 , the lemma is proved . now , according to the discussion given above and main - algorithm , we can present the detail parameterized algorithm for the para - rmaf problem , which is presented in figure [ algrooted - para ] . 4= @xmath128 we consider the correctness and the complexity of the algorithm . to make our discussion more specific , we denote by @xmath129 the original input to the algorithm , and initiate with @xmath130 for @xmath131 . the algorithm is a branch - and - search process , in which step 7 , step 8.1 , and step 8.2 contain branches . the execution of the algorithm can be depicted by a search tree @xmath132 whose leaves correspond to conclusions or solutions generated by the algorithm based on different branches . each internal node of the search tree @xmath132 corresponds to a branch in the search process at step 7 , or step 8.1 , or step 8.2 based on an instance of the problem . the root of the tree @xmath132 is on the instance that is the original input to the algorithm . we will call a path from the root to a leaf in the search tree @xmath132 a _ computational path _ in the process , which corresponds to a particular sequence of executions in the algorithm that leads to a conclusion or solution . the algorithm returns an agreement forest for the original input if and only if there is a computational path that outputs the forest . we first study the correctness of the algorithm . according to step 1 , the algorithm is correct when @xmath133 . therefore , we will assume that @xmath134 and that the algorithm is correct when the input contains no more than @xmath135 @xmath24-forests . if ord@xmath136 , then an maf @xmath33 for @xmath137 , which is a subforest of @xmath31 , must have ord@xmath138 . thus , the instance @xmath139 is a ` no ' . step 2 is correct . by lemma [ reduction rule 1 ] , step 3 is also correct . if @xmath32 has no mss , then by lemma [ simpleroot ] , the unique maximal - af @xmath33 for @xmath31 and @xmath32 can be constructed in linear time . since @xmath33 is the unique maximal - af for @xmath31 and @xmath32 , by lemma [ maximal - maf ] , the instances @xmath139 and @xmath140 have the same set of mafs . by our induction , the algorithm works correctly on @xmath140 . thus , step 4 is correct . if the instance @xmath139 reaches step 5 , then the @xmath24-forest @xmath32 has an mss @xmath92 , and none of labels in @xmath92 is a single - vertex tree in @xmath31 . steps 6 - 8 are applied on the @xmath24-forests @xmath31 and @xmath32 recursively ( during the recursion , step 3 may also be applied ) . these steps remove edges in @xmath31 and @xmath32 thus reduce the sizes of the forests ( step 6 does not remove edges , but it reduces the size of @xmath31 and @xmath32 without changing their essential structures ) . thus , the steps keep the situation , recursively , that @xmath31 and @xmath32 are subforests of @xmath141 and @xmath142 , respectively . this means that during the process of these steps , every agreement forest for @xmath40 remains an agreement forest for the original @xmath143 . these steps continue until the condition in either step 2 or step 4 is met . by the discussion above , step 2 or step 4 then will return a correct solution to the instance @xmath139 , which is either an answer ` no ' , or an agreement forest @xmath39 for @xmath40 with ord@xmath43 , which is also a solution to the original input @xmath129 . thus , no computational path in the algorithm can output an @xmath24-forest that is not a solution to the original input @xmath129 . in particular , if the original input @xmath129 is a ` no ' for the para - rmaf problem , then the algorithm alg - para - rmaf must return an answer ` no ' . on the other hand , suppose that @xmath129 is a ` yes ' and @xmath143 has an maf @xmath39 with ord@xmath43 . let @xmath33 be the maximal - af for @xmath141 and @xmath142 that has @xmath39 as a subforest ( note ord@xmath144 ) . by lemma [ rootlem ] , there is a computational path that starts with @xmath145 and @xmath146 , and applies steps 6 - 8 recursively until @xmath31 and @xmath32 satisfy the condition of step 4 . step 4 then constructs the maximal - af @xmath33 for @xmath141 and @xmath142 . by lemma [ maximal - maf ] and our induction , the recursive call in step 4 will return an agreement forest @xmath28 for @xmath143 with ord@xmath83 . therefore , the algorithm also works correctly in this case . this completes the proof of the correctness of the algorithm . now we consider the complexity of the algorithm . because of step 7 , step 8.1 , and step 8.2 , each branch in the search tree @xmath132 can make at most three ways . moreover , by examining steps 7 , 8.1 , and 8.2 , it is easy to verify that between two consecutive branches in a computational path , the value ord@xmath147 is increased by at least @xmath4 . since the algorithm will stop at step 2 when ord@xmath136 , each computational path in the search tree @xmath132 can go through at most @xmath12 branches . as a consequence , the number of leaves in the search tree @xmath132 is bounded by @xmath148 . it takes time @xmath149 to judge whether two labels are in the same connected component , where @xmath14 is the size of label - set @xmath24 . thus , it is easy to verify that between two consecutive branches , the computational path takes time @xmath150 , where @xmath151 is the number of @xmath24-forests in the original input instance . summarizing all these together , we conclude that the algorithm alg - para - rmaf@xmath139 has its running time bounded by @xmath0 . the para - rmaf problem can be solved in time @xmath0 , where @xmath14 is the size of label - set @xmath24 and @xmath151 is the number of @xmath24-forests in the input instance . the discussion for the para - umaf problem on the instance @xmath152 is similar to that for the para - rmaf problem . however , since unrooted @xmath24-forests preserve no ancestor - descendant relation , there is a little difference . two leaves in an unrooted @xmath24-forest are _ siblings _ if either they are connected by an edge or they have a common neighbor . a _ sibling set _ is a set of leaves that are siblings . a sibling set @xmath92 is _ maximal _ mss ) if either @xmath92 is the label set of a single - edge tree or the common neighbor of @xmath92 has degree at most @xmath95 . an unrooted @xmath24-forest with no mss has an even simpler structure : all its connected components are single - vertex trees . thus , we have the following lemma , which is similar to lemma [ simpleroot ] . [ simpleunroot ] let @xmath31 and @xmath32 be two unrooted general @xmath24-forests . if @xmath32 has no mss , then @xmath31 and @xmath32 have the unique maximal - af which can be constructed in linear time . thus , in the following discussion , we will assume that @xmath32 has an mss @xmath92 that none of labels in @xmath92 is a single - vertex tree in @xmath31 . and , we will assume that labels @xmath98 and @xmath99 belong to @xmath92 . in the following , we consider all possible cases for labels of @xmath92 in @xmath31 . let @xmath39 be a fixed maximal - af for @xmath31 and @xmath32 . * all labels in @xmath92 consist an mss in @xmath31 . in this case , we also treat @xmath92 as an un - decomposable structure . * step 1 . * group all labels in @xmath92 ( and their common neighbor if @xmath92 is not the label set of a single - edge tree ) into an un - decomposable structure , and mark the unit with the same label in @xmath31 and @xmath32 . to implement step 1 , if @xmath92 is the label set of a single - edge tree , then combine the labels into a single vertex that labeled by @xmath100 ; otherwise , simply remove all labels in @xmath92 and label the common neighbor of @xmath92 with @xmath100 . this step not only changes the structures of @xmath31 and @xmath32 , but also replaces the label - set @xmath24 with a new label - set @xmath103 . if we also apply this operation in the maximal - af @xmath39 , then the new @xmath39 remains a maximal - af for the new @xmath31 and @xmath32 . * case 2 . * all labels in @xmath92 are siblings in @xmath31 . obviously , the common neighbor @xmath93 of @xmath92 in @xmath31 has degree not less than @xmath153 . let @xmath154 be the vertex set that contains all vertices that are neighbors of @xmath93 , except the labels in @xmath92 . obviously , @xmath155 . let @xmath156 be the edge that between @xmath157 and @xmath93 in @xmath31 , and let @xmath158 be the edge that between @xmath159 and @xmath93 in @xmath31 . there are three situations for @xmath98 and @xmath99 in @xmath39 . situation 1 . @xmath98 is a single - vertex tree in @xmath39 . thus , removing the edge incident to @xmath98 in @xmath31 and @xmath32 keeps @xmath39 still a maximal - af for @xmath31 and @xmath32 . situation 2 . @xmath99 is a single - vertex tree in @xmath39 . thus , removing the edge incident to @xmath99 in @xmath31 and @xmath32 keeps @xmath39 still a maximal - af for @xmath31 and @xmath32 . situation 3 . neither @xmath98 nor @xmath99 is a single - vertex tree in @xmath39 . because @xmath98 and @xmath99 are siblings in @xmath32 , so @xmath98 and @xmath99 are siblings in @xmath39 . moreover , in this situation , we have the following lemma . [ allsiblings - unrooted ] let @xmath31 and @xmath32 be two unrooted general @xmath24-forests , and let @xmath92 be an mss of @xmath32 that all labels in @xmath92 are siblings in @xmath31 . for any maximal - af @xmath28 for @xmath31 and @xmath32 , if there are two labels in @xmath92 that are siblings in @xmath28 , then all labels in @xmath92 consist an mss in @xmath28 . suppose that labels @xmath98 and @xmath99 belong to @xmath92 and @xmath98 and @xmath99 are siblings in @xmath28 . at first , we show that all labels in @xmath92 are siblings in @xmath28 . case ( i ) : @xmath107 . then , @xmath108 . obviously , this case holds true . case ( ii ) : @xmath109 . suppose that label @xmath110 . if @xmath98 and @xmath111 are in different connected components in @xmath28 , then because @xmath98 and @xmath111 are siblings in @xmath32 , so at least one of @xmath112 and @xmath113 can not be in @xmath28 , where @xmath112 and @xmath113 are the edges that incident to @xmath98 and @xmath111 in @xmath32 , respectively . therefore , at least one of @xmath98 and @xmath111 is a single - vertex tree in @xmath28 . since @xmath98 and @xmath99 are in the same connected component in @xmath28 , @xmath98 is not a single - vertex tree in @xmath28 , thus , @xmath111 is a single - vertex tree in @xmath28 . by attaching the single - vertex tree @xmath111 to the common neighbor of @xmath98 and @xmath99 in @xmath28 ( if @xmath98 and @xmath99 are two labels of a single - edge tree in @xmath28 , then subdividing the edge between @xmath98 and @xmath99 by a new vertex and attaching @xmath111 to the new vertex ) , we could get an agreement forest for @xmath31 and @xmath32 that consists of fewer trees , which contracts the fact that @xmath28 is a maximal - af for @xmath31 and @xmath32 . thus , @xmath98 and @xmath111 must be in the same connected component in @xmath28 . then , because @xmath98 and @xmath111 are siblings in @xmath32 , so @xmath98 and @xmath111 are siblings in @xmath28 . therefore , all labels in @xmath92 are siblings in @xmath28 . now we show that the labels in @xmath92 consist an mss . the common neighbor of @xmath92 in @xmath32 corresponds to the common neighbor of @xmath92 in @xmath28 . because the common neighbor of @xmath92 in @xmath32 has degree at most @xmath95 , so the common neighbor of @xmath92 in @xmath28 also has degree at most @xmath95 . thus , the labels in @xmath92 consist an mss in @xmath28 . by lemma [ allsiblings - unrooted ] , in situation 3 , all labels in @xmath92 are siblings in @xmath39 . if the common neighbor of @xmath92 in @xmath32 has degree @xmath94 , then the common neighbor of @xmath92 in @xmath39 has degree @xmath94 , edges @xmath156 and @xmath158 in @xmath31 can not be in @xmath39 , both of them should be removed . if the common neighbor of @xmath92 in @xmath32 has degree @xmath95 , then the common neighbor of @xmath92 in @xmath39 has degree at most @xmath95 . but the common neighbor of @xmath92 in @xmath31 has degree at least @xmath153 , thus , at least one of @xmath156 and @xmath158 in @xmath31 could not be in @xmath39 . however , the subtrees in unrooted @xmath24-forests do not preserve any ancestor - descendant relation , we can not decide which one of @xmath156 and @xmath158 in @xmath31 should be removed . therefore , we can branch by removing @xmath156 or @xmath158 . in situation 3 , whether or not the common neighbor of @xmath92 in @xmath32 has degree @xmath94 , branching by removing @xmath156 or @xmath158 always is right . summarizing the above analysis , we can apply the following step . one of these following branches keeps @xmath39 a maximal - af for the new @xmath31 and @xmath32 . * step 2 . * ( branch-1 ) remove the edge incident to @xmath98 in both @xmath31 and @xmath32 ; + ( branch-2 ) remove the edge incident to @xmath99 in both @xmath31 and @xmath32 ; + ( branch-3 ) remove the edge @xmath156 in @xmath31 ; + ( branch-4 ) remove the edge @xmath158 in @xmath31 . * some labels in @xmath24 are not siblings in @xmath31 . w.l.o.g . , we assume @xmath98 and @xmath99 are not siblings in @xmath31 . * subcase 3.1 . * @xmath98 and @xmath99 are not in the same connected component in @xmath31 . again that at least one of @xmath98 and @xmath99 must be a single - vertex tree in @xmath39 . we can apply the following step . * step 3.1 . * ( branch-1 ) remove the edge incident to @xmath98 in both @xmath31 and @xmath32 ; + ( branch-2 ) remove the edge incident to @xmath99 in both @xmath31 and @xmath32 . * subcase 3.2 . * @xmath98 and @xmath99 are in the same connected component in @xmath31 . let @xmath160 be the unique path that connects @xmath98 an @xmath99 in @xmath31 , @xmath155 . there are also three situations for @xmath98 and @xmath99 in @xmath39 , which are the same as that for case 2 . the first two situations that either @xmath98 or @xmath99 is a single - vertex tree in @xmath39 again cause removing the edge incident to @xmath98 or @xmath99 in both @xmath31 and @xmath32 . for situation 3 : neither @xmath98 nor @xmath99 is a single - vertex tree in @xmath39 , we have to analyze in detail . because @xmath98 and @xmath99 are siblings in @xmath32 , so @xmath98 and @xmath99 are siblings in @xmath39 . if @xmath98 and @xmath99 are connected by an edge in @xmath39 , then all internal vertices in @xmath126 should be removed by the forced contraction . that is , all the edges that not on the path @xmath126 but incident to a internal vertex in @xmath126 can not be in @xmath39 , thus , all these edges should be removed . if @xmath98 and @xmath99 have a common neighbor in @xmath39 , then only one internal vertex in @xmath126 can be kept , and all the other internal vertices in @xmath126 should be removed by the forced contraction . since the subtrees in unrooted @xmath24-forests do not preserve any ancestor - descendant relation , we do not know which one of the internal vertices in @xmath126 should be kept . on the other side , we know that at least one of @xmath161 and @xmath162 should be removed by the forced contraction . that is , either the edges that incident to @xmath161 but not on @xmath126 or the edges that incident to @xmath162 but not on @xmath126 should be removed . therefore , we can branch by removing the edges that incident to @xmath161 but not on @xmath126 or the edges that incident to @xmath162 but not on @xmath126 . in this situation , whether or not @xmath98 and @xmath99 are connected by an edge in @xmath39 , this branching way is always right . summarizing the above analysis , we can apply the following step . one of these branches keeps @xmath39 a maximal - af for the new @xmath31 and @xmath32 . * step 3.2 . * ( branch-1 ) remove the edge incident to @xmath98 in both @xmath31 and @xmath32 ; + ( branch-2 ) remove the edge incident to @xmath99 in both @xmath31 and @xmath32 ; + ( branch-3 ) the edges that incident to @xmath161 but not on @xmath126 in @xmath31 ; + ( branch-4 ) the edges that incident to @xmath162 but not on @xmath126 in @xmath31 . for two given unrooted general @xmath24-forests @xmath31 and @xmath32 , if we iteratively apply the above process , branching accordingly based on the cases , then the process will end up with a pair @xmath127 in which @xmath32 contains no mss . when this occurs , we again apply the following step : * final step . * if @xmath32 contains no mss , then construct the ( unique ) maximal - af @xmath39 for @xmath31 and @xmath32 , and convert @xmath39 into an agreement forest for the original @xmath31 and @xmath32 . the above analysis finally gives the following conclusion , whose proof is exactly the same as that of lemma [ rootlem ] in subsection 4.1 . [ unrootlem ] let @xmath31 and @xmath32 be two unrooted general @xmath24-forests . if we apply steps 1 - 3.2 iteratively until @xmath32 contains no mss , then for every maximal - af @xmath39 for the original @xmath31 and @xmath32 , at least one of the branches in the process produces the maximal - af @xmath39 in its final step . now we are ready for giving the detail parameterized algorithm for the para - umaf problem , which is presented in figure [ algunrt - para ] . 4= @xmath128 similar to the one for para - rmaf , the algorithm for the para - umaf problem is a combination of the analysis given in section 4.2 and the main - algorithm . comparing the analysis for para - rmaf given in section 4.1 and the analysis for para - umaf given in section 4.2 , we can see that they only differ for case 2 and case 3.2 : case 2 and case 3.2 in section 4.1 branches into three ways while case 2 and case 3.2 in section 4.2 branch into four ways . therefore , the two algorithms only need to differ in step 7 and step 8.2 . the proof of the correctness for the algorithm alg - para - umaf proceeds in exactly the same way , based on the analysis in section 4.2 , as that for the algorithm alg - para - rmaf , which was based on the analysis in section 4.1 . for the computational complexity , since step 7 and step 8.2 of the algorithm alg - para - umaf branches into four ways , the search tree @xmath132 for the algorithm alg - para - umaf has four - way branches . therefore , we can conclude that the number of leaves in the search tree @xmath132 is bounded by @xmath163 . all other analysis is the same as that for the algorithm alg - para - rmaf . as a result , we conclude that the algorithm alg - para - umaf runs in time @xmath2 . the para - umaf problem can be solved in time @xmath2 , where @xmath14 is size of the label - set @xmath24 and @xmath151 is the number of @xmath24-forests in the input instance . in this section , we will present the approximation algorithms for app - rmaf and app - umaf first of all , we give several related definitions , which apply for both app - rmaf and app - umaf . let @xmath28 be an @xmath24-forest ( either rooted or unrooted ) , and let @xmath49 be a subset of edges in @xmath28 . because of forced contraction , we have that ord@xmath164 . an edge - subset @xmath47 of @xmath28 is an _ essential edge - set _ ee - set ) if ord@xmath165 . let @xmath49 be an arbitrary edge - subset of @xmath28 . define the _ essential subset _ of @xmath49 , denoted by @xmath166 , to be a subset of @xmath49 that is an ee - set of @xmath28 such that @xmath167 . obviously , if @xmath49 itself is an ee - set , then @xmath168 . note that for an edge - subset @xmath49 of @xmath28 which is not an ee - set , there maybe more than one essential subset of @xmath49 . our approximation algorithm for app - maf ( app - rmaf or app - umaf ) consist of a sequence of meta - steps " . an _ edge - removal meta - step _ ( or simply meta - step ) of an algorithm is a collection of consecutive computational steps in the algorithm that on an instance @xmath169 of app - maf removes certain edges in the forests in @xmath169 ( and applies the forced contraction ) . our approximation algorithms for app - maf ( app - rmaf or app - umaf ) have the following general framework . 4= @xmath128 the performance of the approximation algorithm alg - app - maf heavily depends on the quality of the meta - steps we employ in step 3 of the algorithm . thus , we introduce the following concept that measures the quality of a meta - step , where @xmath170 is an arbitrary real number . * definition - r . * let @xmath171 be an instance of app - maf ( app - rmaf or app - umaf ) , and let @xmath172 be an edge - removal meta - step that removes a set @xmath173 of edges in the forests in @xmath174 . meta - step @xmath172 keeps a _ @xmath35 if the set @xmath173 contains a subset @xmath175 of edges in @xmath31 such that no edge in @xmath176 is in any agreement forest for @xmath177 , and for each agreement forest @xmath28 for @xmath174 , there always exists an ee - set @xmath178 of @xmath31 , @xmath179 , @xmath180 , and no edge in @xmath178 is in @xmath28 . * remark 1 . * @xmath175 contains all edges in @xmath173 that in @xmath31 . @xmath175 may not be an ee - set of @xmath31 , but @xmath178 should be an ee - set of @xmath31 . * remark 2 . * by definition , if an edge - removal meta - step removes only edges that not in any agreement forest for the instance , then this meta - step keeps ratio @xmath35 for any @xmath170 . define an edge - removal meta - step is _ safe _ if it keeps ratio @xmath35 for any @xmath170 . define the order of an maf for the instance @xmath181 of app - maf ( app - rmaf or app - umaf ) to be the _ optimal order _ for the instance , denoted @xmath182 . [ equation ] let @xmath171 be an instance of app - maf ( app - rmaf or app - umaf ) , and let @xmath172 be an edge - removal meta - step on @xmath174 producing instance @xmath183 . if @xmath172 keeps ratio @xmath35 , then @xmath184 . let @xmath28 be a fixed maf for @xmath171 . according to definition - r , there exists an edge - set @xmath185 that no edge in @xmath186 is in any agreement forest for @xmath177 . thus , instances @xmath183 and @xmath177 have the same collection of solutions . because @xmath187 , so @xmath183 and @xmath188 also have the same collection of solutions . @xmath189 is the same as the optimal order for @xmath188 . since @xmath28 is an agreement forest for @xmath181 , in order to construct an agreement forest for @xmath188 by removing edges from @xmath28 , we just need removing the edges from @xmath28 that are not in @xmath190 to make the new @xmath28 be a subforest of @xmath190 . here , we denote by @xmath191 the subset of @xmath192 in which the edges are in @xmath28 , and denote by @xmath193 the subset of @xmath192 in which the edges are not in @xmath28 . obviously , @xmath194 is an agreement forest for @xmath188 . in the following , we analyze the order of @xmath194 detailly . according to definition - r , we have that for @xmath28 , there exists an ee - set @xmath178 of @xmath31 , @xmath179 , @xmath195 , and no edge in @xmath178 is in @xmath28 . note that we can easily get an essential subset of @xmath185 that contains @xmath178 , thus , we can assume that @xmath196 . therefore , we have that @xmath197 and @xmath198 . because @xmath199 , so @xmath200 . therefore , @xmath201 . since @xmath202 and @xmath203 , we have @xmath184 . [ ratio ] let @xmath204 be an instance of app - maf ( app - rmaf or app - umaf ) , and let @xmath57 be an arbitrary real number . if each meta - step in step 3 of algorithm alg - app - maf keeps ratio not greater than @xmath205 and that the algorithm alg - app - maf halts on the instance @xmath174 , then the output of alg - app - maf is an agreement forest for @xmath174 whose order is at most @xmath205 times the optimal value for @xmath174 . suppose the sequence of meta - steps in the algorithm is @xmath206 , where for each @xmath38 , @xmath207 , meta - step @xmath208 removes an edge - set @xmath209 from the instance @xmath210 produces an instance @xmath211 . by the judgement condition in step 3 , we can make sure that @xmath212 is a subforest of @xmath213 . thus , @xmath212 is an maf for @xmath214 , ord@xmath215 . because each meta - step @xmath208 in @xmath92 , @xmath207 , removes certain edges from forests in @xmath216 , so every agreement forest for @xmath217 is also an agreement forest for @xmath216 . therefore , the forest @xmath212 returned by step 4 is also an agreement forest for the original input instance @xmath218 . for each meta - step @xmath219 , @xmath220 , by lemma [ equation ] , there is @xmath221 . note that , @xmath222 . therefore , for each meta - step @xmath219 , we have the inequality @xmath223 . then , we add up these inequalities for all meta - steps in @xmath92 , and get @xmath224 , where @xmath225 . from this , we can easily get @xmath226 , which proves the theorem . we develop meta - steps for app - rmaf in this subsection . thus , all x - forests considered in this subsection are rooted . as given in the algorithm alg - app - maf ( see figure [ alg - app ] ) , for each execution of step 3 in the algorithm , we are given a fixed integer @xmath227 and an instance @xmath204 of app - rmaf , which is a collection of rooted x - forests , with @xmath31 is a subforest of @xmath228 , and , as long as @xmath229 , meta - steps are applied on @xmath31 and @xmath37 . in the following , we show how these meta - steps are constructed based on different structures of @xmath31 and @xmath37 . let @xmath39 be a fixed maf for @xmath204 , and let @xmath33 be a maximal - af for @xmath31 and @xmath37 that contains @xmath39 . since @xmath39 is a subforest of @xmath33 , if an edge @xmath45 of @xmath31 is not in @xmath33 , then @xmath45 is also not in @xmath39 . an execution of reduction rule 1 on @xmath31 and @xmath37 can be regarded as an edge - removal meta - step . by lemma [ reduction rule 1 ] and remark 2 of definition - r , we can easily get the following lemma . [ reduction rule 1 safe ] reduction rule 1 is safe . by lemma [ simpleroot ] , if @xmath37 has no mss , then @xmath31 and @xmath37 have an unique maf @xmath28 , which either is isomorphic to @xmath37 or consists of a collection of single - vertex trees . and by a series of executions of reduction rule 1 on @xmath31 and @xmath37 , there is @xmath230 , which satisfies the judgement condition of step 3 in algorithm alg - app - maf . therefore , in the following discussion , we will assume that @xmath37 has an mss @xmath92 . note that the instances in our discussion are strongly reducible , so none of labels in @xmath92 is a single - vertex tree in @xmath31 . , we will assume that labels @xmath98 and @xmath99 belong to @xmath92 . * case 1 . * all labels in @xmath92 consist an mss in @xmath31 . * meta - step 1 . * group all labels in @xmath92 and their parent into an un - decomposable structure , and mark the unit with the same label in @xmath31 and @xmath37 . the implementation of meta - step 1 is the same as that of step 1 for para - rmaf . this meta - step can be regarded as a special meta - step that does not remove any edges in the instance . instead , it groups certain structures in some x - forests into un - decomposable units . using the notation in definition - r , @xmath231 . and , we have [ meta - step 1 ] meta - step 1 is safe . this meta - step may lead a subtle problem in the following discussion . because meta - step 1 changes the label - sets of @xmath31 and @xmath37 , so the label - sets of @xmath31 and @xmath37 are different from the label - sets of the other forests in the instance . thus , there is ambiguity of the sentence there exists an maf for the new instance " . note that this operation is just simply for notational convenience . we still can construct an maf for the new instance if we expand " these combined leaves in @xmath31 and @xmath37 . therefore , in the following discussion , we can simply say that there exists an maf @xmath39 for the new instance , although the label - sets of the forests in the instance are different . note again that in the maf @xmath39 for the instance , there maybe no such an un - decomposable structure , because some labels in @xmath92 maybe in different connected components in @xmath39 . however , the maximal - af @xmath33 for @xmath31 and @xmath37 that contains @xmath39 must have such an un - decomposable structure . therefore , in the following discussion , we can assume that the maximal - af @xmath33 for @xmath31 and @xmath37 that contains @xmath39 has been applied all possible group " operations so that @xmath31 , @xmath37 , and @xmath33 have the same new label - set . similarly , even though @xmath39 and @xmath33 have different label - sets , we can also simply say that @xmath33 contains @xmath39 , because we just need to expand these combined leaves in @xmath33 . let @xmath232 be the forest that getting by expanding these combined leaves in @xmath33 . obviously , @xmath232 and @xmath39 have the same label - set , and @xmath39 is a subforest of @xmath232 . therefore , for any edge @xmath45 that in @xmath31 or @xmath37 , if @xmath45 is not in @xmath33 , then @xmath45 is not in @xmath232 . thus , @xmath45 is also not in @xmath39 . * case 2 . * all labels in @xmath92 are siblings in @xmath31 . let @xmath104 be the common parent of @xmath92 in @xmath31 . in this case , there must exist a vertex @xmath30 in @xmath31 which is a child of @xmath104 but not belongs to @xmath92 ( @xmath30 maybe a non - leaf ) . let @xmath45 be the edge that between @xmath104 and @xmath30 in @xmath31 . * meta - step 2 . * remove the edges that incident to @xmath98 and @xmath99 in @xmath31 and @xmath37 , and remove edge @xmath45 . [ meta - step 2 ] meta - step 2 keeps ratio @xmath1 . let @xmath112 and @xmath233 be the edges that incident to @xmath98 and @xmath99 in @xmath31 , respectively , and let @xmath234 and @xmath235 be the edges that incident to @xmath98 and @xmath99 in @xmath37 , respectively . using the notations in definition - r , we have @xmath236 and @xmath237 . in the @xmath24-forest @xmath238 , labels @xmath98 and @xmath99 are single - vertex trees . thus , every maximal - af for @xmath238 and @xmath37 have @xmath98 and @xmath99 as single - vertex trees , no edges in @xmath239 can be in any maximal - af for @xmath238 and @xmath37 . because every agreement forest for @xmath177 must be a subforest of a maximal - af for @xmath238 and @xmath37 , so no edges in @xmath239 can be in any agreement forest for @xmath177 . there are three situations for @xmath98 and @xmath99 in the maximal - af @xmath33 for @xmath31 and @xmath37 which contains @xmath39 . situation 1 . @xmath98 is a single - vertex tree in @xmath33 . thus , @xmath112 is not in @xmath33 , and @xmath112 is also not in @xmath39 . therefore , we can pick @xmath240 as the set @xmath241 , which satisfies : @xmath242 , @xmath243 . note that @xmath244 is not greater than @xmath245 . moreover , since @xmath31 is irreducible and @xmath98 is not a single - vertex tree in @xmath31 , the set @xmath241 is an ee - set of @xmath31 . therefore , for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 2 to keep a ratio @xmath1 . situation 2 . @xmath99 is a single - vertex tree in @xmath33 . thus , @xmath233 is not in @xmath33 , and @xmath233 is also not in @xmath39 . then similarly we let @xmath246 and can verify that for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 2 to keep a ratio @xmath1 . situation 3 . @xmath98 and @xmath99 are siblings in @xmath33 . then , by lemma [ allsiblings - rooted ] , the labels of @xmath92 consist an mss in @xmath33 . in order to make the labels of @xmath92 consist an mss in @xmath31 , edge @xmath45 should be removed . that is , in this situation , edge @xmath45 is not in @xmath33 , and @xmath45 is also not in @xmath39 . thus , in this situation , we let @xmath247 , and verify easily that for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 2 to keep a ratio @xmath1 . this verifies that the set @xmath175 satisfies all conditions in definition - r to make meta - step 2 to keep a ratio @xmath1 . thus , meta - step 2 keeps ratio @xmath1 . * some labels in @xmath92 are not siblings in @xmath31 . w.l.o.g . , we assume @xmath98 and @xmath99 are not siblings in @xmath31 . * subcase 3.1 . * @xmath98 and @xmath99 are in different connected components in @xmath31 . * meta - step 3.1 . * remove the edges incident to @xmath98 and @xmath99 in both @xmath31 and @xmath37 . [ meta - step 3.1 ] meta - step 3.1 keeps ratio @xmath23 . let @xmath112 and @xmath233 be the edges that incident to @xmath98 and @xmath99 in @xmath31 , respectively , and let @xmath234 and @xmath235 be the edges that incident to @xmath98 and @xmath99 in @xmath37 , respectively . using the notations in definition - r , we have @xmath248 and @xmath249 . in the @xmath24-forest @xmath238 , labels @xmath98 and @xmath99 are single - vertex trees . thus , every maximal - af for @xmath238 and @xmath37 have @xmath98 and @xmath99 as single - vertex trees , no edges in @xmath239 can be in any maximal - af for @xmath238 and @xmath37 . because every agreement forest for @xmath177 must be a subforest of a maximal - af for @xmath238 and @xmath37 , so no edges in @xmath239 can be in any agreement forest for @xmath177 . there are two situations for @xmath98 and @xmath99 in the maximal - af @xmath33 for @xmath31 and @xmath37 which contains @xmath39 . situation 1 . @xmath98 is a single - vertex tree in @xmath33 . thus , @xmath112 is not in @xmath33 , and @xmath112 is also not in @xmath39 . therefore , we can pick @xmath240 as the set @xmath241 , which satisfies : @xmath242 , @xmath250 . moreover , since @xmath31 is irreducible and @xmath98 is not a single - vertex tree in @xmath31 , the set @xmath241 is an ee - set of @xmath31 . therefore , for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 3.1 to keep a ratio @xmath23 . situation 2 . @xmath99 is a single - vertex tree in @xmath33 . thus , @xmath233 is not in @xmath33 , and @xmath233 is also not in @xmath39 . then similarly we let @xmath246 and can verify that for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 3.1 to keep a ratio @xmath23 . this verifies that the set @xmath175 satisfies all conditions in definition - r to make meta - step 3.1 to keep a ratio @xmath23 . thus , meta - step 3.1 keeps ratio @xmath23 . * subcase 3.2 . * @xmath98 and @xmath99 are in the same connected component in @xmath31 . let @xmath251 be the path that connects @xmath98 and @xmath99 in @xmath31 , and let @xmath121 be the least common ancestor of @xmath98 and @xmath99 in @xmath31 , @xmath252 . let @xmath123 be the edge set that contains all edge that incident to @xmath124 , @xmath116 , @xmath253 , but not on the path @xmath126 , and let @xmath45 be an arbitrary edge of @xmath123 . * meta - step 3.2 . * remove the edges incident to @xmath98 and @xmath99 in both @xmath31 and @xmath37 , and remove edge @xmath45 . [ meta - step 3.2 ] meta - step 3.2 keeps ratio @xmath1 . let @xmath112 and @xmath233 be the edges that incident to @xmath98 and @xmath99 in @xmath31 , respectively , and let @xmath234 and @xmath235 be the edges that incident to @xmath98 and @xmath99 in @xmath37 , respectively . using the notations in definition - r , we have @xmath236 and @xmath237 . in the @xmath24-forest @xmath238 , labels @xmath98 and @xmath99 are single - vertex trees . thus , every maximal - af for @xmath238 and @xmath37 have @xmath98 and @xmath99 as single - vertex trees , no edges in @xmath239 can be in any maximal - af for @xmath238 and @xmath37 . because every agreement forest for @xmath177 must be a subforest of a maximal - af for @xmath238 and @xmath37 , so no edges in @xmath239 can be in any agreement forest for @xmath177 . there are three situations for @xmath98 and @xmath99 in the maximal - af @xmath33 for @xmath31 and @xmath37 which contains @xmath39 . situation 1 . @xmath98 is a single - vertex tree in @xmath33 . thus , @xmath112 is not in @xmath33 , and @xmath112 is also not in @xmath39 . therefore , we can pick @xmath240 as the set @xmath241 , which satisfies : @xmath242 , @xmath243 . moreover , since @xmath31 is irreducible and @xmath98 is not a single - vertex tree in @xmath31 , the set @xmath241 is an ee - set of @xmath31 . therefore , for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 3.2 to keep a ratio @xmath1 . situation 2 . @xmath99 is a single - vertex tree in @xmath33 . thus , @xmath233 is not in @xmath33 , and @xmath233 is also not in @xmath39 . then similarly we let @xmath246 and can verify that for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 3.2 to keep a ratio @xmath1 . situation 3 . @xmath98 and @xmath99 are siblings in @xmath33 . in order to make labels @xmath98 and @xmath99 be siblings in @xmath31 , all the edges in @xmath123 should be removed . that is , in this situation , all the edges in @xmath123 are not in @xmath33 , so all the edges in @xmath123 are not in @xmath39 . here , we just remove the edge @xmath45 in @xmath123 , so we let @xmath247 , and verify easily that for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 3.2 to keep a ratio @xmath1 . this verifies that the set @xmath175 satisfies all conditions in definition - r to make meta - step 3.2 to keep a ratio @xmath1 . thus , meta - step 3.2 keeps ratio @xmath1 . for two rooted x - forests @xmath31 and @xmath37 , if we iteratively apply the above process based on the cases , then the process will end up that @xmath31 is isomorphic to @xmath37 . ( to here , we also expand " these combined labels in @xmath31 and @xmath37 . ) now , combining the general framework given in figure [ alg - app ] and the meta - steps given above , we are ready to present our approximation algorithm for app - rmaf . 4= @xmath128 [ alg - app - rmaf - ratio ] algorithm alg - app - rmaf is @xmath1-approximation algorithm for the app - rmaf problem that runs in time @xmath254 , where @xmath14 is the size of the label - set @xmath24 and @xmath151 is the number of forests in the input instance . by lemmas [ meta - step 1 ] , [ meta - step 2 ] , [ meta - step 3.1 ] , and [ meta - step 3.2 ] , each of the meta - steps keeps a ratio bounded by @xmath1 . by theorem [ ratio ] , if the algorithm alg - app - maf uses these meta - steps in step 3 , and halts on an instance @xmath174 of app - rmaf , then the algorithm alg - app - rmaf produces an agreement forest for the instance @xmath174 whose order is bounded by @xmath1 times the optimal value for @xmath174 . therefore , to show that the algorithm alg - app - rmaf is a @xmath1-approximation algorithm for the app - rmaf problem , it suffices to show that on any instance @xmath174 of app - rmaf , the algorithm alg - app - rmaf runs in time @xmath255 . by the above discussion , as long as @xmath229 , at least one of the above meta - steps is applicable . let @xmath256 . then , the number of vertices plus the number of edges in an x - forest is @xmath149 . by the algorithm alg - app - rmaf , a meta - step in step 3 is applied on @xmath31 and @xmath37 only when @xmath229 . under the condition @xmath229 , it is easy to verify that each of the meta - steps 2 , 3.1 , 3.2 , and reduction rule 1 removes at least one edge in @xmath257 . therefore , the total number of times these meta - steps can be applied is bounded by @xmath149 . now consider meta - step 1 . initially , each vertex in @xmath31 and @xmath37 is an un - decomposable unit . thus , the total number of un - decomposable units in @xmath257 is @xmath149 . each application of meta - step 1 groups three un - decomposable units into a single un - decomposable unit , in each of @xmath31 and @xmath37 . therefore , meta - step 1 can be applied at most @xmath149 times . summarizing the above discussion , we conclude that if the algorithm alg - app - rmaf uses these meta - steps in step 3 , then the number of times meta - steps are applied in each execution of step 3 is @xmath149 . moreover , it is not very difficult to see that with careful implementation of the data structure representing x - forests , the running time of each of the meta - steps , can be bounded by @xmath149 . therefore , the running time of the algorithm is @xmath254 , where @xmath256 and @xmath151 is the number of x - forests in the input instance . in this subsection , we develop meta - steps for app - umaf . let @xmath39 be a fixed maf for the instance @xmath171 of app - umaf , and let @xmath33 be a maximal - af for @xmath31 and @xmath37 that contains @xmath39 . by lemma [ simpleunroot ] , if @xmath37 has no mss , then @xmath31 and @xmath37 have an unique maf , which consists of a collection of single - vertex trees . and by a series of executions of reduction rule 1 on @xmath31 and @xmath37 , we can get that @xmath258 , which satisfies the judgement condition of step 3 in algorithm alg - app - maf . therefore , in the following discussion , we will assume that @xmath37 has an mss @xmath92 . note that none of labels in @xmath92 is a single - vertex tree in @xmath31 . , we will also assume that labels @xmath98 and @xmath99 belong to @xmath92 . * case 1 . * all labels in @xmath92 consist an mss in @xmath31 . * meta - step 1 . * group all labels in @xmath92 ( and their common neighbor if @xmath92 is not the label set of a single - edge tree ) into an un - decomposable structure , and mark the unit with the same label in @xmath31 and @xmath32 . the implementation of meta - step 1 is the same as that of step 1 for para - umaf . this meta - step also can be regarded as a special meta - step that does not remove any edges in the instance . and , we have [ unrooted meta - step 1 ] meta - step 1 is safe . * case 2 . * all labels in @xmath92 are siblings in @xmath31 . obviously , the common neighbor @xmath93 of @xmath92 in @xmath31 has degree greater than @xmath259 . let @xmath156 and @xmath260 be two arbitrarily edges in @xmath31 that incident to @xmath93 but not incident to the labels in @xmath92 . * meta - step 2 . * remove the edges that incident to @xmath98 and @xmath99 in both @xmath31 and @xmath32 , and remove edges @xmath156 and @xmath260 . [ unrooted meta - step 2 ] meta - step 2 keeps ratio @xmath3 . let @xmath112 and @xmath233 be the edges that incident to @xmath98 and @xmath99 in @xmath31 , respectively , and let @xmath234 and @xmath235 be the edges that incident to @xmath98 and @xmath99 in @xmath37 , respectively . using the notations in definition - r , we have @xmath261 and @xmath262 . in the @xmath24-forest @xmath238 , labels @xmath98 and @xmath99 are single - vertex trees . thus , every maximal - af for @xmath238 and @xmath37 have @xmath98 and @xmath99 as single - vertex trees , no edges in @xmath239 can be in any maximal - af for @xmath238 and @xmath37 . because every agreement forest for @xmath177 must be a subforest of a maximal - af for @xmath238 and @xmath37 , so no edges in @xmath239 can be in any agreement forest for @xmath177 . there are three situations for @xmath98 and @xmath99 in the maximal - af @xmath33 for @xmath31 and @xmath37 which contains @xmath39 . situation 1 . @xmath98 is a single - vertex tree in @xmath33 . thus , @xmath112 is not in @xmath33 , and @xmath112 is also not in @xmath39 . therefore , we can pick @xmath240 as the set @xmath241 , which satisfies : @xmath242 , @xmath263 . note that @xmath244 is not greater than @xmath264 . moreover , since @xmath31 is irreducible and @xmath98 is not a single - vertex tree in @xmath31 , the set @xmath241 is an ee - set of @xmath31 . therefore , for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 2 to keep a ratio @xmath3 . situation 2 . @xmath99 is a single - vertex tree in @xmath33 . thus , @xmath233 is not in @xmath33 , and @xmath233 is also not in @xmath39 . then similarly we let @xmath246 and can verify that for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 2 to keep a ratio @xmath3 . situation 3 . @xmath98 and @xmath99 are siblings in @xmath33 . by lemma [ allsiblings - unrooted ] , all labels of @xmath92 are siblings in @xmath33 . because the labels of @xmath92 consist an mss in @xmath37 , so the labels of @xmath92 consist an mss in @xmath33 . in order to make the labels of @xmath92 consist an mss in @xmath31 , the degree of @xmath93 can not be greater than @xmath259 . that is , at least one of edges @xmath156 and @xmath260 in @xmath31 should be removed . if @xmath156 is not in @xmath33 , then @xmath156 is also not in @xmath39 . let @xmath265 , we can easily verify that for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 2 to keep a ratio @xmath3 . the same argument is applied to the case that @xmath260 is not in @xmath33 . this verifies that the set @xmath175 satisfies all conditions in definition - r to make meta - step 2 to keep a ratio @xmath3 . thus , meta - step 2 keeps ratio @xmath3 . * some labels in @xmath92 are not siblings in @xmath31 . , we will assume that labels @xmath98 and @xmath99 are not siblings in @xmath31 . * subcase 3.1 . * @xmath98 and @xmath99 are in different connected components in @xmath31 . * meta - step 3.1 . * remove the edges incident to @xmath98 and @xmath99 in both @xmath31 and @xmath37 . [ unrooted meta - step 3.1 . ] meta - step 3.1 keeps ratio @xmath23 . let @xmath112 and @xmath233 be the edges that incident to @xmath98 and @xmath99 in @xmath31 , respectively , and let @xmath234 and @xmath235 be the edges that incident to @xmath98 and @xmath99 in @xmath37 , respectively . using the notations in definition - r , we have @xmath248 and @xmath249 . in the @xmath24-forest @xmath238 , labels @xmath98 and @xmath99 are single - vertex trees . thus , every maximal - af for @xmath238 and @xmath37 have @xmath98 and @xmath99 as single - vertex trees , no edges in @xmath239 can be in any maximal - af for @xmath238 and @xmath37 . because every agreement forest for @xmath177 must be a subforest of a maximal - af for @xmath238 and @xmath37 , so no edges in @xmath239 can be in any agreement forest for @xmath177 . there are two situations for @xmath98 and @xmath99 in the maximal - af @xmath33 for @xmath31 and @xmath37 which contains @xmath39 . situation 1 . @xmath98 is a single - vertex tree in @xmath33 . thus , @xmath112 is not in @xmath33 , so @xmath112 also is not in @xmath39 . therefore , we can pick @xmath240 as the set @xmath241 , which satisfies : @xmath242 , @xmath250 . moreover , since @xmath31 is irreducible and @xmath98 is not a single - vertex tree in @xmath31 , the set @xmath241 is an ee - set of @xmath31 . therefore , for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 3.1 to keep a ratio @xmath23 . situation 2 . @xmath99 is a single - vertex tree in @xmath33 . thus , @xmath233 is not in @xmath33 , so @xmath233 also is not in @xmath39 . then similarly we let @xmath246 and can verify that for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 3.1 to keep a ratio @xmath23 . this verifies that the set @xmath175 satisfies all conditions in definition - r to make meta - step 3.1 to keep a ratio @xmath23 . thus , meta - step 3.1 keeps ratio @xmath23 . * subcase 3.2 . * @xmath98 and @xmath99 are in the same connected component in @xmath31 . let @xmath251 be the path that connects @xmath98 and @xmath99 in @xmath31 . and let @xmath123 be the edge set that contains all edge that incident to @xmath124 , @xmath116 , but not on the path @xmath126 . obviously , @xmath266 . let @xmath156 and @xmath260 be two arbitrary edges in @xmath123 . * meta - step 3.2 . * remove the edges that incident to @xmath98 and @xmath99 in both @xmath31 and @xmath37 , and remove edges @xmath156 and @xmath260 in @xmath31 . [ meta - step 3.2 . ] meta - step 3.2 keeps ratio @xmath3 . let @xmath112 and @xmath233 be the edges that incident to @xmath98 and @xmath99 in @xmath31 , respectively , and let @xmath234 and @xmath235 be the edges that incident to @xmath98 and @xmath99 in @xmath37 , respectively . using the notations in definition - r , we have @xmath261 and @xmath262 . in the @xmath24-forest @xmath238 , labels @xmath98 and @xmath99 are single - vertex trees . thus , every maximal - af for @xmath238 and @xmath37 have @xmath98 and @xmath99 as single - vertex trees , no edges in @xmath239 can be in any maximal - af for @xmath238 and @xmath37 . because every agreement forest for @xmath177 must be a subforest of a maximal - af for @xmath238 and @xmath37 , so no edges in @xmath239 can be in any agreement forest for @xmath177 . there are three situations for @xmath98 and @xmath99 in the maximal - af @xmath33 for @xmath31 and @xmath37 which contains @xmath39 . situation 1 . @xmath98 is a single - vertex tree in @xmath33 . thus , @xmath112 is not in @xmath33 , and @xmath112 is also not in @xmath39 . therefore , we can pick @xmath240 as the set @xmath241 , which satisfies : @xmath242 , @xmath263 . note that @xmath244 is always less than @xmath264 . moreover , since @xmath31 is irreducible and @xmath98 is not a single - vertex tree in @xmath31 , the set @xmath241 is an ee - set of @xmath31 . therefore , for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 3.2 to keep a ratio @xmath3 . situation 2 . @xmath99 is a single - vertex tree in @xmath33 . thus , @xmath233 is not in @xmath33 , and @xmath233 is also not in @xmath39 . then similarly we let @xmath246 and can verify that for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 3.2 to keep a ratio @xmath3 . situation 3 . @xmath98 and @xmath99 are siblings in @xmath33 . thus , @xmath98 and @xmath99 either are connected by an edge or have a common neighbor in @xmath33 . if @xmath98 and @xmath99 are connected by an edge in @xmath33 , then all edges in @xmath123 should be removed . here , we just remove edges @xmath156 and @xmath260 , and let @xmath267 , we can easily verify that for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 3.2 to keep a ratio @xmath3 . if @xmath98 and @xmath99 have a common neighbor in @xmath33 , then at least one of @xmath156 and @xmath260 should be removed . if @xmath156 is not in @xmath33 , then @xmath156 is also not in @xmath39 . let @xmath265 , we can verify that for the agreement forest @xmath39 , the set @xmath241 satisfies all conditions in definition - r to make meta - step 3.2 to keep a ratio @xmath3 . the same argument is applied to the case that @xmath260 is not in @xmath33 . this verifies that the set @xmath175 satisfies all conditions in definition - r to make meta - step 3.2 to keep a ratio @xmath3 . thus , meta - step 3.2 keeps ratio @xmath3 . for two unrooted general x - forests @xmath31 and @xmath37 , if we iteratively apply the above process based on the cases , then the process will end up that @xmath31 is isomorphic to @xmath37 . ( to here , we also expand " these combined labels in @xmath31 and @xmath37 . ) the approximation algorithm alg - app - umaf for app - umaf proceeds similarly with the alg - app - rmaf for app - rmaf . due to limit space , we will not present the details of alg - app - umaf here . for app - umaf , we have the following theorem , whose proof is similar to that for theorem [ alg - app - rmaf - ratio ] . [ alg - app - umaf - ratio ] algorithm alg - app - umaf is @xmath3-approximation algorithm for the app - umaf problem that runs in time @xmath254 , where @xmath14 is the size of the label - set @xmath24 and @xmath151 is the number of forests in the input instance . we have implemented our algorithms alg - para - rmaf and alg - app - rmaf in c++ , obtain programs pmaf and amaf , respectively . let @xmath171 be an instance of maf . the program pmaf searches for the optimal value for the instance @xmath174 by starting with a lower bound @xmath12 of opt@xmath268 and increasing @xmath12 until it can return an agreement forest for @xmath174 whose order is not greater than @xmath12 . firstly , we use the program amaf to get an agreement forest whose order is @xmath269 . according to the theoretical analysis given above , we have that @xmath269 is not greater than @xmath1 times the optimal value for the instance . therefore , @xmath270 , and the program pmaf can start with @xmath271 . we test our programs for both simulated and biological data on a 3.06ghz pentium(r ) dual - core cpu system with 2 gb of ram running windows xp . the simulated data are generated by using the following three - stage approach . firstly , generating a random rooted binary phylogenetic tree with @xmath14 labels . our way of generating random rooted binary phylogenetic trees is the same as that in @xcite . we use the integers from @xmath4 to @xmath14 to represent the @xmath14 irrelevant labels . at first , a bipartition on the @xmath14 integers is created by randomly cutting the list into two non - empty parts . this bipartition represents the edges adjacent to the root node of the tree being constructed . then , each of the two induced partitions is randomly split into two lists to create a further bipartition of these sets . new bipartitions are then created recursively by cutting elements of previously created bipartitions into two sets until the bipartitions only consists of singleton elements . thus , the tree is created by starting at the root and creating bipartitions ( edges ) until the leaf nodes are reached . the labels of the leaves are the singleton elements in the bipartitions , respectively . to maintain the consistency with the previous definition of phylogenetic trees , a new leaf labeled with @xmath34 would be attached to the root of the tree . and the leaf @xmath34 will be regarded as the new root of this tree . secondly , transforming the binary tree to a general tree . we randomly choose some internal edges in the binary tree , whose two ends are both internal vertices , to contract . the contracting operation is applied by removing the two endpoints @xmath272 and @xmath30 of the edge and introducing a new vertex which is adjacent to every vertex that is adjacent to at least one of @xmath272 and @xmath30 . the number of edges that are contracted is also random . finally , transforming the original tree into other resulting trees by using a known number of spr operations . suppose @xmath273 resulting trees are constructed by applying @xmath274 spr operations on the original tree @xmath275 , respectively . then the instance which consists of the original tree @xmath275 and the @xmath273 resulting trees @xmath276 has an maf whose order could not be greater than @xmath277 . note that there are two reasons for that the order of the maf for the instance could not be greater than @xmath277 : ( 1 ) . the sequence of spr operations we use to obtain @xmath67 from @xmath275 , @xmath278 , may not be the shortest such sequence , that is , the order of the maf for @xmath275 and @xmath67 maybe less than @xmath279 ; ( 2 ) . the construction of @xmath67 and @xmath280 may use a same spr operation , @xmath281 , that is , the order of the maf for @xmath67 and @xmath280 maybe less than @xmath282 . in the following discussion , we denote by t@xmath283 a set of instances of maf , each instance of which consists of @xmath151 phylogenetic trees with @xmath14-leaves . we run our program pmaf on the simulated data set t@xmath283 , where @xmath284 and @xmath285 . for each instance set t@xmath283 , @xmath284 , @xmath285 , it contains at least @xmath286 instances . figure [ average - time ] shows the mean running time of pmaf on these simulated instances with the given order of maf . the slopes of the curves in the figure are always between @xmath4 and @xmath287 , which indicating that the actual running time of the algorithm is between its worst - case running time of @xmath0 and its best running time @xmath288 ( in the best case , all branches in the search tree of the algorithm for the instance always make two ways ) . the running time of amaf on each instance of the simulated data set t@xmath283 , where @xmath284 and @xmath285 , is always less than one second . figure [ worst - ratio ] shows the worst approximation ratio of amaf for the simulated data set t@xmath283 , where @xmath284 and @xmath285 . the top points of the curves in the figure are always not greater than @xmath1 , which indicating that the actual approximation ratio of the algorithm is less than @xmath1 . moreover , as can be seen from the figure , when the order of maf is greater than @xmath289 , there is a down trending for the ratio of the algorithm with the order of maf increasing . this is because as the increasing of the order of maf , the number of right " edges in a tree , which are the edges that are not in the maf for the instance , is increasing . thus , the probability of removing right " edges by the algorithm is increasing . therefore , the ratio of the algorithm trends down . we run pmaf on the protein tree data set @xcite . the protein tree data set consists of 22,437 binary protein trees , each constructed from a set of proteins covering from 4 to 144 microbial genomes . among these trees , there are @xmath290 protein trees covering all 144 microbial genomes . thus , the @xmath290 trees have the same label - set , and it is of biological meaning to compare these trees . we create extensive instances for the @xmath290 protein trees . the instances are created by the following way . firstly , create a random label - set @xmath291 with fixed size not greater than @xmath292 . secondly , construct the subtrees with label - set @xmath291 of the @xmath290 protein trees , respectively . finally , randomly choose a fixed number of subtrees to compare . + figure [ mean - order ] shows the mean order of mafs for the instances of protein trees . the number of instances we tested in each instance set t@xmath283 is not less than @xmath293 , where @xmath294 and @xmath285 . as can be seen from the figure [ mean - order ] , we can get that the mean order of mafs for the instances are increase as the number of trees in the instances increase , which indicates that the reticulation has influenced the evolutionary history of these parts of the genomes that being compared . therefore , it makes perfect sense to study the maf problem on multiple trees . we also run pmaf and amaf on the poaceae data set from the grass phylogeny working group @xcite . the dataset contains sequences for six loci : internal transcribed spacer of ribosomal dna ( its ) ; nadh dehydrogenase , subunit f ( ndhf ) ; phytochrome b ( phyb ) ; ribulose 1,5-biphosphate carboxylase / oxygenase , large subunit ( rbcl ) ; rna polymerase ii , subunit @xmath295 ( rpoc2 ) ; and granule bound starch synthase i ( waxy ) . when comparing these trees , only shared taxa of the set of trees are kept . .comparing the trees of poaceae data set . [ cols="^,^,^,^",options="header " , ] tabel [ poaceae ] shows our experimental results on the poaceae data set . as can be seen from the table , the ratios outputted by amaf are small , because the size of instances are very small and the order of an maf for these instances are large , relative to their sizes . + the protein tree data set and poaceae data set above , are all binary tree . in order to test our programs on biological data sets of general phylogenetic trees , we introduce the multifurcations in the @xmath290 protein trees which has been discussed above by collapsing each bipartition with a bootstrap support value of less than 0.8 . the way for introduction of multifurcations is the same as the way in @xcite . figure [ mean - nonbinary - order ] shows the mean order of mafs for the instances of general protein trees . the number of instances we tested in each instance set t@xmath283 is not less than @xmath293 , where @xmath294 and @xmath285 . in this paper , we presented the first group of parameterized algorithms for the maximum agreement forest problem on multiple general phylogenetic trees : one for rooted trees that runs in time @xmath0 ; and the other for unrooted trees that runs in time @xmath2 . we also presented the first group of approximation algorithms for the maximum agreement forest problem on multiple general phylogenetic trees : one for rooted trees with ratio @xmath1 , and the other for unrooted trees with ratio @xmath3 . extensive experiments on simulated data and biological data show that our programs pmaf and amaf are able to calculate the orders of mafs for the constructed instances . in particular , when the order of maf is small , pmaf can return the order of maf for the instance quickly . further improvements on the algorithm complexity of parameterized algorithms and the ratio of polynomial - time approximation algorithms for the maximum agreement forest problem on multiple general phylogenetic trees are certainly desired . however , the improvement seems require new observations in the graph structures of the phylogenetic trees . f. hodson , d. kendall , and p. tauta , _ the recovery of trees from measures of dissimilarity . in mathematics in the archaeological and historical sciences _ , edinburgh , u.k . : edinburgh university press , pp . 387 - 395 , 1971 . m. baroni , s. grnewald , v. moulton , and c. semple , `` bounding the number of hybridisation events for a consistent evolutionary history , '' _ journal of mathematical biology _ , vol . 171 - 182 , 2005 . j. fehrer , b. gemeinholzer , j. chrtek jr , et al , `` incongruent plastid and nuclear dna phylogenies reveal ancient intergeneric hybridization in pilosella hawkweeds ( hieracium , cichorieae , asteraceae ) , '' _ molecular phylogenetics and evolution _ 347 - 361 , 2007 . o. paun , c. lehnebach , j. t. johansson , et al , `` phylogenetic relationships and biogeography of ranunculus and allied genera ( ranunculaceae ) in the mediterranean region and in the european alpine system , '' _ taxon _ , vol . 4 , pp . 911 - 932 , 2005 . coyne , s. elwyn , s.y . kim , and a. llopart , `` genetic studies of two sister species in the drosophila melanogaster subgroup , d. yakuba and d. santomea , '' _ genetical research _ , vol . 11 - 26 , 2004 . kliman , p. andolfatto , j.a . coyne , f. depaulis , m. kreitman , a.j . berry , j. mccarter , j. wakeley , and j. hey , `` the population genetics of the origin and divergence of the drosophila simulans complex species , '' _ genetics _ , vol . 1913 - 1931 , 2000 . k. takahashi , y. terai , m. nishida , and n. okada , `` phylogenetic relationships and ancient incomplete lineage sorting among cichlid fishes in lake tanganyika as revealed by analysis of the insertion of retroposons , '' _ molecular biology and evolution _ 18 , pp . 2057 - 2066 , 2001 . rodrigues , m.f . sagot , and y. wakabayashi , `` some approximation results for the maximum agreement forest problem , '' proc . approximation , randomization , and combinatorial optimization : algorithms and techniques ( approx 2001 and random 2001 ) , pp . 159 - 169 , aug . 2007 . e. rodrigues , m. sagot , and y. wakabayashi , `` the maximum agreement forest problem : approximation algorithms and computational experiments , '' _ theoretical computer science _ , vol . 374 , no . 1 - 3 , pp . 91 - 110 , 2007 . j. chen , j.h . fan , and s.h . sze , `` parameterized and approximation algorithms for the maf problem in multifurcating trees , '' proc . graph - theoretic concepts in computer science ( wg 2013 ) , pp . 152 - 164 , jan . 2013 . o. paun , c. lehnebach , j.t . johansson , et al , `` phylogenetic relationships and biogeography of ranunculus and allied genera ( ranunculaceae ) in the mediterranean region and in the european alpine system , '' _ taxon _ , vol . 911 - 932 , 2005 .

the gauge / gravity correspondence provides the best understood example of a complete theory of quantum gravity , that is , a precisely defined quantum system which is approximated by general relativity in the appropriate limit @xcite . an aspect of this is that gravity is a non - renormalizable effective theory , whose ultraviolet completion is provided by string theory . however for the purposes of this note , it will be surprising infrared , albeit nonperturbative , constraints that will play the central role . in recent years , there has been considerable progress in elucidating the construction of bulk gravity observables in the dual cft language . some surprising phenomena have been found , particularly in situations involving causal horizons in the bulk . in particular , contradictions appear to arise between the existence of linear operators associated to bulk observables behind horizons and the semiclassical analysis around the expected smooth spacetimes . a paradigmatic example is the information paradox of black hole evaporation @xcite . in the asymptotically ads context , the evolution is obviously unitary by the duality with a unitary conformal field theory . thus the paradox is recast as a problem in the existence of cft operators that obey the expected bulk evolution equations and commutation relations in the weakly curved region behind the horizon that is relevant to the hawking process . these arguments were sharpened in the work of @xcite , which demonstrated that exponentially small corrections in the planck expansion can not resolve the paradox . in particular , there can not exist cft operators that approximate the naively expected behavior of local bulk fields behind the horizon to within exponential accuracy in typical states of the black hole . it was suggested in @xcite that this implies a sharp breakdown of the bulk effective theory even in certain regimes with low curvatures - that most black hole microstates have firewalls near the horizon . the fascinating work of papadodimas and raju @xcite , and related work of @xcite and @xcite , showed how to explicitly construct bulk operators behind horizons in @xmath0 perturbation theory around a given approximately thermal pure state , using an analog of the kms relation . the striking feature of their construction is that it depends on the background state , so that it is impossible to obtain a linear operator associated to the bulk observable that is valid on all microstates . this can then evade the amps(s ) paradoxes . in the present work , i will focus for simplicity on the situation of a two sided black hole , where similar issues also arise @xcite . given that these interesting puzzles only appear when considering semiclassically distinct configurations , rather than in perturbation theory around a fixed background , it is appropriate to use a non - perturbative framework to describe the effective bulk physics . to that end , in this work , i will investigate the relation between the bulk hartle - hawking wavefunction @xcite and the dual qft states . the asymptotically ads arena is furthermore a promising one to make the hartle - hawking formalism more precise . the main new feature that can be seen in such an analysis is that the observables whose existence as linear operators on the full hilbert space would lead to inconsistencies are already ill - defined as operators on the wavefunction of the bulk effective theory . in particular , they are not non - perturbatively invariant under spacetime diffeomorphisms . i will show that they fail to obey certain non - perturbative consistency conditions analogous to the wheeler - dewitt equation @xcite that implements the hamiltonian constraints in perturbation theory . one may define more refined , gauge fixed versions of these bulk operators that are non - perturbatively diffeomorphism invariant . these involve additional choices , and only agree with the naive observables on a small subspace of the full bulk effective theory hilbert space . for these operators , it appears that no paradoxes arise , and they should be given by linear operators on the exact hilbert space . this paper is organized as follows . in section two , i sketch the basic logic of this work , and review the apparent obstruction to a linear map between certain bulk observables and cft operators , in the context of eternal ads - schwarzschild . in section three , i argue using the hartle - hawking formalism that such observables fail to be nonperturbatively diffeomorphism invariant . in section four , i describe appropriately gauge fixed versions of bulk operators defined relationally to the boundary . finally , in section five , i discuss possible implications for the information paradox and speculate about the appropriate framework for describing bulk measurements . the relation between the hilbert spaces of uv theories and their ir effective theories is , in general , one of projection . thus one expects that the states of the ir hilbert space can be mapped to states of the uv theory . they only span a subspace , since short distance modes remain in an adiabatic vacuum . consider the wilsonian renormalization group flow in a quantum field theory on a @xmath1 dimensional spatial lattice , obtained by passing to an effective description with twice the lattice spacing . the hilbert space of a @xmath2 block must be replaced by the hilbert space of a single lattice site , of dimension @xmath3 @xcite . this is done by a projection map , @xmath4 from the @xmath5 dimensional space to an @xmath3 dimensional space . the cokernel of this map should be interpreted as states in the microscopic theory in which no short distance modes are excited . clearly information is lost during each step of the rg block transformation , and the projection must be chosen judiciously to avoid projecting out physically important states . the hilbert space of the long distance effective theory , @xmath6 , is the subspace of the full hilbert space of the microscopic theory in which only long wavelength degrees of freedom are excited . this should be sharply contrasted with the subspace of states with total energy below some cutoff ( in an ir finite system with a discrete spectrum ) . the latter would contain states with small numbers of ultra high energy quanta , which are not part of the effective theory , and would fail to contain coherent states of long wavelength modes with large total energy . for the present purpose of examining nonperturbative constraints in the bulk , it is important that @xmath6 does contain such coherent states , the quantum analogs of low curvature solutions of the non - linear classical equations of motion . in standard constructions , there is a fixed projection in the above definition of @xmath6 , which , for example , might select states of lowest energy density . an alternative possibility is the use of state dependent projections , in other words , a non - linear map . for example , in the mera block transformation , the hilbert space is projected onto the subspace with maximal entanglement with the rest of the system @xcite . this type of rg flow for states has proven to have significant computational utility in constructing ground states of certain many body systems . however the physical meaning of the state dependence of the resulting effective theory observables is unclear . in particular , if it were significant in some situations , it would seem to violate the linearity of quantum mechanics , which we certainly do not expect in long distance effective observables in condensed matter systems . it is therefore important in the gravitational context to ask what a fundamental state dependence of bulk observables would imply for the hilbert space of the effective bulk theory . in particular , if the theory in ads is dual to a large @xmath7 conformal field theory , what is the relation between the exact cft hilbert space and the bulk effective theory hilbert space ? it is essential to describe the bulk theory in a non - perturbative framework , since the constraints and paradoxes under discussion that motivate state dependent constructions only appear when comparing states that differ non - perturbatively . perturbative treatments only refer to the behavior of observables in the subspace that can be obtained by acting with parametrically less than @xmath7 single trace operators on a given state with a known gravity dual . the cft expressions for bulks fields around the ads vacuum @xcite and in other situations where they can be constructed using the bulk evolution equations fall into this category . even if no causal reconstruction is possible , all constructions clearly lead to linear operators @xcite in such code subspaces . the question at hand is how these can be patched together over the full space of configurations described by the bulk effective theory . state dependence of bulk observables in the emergent bulk gravity theory would seem to imply that there does not exist a linear map between the full bulk low energy effective hilbert space and the exact hilbert space . such maps would then only exist for subspaces of the bulk hilbert space defined perturbatively around a given state . a scenario of this type would be that a mera construction of the cft states determines the bulk state . since the projection in mera is done on maximally entangled subspaces , the result would be a non - linear map on the hilbert space . i will argue instead that the observables in question fail to be non - perturbatively diffeomorphism invariant . in other words , such observables do not exist as linear operators even in the bulk gravity effective theory . thus there is no contradiction with the existence of a linear map , @xmath8 , between all physical , gauge invariant states of the low energy bulk theory and states of the cft . the existence of such a linear map is completely natural , given the ads / cft duality of the quantum systems . furthermore , it appears that there is no need to restrict @xmath9 to be smaller than the expected domain of validity of the bulk effective theory . these situations must be examined in a non - perturbative long wavelength bulk effective description that appropriately imposes the constraints of temporal diffeomorphism invariance , such as the wheeler - dewitt wavefunction . that formalism can be made more precise in the asymptotically ads context , as i will discuss in section 3 . it is non - perturbative because it allows discussion of configurations that are significantly different than the vacuum ( for example , states related to solutions of the classical non - linear equations ) . it is a low energy effective formalism in the bulk because short distance modes are not excited . the hartle - hawking wavefunction is defined by a euclidean path integral up to a slice with on which the spatial metric and other bulks fields , collectively denoted by @xmath10 , are fixed . this data should be understood as defined up to spatial diffeomorphisms . the temporal components of the metric at the slice , the lapse and the shift , are integrated over . this is because they appear in the action without any time derivatives . given a state @xmath11 , this path integral defines the wavefunction , @xmath12 , giving a linear functional on states of the bulk hilbert space , @xmath13 . an important feature of the ads arena is that adjusting asymptotic sources in the euclidean path integral produces many states , as opposed to the situation in cosmological global de sitter spacetime , where only the hartle - hawking no boundary state @xcite is naturally defined by a euclidean path integral . the states obtained by euclidean path integrals with sources for single trace operators generate the bulk long distance effective hilbert space , @xmath9 . by the standard relationship between the path integral and the hilbert space in quantum mechanics , the inner product between two states @xmath11 and @xmath14 is obtained by performing the euclidean gravity path integral with the asymptotic sources that defined the two states . similarly , the overlap @xmath15 between two of the hartle - hawking kets is given by the euclidean path integral with intrinsic metric @xmath10 and @xmath16 on two spatial slice boundaries . in the asymptotically ads context , the two noncompact slices must asymptotically join at the ads boundary . a crucial feature is that because no gauge for temporal diffeomorphisms has been fixed , the overlaps between the kets , @xmath17 , are generally nonzero , and the kets do not form a basis : they are overcomplete . this simply reflects the fact that there are many spatial slices of the same spacetime geometry . the ket states @xmath18 are thus not linearly independent . for this reason , the matrix elements of any gauge invariant operator @xmath19 must obey certain constraints . in the perturbative analysis , for @xmath20 , hartle and hawking showed that these conditions imply that the @xmath21 satisfy the wheeler - dewitt equation , in other words , @xmath22 commutes with the hamiltonian constraints . the euclidean path integral additionally allows one in principle to determine the overlap between arbitrarily different kets , which would be challenging to understand by integrating the wheeler - dewitt equation . as i will explain in more detail in section 3 , the resulting conditions are nonperturbative analogs of the hamiltonian constraints . these will be the key to the present analysis . one way of understanding the state dependence required in the constructions of bulk operators is that naively distinct bulk states are not linearly independent . in other words , the code subspaces based on different background states have unexpected overlaps . for example , using the duality with the exact cft , @xcite demonstrated that very long time evolution of the two sided ads black hole results in states whose overlap does not decay to 0 , but rather remains finite , albeit exponentially small in @xmath7 . the main point of this note is that this overlap of seemingly distinct states has the same origin as the lack of independence of the hartle - hawking kets . for this reason it is already a feature of the bulk effective theory hilbert space , and the associated constraints on operators are imposing non - perturbative diffeomorphism invariance . to illustrate the idea , consider the superselection sector of quantum gravity with two asymptotically ads@xmath23 boundaries . the spacetime geometry may be connected or not , depending on the state . however , the full quantum system consists of two decoupled copies of a large @xmath7 cft . the cfts are non interacting because the hamiltonian density in gravity is a total derivative , hence the total hamiltonian consists of two decoupled boundary terms , @xmath24 , even when the spacetime is connected @xcite . a particularly straightforward example of a non - perturbative ir paradox appears in this context , as pointed out by marolf and wall @xcite . consider a right framed bulk observable , such as a field operator at a position defined by a fixed proper distance , relative to empty ads , along a geodesic that extends from a point on the right boundary at a given direction . in any factorized state , this observable should be represented by a purely right cft operator . after all in that case the left cft is neither interacting nor entangled with the right one , so it seems that from the right point of view it can not have any effect . on the other hand , the thermofield double state is dual to the eternal ads - schwarzschild geometry @xcite , described by the metric @xmath25 the most salient feature of that spacetime is the einstein - rosen bridge connecting the two sides . thus the right framed bulk observable can enter the causal domain of the left boundary , and the observable can not commute with all left operators in that state . this contradicts linearity of the bulk observable , because the thermofield double state is a linear combination of factorized states , as we know from the microscopic cft description , @xmath26 in @xcite , it is suggested that the ads / cft duality is incomplete in this situation and must be supplemented by a choice of connected or disconnected dictionary . in this work , i will argue instead that the bulk observables do not define non - perturbatively diffeomorphism invariant operators on the bulk effective hilbert space . one must make additional choices in specifying the observable to make it gauge invariant , which determine whether it acts non - trivially on the left cft . these play the role of the choice of dictionary , but we will see that they are already required to have a well - defined action on the hartle - hawking states in the bulk effective theory . it is important to note that small corrections to the bulk gravity behavior of these operators can not resolve this paradox . in particular , using arguments similar to those in @xcite , the works @xcite demonstrated that any linear operator must have order 1 disagreement with the expected properties of a right framed bulk field behind the horizon on most states that are dual to time shifted versions of the eternal black hole . the evolution of the thermofield double state by the left hamiltonian is simply the application of a large diffeomorphism that acts nontrivially at the left boundary . this makes it clear that such states , @xmath27 , should remain in the domain of validity of the bulk effective theory , even for long time scales @xmath28 . this time shifted state is geometrically the same spacetime manifold as the eternal black hole , merely with a different origin of time on the left boundary . thus any observables that are truly defined only relative to the right boundary should be unaffected , and the region near the horizon has low curvature for all @xmath29 . a very simple way to see that exponentially small corrections to the observables ca nt help is to consider the bulk observable , @xmath30 , that counts the number of connected components of space . suppose it has an expectation value of @xmath31 on the states @xmath32 in which the two boundaries are connected by an einstein - rosen bridge , and an expectation value of @xmath33 on factorized states , in which the spacetime consists of disconnected left and right pieces . using the microscopic cft description of @xmath32 as @xmath34 one obtains a contradiction , because averaging the result for @xmath32 over a sufficiently long interval of the parameter @xmath35 approximately projects onto individual energy eigenstates . in more detail , if @xmath36 , where @xmath37 is exponentially small in the black hole entropy , then @xmath38 it is impossible for the phases in the second term to add coherently for all @xmath35 over a interval of order @xmath39 in such a way that the double sum over the exponentially large number of states of the exponentially small quantities @xmath40 is of order 1 . therefore , for most @xmath35 , the result will be the average value of 2 up to exponentially small corrections , in contradiction with its expected value of @xmath41 . a sketch of the argument in @xcite for the more interesting operators that measure a field at bulk points defined relationally to the right boundary is as follows . consider such an observable , for example the particle number operator , @xmath7 , constructed out of the right framed field operator , which has an @xmath42 expectation value in the thermofield double state . it will then also have an @xmath42 expectation value in the time shifted states . then one can average over exponentially long times to obtain @xmath43 the second term becomes small when @xmath35 is large enough , which proves that the first term is also of order @xmath44 . this is true for all temperatures @xmath45 . performing a legendre transform , one can approximately project on to factorized states . it is then easy to show that this implies that the right framed operator must have a small expectation value in all factorized states of the same energy . this is inconsistent with the bulk predictions . in the next sections , i will analyze this situation in the hartle - hawking formalism . it is applicable here because all of the time shifted states are well - described by wavefunctionals in the low curvature regime . in particular , one can find nice slices even in these exponentially time shifted states which have small intrinsic and extrinsic curvatures , if the black hole is large in planck units . before turning to the more detailed analysis of these observables in the bulk hartle - hawking framework , it is interesting , although simple , to see that the nonzero overlap between a factorized like such as the vacuum @xmath46 and the thermofield double state is already calculable in the gravity path integral . it is precisely the lack of orthogonality between these states with connected versus disconnected semiclassical descriptions which led to the paradox . the euclidean gravity saddle that contributes to @xmath47 is just euclidean ads@xmath23 . note that in the cft language , the vacuum caps that produce the state @xmath48 and the cylinder that produces @xmath49 are conformally flat , so the overlap partition function is not only topologically but in fact conformally equivalent to @xmath50 . the denominator is just the usual normalization of factor for the states produced by these path integrals . the inner product between the vacuum and the thermofield double state in the gravitational path integral is shown in on the left in figure 1 . it can be interpreted as an amplitude that the thermofield double state is a disconnected geometry . that this is non - vanishing is not surprising , given that the disconnected configuration can even be dominant over the black hole ( below the hawking - page phase transition ) . on the other hand , one can equally well glue together the bulk configurations shown on the right in figure 1 . here it appears to be an amplitude that the vacuum is a connected spacetime . a important aspect is that it would be completely wrong to add together these `` two '' possibilities . they are just different depictions of the same saddle , as can be seen in figure 2 , which is an alternative , but topologically equivalent , depiction of the two slicings of figure 1 . this is precisely because @xmath30 , the number of connected components of space , is not a nonperturbatively diffeomorphism invariant operator . it depends on the choice of slice , and so fails to obey the hamiltonian constraints non - perturbatively . as explained above , a paradox very similar to that described in @xcite appeared for the putative observable @xmath30 . by the ads / cft duality , it should be 2 on factorized states , and 1 on the thermally entangled states , which contradicts linearity . what we see now is that @xmath30 is simply not gauge invariant , so this indicates no obstruction to a linear map between @xmath9 and @xmath51 . . the black hole state is produced by the annulus asymptotic boundary in the middle of the spheres , while the vacuum is produced by the cap boundaries at the top and bottom of the spheres . ] the hartle - hawking kets with different topologies are not independent , and no complete linear operator @xmath30 can act on them with different eigenvalues . the nonzero overlap between these kets implies topology ambiguity , not dynamical change of the topology . a state that `` definitely '' has a given topology still has a nonzero probability to have a different one this is particularly clear in the ads context , since the global hamiltonian makes no appearance in this discussion . if a related non - perturbatively gauge fixed operator was defined , then its dynamical evolution could be discussed . this fact is an important point of consistency in the proposal that entanglement encodes spatial connection @xcite by nontraversable wormholes @xcite , since entanglement is not measurable by any operator . the nonexistence of a linear operator that measures the spatial topology has been pointed out by @xcite and in related work of @xcite , based on the duality with the cft . this is true even without the presence of horizons , as shown the examples of @xcite . now we see that this is already a feature of the bulk effective theory . it is more subtle to see that the geodesic defined operators are not gauge invariant beyond perturbation theory . as explained in section 4 , it requires extra data to define them , so that the geodesic dressing lies in a space - like slice . this is necessary in order to have a straightforward action on the bulk hilbert space . the end result is that appropriately diffeomorphism invariant versions of these observables do not lead to paradoxes . consider the collection of states that can be produced by performing a cft path integral with sources for single trace operators . this class encompasses all states that are accessible in the bulk effective theory . given any such state , we want to find its description as a wavefunctional @xmath12 . here @xmath10 represents the data of the bulk metric and fields on a spatial slice . they are required to obey ads asymptotic conditions that are described below , and we should consider @xmath10 to label equivalence classes with respect to the redundancies generated by spatial gauge transformations obeying the appropriate boundary fall off conditions . the hartle - hawking prescription @xcite is to perform the euclidean path integral with sources on the ads boundary , up to a slice with data @xmath10 , @xmath52 up to a normalization constant , where the time cut boundary of space is @xmath53 , and the ads asymptotics , @xmath54 , are schematically given by the sources @xmath55 that define the state . in other words , one only integrates over metrics and bulk fields that obey the appropriate fefferman - graham fall - off conditions . writing the metric near the slice as @xmath56 where @xmath57 are coordinates on the slice , one integrates over the lapse , @xmath7 , and shift , @xmath58 , at the slice , keeping only @xmath59 fixed . one similarly integrates over all over bulk fields , with ads asymptotics given by sources and fixed on the slice as @xmath60 to obtain the wavefunctional @xmath61 . this formalism is general , but for concreteness , one may consider gravity described by the einstein - hilbert action , @xmath62 , where in ads @xmath63 , and trace of the extrinsic curvature @xmath64 appears in the boundary term . more precisely , the ads asymptotic boundary should be cut off , and the action obtained as a limit after subtraction of the appropriate counter terms , as discussed in this context of producing lorentzian states from euclidean sources on a cap in @xcite in ads , the spacetime metric is required to have the asymptotic form @xmath65 where @xmath66 are the boundary spacetime coordinates , @xmath67 is the ads radius , and the subleading terms vanish in @xmath68 limit . this implies that the slice metric @xmath69 . moreover , the @xmath70 shift must obey @xmath71 at the boundary . only those spatial diffeomorphisms that vanish as @xmath68 are redundancies , and @xmath72 is a functional of such equivalence classes of @xmath10 . the asymptotic vanishing of @xmath73 implies that the adm hamiltonian is a nontrivial boundary operator , rather than a constraint . the path integral above should be understood as being evaluated in bulk perturbation theory around its saddles . it is still non - perturbative in that one may consider states and @xmath10 that differ classically , in other words , at leading order . one should only allow the metrics , @xmath10 , on the slice which have no planck scale features . more precisely , one must perform the path integral in a bulk effective theory with cutoff @xmath74 , and the slice metric must live in the cutoff configuration space , with no wavelengths smaller than @xmath75 . if there is no bulk saddle with small curvatures that contributes , the result will be close to 0 . some cft states may have small amplitude on all @xmath10 , since they contain planckian or string scale objects in the bulk . as discussed above , contributions from subleading euclidean saddles will lead to exponentially small corrections to the hartle - hawking kets , which will not significantly affect the conditions of non - perturbative diffeomorphism invariance . the bulk effective theory hilbert space is spanned by states that can be produced by performing such euclidean path integrals with sources , as can be seen by analytic continuation of lorentzian bulk configurations ( in general the sources will be complex ) . the hartle - hawking path integral then defines a map from @xmath76 . using the hilbert space inner product defined on the states @xmath77 , one identifies @xmath78 with a state ( strictly speaking , a limit of states since these may have divergent normalizations ) . as usual , the relation between the exact hilbert space and a low energy effective hilbert space is by projection , @xmath79 . thus we can also think of @xmath78 as living in @xmath80 , by applying the adjoint of the projection map . this set of kets is overcomplete , and the matrix of their inner products , @xmath81 , given by the euclidean path integral between slices with intrinsic data @xmath10 and @xmath16 , are generally nonzero . for this reason is not entirely trivial to determine the inner product on the wavefunctionals @xmath82 , in other words to find a @xmath83 such that @xmath84 . the kernel @xmath85 would simply be the inverse of @xmath3 if the set was linear independent . instead , there are many matrices @xmath85 that satisfy @xmath86 . one such kernel is a differential operator in the space of @xmath10 , representing the inner product as @xmath87,\ ] ] where the product is over all points in the spatial slice @xcite . this results in an integration over spatial metrics that excludes the local weyl factor , which is a timelike direction in the superspace of metrics . in this work , i will not need to use the inner product on the wavefunctionals . an important caveat to this path integral based approach to the bulk effective theory is that the full rules for summing over different spacetime topologies in the gravity partition function are not known . for example , in a calculation in ads@xmath23 , with asymptotic boundary @xmath50 , one might including a sum over topologies with extra handles . cutting these with a time slice could include a disconnected compact spatial region . one of the puzzles is that in the system with two asymptotic boundaries , the cft implies that correlation functions in the vacuum , @xmath88 , factorize between the left and right . however , if connected spacetime saddles contributed , this factorization would appear mysterious on the gravity side . this issue arose in @xcite in the attempt to define the partition function of pure gravity in three dimensions and find its cft dual . note that in the example of the vacuum correlators for a 2d cft on a pair of circles , there is no such saddle , since no smooth hyperbolic 3-manifold may have two disconnected @xmath89 boundaries . however , as discussed in @xcite , such euclidean saddles do exist with disconnected higher genus boundaries . in that work , it was proposed that the gravity path integral simply does not include them . in spite of this gap in understanding , the discussion of the present paper is unaffected . the reason is that the issue at hand is about the different topologies of slices that cut a given spacetime saddle , and their ket overlaps . thus the question of whether other spacetime saddles should be summed over is not important for the characterization of non - perturbative diffeomorphism constraints . the wheeler - dewitt equation @xmath90 where @xmath91 is the intrinsic curvature of the slice can be derived @xcite from the lack of independence of the kets @xmath78 , which implies that @xmath92 is a redundant data . the differential equation results from expansion of the kets under small variation , @xmath93 . there are also non - perturbative constraints that would be hard to see by integrating the wheeler - dewitt equation . in particular , there is a conceptually identical lack of independence between the data with different topologies of the spatial metric , @xmath10 . the hartle - hawking procedure defines wavefunctions @xmath94 , @xmath95 , ... which each possible spatial topology , including the number of connected components . however the hilbert space does not consist of separate factors , since the kets are not linearly independent . unlike in gauge theory , in gravity it is impossible to fix the gauge even in perturbation theory by imposing a condition on the spatial @xmath10 . this is because the wheeler - dewitt equation is second order in @xmath96 . furthermore , standard gauges do not actually fix the gauge nonperturbatively . in particular , the diffeomorphism group for different topologies is different . in the two sided system , the difference in the group of diffeomorphisms is associated to the existence of the gravitational analog of the wilson line operators discussed in @xcite . consider the example of spacetimes with a pair of asymptotically ads boundaries . the ads / cft duality implies that the kets with disconnected topologies @xmath97 in fact span the entire hilbert space . this is hard to see explicitly in the bulk effective theory analysis , since obtaining the formula for connected kets in terms of disconnected kets involves partially diagonalizing the matrix of overlaps , @xmath98 . the result will depend on the details of the cutoff scheme @xmath99 . for a different class of supersymmetric horizonless geometries @xcite in global ads@xmath100 , it is possible to see the linear relations explicitly , using the explicit description of the dual cft microstates @xcite . without this exact dictionary protected by supersymmetry , the bulk effective theory analysis would in general lead to a linear relation that was cutoff dependent . nevertheless , the matrix , @xmath101 , of kets overlaps is a uv safe quantity . the states @xmath102 are overcomplete ( here @xmath10 represents a metric on the topological space labelled by the first argument ) , so it is mathematically inconsistent to define a linear operator by independently specifying its action on each of these states . but given a linear operator @xmath103 , one can compute its matrix elements @xmath104 . then these must obey some relations . by considering small changes in @xmath10 with a fixed topology , one can show that @xmath22 must commute with the hamiltonian constraints . similarly , by considering different topologies , one can show that @xmath105 obeys further conditions . these are nonperturbative constraint equations . a collection of such matrix elements that do nt satisfy these constraints do not give a mathematically consistent linear operator ; it is not gauge invariant under temporal diffeomorphisms . the ads / cft duality implies that disconnected topology states span the hilbert space of the two sided ads system . therefore defining the action of an operator @xmath103 on those states _ determines _ its action on kets with other topologies . it will be perfectly well defined on the thermofield double state . what is disallowed is to say that one defines an operator like @xmath30 , the number of connected components of space , which obeys @xmath106 . this is just as bad as writing a set of matrix elements that fail to obey the hamiltonian constraints . therefore there is no candidate operator in the bulk effective theory that counts the number of disconnected components of the spacetime . the euclidean gravity path integral implies that naive basis states associated to different spatial topologies are not linearly independent - they are related by the constraint equations . therefore the number of spatial components is not non - perturbatively diffeomorphism invariant . there is no associated operator on the full bulk effective hilbert space that satisfies the non - perturbative constraint equations . it is interesting to contrast this fact with the structure of the classical phase space of general relativity . the latter has distinct components with different spatial topology , since there are no non - singular topology changing solutions . thus naive quantization of this phase space would seem to imply that they give rise to orthogonal subspaces of the hilbert space . moreover , the topology of space at a given asymptotic boundary time appears to be a diffeomorphism invariant property of nonsingular spacetimes . similarly , there is no straightforward lorentzian interpretation of the euclidean instanton that describes the overlap of the factorized vacuum with the eternal two sided black hole . one perspective here is that semiclassical quantization and the path integral formulations of the bulk effective gravity theory disagree , and comparison with the exact cft result is what that tells us that the path integral method is the correct procedure in this situation . of course , the disagreement is non - perturbative in the @xmath107 expansion . however , it is more precise to say that lorentzian semiclassical quantization is agnostic about these issues , since the relevant solutions always involve singularities , which are outside the domain of validity of the bulk effective theory . analogously , the infinitesimal version of the hartle - hawking ket overlaps @xmath108 are described by the hamiltonian constraint equations with a straightforward lorentzian interpretation , but it not easy to integrate up to see the overlap between kets with different topologies . this is because the slice has to become singular somewhere along any deformation which changes its topology , and again one exits the regime of validity of the long distance theory . the fact that using the euclidean path integral one may directly define and calculate the overlap of non - perturbatively distinct kets is a main virtue of the hartle - hawking formalism . in classical gravity in asymptotically ads spacetimes , one may define relational observables , for example the value of a bulk field at some regularized distance , @xmath70 , along a geodesic that extends transversely into the bulk from a fixed boundary point , @xmath109 . these are fully gauge invariant , since ads boundary conditions require that diffeomorphisms vanish at the boundary . in perturbation theory around empty ads in poincare patch , one may think of such an observable @xmath110 as the bulk field in fefferman - graham coordinates . in general , this is not a good description since fefferman - graham is not a good gauge around general backgrounds . the geodesics from different boundary points can intersect , causing this coordinate system to break down . in the quantum path integral , one can insert such an observable , since it is well - defined for any off - shell metric configuration . however , it is a novel feature of gravity that this does not automatically imply the existence of a well - defined operator on the physical hilbert space . in quantum field theory , one can perform the path integral up to some time , act with an operator , and then continue the path integral , so there is a direct relation between insertions and operators . in gravity , there is no canonical choice of time slice , so this procedure does not work . for example , insertion of @xmath110 into the path integral defining the wavefunctional @xmath12 is not well - defined , since for some metrics , the geodesic will exit the slice @xmath10 . one needs to appropriately gauge fix the bulk operators nonperturbatively , so that they are compatible with the slices . if it was possible to pick a non - perturbatively good gauge for diffeomorphisms , then there would be a simple relationship between insertions in the path integral in that gauge and diffeomorphism invariant operators . one of the essential points is that no such gauge exists , as can be seen by the fact that the causal relations between observables depend on the spacetime . another way of describing the same feature is that there are multiple ways of preparing the same state from a euclidean path integral . for example , the thermofield double state can be produced by a path integral with cylinder boundary or via the microscopic relation @xmath111 . in general , an insertion of a geodesically defined observable into a path integral involving one or the other of these need not give the same result . therefore there is no well - defined action on states . it is possible to define closely related operators that are non - perturbatively diffeomorphism invariant . but additional choices are required , and the resulting operators do not have the expected action of all states . the following is one such prescription . consider kets @xmath112 defined by doing the path integral only over metrics that have the fefferman - graham form along a single geodesic originating at the boundary at a point @xmath113 and terminating at a point in the interior at a fixed regulated distance , @xmath70 , defined with respect to empty ads . in other words , one restricts the integration over the radial component of the shift to vanish along this geodesic between the ads boundary and the point defined by @xmath70 , in addition to fixing the spatial gauge on the slice to have the fefferman - graham form along that geodesic . the path integral is thus only over @xmath114 dimensional metrics whose restriction to the @xmath115 slice is given by @xmath116 where the index @xmath117 runs over all spatial directions , @xmath70 and the boundary spatial directions , @xmath118 . this gives a well - defined linear functional on the hilbert space , @xmath119 . note that there is no integration over metrics in which this geodesic is chronal ( ie . no two points on this space - like geodesic may time - like related ) , so these kets will span a smaller subspace than the hartle - hawking kets . these kets are still not all linearly independent , but ones with different values of the fields along the special geodesic will be orthogonal . this is because there is no integration over the shift along the geodesic , thus in the path integral that computes @xmath120 will have the marked geodesic in the two slices identified . therefore one can define a bulk field operator along the geodesic using these kets as @xmath121 this is well - defined because the definition of the restricted kets implies that they are orthogonal if the values of @xmath122 at the point @xmath123 are different . in this way , one can define a nonperturbatively gauge fixed version of the bulk field . the action of this operator on other states is determined by the above definition , supplemented by the condition that it annihilates the orthogonal complement of the full hilbert space . note that in the eternal black hole , this particular observable never enters the region behind the horizon ( in the classical approximation ) , since the geodesic is exactly radial and , if long enough , would pass through the bifurcation surface directly into the left causal wedge . that is sufficient to phrase the paradox @xcite reviewed in section 2 . to reach points in the upper quadrant of the penrose diagram , one needs a geodesic that begins with a general slope in the @xmath124 plane at its origin point on the right boundary . this is easily obtained , since the action of the special conformal generator , @xmath125 that has @xmath113 as a fixed point changes this slope . therefore , one may simply consider the operator @xmath126 . the crucial fact is that the geodesic gauge fixed kets only span the hilbert space of states in which the marked geodesic lies along a spatial slice containing the entire @xmath115 slice of the boundary . in the two sided asymptotically ads context , that includes the left boundary opposite to the right one where the point @xmath113 lies . since the bulk operator is defined in terms of these states , it includes the projection operator onto the subspace which they span . such an operator is explicitly left time dependent , so it will fail to exactly commute with @xmath127 . consider the commutator of left and right boundary framed geodesic observables . this would seem to vanish in all factorized states , since the operators would only act on the left and right hilbert spaces respectively . however , the appropriately gauged fixed operator acts on both boundaries . one can compute @xmath128 |\textrm{tfd}{\rangle}$ ] by finding the non - analytic part of the euclidean path integral with a pair of geodesic insertions , as a function of @xmath109 . this corresponds to the insertion of the geodesics in the geometry shown in figure 2 , and it is clear that they may cross in the interior for sufficiently large @xmath129 . therefore the commutator of the gauge fixed operators is nonzero even when acting on the factorized vacuum . it is exponentially small , since it is suppressed by the exponential of the action of the saddle shown in figure 2 . note that the slice condition for the geodesics is obeyed in the euclidean saddle in this situation , so that the action of the gauge fixed operators is indeed given by a straightforward insertion here . furthermore , the action of @xmath130 on the time shifted states , @xmath131 , is very different than the naive ungauge fixed observable . the time shifted states are described by the same geometry , but with a shifted origin of time on the left boundary . therefore for sufficiently large @xmath35 , the left @xmath115 slice will no longer lie on a spatial slice with entirety of the right framed marked geodesic . thus the operator @xmath130 as defined above will have a completely different action on those states . in the classical limit , it would simply annihilate them . this resolves the paradox . to consistently define a bulk effective theory operator , one must refine the standard description to make the operator obey the hamiltonian constraints nonperturbatively . this involves making various choices ; the above is one example . the gauge fixed operators are perfectly well - defined , and no paradox arises for them . on the other hand , their action on most of the exponentially time shifted states differs significantly from the naive description . in this work , i have argued that bulk observables that have been shown to lack reconstructions as linear cft operators over the entire range of the bulk effective theory also fail to be nonperturbatively diffeomorphism invariant in the bulk description . this resolves the discrepancy with the gauge / gravity duality , and implies that there is no obstruction to a linear map from the hilbert space of the bulk effective theory over its entire expected range of validity to the boundary conformal field theory hilbert space . a simple example is the connectedness of space in the two sided asymptotically ads system . it is a well - defined observable in classical general relativity . however , using the description of connected spacetimes as entangled combinations of microstates of the dual cft , one can show that there can not exist a linear operator @xmath30 that measures the connectedness @xcite . the solution of the puzzle is that the connectedness of space does not satisfy the hamiltonian constraints nonperturbatively , and so does not exist as an operator even in the bulk gravity theory . the bulk physics is often described in terms of boundary relational operators , like a field value at a location determined by some proper distance along a geodesic extending transversally from a given point on the boundary . such geodesically specified observables are well - defined in classical gravity . furthermore , in perturbation theory around any given configuration , there is a well - defined quantum operator , which can be represented by a boundary cft operator in the associated code subspace of states that are perturbative excitations around that fixed state . for example , such bulk observables can be constructed in terms of cft operators in perturbation theory around exponentially time shifted versions of the eternal ads black hole in the two sided ads system @xcite . these states are all described by the same spacetime manifold , with a shift of the origin of time on the left boundary , and thus are in the domain of validity of the long distance bulk effective theory . however , from the microscopic description of the eternal black hole state as the thermofield double entangled sum of factorized cft microstates @xmath132 , one can show that no linear cft operator can agree with the expected bulk matrix elements in all of the code subspaces @xcite . the resolution is similarly that the usual spacetime path integral formulation of the geodesic defined operators is ambiguous . moreover , if one defined an operator using bulk perturbation theory around different configurations whose wavefunction is peaked on a ket @xmath78 , then the total object would fail to satisfy the hamiltonian constraints non - perturbatively . there exist better ways to define such an operator in the bulk theory , that involve a partially gauge fixed set of kets . these make it well - defined and gauge invariant , but then it does not agree with the results of bulk perturbation theory around all states of the bulk effective theory . in particular , in the operators defined in section 4 , there is a projection on to states in which the geodesically specified bulk point is on a spatial slice with the opposite boundary . this results in an order 1 disagreement with the matrix elements of the naive observable in perturbation theory around most time shifted states . in this way , after making such gauge choices to properly define the bulk operator , there is no obstruction to its action as a linear operator in the cft . it would be extremely interesting to explore the role of similar constraints in the black hole information paradox , for black holes given by a pure state in global ads . it is in principle straight forward to express any perturbative bulk calculation in the language of hartle - hawking wavefunctionals , as relations acting on the collection of kets on which the configuration is peaked . presumably , repeating the hawking argument in this framework simply will not result in a contradiction . but that in itself gives little information , since the expansion of the bulk kets in terms of states corresponding to boundary cft local operators is very complicated , and can not be computed in the bulk effective theory . one surprising fact is that the entire black hole formation and collapse process in ads can be captured by spatial slices from a fixed time , @xmath133 , on the boundary , by taking any boundary spatial slice which is spacelike to the end of the horizon where the black hole fully evaporates . this description of the hawking process is complimentary to the more standard picture in which one takes gauge fixed spatial slices whose time evolution is tied to the boundary time . then the process looks like the formation and evaporation of a locally thermalized configuration in the dual cft , such as a region of deconfined phase in a large @xmath7 gauge theory . in the cft description , it is clear that the time evolution is unitary . but the bulk description must already encode the evaporation at a fixed time in the boundary ( for which unitarity is tautological , since there is no evolution under the global hamiltonian ) . from that bulk perspective , one has pure constraint equation evolution , and in principle there is a euclidean path integral calculation of the overlaps between the initial kets and the final ones . if the black hole is formed by the collapse of low energy density matter , then the initial wavefunctional is peaked around kets that are in the domain of validity of the bulk effective theory . similarly , the outgoing hawking quanta are individually of low energy , and that state will also be in the domain of validity of the effective theory . the amplitude between such kets at early and late slices ( ending at the same boundary time @xmath133 ) will be exponentially small , since they are connected by some nontrivial euclidean saddle . however , there is an exponentially large number of relevant late time kets , describing the outgoing hawking quanta , and one expects that the early and late sets of kets are linearly dependent ; that is the pure constraint equation evolution . finding the linear relations again involves partly diagonalizing a matrix of overlaps , which will probably exit the domain of validity of the effective theory . so , unsurprisingly , one does not expect to be able to calculate the exact unitary map from the infalling matter state to the final outgoing state of hawking quanta from the bulk effective theory alone . an intriguing possibility is that by taking into account more planckian configurations , associated to er = epr wormholes connecting the black hole interior to the outgoing hawking quanta @xcite , one can use the linear dependence between kets of the resulting different topologies to obtain a more detailed understanding . the discussions in this work about the restrictions of non - perturbative gauge invariance on bulk operators still leaves unresolved the physical question of measurements in the gravitational bulk . it is well known that there are no local diffeomorphism invariant operators . this certainly does not imply that what we actually measure in gravity are nonlocal observables , in the sense that there is a projection onto eigenspaces of nonlocal operators . the hamiltonian that describes the measurement process is always the integral of a local density , and the physical measurement process can not , for example , change the adm energy encoded by gravitational fields at spatial infinity . the quantum description of a measurement involves coupling the system that one wishes to measure to an apparatus . this interaction can be written schematically as @xmath134 . decoherence processes in the apparatus result in a density matrix that is very well approximated by one that is diagonal in the eigenbasis of @xmath135 . this approximation can be made parametrically good in the limit of a large apparatus , so the result is equivalent to projection on to an eigenstate of @xmath135 , with probabilities given by the born rule . in gravity , there is no canonical way of separating the interaction hamiltonian into diffeomorphism invariant system and apparatus operators , since for a local measurement , it must be defined relationally between them . a useful analogy is an interaction of the form @xmath136 in electrodynamics , where @xmath137 is a wilson line connecting the apparatus and system such that the expression is gauge invariant . there is no canonical way to separate out a gauge invariant @xmath135 - one must add a wilson line to infinity or otherwise pick a gauge . moreover , even if one makes such a choice , the dynamics is never well approximated by projection on to eigenstates of @xmath135 , since that would affect the electromagnetic field far away . in electrodynamics , the resolution is that one does nt call such interactions measurements - rather it is an exchange of an electron between the system and apparatus . however in the bulk gravitational theory , all observations are of this form , and we do consider them to be measurements in practice . therefore it is necessary to examine the physical measurement and decoherence process directly ; there is a irreducible obstruction to replacing it with projection on to eigenstates of an operator . from this perspective , the problem of finding cft descriptions of bulk observations is a dynamical one . at large @xmath7 in certain quantum field theories , one should find that extra non - local operators decohere , ie . that in appropriate states which include a measuring apparatus , there is decoherence in the eigenbasis of such an operator . the only coarse - graining that is intrinsically defined by the system is that given by time averaging . given a short time scale , @xmath138 , one can construct the time dependent density matrix , @xmath139 where @xmath140 is the time evolution operator . then an operator is decohered in the state @xmath141 if it approximately commutes with @xmath142 for a long time @xmath143 . this , _ a priori _ , depends non - linearly on the state . the novelty in strongly interacting theories with holographic duals is that , at least around certain states , additional non - local operators can be made to decohere . these correspond to approximately local bulk operators . for this lorentzian question , it is possible that mera - like constructions of the states , and the resulting state dependent operators will play an important role . but it is not clear what is the correct abstract framework to discuss such observables . alternatively , it is possible that decoherence simply does not occur in these situations , since there is no factorized subsystem that can be associated to the observer non - perturbatively . this suggests the necessity of a new framework of a different type . i would like to thank xi dong , monica guica , daniel hawlow , anton kapustin , aitor lewkowycz , don marolf , kyriakos papadodimas , eric perlmutter , suvrat raju , steve shenker , herman verlinde , aron wall , xi yin , and especially juan maldacena for stimulating and helpful discussions . i also thank noah jafferis for creating the figures . this work was supported in part by nsfcareer grant phy-1352084 and by a sloan fellowship . j. m. maldacena , `` the large n limit of superconformal field theories and supergravity , '' int . j. theor . phys . * 38 * , 1113 ( 1999 ) [ adv . * 2 * , 231 ( 1998 ) ] doi:10.1023/a:1026654312961 [ hep - th/9711200 ] . s. s. gubser , i. r. klebanov and a. m. polyakov , `` gauge theory correlators from noncritical string theory , '' phys . b * 428 * , 105 ( 1998 ) doi:10.1016/s0370 - 2693(98)00377 - 3 [ hep - th/9802109 ] . e. witten , `` anti - de sitter space and holography , '' adv . * 2 * , 253 ( 1998 ) [ hep - th/9802150 ] . s. w. hawking , `` breakdown of predictability in gravitational collapse , '' phys . d * 14 * , 2460 ( 1976 ) . doi:10.1103/physrevd.14.2460 s. d. mathur , `` the information paradox : a pedagogical introduction , '' class . quant . * 26 * , 224001 ( 2009 ) doi:10.1088/0264 - 9381/26/22/224001 [ arxiv:0909.1038 [ hep - th ] ] . a. almheiri , d. marolf , j. polchinski and j. sully , `` black holes : complementarity or firewalls ? , '' jhep * 1302 * , 062 ( 2013 ) doi:10.1007/jhep02(2013)062 [ arxiv:1207.3123 [ hep - th ] ] . a. almheiri , d. marolf , j. polchinski , d. stanford and j. sully , `` an apologia for firewalls , '' jhep * 1309 * , 018 ( 2013 ) doi:10.1007/jhep09(2013)018 [ arxiv:1304.6483 [ hep - th ] ] . d. marolf and j. polchinski , `` gauge / gravity duality and the black hole interior , '' phys . lett . * 111 * , 171301 ( 2013 ) doi:10.1103/physrevlett.111.171301 [ arxiv:1307.4706 [ hep - th ] ] . j. b. hartle and s. w. hawking , `` wave function of the universe , '' phys . d * 28 * , 2960 ( 1983 ) . doi:10.1103/physrevd.28.2960 k. papadodimas and s. raju , `` local operators in the eternal black hole , '' phys . lett . * 115 * , no . 21 , 211601 ( 2015 ) doi:10.1103/physrevlett.115.211601 [ arxiv:1502.06692 [ hep - th ] ] . k. papadodimas and s. raju , `` remarks on the necessity and implications of state - dependence in the black hole interior , '' phys . d * 93 * , no . 8 , 084049 ( 2016 ) doi:10.1103/physrevd.93.084049 [ arxiv:1503.08825 [ hep - th ] ] . e. verlinde and h. verlinde , `` black hole entanglement and quantum error correction , '' jhep * 1310 * , 107 ( 2013 ) doi:10.1007/jhep10(2013)107 [ arxiv:1211.6913 [ hep - th ] ] . e. verlinde and h. verlinde , `` passing through the firewall , '' arxiv:1306.0515 [ hep - th ] . e. verlinde and h. verlinde , `` black hole information as topological qubits , '' arxiv:1306.0516 [ hep - th ] . e. verlinde and h. verlinde , `` behind the horizon in ads / cft , '' arxiv:1311.1137 [ hep - th ] . m. guica and s. f. ross , `` behind the geon horizon , '' class . grav . * 32 * , no . 5 , 055014 ( 2015 ) doi:10.1088/0264 - 9381/32/5/055014 [ arxiv:1412.1084 [ hep - th ] ] . d. marolf and a. c. wall , `` eternal black holes and superselection in ads / cft , '' class . grav . * 30 * , 025001 ( 2013 ) doi:10.1088/0264 - 9381/30/2/025001 [ arxiv:1210.3590 [ hep - th ] ] . k. skenderis and b. c. van rees , `` real - time gauge / gravity duality , '' phys . * 101 * , 081601 ( 2008 ) doi:10.1103/physrevlett.101.081601 [ arxiv:0805.0150 [ hep - th ] ] . k. skenderis and b. c. van rees , `` real - time gauge / gravity duality : prescription , renormalization and examples , '' jhep * 0905 * , 085 ( 2009 ) doi:10.1088/1126 - 6708/2009/05/085 [ arxiv:0812.2909 [ hep - th ] ] . b. s. dewitt , `` quantum theory of gravity . 1 . the canonical theory , '' phys . rev . * 160 * , 1113 ( 1967 ) . doi:10.1103/physrev.160.1113 s. r. white and r. m. noack , `` real - space quantum renormalization groups , '' phys . lett . * 68 * , 3487 ( 1992 ) . doi:10.1103/physrevlett.68.3487 l. p. kadanoff , `` scaling laws for ising models near t(c ) , '' physics * 2 * , 263 ( 1966 ) . g. vidal , `` entanglement renormalization , '' phys . rev . lett . * 99 * , no . 22 , 220405 ( 2007 ) doi:10.1103/physrevlett.99.220405 [ cond - mat/0512165 ] . t. banks , m. r. douglas , g. t. horowitz and e. j. martinec , `` ads dynamics from conformal field theory , '' hep - th/9808016 . a. hamilton , d. n. kabat , g. lifschytz and d. a. lowe , `` holographic representation of local bulk operators , '' phys . d * 74 * , 066009 ( 2006 ) doi:10.1103/physrevd.74.066009 [ hep - th/0606141 ] . a. almheiri , x. dong and d. harlow , `` bulk locality and quantum error correction in ads / cft , '' jhep * 1504 * , 163 ( 2015 ) doi:10.1007/jhep04(2015)163 [ arxiv:1411.7041 [ hep - th ] ] . x. yin , `` on non - handlebody instantons in 3d gravity , '' jhep * 0809 * , 120 ( 2008 ) doi:10.1088/1126 - 6708/2008/09/120 [ arxiv:0711.2803 [ hep - th ] ] . j. m. maldacena , `` eternal black holes in anti - de sitter , '' jhep * 0304 * , 021 ( 2003 ) doi:10.1088/1126 - 6708/2003/04/021 [ hep - th/0106112 ] . m. van raamsdonk , `` comments on quantum gravity and entanglement , '' arxiv:0907.2939 [ hep - th ] . m. van raamsdonk , `` building up spacetime with quantum entanglement , '' gen . grav . * 42 * , 2323 ( 2010 ) [ int . j. mod . phys . d * 19 * , 2429 ( 2010 ) ] doi:10.1007/s10714 - 010 - 1034 - 0 , 10.1142/s0218271810018529 [ arxiv:1005.3035 [ hep - th ] ] . n. bao , j. pollack and g. n. remmen , `` wormhole and entanglement ( non-)detection in the er = epr correspondence , '' jhep * 1511 * , 126 ( 2015 ) doi:10.1007/jhep11(2015)126 [ arxiv:1509.05426 [ hep - th ] ] . d. berenstein and a. miller , `` topology and geometry can not be measured by an operator measurement in quantum gravity , '' arxiv:1605.06166 [ hep - th ] . d. berenstein and a. miller , `` superposition induced topology changes in quantum gravity , '' arxiv:1702.03011 [ hep - th ] . a. almheiri , x. dong and b. swingle , `` linearity of holographic entanglement entropy , '' arxiv:1606.04537 [ hep - th ] . d. harlow , `` wormholes , emergent gauge fields , and the weak gravity conjecture , '' jhep * 1601 * , 122 ( 2016 ) doi:10.1007/jhep01(2016)122 [ arxiv:1510.07911 [ hep - th ] ] . m. guica and d. l. jafferis , `` on the construction of charged operators inside an eternal black hole , '' arxiv:1511.05627 [ hep - th ] . h. lin , o. lunin and j. m. maldacena , `` bubbling ads space and 1/2 bps geometries , '' jhep * 0410 * , 025 ( 2004 ) doi:10.1088/1126 - 6708/2004/10/025 [ hep - th/0409174 ] .

orientable 3-manifolds can be easily described by their heegaard splittings . the manifold is presented as a union of two handlebodies , glued along the boundary by a homeomorphism of a splitting surface to itself . any 3-manifold possesses a heegaard splitting , though not a unique one . however , one can pass from one heegaard splitting to another by a sequence of some standard simple operations . see , e.g. , @xcite . a standard simple way to describe a heegaard splitting of a 3-manifold is by its heegaard diagram . a heegaard diagram consists of two families of disjoint simple closed curves on a splitting surface . the families intersect , producing an immersed family of curves on the surface . there is a simple combinatorial way to encode immersed curves in a plane , namely a gauss diagram . the curve is presented as a circle in the plain with double points joined with a chord . one can also encode the information on the overpasses and underpasses , changing these diagrams into diagrams for knots and links . this presentation of plane curves and knot diagrams proved useful for computational purposes , especially for finite type knot and link invariants . the same construction can be used to describe immersed curves on a surface . a manifold is described by curves on the surface and the curves are described by gauss diagrams . it seems natural to introduce a new notion of gauss diagrams of 3-manifolds , which appears to be a simple combinatorial way to describe 3-manifolds . in section [ gd ] we introduce the notion of a gauss diagram corresponding to a heegaard diagram of a closed orientable manifold . we show that the splitting surface and the entire manifold can be reconstructed from it . the equivalence of gauss diagrams is defined and we prove that a closed orientable 3-manifold is determined by the equivalence class of its gauss diagram . next we show how to compute the fundamental group and the first homology group of a manifold directly from a gauss diagram , and how to distinguish homology spheres . a picture looking like a gauss diagram does not necessary come from a closed orientable manifold . in section [ abstgd ] we consider all diagrams looking like a gauss diagram , calling them abstract gauss diagrams . we show that this construction allows to describe oriented manifolds with boundary . later we formulate the conditions for a gauss diagram to represent a closed orientable manifold and a complement of a knot in some closed orientable manifold . the author would like to express her deep gratitude to her scientific advisors prof . y. moriah and especially prof . m. polyak from the math . department at the technion , haifa , israel for their help , support and patience during the research . the author also thanks the technion for financial support . let @xmath0 be a closed orientable 3-manifold . a heegaard splitting @xmath1 of @xmath0 is a union of two handlebodies @xmath2 , glued along their boundary @xmath3 by a homeomorphism . the genus @xmath4 of @xmath3 is called the genus of the heegaard splitting . one of @xmath2 can be always taken to be standard . the main property of a handlebody @xmath5 is that there exists a set of discs @xmath6 , called a meridional disc system of @xmath5 , such that @xmath7 is a union of 3-balls . one can always find a minimal set of this sort , which means that @xmath7 is a single ball . since there is a unique way to fill in the ball , the handlebody can be easily reconstructed , if a meridional disc system is specified . this means that it is sufficient to specify the boundaries of the meridional discs on the surface . let @xmath8 be minimal meridional disc systems of @xmath2 respectively . the boundaries of these discs are two families @xmath9 of disjoint simple closed curves on @xmath3 , from which the heegaard splitting can be reconstructed by gluing 2-handles along the curves of @xmath9 and filling in 3-balls . the pair @xmath10 is called a _ heegaard diagram _ of @xmath0 . we think of both families as an immersed family of curves on the surface . in order to describe them in a more simple , combinatorial way , one can use gauss diagrams , in a way similar to the plane curves . gauss diagram _ is the immersing collection of circles with the preimages of each double point connected with a chord . each chord is given a sign depending on whether the frame of tangents to the branches of the curve in this point coincides with the orientation of the plane . one can also associate each circle a gauss word by traveling once around the circle and recording the chords with signs . gauss words were introduced by gauss in @xcite , and the planarity problem was extensively studied . see , e.g. @xcite , @xcite , @xcite . carter proved in @xcite that if two sets of immersed curves fill the surface in the sense that the complement regions are discs , then they are stable geotopic if and only if they have equivalent gauss paragraphs . the notion of gauss diagrams for knots and links was introduced by polyak and viro in @xcite . they used gauss diagrams enhanced with the information on the overpasses and underpasses . the equivalence classes of abstract gauss diagrams were called virtual knots by kauffman in @xcite , since only part of the diagrams were realizable . a _ gauss diagram _ @xmath11 consists of two families @xmath12 of @xmath4 disjoint oriented circles @xmath13 , a family @xmath14 of chords joining the circles of @xmath15 to those of @xmath16 , and the signs @xmath17 of the chords respectively . let @xmath18 be a heegaard diagram of a closed orientable 3-manifold @xmath0 . we associate to it a gauss diagram @xmath19 in a following way : each closed curve @xmath20 can be regarded as an image of an immersion @xmath21 . let @xmath12 be two families of the preimages of @xmath9 . the curves of different families intersect , and the intersection points have two preimages , one in each family . we join these points with a chord , joining a circle in @xmath15 with a circle in @xmath16 . no chords join the circles of the same family . each chord is given a sign of the intersection , depending on whether the orientation given by the frame ( tangent to @xmath22 in @xmath23 , tangent to @xmath24 in @xmath23 ) coincides with the orientation of @xmath3 , see figure [ signs ] . this construction depends on the choice of the orientation of the curves of @xmath9 . @xmath25 the choice of orientation of curves motivates the following notion of the @xmath26-equivalence : an @xmath26-move is reversing the orientation of a circle together with reversing the signs of all chords ending on this circle . two gauss diagrams @xmath27 and @xmath28 are _ @xmath26-equivalent _ if one can be obtained from the other by a sequence of @xmath26-moves . a gauss diagram can be viewed as a picture with two rows of circles joined with chords . for convenience , we assume the following conventions , when drawing a gauss diagram : the circles of @xmath15 are placed in the top row and are oriented counterclockwise , while the circles of @xmath16 , placed in the bottom row , are oriented clockwise . given a gauss diagram associated to some closed orientable 3-manifold , one might think that only the information on the intersections is available and the information on the surface is lost . it turns out , however , that this information is enough to reconstruct the entire manifold . see theorem [ gdthm ] . this result is based on a construction of a surface , associated to a gauss diagram . first we construct a ribbon graph from a gauss diagram , then glue discs with holes to the boundary components of this graph . let @xmath29 be a gauss diagram . contract all chords to points , obtaining an oriented graph @xmath30 . each vertex and each edge of this graph is equipped with a sign . the signs of the vertices are the signs of the chords they come from , where the signs of the edges signify the family @xmath12 the edge belongs to . this graph will be the core of the ribbon graph . consider two families of @xmath4 oriented annuli as a product of @xmath12 with @xmath31 $ ] . let @xmath32 denote the ends of a chord @xmath33 in the circles @xmath34 in @xmath12 respectively . let @xmath35 be small neighborhoods of @xmath32 , their direction induced by the orientation of @xmath34 . we identify the square @xmath36 with @xmath37 by a homeomorphism @xmath38 such that @xmath39 coincides with the positive @xmath40 direction in @xmath36 and the orientation is preserved . we identify ( by an orientation - preserving homeomorphism @xmath41 ) @xmath36 with @xmath42 such that @xmath43 coincides with the positive @xmath44 direction for a positive chord and with negative @xmath44 direction for a negative chord . a homeomorphism @xmath45 identifies the small squares in these annuli at the neighborhood of each @xmath32 cross - like , as in figure [ glue ] . using such a homeomorphism for all chords , one obtains a ribbon graph @xmath46 with core @xmath30 . we will call it _ a ribbon graph associated to the gauss diagram_. a ribbon graph can be viewed as a union of ribbons with bases glued to small squares , called coupons by reshetikhin and turaev in @xcite . we will treat ribbon graphs in a similar way . [ ribbon ] let @xmath18 be a heegaard diagram of some closed orientable 3-manifold @xmath0 , and let @xmath47 be a regular neighborhood of the union @xmath48 in @xmath3 . let @xmath19 be a gauss diagram associated to @xmath18 . then @xmath49 and @xmath50 are homeomorphic as pairs . we have two ribbon graphs @xmath47 and @xmath46 . by construction of @xmath46 we have the same number of coupons and ribbons in both graphs . one can identify all the coupons according to the order of intersections in one of the families @xmath9 , say @xmath51 . when identifying the coupons , one has to take care of the identification of the ribbon bases . however , the gauss diagram provides the exact order to do it , since the cyclic order of the edges joining the vertex is known . since the order of the identification of the coupons is given by the order of edges in @xmath51 , meaning that we come to the coupon by a specific edge , the identification of the ribbon bases is unique . since both ribbon graphs are oriented as surfaces and the identification goes along annuli , this can be done in orientation - preserving way . thus the bases of the ribbons corresponding to the edges of @xmath52 are identified . the cores of these ribbons are identified by the obvious homeomorphism , hence these ribbons can be also identified now , in the orientation - preserving way . a ribbon graph @xmath46 is an oriented surface with boundary . as reshetikhin and turaev mention in @xcite , we can think that the bands of an oriented ribbon graph have two sides and we always see only one of them . we assume that this side is such that the boundary arc whose direction coincides with the orientation of the edge is always seen on the right side of the edge , as in figure [ colorglue ] . if our gauss diagram is associated to some heegaard diagram @xmath18 of a closed orientable manifold , this graph is a ribbon graph on the surface @xmath3 . each family @xmath9 separates @xmath3 into a sphere with holes , hence the complement of this graph in the surface is a collection of discs with holes . we would like to specify the sequences of edges the boundaries of these discs are glued along . we follow the construction of j.s . carter in @xcite for gauss paragraphs , adapting his construction to the case of gauss diagrams . if one is approaching the vertex , while going along a boundary component , he has to turn right - see figure [ crossing ] . in terms of gauss diagrams , a turn in a vertex of @xmath30 means that if one approaches a chord , going along a curve of the top family @xmath15 , he has to descend using a chord and continue along the curve of @xmath16 , and vice versa . this rule can be applied to any gauss diagram , not necessary to the one associated to some heegaard diagram of some manifold . right turn rule " then means that for the positive chord one has to cross it on ascending path and to continue along it on the descending path . see figure [ negpos_fig]a . approaching the negative chord , one ascends along the chord and crosses it on his way downwards . see figure [ negpos_fig]b . a sequence of edges corresponding to the boundary component is called a _ cycle _ in a gauss diagram . we are going to glue discs with holes to boundary components . each edge of the gauss diagram @xmath29 produces two arcs in the boundary of @xmath46 . we would like to know directly from @xmath29 which boundary component contains each of these arcs . it proves useful for computational purposes to include in the gauss diagram the information which pair of cycles the edge belongs to . an easy way to encode it is using the elements of a finite set @xmath53 . elements of @xmath53 are in one - to - one correspondence with boundary components of @xmath46 . we associate to each edge of @xmath29 an ordered pair @xmath54 . the first element corresponds to the boundary component having the same direction as the edge itself and the second one - to the opposite one . in these terms the right turn rule " will look as follows . let @xmath55 , @xmath56 be the pairs assigned to incoming and outgoing edges of @xmath57 for some chord @xmath23 , @xmath58 - the pairs assigned to incoming and outgoing edges of @xmath59 respectively . see figure [ colorglue ] . for positive intersection we have : @xmath60 , see figure [ colorglue]a , and for negative one we have : @xmath61 , see figure [ colorglue]b . from now on we incorporate this information into gauss diagram , without changing the notation . we would like to construct a closed surface from @xmath46 , hence we glue discs with holes to it . the result of this construction is not unique . if @xmath62 , we would like to obtain the surface homeomorphic to @xmath3 and to get a heegaard diagram producing the same manifold . to do this one has to know which boundary components will be glued to the same disc with holes . this leads us to the notion of a decorated gauss diagram . consider an equivalence relation on the set @xmath53 of all the cycles of @xmath19 , where two cycles will be equivalent if they correspond to the boundary components of the same disc with holes in @xmath63 . it is convenient to define this equivalence relation as a coloring of the elements of @xmath53 by the colors @xmath64 of the corresponding discs with holes . a _ decorated gauss diagram _ @xmath65 is a gauss diagram with a color @xmath66 associated to each cycle , where the colors are members of a finite set @xmath67 . we associate to a heegaard diagram @xmath18 a decorated gauss diagram @xmath68 . the only thing to do is to define the coloring of the cycles . all boundary components of the same connected component of @xmath63 have the same color . to construct the surface associated to @xmath69 , for each color pick a sphere with a number of holes equal to the number of cycles having this color and glue it to all these cycles . this surface will be denoted by @xmath70 and will be called _ the surface associated to the decorated gauss diagram_. the manifold associated to it can be constructed by gluing discs to @xmath12 and capping the spheres with 3-balls . [ surfdeco ] let @xmath10 be a heegaard diagram of a closed orientable 3-manifold @xmath0 . the surface @xmath70 associated to @xmath71 is homeomorphic to @xmath3 and the associated manifold is homeomorphic to @xmath0 . since each of the families @xmath9 separates @xmath3 into discs with holes , the surface @xmath72 consists of discs with holes , too . by proposition [ ribbon ] , we have a homeomorphism @xmath73 . consider all the cycles of @xmath71 colored by the same color @xmath66 . all of them correspond to the boundary components of @xmath74 , whose images under @xmath75 are glued to the same disc with holes in @xmath72 . the discs with holes glued to the boundary components of @xmath74 and the components of @xmath72 are homeomorphic . the homeomorphism is defined by the image of @xmath75 on the boundaries . since the heegaard diagram was reconstructed , the manifold obtained by gluing discs along the core of @xmath74 will be homeomorphic to @xmath0 . if @xmath76 is a union of discs , then all cycles of @xmath71 have different colors . this means that a manifold can be reconstructed from a non - decorated gauss diagram , and we can omit the decorations . suppose that on the surface @xmath3 there are two curves belonging to different families @xmath9 , which can be isotoped along @xmath3 , such that a pair of cancelling intersections appears , in a manner similar to the second reidemeister move in knots . see figure [ reidmove ] . in a gauss diagram this will look like adding a pair of adjacent chords with opposite signs . consider a decorated gauss diagram @xmath69 . let @xmath77 be two edges , belonging to two cycles of the same color . let @xmath78 be the respective colorings of @xmath79 , and let @xmath80 be points of @xmath81 , @xmath82 be points of @xmath83 . the chord joining @xmath84 with @xmath85 will be positive and the chord joining @xmath86 with @xmath87 will be negative . going along @xmath88 induces an order on the points of an edge . we write @xmath89 if @xmath23 appears before @xmath90 . the signs and the order of the chords in @xmath88 depends on the colors of the edges : * if @xmath91 then @xmath92 and @xmath93 * if @xmath94 then @xmath95 and @xmath93 * if @xmath96 then @xmath92 and @xmath97 * if @xmath98 then @xmath95 and @xmath97 . this rule explains which sides of the bands meet when two chords appear while two cancelling intersections are created . see figure [ reidglue ] for an example of two bands meeting with @xmath98 . we will call this change in a decorated gauss diagram an _ @xmath99-move_. during an @xmath99-move three new edges @xmath100 are created instead of each of @xmath88 , so that @xmath101 have a common chord , and @xmath102 too . see figure [ reidglue ] . the colorings of these edges are given by the following . if @xmath88 belonged to the same cycle of color @xmath103 , then there is a partition of the set of all cycles colored by @xmath103 into two sets , colored into two new colors that did not exist in the color set of @xmath29 . two new cycles , containing @xmath104 and @xmath105 respectively , belong to different sets . a new cycle , consisting of two edges @xmath106 has an entirely new color @xmath107 . if @xmath88 belonged to two different cycles of color @xmath103 , then two new cycles appear . one of them contains all the edges of the old cycles and two pairs @xmath108 and has the same color @xmath103 . the other one consists only of @xmath106 and has a new color @xmath107 . the edges @xmath104 and @xmath105 have the same coloring pairs as @xmath88 , save that @xmath103 can be replaced by a new color , as mentioned above . the edges @xmath109 have a new coloring , obtained from the coloring of @xmath88 by the following changes : @xmath103 is replaced by @xmath107 , the second color in the pairs is switched between @xmath81 and @xmath83 , and the colors inside each pair are also switched . e.g. , if @xmath81 is colored by @xmath110 and @xmath83 - by @xmath111 , as in figure [ reidglue ] , the edge @xmath112 is colored by @xmath113 and @xmath114 - by @xmath115 . the opposite move to an @xmath99-move is called _ @xmath116-move_. two decorated gauss diagrams are called _ @xmath99-equivalent _ if one can be obtained from the other by a sequence of @xmath99 and @xmath116-moves . @xmath99-equivalent diagrams have the same heegaard splitting . an @xmath99-move reminds us the second reidemeister move in knots . there is no equivalent to the third reidemeister move , since three curves intersecting would produce an intersection between the curves of the same family , which is forbidden . a ribbon graph @xmath74 may be connected , and may be not . this leads us to the notion of @xmath99-connected diagram . let @xmath69 be an abstract decorated gauss diagram . consider a graph , whose vertices are connected components of @xmath69 . two vertices are connected with an edge if and only if the sets of colors appearing on these connected components have a color in common . we call the diagram @xmath69 _ @xmath99-connected _ if and only if this graph is connected . the surface @xmath70 is connected if and only if @xmath69 is @xmath99-connected . the connected components of @xmath74 are exactly the connected components of @xmath69 itself . two components will have the same disc with holes glued to both of them if and only if they have a color in common . hence the connected components of the graph defined above correspond to connected components of @xmath70 . [ requiv ] any decorated gauss diagram @xmath69 is @xmath99-equivalent to a gauss diagram with the number of colors used on it equal to the number of cycles . the resulting diagram is connected if and only if @xmath69 is @xmath99-connected . consider a decorated gauss diagram @xmath69 . let @xmath117 be a non - simply connected component of @xmath118 . it is a disc with holes and each of its boundary components contains arcs of both families . perform an @xmath99-move between two different boundary components of @xmath117 . such a move produces two components , such that one of them is a disc with a new color . the other one has one hole less than the initial , thus lowering the number of cycles colored by the same color . see figure [ reidemeister ] . an @xmath99-move does not change the number of boundary components of any other component of @xmath118 . the resulting diagram will be @xmath99-equivalent to the previous one . one can continue this process till @xmath117 turns into a disc . remove all non - simply - connected components of @xmath118 by the @xmath99-moves . this will give a decorated gauss diagram with all cycles colored differently . if the gauss diagram @xmath71 is not connected , we assume for convenience that it has two connected components . if @xmath69 is @xmath99-connected , there are two cycles colored by the same color in different components of @xmath71 . an @xmath99-move between these cycles will change the diagram into a connected one . [ connexist ] for any heegaard diagram @xmath10 of a closed connected orientable 3-manifold there is a connected gauss diagram @xmath29 with all cycles colored in different colors @xmath99-equivalent to @xmath71 . moreover , there is a heegaard diagram isotopic to @xmath10 , such that @xmath29 is associated to it . to compute the genus of the surface associated to an abstract gauss diagram , we will use the notion of the _ excess of a coloring _ , which is the difference @xmath119 between the number of cycles and number of colors . obviously , if all cycles have different colors , @xmath119 vanishes . [ genus ] the genus of a surface @xmath70 associated to an abstract decorated gauss diagram @xmath69 is @xmath120 , where @xmath121 is the number of cycles and @xmath122 is the number of chords in @xmath69 . this can be done by computing the euler characteristic of @xmath70 . the surface @xmath70 is constructed by gluing discs with holes to the boundary components of a ribbon graph @xmath74 . the number of vertices of @xmath74 is the number of chords @xmath122 . the number of edges of @xmath74 equals twice the number of chords . we do not count circles that have no chord attached to them , since they produce annuli in @xmath74 , and their euler characteristic is @xmath123 . thus @xmath124 . glue discs with holes to @xmath74 . for each of them , the number of the boundary components equals the number @xmath125 of cycles colored by @xmath66 . the total number of these discs with holes is the number of colors @xmath126 . the euler characteristic of each such disc with holes is @xmath127 . summarizing , we have : @xmath128 the sum over @xmath125 equals the number of cycles @xmath121 , giving @xmath129 . the genus is given by @xmath130 . a surface associated to a non - decorated gauss diagram @xmath29 is a surface obtained by gluing discs to all boundary components of @xmath46 . in other words , this is the surface associated to a decorated gauss diagram with all cycles colored in different colors . [ gnondeco ] the genus of the surface associated to the connected non - decorated gauss diagram is @xmath131 . let @xmath10 be a heegaard diagram of some closed orientable 3-manifold @xmath0 . assume that @xmath132 is connected . let @xmath4 be the number of curves in each family @xmath9 . if @xmath133 , then @xmath134 is a collection of discs , otherwise @xmath10 is reducible . if @xmath133 , the surface associated to a non - decorated gauss diagram is homeomorphic to @xmath3 . the boundaries of the meridional disc system are also reconstructed , hence @xmath3 is separated into discs by the curves of @xmath135 . if @xmath136 , then cutting @xmath137 along the curves of the diagram produced surfaces other than discs . thus any essential closed curve in such a non - simply connected component bounds a disc in both handlebodies of the heegaard splitting . this means that the heegaard splitting is reducible . we introduce now two types of moves of gauss diagram motivated by some equivalences of heegaard splittings . suppose that two curves @xmath138 on the heegaard diagram @xmath10 , belonging to the same family @xmath9 , say @xmath51 , can be joined together with a simple curve @xmath139 disjoint from all other curves of @xmath9 . let @xmath140 be the boundary components of a thin collar @xmath141 $ ] of @xmath139 in @xmath134 . set @xmath142 by a slight push make this curve disjoint from @xmath143 . this does not change the handlebodies of the splittings , only the meridional discs defining them . the heegaard diagram @xmath144 is obtained from the diagram @xmath10 by a handle slide of @xmath145 along @xmath143 . see figure [ handlesliding ] . this process changes the gauss diagram of @xmath0 , with a small section of @xmath146 being substituted by a curve equal to @xmath147 without a small segment . all the curves that intersected @xmath143 now intersect @xmath148 also , which means that we have the copies of all chords that joined @xmath147 now joining @xmath149 . the signs of the intersections ( and the chords ) are the same , if the direction of the segment removed coincides with the direction of the added segment . if not , one has to change the direction of the added segment , and reverse the signs for all chords joining it , as in the case of @xmath26-equivalence . clearly , the move is allowed only if the segments removed from both cycles are of the same color . we say that the resulting diagram is obtained from the initial one by an _ @xmath5-move_. the opposite of this move will be called _ @xmath150-move_. the change of the colors in the case of handle slide is illustrated in figure [ colorslide ] . the color of the region is the color of the cycle bounding it . let @xmath151 be the colorings of the edges @xmath152 of @xmath147 , and let @xmath139 join @xmath153 and @xmath154 . let @xmath155 be the coloring of @xmath156 . the edges @xmath152 , save @xmath153 , will now be parts of new small cycles , consisting only of @xmath157 edges each and having new colors @xmath158 . the edges @xmath159 will have colors @xmath160 . the edge @xmath153 will have the coloring @xmath161 , if the directions of the added and removed segments coincide . in this case also the edges @xmath162 added to @xmath146 have colors @xmath163 and belong to the same new cycles . the edges @xmath162 have colors @xmath164 . the remaining two edges of these new cycles have the colors @xmath165 or @xmath166 , depending on the signs of their ends . we described already the colors of all @xmath162 save the first and the last ones . their coloring equals that of @xmath156 if @xmath139 joined two edges of different cycles . if not , the color @xmath167 splits and the coloring is as in figure [ colorslide ] . if the direction of the added segment is opposite to that of the removed one , the colorings of the edges of the added segment switch . [ hgs ] @xmath5-move does not change the genus of @xmath70 . consider the slide of @xmath168 along @xmath169 . let @xmath170 be the number of chords joining @xmath169 . then the total number of chords becomes @xmath171 . the number of cycles and colors depend on whether the edges joined by @xmath139 belonged to the same cycle or not . if @xmath139 connected two edges of the same cycle , two cycles are created instead of it . their colors are different , and the set of the cycles colored by the initial color is partitioned into two sets , each with a new color . additional @xmath172 new @xmath157-edged cycles are also created , each with a separate color . hence both the number of colors and the number of cycles increases by @xmath170 , and @xmath119 does not change . the formula of proposition [ genus ] for the changed diagram @xmath173 gives : @xmath174 . see also figure [ colorslide ] . if @xmath139 connected two edges of two different cycles of the same color , these cycles are now replaced by a single one . hence the number of colors increases only by @xmath172 , while the number of cycles increases by @xmath175 . the formula of the proposition [ genus ] then gives us : @xmath176 . another basic operation with heegaard splittings is stabilization . this operation consists of adding a pair of cancelling handles . see figure [ stabilization ] . one can think of it as a connected sum of @xmath0 with @xmath177 , where the ball of connection is a small neighborhood of a point on the splitting surface @xmath3 disjoint from the curves of @xmath9 , and @xmath177 is thought as having a standard genus 1 splitting @xmath178 . the boundaries of the meridional discs of the solid tori @xmath178 intersect in one point and are disjoint from @xmath9 . this means that in terms of gauss diagrams the stabilization can be viewed as adding a circle to each family @xmath12 . these two circles are joined with one chord ( the sign depends on the orientation of the boundaries of the meridional discs of @xmath178 and can be chosen arbitrarily ) , and no chord joins them to the other circles . they produce a single cycle whose color is one of the existing colors . the color represents the place where the stabilization was made . we call this operation an _ @xmath179-move_. there is , certainly , the inverse operation of removing two cancelling handles , disjoint from all others . this looks like removing a pair of circles joined by a single chord and not connected to any other circle of a gauss diagram . _ @xmath180-move_. [ sgs ] @xmath179-move increases the genus of @xmath70 by @xmath181 . let @xmath173 be a stabilization of @xmath69 . a single cycle was added to @xmath69 , and no new colors . hence @xmath119 increased by @xmath181 . we also have one chord more . substituting to the formula of the proposition [ genus ] for @xmath182 we have : @xmath183 . two gauss diagrams are called _ @xmath184-equivalent _ if one can be obtained from the other by a sequence of @xmath179-moves , @xmath5-moves and their opposites . one of the problems of heegaard splittings is that a 3-manifold can be presented by different heegaard splittings ( possibly of a different genus ) . moreover , the same heegaard splitting can be presented by different heegaard diagrams . the @xmath184-equivalence reflects this phenomenon on the level of gauss diagrams . two gauss diagrams are _ equivalent _ if one can be obtained from the other by a sequence of @xmath185-moves or their opposites . [ gdthm ] a closed connected orientable 3-manifold is in one - to - one correspondence with the equivalence class of its gauss diagram . a well known theorem by singer says that one can pass from one heegaard splitting to any other heegaard splitting of the same closed manifold by a consequent use of handle slides and stabilizations . see @xcite . on the level of gauss diagrams it means that the gauss diagrams associated to these heegaard splittings are @xmath184-equivalent . now let @xmath10 be a heegaard diagram of some closed orientable 3-manifold m. by corollary [ connexist ] , there is a connected gauss diagram @xmath29 with all cycled colored differently , @xmath99-equivalent to @xmath132 , and the curves of the heegaard diagram can be isotoped so that @xmath29 is the diagram associated to @xmath10 after the changes . we will not change the notation for the isotoped @xmath9 . construct the surface associated to @xmath29 . by proposition [ ribbon ] , a ribbon graph associated to @xmath132 is homeomorphic to a regular neighborhood of @xmath9 . consider the boundary components of this neighborhood . each of them is a simple closed curve disjoint from the curves of @xmath9 , homotopic to the image of a cycle in @xmath132 under this homeomorphism , and all cycles of @xmath132 are presented by these boundary components . if this curve is essential in @xmath3 , compress along it . continue for all cycles . take only those components of the resulting surface that contain curves of @xmath9 . this surface is homeomorphic to the surface @xmath137 . the complement of the ribbon graph in @xmath137 is a collection of discs . in the surgered surface containing @xmath9 , the complement of @xmath9 is also a collection of discs , the boundary of each is a cycle in the gauss diagram . a homeomorphism is constructed by taking the corresponding discs to each other and the correspondence is determined by the boundary . thus any two heegaard diagrams with equivalent gauss diagrams can be obtained from a surface associated to their common gauss diagram by a connected sum with unknotted tori and handle slides . the algorithm of section [ cell complex ] gives us the possibility to reconstruct the surface from the gauss diagram , but it is hard to visualize it . the following algorithm allows us to reconstruct the picture of the heegaard diagram in an easy and quite visual way , provided that a gauss diagram comes from some heegaard diagram of a closed orientable 3-manifold m and is connected . we also need to assume that @xmath186 , where @xmath187 is given by the formula of corollary [ gnondeco ] . consider such a gauss diagram @xmath29 . since one of the families @xmath9 of curves on the surface can be taken to be standard , we start with an oriented disc with @xmath188 holes . let the boundary have the orientation induced by the orientation of the disc . split the set of the boundary components in pairs . choose an orientation - reversing homeomorphism between the curves and identify them in pairs . this gives us a surface @xmath179 of genus @xmath4 . see figure [ gdv ] , where the boundary components already have reversed orientations , where necessary . after the identification we refer to the identified curves as @xmath189 . the curves @xmath190 can be considered as a map from a gauss diagram to a heegaard diagram , defining the closed curves belonging to one of the families . to get a map from @xmath16 to @xmath179 for all segments of the circles of @xmath16 between the endpoints of the chords draw disjoint curves in @xmath179 between the points corresponding to the upper ends of the chords intersecting this segment . when choosing the copy of @xmath190 that should contain the final point of the edge of @xmath191 , one has to take the chord sign into account . for instance , taking the orientations of @xmath192 s as in figure [ gdv ] , @xmath193 has to join the counterclockwise copy of @xmath194 for a negative intersection and the clockwise copy of @xmath194 for a positive one . the concatenation of these curves after identifying the @xmath192 s will be a closed curve in @xmath179 corresponding to a circle of @xmath16 . the gauss diagram associated to the heegaard diagram obtained above is the same as @xmath29 . by theorem [ gdthm ] , the manifold given by this heegaard diagram is homeomorphic to @xmath0 . the fundamental group of a closed orientable 3-manifold @xmath0 and its first homology group can be computed easily from a gauss diagram @xmath29 , associated to some heegaard splitting of @xmath0 . for each circle @xmath195 of @xmath16 we write a word @xmath196 in letters @xmath197 in the following way : let @xmath198 be the sequence of chords with signs @xmath199 joining @xmath195 to the circles of @xmath15 . suppose a chord @xmath200 joins @xmath201 to @xmath202 . for such a chord we write @xmath203 in @xmath196 . thus we get a cyclic word , which is a relation in the presentation of @xmath204 . [ pi1 ] let @xmath10 be a heegaard diagram of some closed orientable 3-manifold @xmath0 . assume that @xmath205 is connected and @xmath186 . then @xmath206 has a presentation @xmath207 , where @xmath208 correspond to the circles of @xmath15 and @xmath196 correspond to the circles of @xmath16 and are obtained as above . we use the algorithm and the notation of section [ gdvisualizing ] . each @xmath208 corresponds to a curve @xmath209 , see figure [ gdv ] . the fundamental group @xmath204 can be obtained from the fundamental group of @xmath179 by adding the relations given by the discs glued to the curves of the heegaard diagram . the curves @xmath210 form the generator set for @xmath211 . when we glue discs to the curves corresponding to @xmath15 , we remain with @xmath209 s only . they will be the generators of @xmath204 . when one wants to count how many times does the curve wind along the handle defined by @xmath209 , he simply counts how many times it crosses some fixed curve dual to @xmath209 , which is @xmath190 . this explains the form of the relations . let @xmath212 denote the algebraic number of chords joining the circle @xmath195 in @xmath16 to a circle @xmath213 in @xmath15 . the first homology group of @xmath0 is the abelian group generated by @xmath214 , where @xmath208 correspond to the circles of @xmath15 . the relations are @xmath215 , where @xmath216 . the algebraic numbers of chords joining the circles can also be written in the intersection matrix @xmath217 . this matrix allows us to determine whether a manifold is a homology sphere . the 3-sphere : : @xmath177 is the simplest example . we represent a gauss diagram of the standard genus @xmath181 splitting of it . see figure [ sphere ] . + lens spaces : : they are a bit more complicated . in figure [ lens ] we present the gauss diagrams of @xmath218 and @xmath219 . the signs of all chords are @xmath220 on both diagrams . the fundamental group is the same , as one should expect : @xmath221 + poincare homology sphere : : figure [ poincare ] represents a heegaard diagram and the associated gauss diagram of the poincare manifold @xmath222 , as it appears in @xcite . the picture presented by rolfsen looks exactly like the one obtained by the algorithm of section [ gdvisualizing ] from a gauss diagram . the fundamental group , according to the proposition [ pi1 ] , is @xmath223 which is the binary icosahedral group , as expected . abelinizing , we get @xmath224 , hence @xmath222 is indeed homology sphere . + another homology sphere : : we consider the example of a homology sphere given by hempel in @xcite . its heegaard diagram is shown in figure [ homsph ] , together with its gauss diagram . one can easily calculate the fundamental group : @xmath225 which gives trivial first homology group . the intersection matrix is in this case @xmath226 and its determinant is @xmath181 , in agreement with the expectations . a natural question that arises when dealing with gauss diagrams of 3-manifolds is : given an abstract gauss diagram , can one find a manifold that this gauss diagram is associated to it ? gauss diagrams for 3-manifolds closely resemble gauss diagrams for knots and links . the equivalence classes of gauss diagrams for links that do not necessarily correspond to a link are called virtual links . we can also call the equivalence classes of gauss diagrams of 3-manifolds virtual manifolds . below we show that a 3-manifold , in general with a boundary , can be associated to any abstract gauss diagram . in section [ bdry ] we give the conditions for an abstract gauss diagram to represent a closed 3-manifold or a component of a knot in some closed orientable 3-manifold . in order to generalize the definition of heegaard splitting to 3-manifolds with boundary , one replaces the handlebodies with the compression bodies . see , e.g. @xcite . a _ compression body _ @xmath5 is a 3-manifold with boundary obtained in a following way : let @xmath3 be a closed surface . take a cylinder @xmath227 $ ] , attach to @xmath228 a collection of 2-handles . if the resulting manifold has any spherical components of the boundary , fill them with 3-balls . the surface @xmath229 is called the outer boundary @xmath230 of the compression body , and the surface @xmath231 is called the inner boundary of @xmath5 . if a boundary has one component only , namely its outer boundary @xmath230 , we get a handlebody . the boundary of a compression body may consist of more than two components . the genus of the outer boundary is the genus of a compression body . unlike the case of a handlebody , the genus does not determine the manifold completely . two compression bodies of the same genus have homeomorphic outer boundaries , but there are no restrictions on the inner boundary , and the manifolds can be completely different . a heegaard splitting of a manifold with boundary is then a union of two compression bodies @xmath2 glued using a homeomorphism of the outer boundaries @xmath232 of the compression bodies . the surface @xmath233 is then called a splitting surface of the heegaard splitting . we also allow now the families @xmath12 to have different number of circles . the number of circles in each family @xmath12 will be denoted by @xmath234 . assume that @xmath69 is an abstract decorated gauss diagram with @xmath4 circles in each family . then one can look at the gauss diagram as representing curves on @xmath70 to which the discs are glued . being more precise , one takes a collar @xmath235 $ ] and attaches 2-handles to it along the curves in @xmath236 and @xmath237 . if a 3-sphere appears in the boundary , it is filled with a 3-ball . then @xmath238 is a heegaard splitting surface for a manifold , possibly with boundary . if one adds to one of the families of a heegaard diagram a curve contractible on the splitting surface , gluing disc along it produces a spheric boundary component . this sphere is filled with a ball and the heegaard splitting does not change . this move is called a bubble - move or a _ @xmath239-move_. in terms of gauss diagrams this looks like the addition to one of the families @xmath12 a circle ( called a _ bubble _ ) without chords . this circle produces two cycles , one colored in a new color and the other in one of the existing colors . the choice of the existing color is the choice of the component of @xmath240 to which the bubble is added . the opposite move to this one , i.e. a removing of such a circle , is called @xmath241-move . let @xmath242 be two abstract decorated gauss diagrams . assume that @xmath243 can be obtained from @xmath244 by an @xmath26-move . we would like to show that @xmath243 can be obtained from @xmath244 by a sequence of @xmath245 and @xmath246-moves . the diagram @xmath244 differs from @xmath243 by the signs of the chords joining one circle , say @xmath57 in @xmath247 . choose an edge in @xmath57 . let @xmath248 be its coloring . perform a @xmath239-move adding a circle @xmath249 to @xmath247 , colored by @xmath250 , where @xmath156 is a new color . perform an @xmath5-move , sliding @xmath249 along @xmath57 . the new circle will have the direction opposite to that of @xmath57 , and the chords will have opposite signs . perform an @xmath150-move on @xmath57 changing it into bubble . remove this bubble with @xmath241-move . the resulting diagram is @xmath243 . see figure [ epsilon ] . let @xmath173 denote the diagram resulting from a gauss diagram @xmath69 after the @xmath239 move . a @xmath239-move adds a circle to one of the families , without any chords joining it . two new cycles and one new color are added , increasing the excess of coloring @xmath119 by @xmath181 and @xmath121 by @xmath251 . hence @xmath252 . we would like to understand which conditions a gauss diagram should satisfy so that a manifold associated to it is closed . if the number of circles in each of @xmath12 equals @xmath253 , we have an honest heegaard splitting of a closed orientable 3-manifold , provided that neither @xmath15 nor @xmath16 separate the surface . see , e.g. , @xcite . for a decorated gauss diagram @xmath69 consider a pair of graphs @xmath254 . the vertices of both graphs are the colors used in @xmath69 . two vertices @xmath255 of , say , @xmath256 have a common edge if there is an edge or a circle without chords in @xmath16 with a coloring @xmath257 or @xmath258 . the following lemma is obvious : let @xmath265 be surfaces obtained from @xmath70 by a surgery on the curves of @xmath12 respectively . each @xmath265 is the inner boundary of the compression body @xmath2 together with several spheres that will be capped . then the euler characteristic of , say , @xmath266 is : @xmath267 this is also the sum of the euler characteristics of the components @xmath268 of @xmath266 , each of which has genus @xmath269 . their number is exactly the number of the components of @xmath270 . by lemma [ separation ] this number is @xmath271 . consequently , @xmath272 and the assertion follows . [ clmfldcond ] if @xmath273 , then the manifold associated to @xmath69 is a closed orientable 3-manifold . if @xmath274 , or vice versa , then the manifold associated to @xmath69 is a complement of a knot in some closed orientable 3-manifold . let @xmath2 be compression bodies obtained from @xmath275 $ ] by gluing 2-handles along the curves of @xmath12 respectively and filling in 3-balls . for a handlebody we need @xmath276 , since @xmath261 has to be empty . to get a single torus as the inner boundary of @xmath277 and a single boundary component of @xmath0 , we need @xmath278 . consider a @xmath239-move . by lemma [ bgs ] , it preserves @xmath253 . when a bubble is added to @xmath15 , a vertex is added to both graphs @xmath263 . in @xmath280 this vertex is joined to another by an edge , and @xmath281 does not change , but in @xmath256 no edge joins it . hence @xmath271 increases by @xmath181 . the number of circles also increases by @xmath181 only in @xmath15 . computing @xmath260 by formula of proposition [ pg ] shows that the values are preserved by @xmath239-move . consider an @xmath179-move . both @xmath234 increase by @xmath181 . no new colors are added , and the edges added are colored by @xmath282 . hence no change is made on @xmath262 . from lemma [ sgs ] we know that @xmath253 increases by @xmath181 . hence @xmath260 are preserved by @xmath179-move . consider an @xmath5-move . this time @xmath253 is preserved ( lemma [ hgs ] ) , and @xmath234 too . according to lemma [ separation ] , @xmath262 is the number of connected components of @xmath240 . assume , for convenience , that the slide occurs in the @xmath15 family . we would like to understand how it affects the connected components of @xmath240 . since only the curves of @xmath15 are affected by the move , there is no change on the connected components of @xmath283 , and @xmath281 is preserved . let @xmath147 slide along @xmath146 . both of them are boundary components of some @xmath284 , which is a connected component of @xmath270 . a handle slide looks on @xmath284 like replacing two components of @xmath285 by their connected sum . assume that @xmath284 is glued along @xmath147 to another component @xmath286 of @xmath270 . a slide of @xmath147 along @xmath146 looks like the addition to @xmath286 of an annulus on a thin band . no new components are created in @xmath270 , and no components are glued together . see figure [ surfslide ] . if both copies of @xmath147 are boundary components of @xmath284 , then an annulus on a thin band is glued to the same @xmath284 . there is no change in the other connected components of @xmath270 . hence @xmath271 is preserved in both cases . we conclude that @xmath260 is invariant under @xmath5-move . consider an @xmath99-move . there is no change in @xmath234 . if the edges , where the intersections appear , belonged to different cycles , then the number of colors increases by one , and the number of cycles does not change this means that the genus of the surface associated to the changed diagram @xmath173 is @xmath287 . in this case the only new color , in both @xmath259 , is a vertex of valency @xmath181 , joining the existing components . see also figure [ reidglue ] . there is no other change in the graphs @xmath259 . if the edges where the intersections appeared belonged to the same cycle , then the number of colors increases by @xmath251 . the number of cycles @xmath121 also increases by @xmath251 . this means that @xmath288 . one of the new colors , in both @xmath259 , is a vertex of valency @xmath181 , joining the existing components , as in the previous case . there are two more new colors replacing one old , and this looks like splitting the vertex . just as in the case of handle slide , these vertices belong to the same component , since in both graphs a path can be found between them . we conclude that @xmath262 is also preserved and @xmath260 is invariant under @xmath99-move . obviously , two equivalent diagrams have the same associated manifold . now assume that we have a gauss diagram @xmath244 , satisfying @xmath273 , that is equivalent to a gauss diagram @xmath243 with @xmath289 . by proposition [ pginvt ] , @xmath290 . by corollary [ clmfldcond ] , the manifold associated to @xmath243 is closed . by theorem [ gdthm ] we are done . it remains to show that any abstract gauss diagram @xmath244 satisfying @xmath273 is equivalent to a gauss diagram @xmath243 with @xmath289 . we can assume that @xmath291 . let @xmath284 be a connected component of @xmath292 . if @xmath284 is not a disc with holes , then @xmath293 together with discs glued to its boundary components is a non - trivial component of @xmath294 . this contradicts the assumption @xmath295 . thus @xmath284 is a disc with holes . since @xmath296 , there is a circle @xmath57 in @xmath285 , that is not glued to any other component of @xmath285 . isotope @xmath57 , probably performing some @xmath99-moves , so that @xmath57 can be connected by arcs disjoint from @xmath16 to all other components of @xmath285 . then @xmath57 can slide along all boundary components of @xmath284 . a handle slide looks on @xmath284 like replacing two components of @xmath285 by their connected sum , hence the number of components of @xmath285 decreases . from the proof of proposition [ pginvt ] we know that handle slides do not change @xmath262 . the handle slides of @xmath57 along the components of @xmath285 change @xmath284 into a disc . since @xmath297 consists of disjoint arcs and bubbles , all these can be removed by @xmath241 and @xmath116-moves . this will change @xmath57 into a bubble . remove the bubble by another @xmath241-move . this will decrease @xmath271 by @xmath181 . thus @xmath244 is equivalent to a diagram with @xmath298 . since we reconstruct @xmath0 from its gauss diagram as a cell complex , there is an obvious way to obtain its fundamental group . a gauss diagram is a graph , which we denote also by @xmath29 for convenience . let @xmath29 be an abstract gauss diagram with all cycles colored differently and @xmath0 the associated manifold . let @xmath299 be the maximal tree subgraph of @xmath29 , containing all chords . the generators of @xmath204 are the edges of @xmath300 , and the relations are defined by the cycles of @xmath29 and the circles of both families . the fundamental group of the ribbon graph @xmath46 coincides with the fundamental group of @xmath29 , which is the free group generated by the edges of @xmath300 . to construct the splitting surface one glues discs to the boundary components of @xmath46 , which means to the cycles . for any cycle write a relation going along it and writing down only the edges in @xmath300 . in the cycles part of the edges are traversed in the direction opposite to their orientation . in this case write the inverse of the generator . gluing 2-handles along the circles of @xmath29 means introducing another set of relations . the circles are also sequences of edges , and only those in @xmath300 should be written . filling in 3-balls does not affect the fundamental group . there is another , simpler way to obtain the fundamental group for closed orientable manifolds , if the number of circles in each family of the gauss diagram equals the genus of the associated surface and no separation occurs . this is the method of proposition [ pi1 ] . the first homology group of @xmath0 is the abelian group generated by the edges of @xmath300 . the relations coming from the circles in the gauss diagram remain unchanged . in some cycles there could be edges appearing twice , with opposite signs . hence in @xmath301 in the relations coming from the cycles one writes only those generators that are colored by two different cycles . thickened torus : : @xmath302 can be represented by a diagram in figure [ torusxi ] , which also shows how the curves are situated on the surface . the count of cycles gives @xmath303 , so @xmath304 , which equals the number of curves in each family . one can also check , using the graphs @xmath259 of section [ bdry ] , that each curve separates the surface into two components . the proposition [ pg ] then implies that both boundary components are tori . + unknot complement : : the diagram of figure [ unknot ] has @xmath157 cycles , which implies genus @xmath305 , and drawing the graphs @xmath259 one can see that the top family does not separate the surface , while the bottom does , and we have two components . let us count the fundamental group . the maximal tree @xmath299 is thickened in the diagram . the generators are the arcs in @xmath300 , as shown in figure [ unknot ] . the fundamental group then is : @xmath306 where the last four relations come from the cycles . this gives us @xmath307 . the manifold is the complement of the unknot in @xmath177 .

the _ gromov product _ with reference point @xmath0 of a metric space @xmath1 is defined by @xmath2 we call a sequence @xmath3 in @xmath4 _ convergent at infinity _ if @xmath5 as @xmath6 ( cf . 8 in @xcite ) . the central aim of this paper is to develop a theory for understanding the large scale geometry on metric spaces and the action of mappings or transformations on metric spaces using convergent sequences at infinity , with reference to the theory of gromov hyperbolic spaces . in fact , we will focus on mappings between metric spaces which send sequences keeping the condition of the convergence at infinity . we will call such mappings _ asymptotically conservative with the gromov product _ ( or _ asymptotically conservative _ , for short ) . see [ subsec : asymptotically_conservative ] for details . _ gromov hyperbolic spaces _ were introduced and developed by m. gromov in his very important paper @xcite . gromov hyperbolic spaces form a large and much studied class of metric spaces , which satisfy certain characteristic properties of hyperbolic geometry . indeed , they include all complete simply connected riemannian manifolds whose sectional curvature is everywhere less than a negative constant . convergent sequences at infinity appear naturally in studying the large scale geometry of gromov hyperbolic spaces . indeed , every gromov hyperbolic space @xmath4 admits a natural ideal boundary @xmath7 , called the _ gromov boundary _ of @xmath4 , and any convergent sequence at infinity in @xmath4 determines a boundary point . the gromov product admits a canonical extension to the gromov boundary , and the extension defines a distance on the gromov boundary ( 1.8b in @xcite . see also 2 in @xcite ) . in a gromov hyperbolic space , if two sequences @xmath8 and @xmath9 are convergent together at infinity in the sense that @xmath10 as @xmath6 , they determine the same ideal boundary point . from other viewpoints , this means that we are unable to distinguish between their asymptotic behaviors in a gromov hyperbolic space . we will call two sequences in an ( arbitrary ) metric space satisfying _ visually indistinguishable_. one simple but remarkable fact is that in a gromov hyperbolic space , the relation visually indistinguishable " defines an equivalence relation on the set of convergent sequences at infinity . indeed , this fact allows us to define the ideal boundary ( gromov boundary ) for every gromov hyperbolic space . however , on an arbitrary metric space , the relation is not an equivalent relation . the _ teichmller distance _ is a natural distance on the teichmller space . in @xcite , s. kerckhoff discovered an elegant geometric interpretation of the teichmller distance via extremal length . indeed , kerckhoff s formula asserts that the teichmller distance is represented by the ratio of extremal lengths of simple closed curves on marked riemann surfaces . following his observation , we often call the geometry on the teichmller distance the _ extremal length geometry on teichmller space_. in @xcite , h. masur and m. wolf observed that teichmller space is not gromov hyperbolic when the base surface is neither a torus with one hole nor a sphere with four holes . in fact , the relation visually indistinguishable " discussed in the previous section does not define an equivalent relation on the set of convergent sequences at infinity in teichmller space . in addition , masur @xcite showed that teichmller space is not negatively curved in the sense of busemann , and y. minsky @xcite observed an interesting phenomenon which implies that teichmller space behaves like a positively curved space in the thin part . thus , it seems to be hard to reap benefit directly from the general theory of negatively curved spaces . on the other hand , the gromov product on teichmller space connects between the analytic aspect and the topological aspect at infinity of teichmller space described as follows ( see also [ subsec : cones_intersectionnumber ] ) . consider two quasiconformal mappings @xmath11 and @xmath12 . it is known that @xmath13 where @xmath14 is the maximal dilatation of @xmath15 . the equality @xmath16 does not hold in general . for instance @xmath17 for a quasiconformal mapping with @xmath18 . in the theory of quasiconformal mappings , the following question is fundamental ( but obscure ) : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ how can we measure the _ insufficiency _ @xmath19 of the composition from @xmath15 and @xmath20 ? _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ for a quasiconformal mapping @xmath21 between riemann surfaces , we denote by @xmath22 the infimum of the maximal dilatations of quasiconformal mappings from @xmath23 to @xmath24 homotopic to @xmath15 . the teichmller distance between marked riemann surfaces @xmath25 and @xmath26 is originally defined as @xmath27 ( cf . [ subsec : teichmuller - space ] ) . hence , the gromov product of @xmath28 and @xmath29 with base point @xmath30 satisfies @xmath31 thus the gromov product measures the insufficiency " of the maximal dilatation arising from the difference of markings . in @xcite , zhong li and yi qi give another treatment of the insufficiency or the gromov product in the analytic aspect . the gromov product of the teichmller distance links between the extremal length geometry and the thurston theory on the geometry of simple closed curves with the intersection number . for two points @xmath28 and @xmath29 in teichmller space , the quantity @xmath32 is recognized as the ( weighted ) _ intersection number _ between @xmath28 and @xmath29 with the base point @xmath0 ( cf . [ subsec : intersection_number_with_base_point ] ) . namely , the pairing extends continuously to a canonical boundary in the extremal length geometry on teichmller space , which is called the _ gardiner - masur boundary _ , and the extension coincides with the weighted intersection number function on the boundary where includes the space of projective measured foliations . in addition , this observation implies that teichmller space has a boundary continuous " property in terms of the gromov product ( cf [ subsec : boundary_continuous ] ) . as we discussed above , the gromov product of the teichmller distance gives a strong connection between * analytic aspect : theory on quasiconformal mappings ( insufficiency of the maximal dilation of the composition ) , and * topological aspect : thurston theory ( geometry on simple closed curves with the intersection number ) . indeed , the above insufficiency coincides with the weight intersection number at infinity of teichmller space . from the seemingly useful property of the gromov product on teichmller space , one may ask : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ what can we know about the large scale geometry of teichmller space using the gromov product ? _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ however , because of masur - wolf s result discussed above , the properties of the gromov product in terms of the large scale geometry of teichmller space are less known . thus , from the general point of view , we need to develop a theory on arbitrary metric spaces by specializing the asymptotic behavior on convergent sequences at infinity . let @xmath4 and @xmath33 be metric spaces . a mapping @xmath34 is said to be _ asymptotically conservative with the gromov product _ ( _ asymptotically conservative _ for short ) if @xmath35 maps every pair of visually indistinguishable sequences in @xmath4 to a pair of such sequences , vice versa ( cf . definition [ def : asymptotically_conservative ] ) . asymptotically conservative mappings are defined in an effort to produce a qualitative version of quasi - isometries with respect to the asymptotic behavior of visually indistinguishable sequences under quasi - isometries between gromov hyperbolic spaces . indeed , any quasi - isometry between gromov hyperbolic spaces is asymptotically conservative ( cf . theorem 7.2h in @xcite ) . we will see in [ sec : asymptotic_conservative ] that the composition of two asymptotically conservative mappings is also asymptotically conservative , and asymptotically conservative mappings are stable under paralleism , like as quasi - isometries between metric spaces . in particular , any mapping parallel to an isometry is asymptotically conservative ( see [ subsec : remarks_asymptotically_conservative ] ) . let @xmath36 be a convergent sequence at infinity . we denote by @xmath37 the set of convergent sequences at infinity which are visually indistinguishable from @xmath36 . though the relation visually indistinguishable " is not an equivalent relation on the set of convergent sequences on any metric space , if two convergent sequences @xmath38 and @xmath39 at infinity satisfy @xmath40 , we ( pretend to ) recognize that two sequences @xmath38 and @xmath39 determine the same ideal point at infinity . two asymptotically conservative mappings @xmath41 and @xmath42 are said to be _ close at infinity _ , if for any sequences @xmath43 and @xmath44 , @xmath45 implies @xmath46 . namely , if @xmath41 and @xmath42 are close at infinity , we can not find any difference between @xmath41 and @xmath42 in the large scale geometry in terms of the asymptotic behavior of convergent sequences at infinity . a mapping @xmath34 is said to be _ invertible _ if there is a mapping @xmath47 such that @xmath48 and @xmath49 are close to the identity mappings on @xmath4 and @xmath33 , respectively . we call such @xmath50 an _ asymptotic quasi - inverse _ of @xmath35 . a quasi - inverse in the coarse geometry is a typical example of asymptotic quasi - inverses ( cf . remark ( 5 ) of [ subsec : remarks_asymptotically_conservative ] ) . let @xmath51 be the set of invertible asymptotically conservative mappings on @xmath4 to itself . we first observe the following theorem ( cf . theorems [ thm : properties_quasimorphisms_identity ] and [ thm : quotient_semigroup ] ) . [ thm : arbitrary_quotient_space ] let @xmath4 be a metric space . the set @xmath51 admits a canonical monoid structure with respect to the composition of mappings . furthermore , the relation closeness at infinity " is a semigroup congruence on @xmath51 and the quotient semigroup @xmath52 is a group . as discussed in the previous section , our mappings are defined after on the model of quasi - isometries between gromov hyperbolic spaces . hence , we may presume that the group @xmath52 has a good property in a sense when @xmath4 is a gromov hyperbolic space . in fact , we shall see the following ( for details , see [ subsec : conclusion_characterization ] ) . [ thm : gromov_hyperbolic ] let @xmath4 be a gromov hyperbolic space . then , the group @xmath52 is canonically identified with the group of bijective mappings on @xmath7 such that they and their inverses are extensions of mappings of @xmath4 . in addition , if @xmath4 is boundary continuous , above bijective mappings of @xmath7 become homeomorphisms . [ [ large - scale - geometry - of - teichmller - space ] ] * large scale geometry of teichmller space * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as mentioned in [ subsec : background_motivation ] , our main interest is to clarify the large scale geometry of teichmller space with the asymptotic behaviors of convergent sequences at infinity . the extened mapping class group @xmath53 of @xmath54 acts isometrically on @xmath55 and we have a group homomorphism @xmath56 since any isometry is an invertible asymptotically conservative mapping , the inclusion @xmath57 is a monoid homomorphism . for details , see [ subsec : action_of_extended_mapping_class_groups ] . in the case of teichmller space , we will prove the following rigidity theorem ( cf . theorem [ thm : induced_automorphism ] ) . [ thm : teichmullerspace ] let @xmath54 be a compact orientable surface of negative euler characteristic such that @xmath54 is neither a torus with one hole nor a sphere with four holes . let @xmath55 be the teichmller space of @xmath54 , endowed with the teichmller distance . let @xmath58 be the complex of curves on @xmath54 . then , there is a monoid epimorphism @xmath59 which descends to an isomorphism @xmath60 satisfying the following commutative diagram @xmath61^{\mathcal{i}_0 } & { \rm isom}({\mathcal{t } } ) \ar[r]^{\mathcal{i } } \ar[dr]^{\mbox{{\tiny group iso } } } & { { \rm ac_{inv}}}({\mathcal{t } } ) \ar[d]^{\mbox{{\tiny proj } } } \ar[dr]^{\xi } & \\ & & { \mathfrak{ac}}({\mathcal{t } } ) \ar[r]^{\mbox{{\tiny group iso } } } & { \rm aut}(\mathbb{x}(s ) ) , & } \ ] ] where @xmath62 is the group of simplicial automorphisms of @xmath58 . theorem [ thm : teichmullerspace ] implies that any invertible asymptotically conservative mapping on @xmath55 is close to an isometry on @xmath55 at infinity . thus , invertible asymptotically conservative mappings are considered as _ coarsifications _ " of isometries which are rigid at infinity in the teichmller theory . when @xmath54 is either a torus with one hole or a sphere with four holes , the teichmller space of @xmath54 is isomorphic to the poincar disk and the characterization for this case follows from theorem [ thm : gromov_hyperbolic ] . by applying the discussion in the proof of theorem [ thm : teichmullerspace ] , we also obtain a hyperbolic characteristic of teichmller space . in fact , we will give a proof of the following folklore result in [ subsec : rough_homothety ] . [ thm : teichmullerspace_homothety ] let @xmath54 be a compact orientable surface of negative euler characteristic . then , there is no @xmath63-rough homothety with asymptotic quasi - inverse on the teichmller space of @xmath54 unless @xmath64 . here , a mapping @xmath65 between metric spaces is said to be a @xmath63-_rough homothety _ if @xmath66 for @xmath67 ( cf . chapter 7 of @xcite ) . one can easily see that any rough - homothety is asymptotically conservative . by definition , any isometry is a @xmath68-rough homothety for all @xmath69 . hence , teichmller space admits @xmath68-rough homotheties for all @xmath69 . theorem [ thm : teichmullerspace_homothety ] implies that there is no non - trivial similarity in teichmller space , like in hyperbolic spaces . though enormous hyperbolic characteristics have been observed in teichmller space , we would like to notice a remarkable fact proven by athreya , bufetov , eskin and mirzakhani in @xcite . indeed , they observed that the volume of the metric balls in teichmller space has exponential growth . thus , their result might imply that a measurable @xmath63-homothety on teichmller space does not exist unless @xmath70 . this paper is organized as follows : in [ sec : asymptotic_conservative ] , we will introduce asymptotically conservative mappings on metric spaces . we first start with the basics for the gromov product , and we next develop the properties of asymptotically conservative mappings . we will prove theorem [ thm : arbitrary_quotient_space ] in [ subsec : monoild_semingroups ] . in [ subsec : remarks_asymptotically_conservative ] , we will give typical examples of asymptotically conservative mappings . in [ sec : gromov_hyperbolic_space ] , we give characterizations of asymptotically conservative mappings on gromov hyperbolic spaces and will prove theorem [ thm : gromov_hyperbolic ] in [ subsec : conclusion_characterization ] . from [ sec : teichmuller_theory ] to [ sec : action_reduced_boundary ] , we devote to prepare for the proofs of theorems [ thm : teichmullerspace ] and [ thm : teichmullerspace_homothety ] . in [ sec : teichmuller_theory ] , we give basic notions of teichmller theory including the definitions of teichmller space , measured foliations and extremal length . in [ sec : thurston_theory_extremal_length ] , we recall our unification theorem for extremal length geometry on teichmller space via intersection number . this unification connects between the geometry on the gromov product on teichmller space and that of the intersection number on the space of measured foliations . we will define cones , the stage where we develop the thurston theory with extremal length , from the garidner - masur compactification . indeed , we can define the intersection number function on the cones . we also give basic properties of the cones and the intersection number . in [ sec : null_sets_section ] , we shall define the null sets for points in the cone , and show that any point of the cone has the same null set with a measured foliations . we will also give a characterization of such measured foliations for points in the cone . in [ sec : action_reduced_boundary ] , we define the reduced gardner - masur compactification and study the action of asymptotically conservative mappings on the reduced gardiner - masur compactification . in [ sec : rigidity_asymptotic_teichmuller ] , we will prove theorems [ thm : teichmullerspace ] and [ thm : teichmullerspace_homothety ] . in the proof of theorem [ thm : teichmullerspace_homothety ] , we mainly treat the case where the teichmller space of @xmath54 has dimension at least @xmath71 , but the proof is modeled on the proof for the case of one dimensional teichmller space . we will also sketch the proof of the one dimensional case . in the last section ( [ sec : appendix ] ) , as an appendix , we shall give an estimation of extremal length . indeed , our result is a refinement of the estimation given by gardiner and masur in @xcite . the author would like to express his hearty gratitude to professor kenichi ohshika for fruitful discussions and his constant encouragements . he also thanks professor yair minsky for informing him about a work by athreya , bufetov , eskin and mirzakhani in @xcite . the author also thanks professor athanase papadopoulos for his kindness and useful comments . in this section , we introduce asymptotically conservative mappings with the gromov product between metric spaces . for completeness , we redefine them . let @xmath1 be a metric space . in what follows , we assume that all metric space is of infinite diameter . as given in the introduction , the _ gromov product _ on @xmath4 is , by definition , @xmath72 for @xmath73 . we call the point @xmath74 in the _ reference point _ of the gromov product ( cf . 1 in @xcite ) . when the metric space in the discussion is clear in the context , we omit to specify the metric space in denoting the gromov product . the following is known for @xmath28 , @xmath29 , @xmath75 , @xmath76 , @xmath77 : @xmath78 a sequence @xmath3 in @xmath4 is said to be _ convergent at infinity _ if the sequence satisfies @xmath79 ( @xmath6 ) for some ( and hence any ) point @xmath80 . in this section , unless otherwise specified , we always assume that every sequence converges at infinity . [ def : visually_indistinguishable ] a sequence @xmath81 in @xmath4 is said to be * visually indistinguishable * , if @xmath4 is gromov hyperbolic , the relation visually indistinguishable " is an equivalent relation on the set of convergent sequences at infinity ( cf . 2.2 in @xcite ) . however , the relation visually indistinguishable " is not an equivalent relation in general . hence , we do not use here the word equivalence " which is commonly used in the study of hyperbolic spaces . ] * at infinity ( visually indistinguishable , for short ) * from a sequence @xmath82 in @xmath4 if @xmath83 from , the above definition is independent of the choice of the reference points @xmath0 of the gromov product on @xmath4 . therefore , in what follows , we may fix reference points of metric spaces for the gromov product , for instance , @xmath80 and @xmath84 etc . for a sequence @xmath85 in @xmath4 , we define @xmath86 for a map @xmath34 and a sequence @xmath85 , we write @xmath87 as usual . [ def : asymptotic ] two sequences @xmath43 and @xmath44 in @xmath4 are said to be * asymptotic * if @xmath45 . [ remark : accompanying_sequence ] the following hold : 1 . the relation visually indistinguishable " is reflexive on the set of convergent sequences at infinity . indeed , @xmath36 converges at infinity if and only if @xmath36 is visually indistinguishable from @xmath36 itself . the relation visually indistinguishable " is symmetric : if @xmath88 , then @xmath89 . the relation visually indistinguishable " is not transitive in general . namely , it is possible that @xmath90 for some sequences @xmath91 with @xmath92 . 4 . for any @xmath88 , any subsequence of @xmath93 is also in @xmath37 . 5 . let @xmath36 and @xmath93 be convergent sequences at infinity . if any subsequence of @xmath93 contains a subsequence which is visually indistinguishable from @xmath36 , then @xmath93 itself is contained in @xmath37 . indeed , we can see that on teichmller space , the relation visually indistinguishable " is not an equivalent relation when the base surface is neither a torus with one hole nor a sphere with four holes ( cf . [ sec : remark_visually_indistinguishable ] ) . we shall give a brief check of the assertion ( 5 ) above . suppose that @xmath94 . then , @xmath95 for some @xmath96 where @xmath85 and @xmath97 . then , we can take @xmath98 and @xmath99 such that @xmath100 for @xmath101 . hence , any subsequence of a subsequence @xmath102 of @xmath93 is not visually indistinguishable from @xmath36 . let @xmath1 and @xmath103 be metric spaces . [ def : asymptotically_conservative ] a mapping @xmath34 is said to be * asymptotically conservative with the gromov product ( asymptotically conservative , for short ) * if for any two sequences @xmath43 and @xmath44 in @xmath4 , @xmath43 is visually indistinguishable from @xmath44 in @xmath4 if and only if @xmath104 is also visually indistinguishable from @xmath105 in @xmath33 . namely , for any sequence @xmath36 in @xmath4 , the following two conditions hold ; 1 . @xmath106 ; 2 . for any sequence @xmath93 in @xmath4 , if @xmath107 , then @xmath88 . let @xmath108 be the set of asymptotically conservative mappings from @xmath4 to @xmath33 . set @xmath109 for simplicity . one can easily see that any isometry is asymptotically conservative ( see also [ subsec : remarks_asymptotically_conservative ] ) . [ remark:1 ] suppose @xmath34 is asymptotically conservative . if a sequence @xmath36 in @xmath4 converges at infinity , then @xmath36 is visually indistinguishable from @xmath36 itself . hence , @xmath110 is visually indistinguishable from @xmath110 itself , which means that @xmath110 also converges at infinity in @xmath33 . the converse is also true . namely , if @xmath110 converges at infinity , so does @xmath36 . [ prop : properties_quasimorphisms ] let @xmath4 , @xmath33 and @xmath111 be metric spaces . for @xmath112 and @xmath113 , @xmath114 . let @xmath36 be a sequence in @xmath4 . then , @xmath115 let @xmath93 be a sequence in @xmath4 with @xmath116 . since @xmath41 is asymptotically conservative , @xmath117 . since @xmath42 is also asymptotically conservative again , we have @xmath88 . a mapping @xmath34 is said to be _ asymptotically - surjective _ if for any @xmath118 in @xmath33 convergent sequence at infinity , there is a sequence @xmath36 in @xmath4 with @xmath119 . let @xmath120 be the set of asymptotically conservative and asymptotically surjective mappings from @xmath4 to @xmath33 . [ def : closeness_at_infinity ] two mappings @xmath41 and @xmath42 from @xmath4 to @xmath33 are said to be * close at infinity * if and only if for any sequences @xmath43 and @xmath44 in @xmath4 , if @xmath43 and @xmath44 are asymptotic , so are @xmath121 and @xmath122 . namely , if @xmath45 , then @xmath46 . in particular , if two mappings @xmath41 and @xmath42 from @xmath4 to @xmath33 are close at infinity , @xmath123 for all @xmath36 in @xmath4 convergent sequence at infinity . [ prop : properties_quasimorphisms_close ] let @xmath4 , @xmath33 and @xmath111 be metric spaces . let @xmath41 , @xmath124 and @xmath42 , @xmath125 be mappings . suppose @xmath126 and @xmath127 are close at infinity for @xmath128 . then , @xmath129 is close to @xmath130 at infinity . let @xmath43 and @xmath44 be sequences in @xmath4 with @xmath45 . by definition , @xmath131 , and hence @xmath132 . [ prop : quasi - surjective_asymptotic ] let @xmath133 . let @xmath43 and @xmath44 be convergent sequences in @xmath4 at infinity . if @xmath134 , then @xmath135 . let @xmath136 . since @xmath35 is asymptotically surjective , there is a sequence @xmath93 in @xmath4 such that @xmath137 . since @xmath35 is asymptotically conservative and @xmath138 , we have @xmath139 and @xmath140 therefore , @xmath141 ( cf . ( 2 ) of remark [ remark : accompanying_sequence ] ) . hence we deduce @xmath142 and @xmath143 . [ prop : gp_close_equivalence_relation ] let @xmath4 and @xmath33 be metric spaces . the relation closeness at infinity " is an equivalent relation on @xmath120 . let @xmath43 and @xmath44 be sequences in @xmath4 . suppose that @xmath43 and @xmath44 are asymptotic , which means that @xmath45 . ( reflexive law)this follows from proposition [ prop : quasi - surjective_asymptotic ] . ( symmetric law)take two mappings @xmath144 . since @xmath41 is close to @xmath42 at infinity , @xmath46 . by interchanging the roles of @xmath43 and @xmath44 , @xmath145 . this means that @xmath42 is close to @xmath41 at infinity . ( transitive law)take three mappings @xmath146 . suppose that @xmath126 is close to @xmath147 at infinity ( @xmath128 ) . then , from , @xmath148 and hence , @xmath41 is close to @xmath149 at infinity . a mapping @xmath150 is said to be _ invertible _ if there is @xmath151 such that @xmath48 and @xmath152 are close to the identity mappings on @xmath4 and @xmath33 respectively . we call such @xmath50 an _ asymptotic quasi - inverse _ of @xmath35 . let @xmath153 be the set of invertible asymptotically conservative mappings . set @xmath154 . notice that for @xmath155 , any asymptotic quasi - inverse @xmath50 of @xmath35 is in @xmath156 , and @xmath35 is an asymptotic quasi - inverse of @xmath50 . [ prop : invertible_means_surjectivity ] let @xmath4 and @xmath33 be metric spaces . then , @xmath157 . let @xmath155 . let @xmath50 be an asymptotic quasi - inverse of @xmath35 . let @xmath118 be a sequence in @xmath33 . set @xmath158 . since @xmath152 is close to the identity mapping on @xmath33 at infinity , @xmath159 which implies @xmath133 . [ prop : composition ] for @xmath160 and @xmath161 , we have @xmath162 . let @xmath127 be an asymptotic quasi - inverse of @xmath126 for @xmath128 . let @xmath163 and @xmath164 be sequences in @xmath111 . suppose that @xmath165 . let @xmath166 for @xmath128 . since @xmath167 is close to the identity mapping on @xmath33 , @xmath168 since @xmath169 is asymptotically surjective , @xmath170 from proposition [ prop : quasi - surjective_asymptotic ] . therefore , @xmath171 and @xmath172 are asymptotic . since @xmath41 is asymptotically surjective again , we have @xmath173 since @xmath174 is asymptotically close to the identity mapping on @xmath111 at infinity . therefore , @xmath175 which means that @xmath176 is close to the the identity mapping on @xmath111 at infinity . by the same argument , we can see that @xmath177 is close to the identity mapping on @xmath4 . therefore , @xmath178 is an asymptotic quasi - inverse of @xmath129 and @xmath162 . [ prop : relation_closeness_at_infinity ] let @xmath144 . suppose that @xmath41 and @xmath42 are close at infinity . if @xmath179 , so is @xmath42 . in particular , in this case , any asymptotic quasi - inverse of @xmath41 is one of @xmath42 . let @xmath169 be an asymptotic quasi - inverse of @xmath41 . let @xmath43 and @xmath44 be sequences in @xmath4 . suppose that @xmath43 and @xmath44 are asymptotic . since @xmath41 and @xmath42 are close at infinity , @xmath180 since @xmath42 is asymptotically surjective , by proposition [ prop : quasi - surjective_asymptotic ] , we have @xmath181 let @xmath182 and @xmath183 be sequences in @xmath33 . suppose that @xmath182 and @xmath183 are asymptotic . since @xmath169 is asymptotically surjective again , @xmath184 since @xmath41 and @xmath42 are close at infinity , we deduce @xmath185 from and , we conclude that @xmath169 is an asymptotic quasi - inverse of @xmath42 , and @xmath161 . the following theorem follows from propositions [ prop : properties_quasimorphisms ] and [ prop : composition ] . [ thm : properties_quasimorphisms_identity ] @xmath186 admits a canonical monoid structure with respect to the composition of mappings . the identity element of @xmath186 is the identity mapping on @xmath4 . in addition , @xmath51 is a submonoid of @xmath186 . let @xmath187 be a semigroup . semigroup congruence _ is an equivalent relation @xmath188 on @xmath187 with the property that for @xmath189 , @xmath190 and @xmath191 imply @xmath192 . then , the congruence classes @xmath193\mid g\in g\}\ ] ] is also a semigroup with the product @xmath194[g_2]=[g_1g_2]$ ] . we call @xmath195 the _ quotient semigroup _ of @xmath187 with the semigroup congruence @xmath188 . we define a relation on @xmath51 by using closeness at infinity . namely , for two @xmath41 and @xmath196 , @xmath41 is equivalent to @xmath42 if @xmath41 is close to @xmath42 at infinity . from propositions [ prop : gp_close_equivalence_relation ] and [ prop : invertible_means_surjectivity ] , this relation is an equivalent relation on @xmath51 . from proposition [ prop : properties_quasimorphisms_close ] , this relation is a subgroup congruence on @xmath51 . we denote the quotient semigroup by @xmath52 . from the definition of asymptotic quasi - inverses and proposition [ prop : relation_closeness_at_infinity ] , we have the following . [ thm : quotient_semigroup ] let @xmath4 be a metric space . then , the quotient semigroup @xmath52 is a group . the identity element of @xmath52 is the congruence class of the identity mapping , and the inverse of the congruence class @xmath197 $ ] of @xmath198 is the congruence class of an asymptotic quasi - inverse of @xmath35 . let @xmath4 and @xmath33 be metric spaces . consider @xmath150 and @xmath151 . if @xmath50 is an asymptotic quasi - inverse of @xmath35 , the mapping @xmath199 induces an isomorphism @xmath200\mapsto [ \omega\circ f\circ \omega ' ] \in { \mathfrak{ac}}(y).\ ] ] ( 1)an asymptotically conservative mapping need not be a quasi - isometry . indeed , let @xmath201 with @xmath202 . let @xmath203 be the reference point . then , any increasing function @xmath204 with @xmath205 is asymptotically conservative . furthermore , as we will see in theorem [ thm : gromov_hyperbolic_ac ] in [ subsec : asymptotically_conservative_hyperbolic ] , when @xmath4 and @xmath33 are gromov hyperbolic , if a mapping @xmath34 is extendable to their gromov boundaries and the extension is injective , @xmath35 is asymptotically conservative . hence , one can easily find an asymptotically conservative mapping which is not a quasi - isometry . ( 2)meanwhile , little is known as to when quasi - isometries become asymptotically conservative . for instance , any rough homothety is asymptotically conservative ( cf . ) . actually , it follows from the following fact that any rough homothety @xmath35 satisfies @xmath206 for some @xmath207 . ( 3)asymptotically conservative mappings are stable with respect to _ paralleism _ , where two mappings @xmath208 between metric spaces are said to be _ parallel _ if and only if @xmath209 ( cf . 1.a in @xcite ) . if a mapping @xmath34 is parallel to an asymptotically conservative mapping @xmath210 , then @xmath35 is also asymptotically conservative . indeed , let @xmath211 . let @xmath212 ( @xmath128 ) be sequences in @xmath4 with @xmath213 . since @xmath214 as @xmath6 , we have @xmath215 and hence @xmath216 ( cf . ) . suppose that @xmath85 and @xmath217 are sequences with @xmath107 . then , @xmath218 and @xmath219 . since @xmath220 is asymptotically conservative , @xmath88 . ( 4)a typical example of asymptotically surjective mappings is a quasi - surjective mapping . here , a mapping @xmath34 from between metric spaces is said to be _ quasi - surjective _ if there is a constant @xmath221 such that for any @xmath222 , there is an @xmath223 with @xmath224 . indeed , let @xmath225 be a convergent sequence in @xmath33 at infinity . take @xmath226 such that @xmath227 . then , for @xmath228 , @xmath229 as @xmath6 and @xmath230 ( cf . ) . therefore , we obtain @xmath231 . by applying the same argument , we get @xmath232 . thus , @xmath35 is asymptotically surjective . ( 5)similarly , a typical example of asymptotic quasi - inverses is a quasi - inverse . here , for a mapping @xmath34 , a mapping @xmath47 is said to be a _ quasi - inverse _ ( in the coarse geometry ) if there is a constant @xmath233 such that @xmath234 indeed , let @xmath43 and @xmath44 be sequences with @xmath45 . let @xmath235 ( @xmath236 ) . then , @xmath237 as @xmath6 , where @xmath238 ( cf . ) . hence , @xmath239 . by applying the same argument , we deduce that @xmath240 . thus , @xmath48 is close to the identity mapping on @xmath4 . similarly , we can see that @xmath152 is close to the identity mapping on @xmath33 . hence , @xmath50 is an asymptotic quasi - inverse of @xmath35 . gromov hyperbolic spaces are an important class of metric spaces , and they are applied in various fields in mathematics . in this section , we shall study the asymptotic geometry of asymptotically conservative mappings on gromov hyperbolic spaces , and give a characterization of the mappings in terms of boundary values . let @xmath241 be a metric space . fix a reference point @xmath80 . we say that @xmath4 is _ gromov @xmath242-hyperbolic _ if @xmath243 for any @xmath244 . if a metric space is gromov @xmath242-hyperbolic for some @xmath245 , we call the metric space _ gromov hyperbolic_. two convergent sequences at infinity are said to be _ equivalent _ if they are visually indistinguishable from each other ( cf . chapter 4 of @xcite and 3.4 of @xcite ) . from , when @xmath4 is gromov hyperbolic , the relation visually indistinguishable " is an equivalent relation on the set of convergent sequences at infinity ( cf . 2.2 in @xcite ) . in other words , for any convergent sequences @xmath43 and @xmath44 at infinity , if @xmath246 , then @xmath45 . namely , for convergent sequences @xmath43 and @xmath44 at infinity , the following are equivalent : 1 . @xmath43 and @xmath44 are equivalent ; 2 . @xmath45 ; 3 . @xmath246 . the set @xmath7 of equivalence classes of convergent sequences at infinity is called the _ gromov boundary _ of @xmath4 . let @xmath4 and @xmath33 be gromov hyperbolic metric spaces . a mapping @xmath34 is said to be _ extendable _ if there is a mapping @xmath247 such that for any @xmath248 and any sequence @xmath36 in @xmath4 representing @xmath249 , @xmath110 converges at infinity and represents @xmath250 . we call @xmath251 the _ extension _ of @xmath35 . [ thm : gromov_hyperbolic_ac ] let @xmath4 and @xmath33 be gromov hyperbolic spaces . let @xmath34 be a mapping . then , @xmath35 is in @xmath108 if and only if @xmath35 is extendable and the extension is injective . suppose @xmath150 . let @xmath248 . let @xmath85 be a representative of @xmath249 . from remark [ remark:1 ] , @xmath110 also converges at infinity . we define @xmath247 by assigning the equivalence class @xmath249 of @xmath36 to the equivalence class @xmath250 of @xmath110 . since @xmath35 is asymptotically conservative , the equivalence class @xmath250 is determined independently of the choice of sequences @xmath36 . hence , the extension @xmath251 is well - defined . let @xmath252 with @xmath253 . for @xmath128 , let @xmath254 be a representative of @xmath255 . since @xmath256 , we have @xmath257 because @xmath35 is asymptotically conservative again . since @xmath4 is gromov hyperbolic , @xmath258 and @xmath259 . this implies that the extension @xmath251 is injective . suppose that @xmath35 admits an injective extension @xmath251 to @xmath7 . let @xmath248 and @xmath36 be a representative of @xmath249 . since @xmath35 is extendable , @xmath110 converges at infinity and represents @xmath250 . let @xmath88 . from the definition of the gromov boundary , @xmath93 also represents @xmath249 . therefore , by the definition of extensions , we have @xmath260 , and @xmath261 since @xmath262 and @xmath110 represent the same boundary point in @xmath263 . therefore , @xmath264 let @xmath93 be a convergent sequence at infinity . let @xmath265 be the equivalence class of @xmath93 . suppose @xmath107 . this means that @xmath266 and @xmath267 . since @xmath251 is injective , we have @xmath268 and @xmath269 hence , @xmath35 is asymptotically conservative . [ prop : gromov_hyperbolic_asymptotic_surjective ] let @xmath34 be a mapping . then , @xmath133 if and only if @xmath35 is extendable and the extension is bijective . suppose @xmath133 . from theorem [ thm : gromov_hyperbolic_ac ] , @xmath35 admits an injective extension @xmath270 . let @xmath271 and @xmath272 be a sequence which represents @xmath273 . since @xmath35 is asymptotically surjective , there is a sequence @xmath93 in @xmath4 such that @xmath274 . let @xmath249 be the equivalence class of @xmath93 . then , the condition @xmath274 implies that @xmath275 . hence @xmath251 is surjective . suppose @xmath35 admits a bijective extension @xmath247 . from theorem [ thm : gromov_hyperbolic_ac ] again , @xmath35 is asymptotically conservative . let @xmath272 be a convergent sequence in @xmath33 at infinity . let @xmath273 be the equivalence class of @xmath272 . since @xmath251 is surjective , there is @xmath248 such that @xmath275 . let @xmath93 be a sequence representing @xmath249 . the condition @xmath275 implies that @xmath274 . hence , @xmath35 is asymptotically surjective . [ prop : closeness_on_hyperbolic_space ] let @xmath4 and @xmath33 be gromov hyperbolic spaces . for @xmath276 , @xmath41 is close to @xmath42 at infinity if and only if their extensions coincide . suppose that @xmath41 and @xmath42 are close at infinity . for @xmath128 , let @xmath277 be the extension of @xmath126 . let @xmath248 and @xmath36 a sequence representing @xmath249 . since @xmath41 and @xmath42 are close at infinity , from , @xmath278 . this means that @xmath279 . conversely , suppose that @xmath280 on @xmath7 . let @xmath43 and @xmath44 be sequences with @xmath45 . this means that they represent the same boundary point , say @xmath248 . from the assumption @xmath279 . since @xmath281 is a representative of @xmath282 for @xmath128 , we deduce that @xmath121 is equivalent to @xmath122 , and hence , @xmath283 since @xmath33 is gromov hyperbolic . thus , we conclude that @xmath41 is close to @xmath42 at infinity . let @xmath4 be a gromov hyperbolic space . let @xmath284 . we define the gromov product between @xmath249 and @xmath273 by @xmath285 where the infimum is taken over all sequences @xmath286 and @xmath287 . notice that the gromov product takes values in @xmath288 $ ] , that @xmath289 if and only if @xmath268 . we also define the gromov product between @xmath223 and @xmath248 by @xmath290 where the infimum is taken over all sequences @xmath286 . one can show that when @xmath4 is gromov @xmath242-hyperbolic , @xmath291 for all @xmath292 ( cf . 2.2.1 in @xcite ) . the gromov boundary admits a _ visual metric _ @xmath293 with the following property : for any @xmath80 , there is @xmath294 such that for @xmath295 , there exist a distance @xmath293 on @xmath7 and positive constants @xmath296 such that @xmath297 for @xmath284 ( cf . theorem 2.2.7 in @xcite ) . we call a gromov hyperbolic space @xmath4 _ boundary continuous _ if the gromov product is continuous on @xmath298 in the sense that for any @xmath299 and sequences @xmath286 and @xmath287 in @xmath4 , @xmath300 as @xmath301 for every @xmath80 ( cf . 3.4.2 in @xcite ) . it is known that every proper @xmath302-space is a boundary continuous gromov hyperbolic space ( cf . proposition 3.4.2 in @xcite ) . [ prop : ac_boundarycontinuous ] let @xmath4 and @xmath33 be boundary continuous gromov hyperbolic spaces . let @xmath34 be a mapping . then , @xmath150 if and only if @xmath35 is extendable and the extension @xmath251 is a continuous injection . suppose @xmath4 and @xmath33 are @xmath242-hyperbolic . from theorem [ thm : gromov_hyperbolic_ac ] , a mapping @xmath34 is in @xmath108 if and only if @xmath35 is extendable and the extension @xmath251 is injective . hence , it suffices to check only that the extension @xmath251 of @xmath150 is continuous . let @xmath248 . let @xmath303 be a convergent sequence to @xmath249 . let @xmath304 be a sequence representing @xmath305 . since @xmath4 and @xmath33 are boundary continuous , for any @xmath306 there is an @xmath307 such that when @xmath308 , @xmath309 let @xmath310 . from , @xmath311 since @xmath312 , by , @xmath313 therefore , from and , we see that @xmath314 represents @xmath249 . since @xmath251 is the extension of @xmath35 , @xmath315 represents @xmath250 . from , @xmath316 by , we get @xmath317 and hence @xmath318 converges to @xmath250 from . thus , we conclude that the extension @xmath251 is continuous . let @xmath319 be the set of self - mappings @xmath35 of @xmath4 with the following properties : 1 . @xmath35 is extendable and the extension @xmath251 is bijective ; 2 . the inverse @xmath251 on @xmath7 is also the extension of a mapping of @xmath4 . notice from proposition [ prop : gromov_hyperbolic_asymptotic_surjective ] that @xmath319 is canonically identified with @xmath51 . we consider the @xmath319 as a monoid . we define an equivalent relation on @xmath319 as follows : we say that two @xmath320 are equivalent if their extensions @xmath321 and @xmath322 satisfy @xmath280 on @xmath7 . from the definition , this equivalent relation is a semigroup congruence . hence , the quotient semigroup @xmath323 is a semigroup . from proposition [ prop : closeness_on_hyperbolic_space ] , we conclude the following characterization . [ thm : acgroup_hyperbolic_group ] let @xmath4 be a gromov hyperbolic space . then the group @xmath52 is canonically isomorphic to @xmath323 . namely , the identification @xmath324 induces an isomorphism @xmath325 . [ [ boundary - continuous - hyperbolic - spaces ] ] * boundary continuous hyperbolic spaces * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + suppose @xmath4 is a boundary continuous gromov hyperbolic space . let @xmath326 be the set of self - mappings @xmath35 of @xmath4 with the following properties : 1 . @xmath35 is extendable and the extension @xmath251 is a homeomorphism ; 2 . the inverse @xmath251 on @xmath7 is also the extension of a mapping of @xmath4 . notice from propositions [ prop : gromov_hyperbolic_asymptotic_surjective ] and [ prop : ac_boundarycontinuous ] that @xmath326 is canonically identified with @xmath51 . we define @xmath327 as above . [ thm : acgroup_hyperbolic_group_boundary_continuous ] let @xmath4 be a boundary continuous gromov hyperbolic space . then the group @xmath52 is canonically isomorphic to @xmath327 . namely , the identification @xmath328 induces an isomorphism @xmath329 . in this section , we recall basics in the teichmller theory . for details on the contents of this section , the reader can refer to @xcite , @xcite , @xcite and @xcite . let @xmath54 be a compact orientable surface . we denote the _ complexity _ of @xmath54 by @xmath330 the euler characteristic of @xmath54 is denoted by @xmath331 . throughout this paper , we always assume that @xmath332 . the _ teichmller space _ @xmath333 of @xmath54 is the set of equivalence classes of marked riemann surfaces @xmath334 where @xmath33 is a riemann surface of analytically finite type and @xmath335 is an orientation preserving homeomorphism . two marked riemann surfaces @xmath336 and @xmath337 are _ teichmller equivalent _ if there is a conformal mapping @xmath338 which is homotopic to @xmath339 . teichmller space @xmath55 has a canonical complete distance , called the _ teichmller distance @xmath340 _ , and defined by the equation in [ subsubsec : analytic_aspect ] . it is known that the teichmller space @xmath333 of @xmath54 is homeomorphic to @xmath341 . throughout this paper , we fix a conformal structure @xmath4 on @xmath54 and consider @xmath30 as the base point of the teichmller space @xmath55 of @xmath54 . for the details about the contents in this section , see for instance @xcite and @xcite . let @xmath342 be the set of homotopy classes of non - trivial and non - peripheral simple closed curves on @xmath54 . consider the set of weighted simple close curves @xmath343 , where @xmath344 is the formal product between @xmath345 and @xmath346 . we embed @xmath347 into the space @xmath348 of non - negative functions on @xmath342 by @xmath349\in { \mathbb{r}_{+}^{\mathcal{s}}}\ ] ] where @xmath350 is the geometric intersection number on @xmath342 . we topologize @xmath348 with the topology of pointwise convergence . the closure @xmath351 of the image of the mapping is called the _ space of measured foliations _ on @xmath54 . the space @xmath348 admits a canonical action of @xmath352 by multiplication . this action leaves @xmath351 invariant . the quotient space @xmath353 is said to be the _ space of projective measured foliations_. by definition , @xmath351 contains @xmath347 as a dense subset . the intersection number function on @xmath347 defined by @xmath354 extends continuously on @xmath355 . it is known that @xmath351 and @xmath356 are homeomorphic to @xmath341 and @xmath357 , respectively . [ [ normal - forms ] ] * normal forms * + + + + + + + + + + + + + + any @xmath358 is represented by a pair @xmath359 of a singular foliation @xmath360 and a transverse measure @xmath361 to @xmath360 . the intersection number @xmath362 , @xmath346 , is obtained as @xmath363 the _ support _ of a measured foliation is , by definition , the minimal essential subsurface containing the underlying foliation . a measured foliation is said to be _ arational _ if it intersects any non - trivial and non - peripheral simple closed curve in its support . according to the structure of the underlying foliation of @xmath187 , any @xmath358 has the _ normal form _ : any measured foliation @xmath358 is decomposed as @xmath364 where @xmath365 is an arational foliation in its support @xmath366 , @xmath367 and @xmath368 are simple closed curves such that each @xmath367 can not be deformed into any @xmath366 and @xmath368 is homotopic to a component of @xmath369 for some @xmath370 ( cf . 2.4 of @xcite ) . in this paper , we call @xmath365 , @xmath367 and @xmath368 an _ arational component _ , an _ essential curve _ , and a _ peripheral curve _ of @xmath187 respectively . [ [ null - sets ] ] * null sets * + + + + + + + + + + + for a measured foliation @xmath187 , we denote by @xmath371 the measured foliation defined from @xmath187 by deleting the foliated annuli associated to the peripheral curves in @xmath187 . the following might be well - known . however , we give a proof of the proposition for completeness : in the following proposition , we define the null set of @xmath187 by @xmath372 notice that this definition coincides with the definition of null sets in the general situation given in later ( cf . [ subset : associated_foliation ] ) . [ prop : measured_foliation_removing_peripehral ] let @xmath373 . then , the following are equivalent . * @xmath374 . * @xmath371 is topologically equivalent to @xmath375 . suppose ( 1 ) holds . we decompose @xmath187 as : @xmath376 since @xmath377 , the decomposition of @xmath378 is represented as @xmath379 where @xmath380 is either topologically equivalent to @xmath365 or is @xmath381 , @xmath382 and the support of @xmath383 is contained in the complement of the support of @xmath187 . in the summation @xmath384 in , @xmath385 rums over all component of @xmath369 . see proposition 3.2 of ivanov @xcite or lemma 3.1 of papadopoulos @xcite . indeed , ivanov in @xcite works under the assumption that each @xmath365 is a stable lamination for some pseudo - anosov mapping on @xmath366 . however , we can easily check that his discussion in the proof of the proposition can be applied to our case . if @xmath386 , there is an @xmath387 with @xmath388 but @xmath389 . since @xmath390 from the assumption , this is a contradiction . hence @xmath391 suppose @xmath392 for some @xmath370 . since @xmath393 is essential , we can find an @xmath346 such that @xmath394 . such an @xmath395 satisfies @xmath396 , which is a contradiction . with the same argument , we can see that @xmath397 . thus , @xmath398 are topologically equivalent . suppose ( 2 ) holds . let @xmath399 . consider the decomposition for @xmath400 instead of @xmath378 , one can easily deduce that @xmath401 . let @xmath4 be a riemann surface and let @xmath402 be a doubly connected domain on @xmath4 . if @xmath402 is conformally equivalent to a round annulus @xmath403 , we define the _ modulus _ of @xmath402 by @xmath404 _ extremal length _ of a simple closed curve @xmath395 on @xmath4 is defined by @xmath405 the extremal length has various definitions ( cf . @xcite and @xcite ) . definition is called the _ geometric definition_. in @xcite , kerckhoff showed that if we define the extremal length of @xmath406 by @xmath407 then the extremal length function @xmath408 on @xmath347 extends continuously to @xmath351 . for @xmath409 and @xmath358 , we define @xmath410 we define the _ unit sphere _ in @xmath351 by @xmath411 the projection @xmath412 induces a homeomorphism @xmath413 . it is known that for any @xmath358 and @xmath409 , there is a unique holomorphic quadratic differential @xmath414 such that @xmath415 namely , the _ vertical foliation _ of @xmath414 is equal to @xmath187 . we call @xmath414 the _ hubbard - masur differential _ for @xmath187 on @xmath416 ( cf . @xcite ) . the hubbard - masur differential @xmath417 for @xmath187 on @xmath418 satisfies @xmath419 in particular , it is known that @xmath420 where @xmath421 is the length of the geodesic representative homotopic to @xmath422 with respect to the singular flat metric @xmath423 . [ [ kerckhoffs - formula ] ] * kerckhoff s formula * + + + + + + + + + + + + + + + + + + + + + as we noted in [ subsec : background_motivation ] , the teichmller distance @xmath340 is expressed by extremal length , which we call _ kerckhoff s formula _ : @xmath424 ( see @xcite ) . since @xmath356 is compact , for any @xmath425 , there are @xmath426 such that @xmath427 in fact , such @xmath400 and @xmath187 are unique up to multiplying by positive constants , and are realized as the horizontal and vertical foliations of a holomorphic quadratic differential which defines a teichmller geodesic connecting @xmath428 and @xmath429 ( cf . @xcite and [ subsec : teichmuller_rays ] ) . [ [ minskys - inequality ] ] * minsky s inequality * + + + + + + + + + + + + + + + + + + + + + minsky @xcite observed the following inequality , which we recently call _ minsky s inequality _ : @xmath430 for @xmath431 and @xmath426 . minsky s inequality is sharp in the sense that for any @xmath431 and @xmath432 , there is a unique @xmath358 up to multiplication by a positive constant such that @xmath433 ( cf . @xcite ) . let @xmath434 and @xmath435\in \mathcal{pmf}$ ] . by the ahlfors - bers theorem , we can define an isometric embedding @xmath436 with respect to the teichmller distance by assigning the solution of the beltrami equation defined by the teichmller beltrami differential @xmath437 for @xmath345 . we call @xmath438 the _ teichmller ( geodesic ) ray _ associated to @xmath435\in \mathcal{pmf}$ ] . notice that the differential depends only on the projective class of @xmath187 . combining teichmller s theorem and the height theorem for holomorphic quadratic differentials ( cf . @xcite ) , we obtain the _ exponential map _ @xmath439,t ) \mapsto r_{g , t}(t)\in { \mathcal{t}}\ ] ] which is a homeomorphism ( see also @xcite ) . in this section , we recall a natural compactification of teichmller space @xmath55 in terms of extremal length geometry , which is called the _ gardiner - masur closure _ or the _ gardiner - masur compactification_. we will also recall the unification of extremal length geometry via intersection number . our unification gives a link between the gromov product on @xmath440 and intersection number on @xmath351 . the thurston theory with respect to the geometry of teichmller space ( with hyperbolic geometry ) is developed with the measured foliations and the intersection number function . the unification allows us to discuss the thurston theory of the geometry of teichmller space with extremal length . let @xmath441 for @xmath431 . we define a continuous function @xmath442 on @xmath351 by @xmath443 for @xmath432 . consider a mapping @xmath444 \in { \mathbb{r}_{+}^{\mathcal{s}}}\\ \psi_{gm}&\colon { \mathcal{t}}\ni y\mapsto [ \mathcal{s}\ni \alpha\mapsto \mathcal{e}_y(\alpha ) ] \in { \mathbb{r}_{+}^{\mathcal{s}}}.\end{aligned}\ ] ] as we noticed in [ subsec : measured_foliations ] , the space @xmath348 is topologized with pointwise convergence and it has a natural action by multiplication by positive constants . let @xmath445 be the quotient mapping of the action . in @xcite , gardiner and masur showed that the mapping @xmath446 is an embedding with compact closure . the closure @xmath447 of the image is called the _ gardiner - masur closure _ or the _ gardiner - masur compactification _ , and the complement @xmath448 is called the _ gardiner - masur boundary_. they also observed that the space @xmath356 of projective measured foliaitons is contained in @xmath449 . in @xcite and @xcite , the author observed that for any @xmath450 , there is a unique non - negative continuous function @xmath451 on @xmath351 such that * the assignment @xmath452 represents @xmath249 ; * @xmath453 ; * for any sequence @xmath454 with @xmath455 , @xmath456 converges to @xmath451 uniformly on any compact set of @xmath351 . we define @xmath457 since @xmath458 , the space of @xmath351 of measured foliations is contained in @xmath459 and @xmath460 . in @xcite , the author established the following _ unification _ of extremal length geometry via the intersection number . [ thm : main_realization ] let @xmath461 be the base point taken as above . there is a unique continuous function @xmath462 with the following properties . * @xmath463 and @xmath464 for any @xmath450 , @xmath431 and @xmath432 . in particular , the projective class of the function @xmath465 is exactly the image of @xmath431 under the gardiner - masur embedding . * for @xmath466 , @xmath467 . * for @xmath466 and @xmath468 , @xmath469 . * for any @xmath470 , @xmath471 * for @xmath472 , the value @xmath473 is equal to the geometric intersection number @xmath474 between @xmath400 and @xmath187 . we define the _ extremal length _ of @xmath475 on @xmath431 by @xmath476 ( cf . corollary 3 in @xcite ) . by applying minsky s inequality and gardiner - masur s work in @xcite , coincides with the original extremal length when @xmath477 . we can see that @xmath478 is continuous on @xmath460 and satsfies @xmath479 for @xmath470 and @xmath475 ( cf . theorem 4 and proposition 7 in @xcite ) . the extremal length also satisfies the following _ generalized minsky inequality _ : @xmath480 for all @xmath431 and @xmath466 ( cf . corollary 4 of 8 in @xcite ) . the intersection number has the following _ subadditive property_. [ lem : subadditivity ] let @xmath472 with @xmath481 . then , for any @xmath475 we have @xmath482 let @xmath431 . then , we have @xmath483 hence , @xmath484 for @xmath485 . hence , the right - hand side of follows from the density of @xmath486 in @xmath460 . we prove the left - hand side of . we first show the case where @xmath400 and @xmath187 are rational . let @xmath487 and @xmath488 where @xmath489 are mutually disjoint and distinct simple closed curves and @xmath490 and @xmath491 . let @xmath431 and @xmath492 , @xmath493 @xmath494 be the characteristic annuli for @xmath495 , @xmath393 and @xmath496 of @xmath497 , where the _ characteristic annulus _ for @xmath346 is the maximal annulus foliated by closed trajectories homotopic to @xmath395 . from the geometric definition of extremal length , we have @xmath498 where is obtained from the extremality of the @xmath499-norm of the jenkins - strebel differentials ( cf . theorem 20.5 in @xcite ) . since @xmath486 is dense in @xmath460 , the above calculation implies @xmath500 for all @xmath475 . hence , the left - hand side of also follows from the density of the set of weighted multicurves in @xmath351 . recall that the embedding @xmath501 of @xmath447 into @xmath460 depends on the base point @xmath0 ( cf . [ subsec : gm_closure ] ) . we define the _ intersection number with base point @xmath0 _ by @xmath502 for @xmath503 ( cf . 8.2 in @xcite ) . since the intersection number is continuous , so is @xmath504 on the product @xmath505 . in particular , we can see that @xmath506,[h])=\frac{i(g , h)}{{{\rm ext}}_{x_0}(g)^{1/2}{{\rm ext}}_{x_0}(h)^{1/2}}\ ] ] for @xmath435,[h]\in \mathcal{pmf}\subset \partial_{gm}{\mathcal{t}}$ ] . the intersection number in the right - hand side of is the original intersection number on the space of measured foliations , as noticed in ( v ) of theorem [ thm : main_realization ] . from theorem [ thm : main_realization ] , teichmller space @xmath55 is _ boundary continuous _ in the sense that the gromov product @xmath507 on @xmath508 extends continuously to @xmath505 with values in the closed interval @xmath288 $ ] ( cf . corollary 1 in @xcite ) . we check that the relation visually indistinguishable " is not an equivalent relation on teichmller space when @xmath509 . this also implies that teichmller space @xmath440 is not gromov hyperbolic as remarked in [ subsec : background_motivation ] . indeed , let @xmath510 with @xmath511 , but @xmath512 . consider sequences @xmath85 , @xmath225 and @xmath217 in @xmath55 with @xmath513 $ ] , @xmath514 $ ] and @xmath515 $ ] in @xmath447 , where the projective classes @xmath516 $ ] , @xmath517 $ ] and @xmath518 $ ] are recognized as points in @xmath449 . then , @xmath519 , [ \beta])=0 \\ i_{x_0}(x_n , z_n ) & \to i_{x_0}([\alpha ] , [ \gamma])=0,\end{aligned}\ ] ] but @xmath520 , [ \gamma])\ne 0.\ ] ] from , these observations imply that @xmath521 but @xmath522 . we define the _ null set _ for @xmath523 by @xmath524 in this section , we will show the following . [ thm : null_set ] for any @xmath525 , there is @xmath526 with @xmath527 . the precise statement is given in [ subsec : proof_of_theorem_null_sets ] . indeed , we will see that measured foliations satisfying the statement in theorem [ thm : null_set ] are essentially unique . the following is known . [ prop : translation_null_space1 ] for @xmath528 , @xmath529 if and only if @xmath530 . in any case , @xmath531 . let @xmath525 . the projective class @xmath435\in \mathcal{pmf}$ ] is said to be an _ associated foliation _ for @xmath532 if there exist @xmath533 , a sequence @xmath534\in \mathcal{pmf}$ ] and @xmath535 such that @xmath536 and @xmath534\to [ g]$ ] as @xmath537 . we call the point @xmath538 the _ base point _ for the associated foliation @xmath435 $ ] . we denote by @xmath539 the set of associated foliations for @xmath532 . we define @xmath540 the set coincides with the set defined as when @xmath477 . [ lem : self - intersection ] the following hold : * @xmath541\in \mathcal{af}(\mathfrak{a})\}\subset { \mathcal{n}_{m\!f}}(\mathfrak{a})$ ] . * for @xmath435\in \mathcal{af}(\mathfrak{a})$ ] , we have @xmath542 and @xmath543 . ( 1)let @xmath435\in \mathcal{af}(\mathfrak{a})$ ] . take @xmath533 , @xmath544\}_{n\in \mathbb{n}}\subset\mathcal{pmf}$ ] , and @xmath535 such that @xmath536 and @xmath545 as @xmath537 . since the projection @xmath546 is continuous , from proposition 1 in @xcite , we can see that @xmath547 converges to @xmath548 which is projectively equivalent to @xmath532 . therefore @xmath549 and @xmath550 . ( 2)let @xmath551 . take @xmath533 , @xmath544\}_{n\in \mathbb{n}}\subset\mathcal{pmf}$ ] , @xmath535 , and @xmath552 as above . from generalized minsky s inequality , we have @xmath553 and @xmath554 . from the definition , @xmath555 and we are done . we define for @xmath528 , @xmath556\in \mathcal{af}(\mathfrak{a})}{\mathcal{n}_{m\!f}}(g ) \subset \mathcal{mf}. \ ] ] from ( 2 ) of lemma [ lem : self - intersection ] , we obtain @xmath557\in \mathcal{af}(\mathfrak{a})}{\mathcal{n}_{m\!f}}(g ) \subset \mathcal{an}(\mathfrak{a}).\ ] ] the following proposition implies that the converse holds for simple closed curves . [ prop : vanish - intersection ] @xmath558 for all @xmath528 . let @xmath559 . let @xmath435\in \mathcal{af}(\mathfrak{a})$ ] with @xmath388 . then , there are @xmath533 , a sequence @xmath544\}_{n\in \mathbb{n}}$ ] converging to @xmath435 $ ] and @xmath535 such that @xmath560 tends to @xmath561 as @xmath537 . let @xmath562 . here , we refer to the argument in theorem 3.1 of @xcite ( see also @xcite ) . let @xmath563 be the critical vertical graph of the holomorphic quadratic differential of @xmath564 . we add mutually disjoint critical vertical segments to @xmath563 emanating from critical points to get a graph @xmath565 whose edges are all vertical . the degree of a vertex @xmath565 is one - prong if it is one of endpoints of an added vertical segment . take @xmath566 sufficiently small such that the @xmath567-neighborhood @xmath568 ( with respect to the @xmath569-metric ) is embedded in @xmath4 . then , as the argument in the proof of theorem 3.1 in @xcite , by shrinking with a factor @xmath570 , we get a canonical conformal embedding @xmath571 such that @xmath572 . since @xmath573 , @xmath395 can be deformed into @xmath568 . hence , by from the geometric definition of extremal length , the conformal embedding @xmath571 induces @xmath574 for some @xmath575 independent of @xmath576 . let @xmath566 . take @xmath577 such that @xmath578 . since @xmath534\to [ g]$ ] , by , there exists an @xmath579 such that @xmath580 and @xmath581 for @xmath582 . it has shown from lemma 1 of @xcite that a function @xmath583 is a non - increasing function for any @xmath432 . hence , we have @xmath584 since @xmath585 , @xmath547 converges to @xmath548 which is projectively equivalent to @xmath532 . therefore , we get @xmath586 and hence @xmath587 . this means that @xmath588 . [ coro : charcterization_vanishing_curves ] let @xmath528 . for any @xmath435\in \mathcal{af}(\mathfrak{a})$ ] , we have @xmath589 in particular , for @xmath435\in \mathcal{af}(\mathfrak{a})$ ] , @xmath590 depends only on @xmath532 . let @xmath435\in \mathcal{af}(\mathfrak{a})$ ] . from ( 2 ) of lemma [ lem : self - intersection ] , and proposition [ prop : vanish - intersection ] , we have @xmath591\in \mathcal{af}(\mathfrak{a})}{\mathcal{n}_{m\!f}}(g))\cap \mathcal{s } = \mathcal{an}(\mathfrak{a})\cap \mathcal{s } \\ & \subset \mathcal{n}(\mathfrak{a})\cap \mathcal{s},\end{aligned}\ ] ] which implies what we wanted . let @xmath528 . let @xmath592 be the minimal essential subsurface of @xmath4 which contains all simple closed curve in @xmath593 . we call @xmath592 the _ vanishing surface _ for @xmath532 . by definition , any component @xmath594 of @xmath592 contains a collection of curves in @xmath593 which fills up @xmath594 . if a component of @xmath592 is an annulus , the core is in @xmath595 . from lemma [ lem : fills_curves ] in appendix , if @xmath346 can be deformed into @xmath592 , then @xmath587 ( see also theorem 6.1 of @xcite ) . it is possible that either @xmath593 or @xmath592 is empty . let @xmath435\in \mathcal{af}(\mathfrak{a})$ ] . we decompose @xmath187 into disjoint measured foliations as in : @xmath596 let @xmath366 be the support of @xmath365 . we may assume that any two elements of @xmath597 are mutually disjoint . [ prop : betas ] let @xmath435\in \mathcal{af}(\mathfrak{a})$ ] . for @xmath346 , the following are equivalent . * @xmath395 is homotopic to a component of @xmath598 . * @xmath395 is homotopic to either an essential curve or a peripheral curve of @xmath187 . _ ( 1 ) implies ( 2)._we first assume that @xmath395 is homotopic to a component of @xmath598 . since @xmath587 , from corollary [ coro : charcterization_vanishing_curves ] , we have @xmath599 , that is , @xmath573 . suppose that @xmath395 is non - peripheral in a component @xmath600 of the complement of the support of @xmath187 . then , there is an @xmath601 which is non - peripheral in @xmath600 satisfying @xmath602 . since @xmath603 , @xmath604 by corollary [ coro : charcterization_vanishing_curves ] this means that @xmath395 can not be homotopic to a component of @xmath598 because @xmath592 contains a regular neighborhood of @xmath605 . this contradicts our assumption . hence , @xmath395 is homotopic to either an essential curve or a component of @xmath369 for some @xmath370 . _ ( 2 ) implies ( 1)._suppose @xmath395 is homotopic to either an essential curve or a component of @xmath369 for some @xmath370 . since @xmath573 , @xmath606 by corollary [ coro : charcterization_vanishing_curves ] therefore , @xmath395 can be deformed into the vanishing surface @xmath592 . suppose to the contrary that @xmath395 is non - peripheral in @xmath592 . then , there is a non - peripheral curve @xmath242 in a component @xmath592 with @xmath607 . since @xmath608 , we have @xmath609 by corollary [ coro : charcterization_vanishing_curves ] . if @xmath242 is a component of some @xmath369 , @xmath610 by lemma 2.14 of @xcite . hence @xmath611 by corollary [ coro : charcterization_vanishing_curves ] again , which contradicts that @xmath612 . hence , @xmath242 is non - peripheral in a component of the complement of the support of @xmath187 . therefore , so is @xmath395 , since @xmath607 . this is also a contradiction . thus , we get the following decomposition theorem ( cf . figure [ fig : decomposition ] ) . [ thm : decomposition ] for any @xmath528 , the reference surface @xmath4 is decomposed into a union of closed essential surfaces with mutually disjoint interiors as @xmath613 and each @xmath366 contains an arational foliation @xmath614 filling on @xmath366 for @xmath615 satisfying the following properties : for all @xmath435\in \mathcal{af}(\mathfrak{a})$ ] . * @xmath616 is a subsurface of the vanishing surface @xmath592 defined by deleting annular components whose core is homotopic to an essential curve of @xmath187 . in addition , @xmath616 is independent of the choice of @xmath435\in \mathcal{af}(\mathfrak{a})$ ] . * each @xmath366 is the support of an arational component @xmath365 of @xmath187 . in addition , an arational component of @xmath187 which contained in @xmath366 is topologically equivalent to @xmath614 for @xmath615 . * @xmath617 is a pair of pants with the property that components of @xmath618 are homotopic to among a loop around a puncture of @xmath4 , an essential curve of @xmath187 , or a peripheral curve of @xmath187 . * @xmath619 is an annulus whose core is homotopic to some essential curve of @xmath187 . conversely , any essential curve of @xmath187 is the core of some @xmath619 . in particular , the decomposition is independent of @xmath435\in \mathcal{af}(\mathfrak{a})$ ] . we say that two measured foliations @xmath620 and @xmath621 are _ topologically equivalent _ if the underlying foliations of @xmath620 and @xmath621 are modified by whitehead operations to foliations with trivalent singularities such that the resulting foliations ( without transversal measures ) are isotopic ( cf . 3.1 of @xcite ) . let @xmath435\in \mathcal{af}(\mathfrak{a})$ ] . we decompose @xmath187 as in . we first notice the following : let @xmath395 be any non - peripheral curve in @xmath622 . since @xmath395 can not be deformed into @xmath592 , @xmath611 , which implies @xmath623 from corollary [ coro : charcterization_vanishing_curves ] . let @xmath600 be a component of @xmath622 . since @xmath592 is essential , @xmath600 can be neither a disc nor a punctured disc whose core is homotopic to a peripheral curve of @xmath4 . [ [ case-1 . ] ] * case 1 . * + + + + + + + + suppose first that @xmath600 is neither an annulus nor a pair of pants . then , @xmath600 contains a curve @xmath395 which is non - peripheral in @xmath600 , that is , @xmath395 can not be deformed into @xmath592 . from corollary [ coro : charcterization_vanishing_curves ] , we have @xmath623 . notice from ( 1 ) of lemma [ lem : self - intersection ] , essential curves and peripheral curves of @xmath187 are deformed into @xmath592 . hence , @xmath395 must intersect an arational component @xmath365 of @xmath187 . we check that @xmath624 in the homotopy sense . we first check @xmath625 . indeed , otherwise , there would be a component @xmath385 of @xmath626 which intersects non - trivially to @xmath366 . this means that @xmath627 and hence @xmath628 from corollary [ coro : charcterization_vanishing_curves ] . since any component of @xmath629 is either in @xmath595 or homotopic to a cusp of @xmath4 , this is a contradiction . notice also that any component @xmath385 of @xmath369 is peripheral in @xmath600 , because otherwise @xmath630 , as we checked in the previous paragraph . thus , we conclude that @xmath631 is a deformation retract . since any arational foliation is minimal in its support , the topological equivalence of arational components with common support follows from theorem 1.1 in @xcite . thus , we deduce @xmath632 hence we conclude ( 2 ) . [ [ case-2 . ] ] * case 2 . * + + + + + + + + + let us return to the proof of theorem [ thm : decomposition ] . if @xmath600 is a pair of pants , since @xmath600 is a component of @xmath622 , any component of @xmath629 is either peripheral in @xmath4 or homotopic to a component of @xmath598 . therefore , the condition ( 3 ) follows from proposition [ prop : betas ] . [ [ case-3 . ] ] * case 3 . * + + + + + + + + + before dealing with the remaining case , we check the following claim , which implies ( 1 ) that @xmath616 is independent of the choice of @xmath435\in \mathcal{af}(\mathfrak{a})$ ] . [ claim : essential_curves ] essential curves of the associated foliation to @xmath532 are uniquely determined in the sense that for any @xmath633,[g_2]\in \mathcal{af}(\mathfrak{a})$ ] , any essential curve of @xmath634 is a curve of @xmath635 , and vice versa . let @xmath636 be an essential curve of @xmath634 . suppose on the contrary that @xmath636 is not an essential curve in @xmath635 . notice from ( 2 ) of lemma [ lem : self - intersection ] and that arational components of @xmath634 and @xmath635 are topologically equivalent since @xmath637 . therefore , there is a curve @xmath346 such that @xmath638 but @xmath639 . namely , such an @xmath395 satisfies @xmath640 but @xmath641 , which is a contradiction since @xmath642 from corollary [ coro : charcterization_vanishing_curves ] . [ [ case-4 . ] ] * case 4 . * + + + + + + + + + finally , we deal with ( 4 ) . suppose that @xmath600 is an annulus . since a component of @xmath629 is a component of @xmath598 , the core @xmath242 of @xmath600 is homotopic to a component of @xmath598 . by proposition [ prop : betas ] , @xmath242 is either an essential curve or a peripheral curve of @xmath187 . suppose @xmath242 is a peripheral curve of @xmath187 . let @xmath643 and @xmath644 be components of @xmath592 with @xmath645 and @xmath646 . it is possible that @xmath647 when the core of @xmath600 is non - separating in @xmath4 . however , it is impossible that either @xmath643 or @xmath644 is an annulus . indeed , if @xmath643 is an annulus , from the minimality of @xmath592 , @xmath643 is absorbed into @xmath644 , which is a contradiction . since @xmath242 is a peripheral curve of @xmath187 and @xmath643 , @xmath644 and @xmath600 are essential in @xmath4 , from the definition , either @xmath643 or @xmath644 intersects non - trivially the support of an arational component of @xmath187 in the sense that either @xmath648 or @xmath649 contains a non - annular subsurface @xmath650 such that a component of @xmath651 is homotopic to @xmath242 . however , from corollary [ coro : charcterization_vanishing_curves ] , the support of any arational component of @xmath187 can not contains any vanishing curve , which is a contradiction . hence , the core @xmath242 of @xmath600 is an essential curve of @xmath187 . conversely , let @xmath636 be an essential curve of @xmath187 which does not appear as the core of an annular component of @xmath592 . notice from the definition that @xmath636 can not be homotopic to a component of @xmath369 for any @xmath615 . proposition [ prop : betas ] asserts that @xmath636 is homotopic to a component of @xmath598 . since any curve intersecting @xmath636 is not contained in @xmath592 and @xmath636 is essential , boundaries of an annulus having @xmath636 as its core should be either a component of @xmath598 or a component of the boundary of some @xmath617 . in either case , @xmath636 is the core of an annular component of @xmath622 . from proposition [ prop : measured_foliation_removing_peripehral ] and theorem [ thm : decomposition ] , we conclude the following . [ coro : uniqueness_associate_foliations ] for @xmath633,[g_2]\in \mathcal{af}(\mathfrak{a})$ ] , @xmath652 and @xmath653 are topologically equivalent . we first check the following . [ prop : zeros_topologically_equivalent ] let @xmath435\in \mathcal{af}(\mathfrak{a})$ ] . let @xmath432 be a measured foliation which is topologically equivalent to an arational component of @xmath187 . then , @xmath654 . take @xmath533 , @xmath534\in \mathcal{pmf}$ ] , and @xmath535 such that @xmath655 and @xmath545 as @xmath537 . let @xmath656 . let @xmath657 be the geodesic current associated to the singular flat structure defined as @xmath658 given by duchin , leininger and rafi in @xcite . suppose on the contrary that @xmath659 . then , by proposition 4 in @xcite , @xmath660 is a stable sequence in the sense that the set of accumulation points of @xmath661 in the space of geodesic currents is contained in @xmath662 ( as geodesic currents ) . in addition , any accumulation point @xmath663 satisfies that @xmath664 for some @xmath665 and any @xmath666 ( see proposition 5 in @xcite ) . let @xmath667 be an arational component of @xmath187 which is topologically equivalent to @xmath400 and @xmath668 be the support of @xmath667 . from , @xmath669 . hence , if @xmath670 has a component @xmath671 whose support intersects @xmath668 , then @xmath671 is topologically equivalent to @xmath667 ( cf . this means that @xmath672 , which contradicts to . in this section , we deal with an inequality for intersection number between two points in the core @xmath460 . [ lem : intersection_number ] let @xmath673 and @xmath435\in \mathcal{af}(\mathfrak{a})$ ] and @xmath674\in \mathcal{af}(\mathfrak{b})$ ] . let @xmath538 and @xmath416 be base points for the associated foliations @xmath435 $ ] and @xmath674 $ ] respectively . then , there is an @xmath675\in \overline{\mathcal{af}(\mathfrak{a})}$ ] in @xmath356 such that @xmath676 where @xmath677 . by definition , there are @xmath544\}_{n\in \mathbb{n}}$ ] , @xmath678\}_{n\in \mathbb{n}}\subset \mathcal{pmf}$ ] and @xmath679 such that * @xmath680,x}(t_n)\to { { \rm proj}}(\mathfrak{a})$ ] and @xmath681,y}(s_n)\to { { \rm proj}}(\mathfrak{b})$ ] as @xmath537 ; * @xmath545 and @xmath682 as @xmath537 . for simplicity , let @xmath683,x}(t_n)$ ] and @xmath684,y}(s_n)$ ] . since @xmath685 and @xmath686 , we deduce @xmath687 where @xmath688 . from the assumption , @xmath689 and @xmath690 converge @xmath561 and @xmath691 . since @xmath692 @xmath693 and @xmath694 converge to @xmath695 and @xmath696 , respectively . this implies the left - hand side of the desired inequality . let @xmath697 , @xmath698 and @xmath699 as above . take @xmath700 such that @xmath701 . fix @xmath702 . let @xmath703 and @xmath704 such that @xmath705 . notice that @xmath706 then , by taking a subsequence ( or by the diagonal argument ) , we may assume that @xmath707 as @xmath708 for each @xmath709 , and @xmath710 converges to @xmath711 . since @xmath712 , @xmath713 $ ] is an associated foliation for @xmath532 with base point @xmath714 . therefore , the limit @xmath675 $ ] is contained in the closure of @xmath539 in @xmath356 . from theorem [ thm : main_realization ] , we deduce @xmath715 since @xmath703 . letting @xmath708 , we conclude @xmath716 thus , if @xmath537 in , we have @xmath717 since @xmath718 , we have the assertion . we are ready to prove theorem [ thm : null_set ] . indeed , we show the following theorem . [ thm : associated_foliations_and_null_sets ] let @xmath525 . then , any @xmath435\in \mathcal{af}(\mathfrak{a})$ ] satisfies @xmath719 . from lemma [ lem : self - intersection ] , @xmath542 . hence , we shall show the converse . we first claim that @xmath720 for @xmath435\in \mathcal{af}(\mathfrak{a})$ ] . we decompose @xmath187 as : @xmath721 let @xmath722 . then , @xmath378 can be decomposed as @xmath723 where @xmath382 , @xmath380 is a measured foliation topologically equivalent to @xmath724 ( possibly @xmath725 ) , and @xmath726 is a measured foliation whose support is contained in the complement of the support of @xmath187 ( cf . @xcite ) . from theorem [ thm : decomposition ] , the support of @xmath726 is contained in @xmath727 and hence @xmath728 . since any component of @xmath369 is deformed into @xmath592 , from lemma [ lem : subadditivity ] and proposition [ prop : zeros_topologically_equivalent ] , we have @xmath729 and hence @xmath730 . the converse follows from lemma [ lem : self - intersection ] . let @xmath731 and take @xmath732 such that such that @xmath701 as @xmath537 . let @xmath675\in \overline{\mathcal{af}(\mathfrak{a})}$ ] as lemma [ lem : intersection_number ] for @xmath532 and @xmath733 . to show that @xmath551 , it suffices to show that @xmath734 from lemma [ lem : intersection_number ] . let @xmath735 , @xmath736 such that @xmath737 . by taking a subsequence , we may assume that @xmath738 . then , @xmath739 $ ] is an associate foliation of @xmath733 . since @xmath731 and @xmath740 , we have @xmath741 . therefore , @xmath742 is decomposed as @xmath743 where @xmath744 is topologically equivalent to @xmath724 , @xmath382 , @xmath383 is a measured foliation whose support is contained in the complement of the support of @xmath187 . since @xmath675\in \overline{\mathcal{af}(\mathfrak{a})}$ ] , from corollary [ coro : uniqueness_associate_foliations ] , we can see that @xmath745 is decomposed as @xmath746 where @xmath747 is topologically equivalent to @xmath724 ( possibly @xmath748 ) and @xmath382 . from and , we have @xmath749 . since @xmath740 , we conclude that @xmath750 and hence @xmath751 as desired . before closing this section , we notice the following expected property . [ coro : topological_equivalence_null_set ] for @xmath373 , the following are equivalent : 1 . @xmath371 and @xmath375 are topologically equivalent ; 2 . @xmath374 ; 3 . @xmath752 . from proposition [ prop : measured_foliation_removing_peripehral ] , the conditions ( 1 ) and ( 2 ) are equivalent . since @xmath753 , ( 2 ) follows from ( 3 ) . hence , we need to show that ( 1 ) implies ( 3 ) . from the symmetry of the topological equivalence , it suffices to show that @xmath754 . let @xmath755 and @xmath756\in \mathcal{af}(\mathfrak{a})$ ] . then , @xmath757 from theorem [ thm : associated_foliations_and_null_sets ] . since @xmath375 is topologically equivalent to @xmath371 , by proposition [ prop : measured_foliation_removing_peripehral ] , we have @xmath758 . hence , by applying theorem [ thm : associated_foliations_and_null_sets ] again , we have @xmath759 and @xmath760 . in this section , we consider the group @xmath761 . let @xmath762 . for @xmath450 , we define the _ accumulation set _ @xmath763 as follows : @xmath764 is contained in @xmath765 if and only if there is a sequence @xmath766 such that @xmath455 and @xmath767 . notice that if a sequence @xmath225 converges to a point in @xmath447 , @xmath118 converges at infinity in the sense of . indeed , suppose @xmath118 converges to @xmath768 . then , from ( iv ) in theorem [ thm : main_realization ] , we have @xmath769 and hence @xmath770 as @xmath6 . for @xmath450 , we define the _ null set _ for @xmath249 by @xmath771 we claim [ prop : accumulation_point_accompany ] let @xmath772 . let @xmath773 be a sequence converging to @xmath255 for @xmath128 . then , the following are equivalent : 1 . @xmath774 ; 2 . @xmath775 . furthermore , @xmath776 if and only if @xmath777 . suppose @xmath778 . then , from ( iv ) in theorem [ thm : main_realization ] , @xmath779 and @xmath775 . conversely , suppose that @xmath775 . then , @xmath780 and @xmath774 . suppose that @xmath776 . let @xmath781 . since @xmath183 converges to @xmath782 , any accumulation point @xmath783 of @xmath93 satisfies @xmath784 and hence @xmath785 from the assumption . this means that any subsequence of @xmath93 contains a convergence subsequence which is contained in @xmath786 . hence , @xmath93 itself is contained in @xmath786 ( cf . ( 4 ) in remark [ remark : accompanying_sequence ] in [ subsec : gromov_product ] ) . suppose that @xmath777 . let @xmath787 . take a sequence @xmath93 converging to @xmath273 . then , @xmath788 as observed above . this means that @xmath789 and @xmath785 . the following lemma will be applied for defining the extension to the reduced gardiner - masur closure in [ subsec : extension_to_the_boundary ] . [ lem : equivalent_1 ] let @xmath790 . let @xmath772 and @xmath791 for @xmath128 . if @xmath776 , then @xmath792 . for @xmath128 , let @xmath254 be a sequence converging to @xmath255 such that @xmath793 converges to @xmath794 . from proposition [ prop : accumulation_point_accompany ] , the assumption @xmath776 implies @xmath134 . since @xmath790 , by propositions [ prop : quasi - surjective_asymptotic ] , we have @xmath795 . therefore , by applying proposition [ prop : accumulation_point_accompany ] again , we obtain @xmath792 . we define @xmath796 for a convergent sequence @xmath36 at infinity , where @xmath797 is the closure of @xmath93 in @xmath447 . [ prop : null_sets_and_accompany ] let @xmath36 be a convergent sequence at infinity . then , there is @xmath358 such that @xmath798).\ ] ] furthermore , the following are equivalent for @xmath783 : 1 . @xmath799)$ ] ; 2 . for any @xmath800 and @xmath801\in \mathcal{af}(\psi_{gm}(p))$ ] , @xmath802)$ ] ; 3 . for any @xmath800 @xmath803 . let @xmath804 be accumulation points of @xmath36 in @xmath447 . let @xmath212 be a subsequence of @xmath36 convergent to @xmath255 for @xmath128 . since @xmath36 converges at infinity , @xmath805 and hence @xmath806 for @xmath800 , fix @xmath801\in \mathcal{af}(\psi_{gm}(p))$ ] . from proposition [ prop : measured_foliation_removing_peripehral ] and theorem [ thm : associated_foliations_and_null_sets ] , the topological equivalence class of @xmath807 depends only on @xmath249 , and @xmath808 for all @xmath809 from . hence , we can find @xmath358 such that 1 . for any @xmath800 , @xmath807 is topologically equivalent to a subfoliation of @xmath187 ; 2 . any component of @xmath371 is topologically equivalent to a component of some @xmath810 , @xmath800 . we shall check that @xmath187 satisfies the desired property . indeed , let @xmath811 be an accumulation point of a sequence @xmath88 . then , by the argument above , we can see that @xmath812 for any @xmath800 . hence , for any @xmath674\in \mathcal{ah}(\psi_{gm}(q))$ ] , @xmath813 for all @xmath814 . from the condition ( 2 ) of @xmath187 , we have @xmath377 and hence @xmath815 by theorem [ thm : associated_foliations_and_null_sets ] . this means that @xmath799)$ ] and @xmath816).\ ] ] conversely , let @xmath799)$ ] . take a sequence @xmath93 in @xmath4 converging to @xmath273 . then , by the condition ( 1 ) of @xmath187 above , we have @xmath812 for all @xmath800 . in other words , any subsequence of @xmath36 contains a subsequence which is visually indistinguishable from @xmath93 . therefore , we have @xmath89 and hence @xmath88 ( cf . remark [ remark : accompanying_sequence ] ) . the last statement follows from the construction of @xmath187 and theorem [ thm : associated_foliations_and_null_sets ] . [ prop : accumulation_points_and_accompany ] let @xmath43 and @xmath44 be convergent sequences at infinity . then , the following are equivalent : 1 . @xmath817 ; 2 . @xmath818 . from the definition , the condition ( 2 ) implies ( 1 ) . suppose the condition ( 1 ) . assume on the contrary that there is @xmath819 . the condition @xmath820 means that there are subsequences @xmath821 of @xmath93 and @xmath822 of @xmath44 such that @xmath823 for all @xmath702 . hence , there are @xmath824 and @xmath825 such that @xmath826 ( cf . ) . hence , by proposition [ prop : null_sets_and_accompany ] , @xmath827 . since @xmath828 from the assumption , @xmath829 . on the other hand , since @xmath830 , @xmath831 . this is a contradiction . two points @xmath503 are equivalent if one of the following holds : * @xmath832 and @xmath268 ; * @xmath833 and @xmath834 . we denote by @xmath835}$ ] the equivalence class of @xmath450 . we abbreviate the equivalence class @xmath836]}$ ] of the projective class @xmath435\in \mathcal{pmf}\subset\partial_{gm}{\mathcal{t}}$ ] as @xmath837}$ ] . we denote by @xmath838 the quotient of @xmath447 under this equivalent relation . let @xmath839 be the quotient map . we always identify @xmath840 with @xmath55 . we call @xmath838 the _ reduced gardiner - masur closure _ of @xmath55 . from the definition , the space @xmath838 contains @xmath55 canonically . we call the complement @xmath841 the _ reduced gardiner - masur boundary _ of @xmath55 . let @xmath790 . we define the _ boundary extension _ @xmath842 to @xmath838 as follows . let @xmath768 . we define @xmath843 by @xmath844 } ) = \begin{cases } { [ [ \omega(p ) ] ] } & ( \mbox{$p\in { \mathcal{t}}$ } ) \\ { [ [ q ] ] } & ( \mbox{$q\in \mathcal{a}(\omega\colon p)$ if $ p\in \partial_{gm}{\mathcal{t}}$ } ) \end{cases}\ ] ] from lemma [ lem : equivalent_1 ] , the extension @xmath842 is well - defined . [ lem : composition ] let @xmath845 . then , the extensions satisify @xmath846 on @xmath847 . let @xmath835}\in { \partial_{gm}^{red}}{{\mathcal{t}}}$ ] . take @xmath848 such that @xmath849 and @xmath850 . by definition , @xmath851})={[[p']]}.\ ] ] on the other hand , we may assume that @xmath852 converges to @xmath853 . then , from the definition , we have @xmath854})={[[q]]}$ ] . since @xmath855 , @xmath856 and hence @xmath857}= { \partial_\infty(\omega_1)}({[[q]]})= { \partial_\infty(\omega_1)}\circ{\partial_\infty(\omega_2)}({[[p]]}).\ ] ] this means that @xmath858 on @xmath859 . [ lem : quasi - inverse ] let @xmath198 and @xmath50 be an asymptotic quasi - inverse of @xmath35 . then , the extensions satisfy @xmath860 on @xmath847 . it suffices to show only that @xmath861 on @xmath847 . let @xmath835}\in { \partial_{gm}^{red}}{{\mathcal{t}}}$ ] . take @xmath848 such that @xmath849 and @xmath862 . notice from the definition that @xmath863})={[[q]]}$ ] . let @xmath864 . we may assume that @xmath865 converges to some point @xmath866 . then , by definition , @xmath856 and hence @xmath867})= { \partial_\infty(\omega')}({[[q]]})={[[p']]}.\ ] ] let @xmath868 . take @xmath869 with @xmath870 . since @xmath50 is an asymptotic quasi - inverse of @xmath35 , @xmath48 is close to the identity mapping on @xmath55 . hence , @xmath871 since @xmath872 converges to @xmath873 , @xmath874 and @xmath875 since @xmath48 is asymptotically conservative by proposition [ prop : properties_quasimorphisms ] , @xmath876 . since @xmath36 converges to @xmath249 , we deduce that @xmath877 and @xmath878 . therefore , @xmath879 . let @xmath878 and @xmath880 a convergent sequence to @xmath881 . since @xmath48 is close to the identity mapping on @xmath55 , @xmath882 since @xmath872 converges to @xmath873 , we have @xmath883 and @xmath868 . we conclude that @xmath884 and hence @xmath885}={[[p]]}$ ] . thus , we obtain that @xmath886})={[[p']]}={[[p]]}$ ] , which is what we wanted . an ordered sequence @xmath887}\}_{k=1}^m$ ] in @xmath859 is said to be an _ _ adherence tower _ _ starting at @xmath888}$ ] if @xmath889 we call the number @xmath890 the _ length _ of the adherence tower . let @xmath835}\in { \partial_{gm}^{red}}{\mathcal{t}}$ ] . we define the _ height _ @xmath891})$ ] of @xmath835}$ ] by @xmath892})=\sup\{\mbox{lengths of adherence towers starting $ { [ [ p]]}$}\}.\ ] ] to prove the following lemma , we define the complexity of measured foliation as follows ( cf . theorem 1 in @xcite ) . for a measured foliation @xmath187 , we set @xmath597 be the supports of the arational components of @xmath187 . we define @xmath893 [ lem : height_boundary_points ] suppose that @xmath54 is neither a torus with one hole nor a sphere with four holes . the height of any @xmath835}\in { \partial_{gm}^{red}}{\mathcal{t}}$ ] satisfies @xmath891})\le { { \rm cx}}(s)$ ] . the equality @xmath891})={{\rm cx}}(s)$ ] holds if and only if the support of any @xmath435\in \mathcal{af}(\psi_{gm}(p))$ ] is a simple closed curve . we first study the associated foliations of points in an adherence tower of length two . let @xmath888},{[[p_2]]}\in { \partial_{gm}^{red}}{\mathcal{t}}$ ] . let @xmath894\in \mathcal{af}(\psi_{gm}(p_i))$ ] for @xmath128 . suppose that @xmath895},{[[p_2]]}\}$ ] is an adherence tower , that is , @xmath896 . from the definition , @xmath897 . from corollary [ coro : topological_equivalence_null_set ] , we see @xmath898 we decompose @xmath652 as in : @xmath899 where @xmath724 is an arational component , and @xmath393 is a ( weighted ) essential curve of @xmath634 . since @xmath900 , the composition of @xmath653 is represented as @xmath901 where @xmath744 is topologically equivalent to @xmath724 , @xmath902 and the support of @xmath903 is contained in the complement of the support of @xmath634 . since @xmath904 , any measured foliation @xmath432 with @xmath905 satisfies @xmath906 . hence , @xmath907 and @xmath908 . moreover , from the assumption @xmath909 implies that @xmath910 . therefore , from and , we have @xmath911 in the lexicographical order in @xmath912 , since @xmath903 in is either an arational component or an essential curve of @xmath635 . let us return to the proof of the lemma . let @xmath913}\}_{i=1}^m$ ] be an adherence tower of length @xmath890 . let @xmath894\in \mathcal{af}(\psi_{gm}(p_i))$ ] . from the argument argument , we have @xmath914 notice that the number of essential curves is at most @xmath915 and the euler - characteristic satisfies the inclusion - exclusion principle ( and the euler characteristic of a circle is zero ) . we should care about the case where the genus of @xmath54 is zero since the negative of the euler characteristic is greater than @xmath915 in this case . since any planer surface does not contain an essential subsurface of euler characteristic @xmath916 which admits an arational foliation , the first coordinate of @xmath917 rises at least by @xmath71 ( if it increases ) . therefore , in any case , we have @xmath918 . suppose that some @xmath365 contains an arational component @xmath919 . let @xmath920 be the support of @xmath919 . since @xmath920 contains a non - peripheral curve ( in @xmath920 ) and any component @xmath920 can not be an essential curves of @xmath921 for @xmath922 , the number of essential curves of @xmath923 is less than @xmath915 . therefore , by the above argument , we have @xmath924 . hence if @xmath925 , each @xmath365 consists of essential curves . in this case , one can easily see that the adherence tower starts with a simple closed curve . the _ extended mapping class group _ @xmath53 of @xmath54 is the group of all isotopy classes of homeomorphisms on @xmath54 . the extended mapping class group @xmath53 acts on @xmath55 isometrically by @xmath926_*(y)=(y , f\circ h^{-1})\in { \mathcal{t}}\ ] ] for @xmath927\in { { \rm mcg}}^*(s)$ ] . hence , we have a group homomorphism @xmath928\to [ h]_*\in { \rm isom}({\mathcal{t}}),\ ] ] where @xmath929 is the group of all isometries of @xmath55 . let @xmath58 be the complex of curves of @xmath54 and @xmath62 be the simplicial automorphisms on @xmath58 . since @xmath53 acts on @xmath58 canonically , we have a ( group ) homomorphism @xmath930 it is known that @xmath931 is an isomorphism if @xmath54 is neither a torus with two holes nor a closed surface of genus @xmath71 , and an epimorphism if @xmath54 is not a torus with two holes ( cf . ivanov @xcite , korkmaz @xcite and luo @xcite ) . the action of any isometry on @xmath55 extends homeomorphically to the gardiner - masur boundary ( cf . we can observe that the extension of the action leaves @xmath932 invariant , and it induces a canonical homomorphism @xmath933 such that the diagram @xmath61^{\mathcal{i}_0 } \ar[dr]_{\mathcal{j } } & { \rm isom}({\mathcal{t } } ) \ar[d]^{\mathcal{j}_1 } \\ & { \rm aut}({\mathbb{x}}(s ) ) & \\ } \ ] ] is commutative ( cf . @xcite ) . the homomorphism @xmath934 is an isomorphism for any @xmath54 with @xmath509 . we first sketch the proof of the injectivity . the following argument is given in @xcite by modifying the argument by n. ivanov in @xcite . indeed , since , any element @xmath935 fixes any point of @xmath342 , by applying the argument of the proof of theorem a in @xcite , we can find a fixed point @xmath936 of @xmath35 . since almost all geodesic rays emanating @xmath937 land at a point in @xmath356 , we can see that @xmath35 is the identity mapping on @xmath55 ( cf . @xcite ) . since @xmath931 is surjective unless @xmath54 is a torus with two holes , so is @xmath934 ( cf . luo @xcite ) . suppose that @xmath54 is a torus with two holes . we consider the quotient map @xmath938 defined by the action of the hyper - elliptic involution on @xmath54 . the quotient map induces the isomorphism from @xmath58 to @xmath939 and an isometry from @xmath940 to @xmath941 ( cf . @xcite and @xcite ) . the reason why is not surjective when @xmath54 is a torus with two holes is that there is no homeomorphism on @xmath54 which sends a non - null - homologous curve to a null - homologous curve , while each curve on @xmath942 is null - homologous . thus , in any case , the homomorphism @xmath934 is surjective . recall that any isometry is an invertible asymptotically conservative mapping . hence , we have a monoid monomorphism @xmath943 our rigidity theorem is given as follows . [ thm : induced_automorphism ] suppose that @xmath54 is neither a torus with one hole nor a sphere with four holes . then , there is a monoid epimorphism @xmath944 with the following properties : * if @xmath945 is an asymptotic quasi - inverse of @xmath946 , @xmath947 ; * @xmath948 as monoid homomorphism . in addition , @xmath949 induces a group isomorphism @xmath950 which satisfies the following commutative diagram : @xmath61^{\mathcal{i}_0 } & { \rm isom}({\mathcal{t } } ) \ar[r]^{\mathcal{i } } \ar[dr]^{\mbox{{\tiny group iso } } } & { { \rm ac_{inv}}}({\mathcal{t } } ) \ar[d]^{\mbox{{\tiny proj } } } \ar[dr]^{\xi } & \\ & & { \mathfrak{ac}}({\mathcal{t } } ) \ar[r]^{\mbox{{\tiny group iso } } } & { \rm aut}(\mathbb{x}(s ) ) . & } \ ] ] we may assume that @xmath54 is not a torus with two holes . indeed , when @xmath54 is a torus with two holes , the quotient map @xmath938 by the hyper - elliptic action induces an isometry between the teichmller spaces of @xmath54 and @xmath942 and an isomorphism between @xmath951 and @xmath952 , where @xmath942 is a sphere with five holes ( cf . @xcite and @xcite ) . hence , we consider @xmath942 instead of @xmath54 in this case . let @xmath953 be the @xmath381-skeleton of @xmath58 . for @xmath954 , we identify the projective class @xmath516 $ ] with a point in the @xmath449 . let @xmath946 . from lemma [ lem : height_boundary_points ] , there is an adherence tower @xmath913}\}_{i=1}^{{{\rm cx}}(s)}$ ] with @xmath888}={[[\alpha]]}$ ] . from lemma [ lem : equivalent_1 ] , @xmath955})\}_{i=1}^{{{\rm cx}}(s)}$ ] is also an adherence tower starting @xmath956 } ) = { \partial_\infty(\omega)}({[[\alpha]]})$ ] . therefore , @xmath957})={[[\beta]]}\ ] ] for some simple closed curve @xmath958 , by lemma [ lem : height_boundary_points ] again . hence , @xmath35 induces a self - mapping @xmath959 of @xmath953 . namely , @xmath959 satisfies @xmath960})={[[h_\omega(\alpha)]]}\ ] ] for all @xmath961 . since @xmath35 admits an asymptotic quasi - inverse @xmath50 , by lemma [ lem : quasi - inverse ] , the self - mappings @xmath959 and @xmath962 satisfy @xmath963 on @xmath953 this means that @xmath959 is bijective . let @xmath964 with @xmath965 . then , @xmath966 and @xmath967 . therefore , @xmath968},{[[g]]}\}$ ] and @xmath969},{[[g]]}\}$ ] are adherence towers . from lemma [ lem : equivalent_1 ] , @xmath970}),{\partial_\infty(\omega)}({[[g]]})\}$ ] and @xmath971}),{\partial_\infty(\omega)}({[[g]]})\}$ ] are also adherence towers . since @xmath959 is bijective and @xmath972})={[[h_\omega(\alpha)]]}$ ] and @xmath973})={[[h_\omega(\beta)]]}$ ] , we have @xmath974 . this means that @xmath975 . from lemma [ lem : composition ] , the mapping @xmath976 is a monoid homomorphism . the condition ( 1 ) in the statement follows from the above argument . we next check the condition ( 2 ) in the statement . let @xmath977 . then , @xmath35 preserves @xmath342 in @xmath978 ( cf . 9 in @xcite ) . from the definition of @xmath959 , for any @xmath346 , @xmath979 coincides with @xmath980 . this means that @xmath981 . we check is an epimorphism . we will prove the injectivity of the homomorphism in the next section . let @xmath982 . suppose that @xmath41 and @xmath42 are close at infinity . this means that for any sequences @xmath8 and @xmath9 in @xmath55 , if @xmath45 , @xmath983 . suppose that for @xmath128 , @xmath984 and @xmath985 for @xmath986 as @xmath537 . from proposition [ prop : accumulation_point_accompany ] , @xmath45 if and only if @xmath888}={[[p_2]]}$ ] . hence , the condition closeness at infinity " implies that @xmath987 on @xmath859 . therefore , by the definition , we have @xmath988 . from lemmas [ lem : composition ] and [ lem : quasi - inverse ] , @xmath949 descends to a homomorphism @xmath989\mapsto \xi(\omega)=h_\omega\in { \rm aut}({\mathbb{x}}(s)).\ ] ] since @xmath54 is not a torus with two holes , @xmath931 is an epimorphism , and so is @xmath949 . the injectivity of ( or ) is proven in the next section . in this section , we shall show that the epimorphism is an isomorphism . we first check the following . [ lem : rigidity_1 ] for @xmath946 , let @xmath975 as theorem [ thm : induced_automorphism ] . 1 . for @xmath346 . consider the projective class @xmath516 $ ] as a point in @xmath449 . then , @xmath990)=\{[h_\omega(\alpha)]\}$ ] ; 2 . if @xmath516\in \mathcal{n}^\psi(p)$ ] , then @xmath991\in \mathcal{n}^\psi(q)$ ] for all @xmath853 . conversely , if @xmath991\in \mathcal{n}^\psi(q)$ ] for some @xmath853 , @xmath516\in \mathcal{n}^\psi(p)$ ] . ( 1)let @xmath36 be a sequence in @xmath55 converging to @xmath516 $ ] in @xmath447 . let @xmath783 be an accumulation point of @xmath110 . we may assume that @xmath992 converges to @xmath273 in @xmath447 . by and , any accumulation point @xmath783 of a sequence @xmath110 satisfies @xmath993)$ ] from lemma [ lem : equivalent_1 ] . namely , @xmath994 for all @xmath995 . by applying theorem 3 in @xcite , we conclude that @xmath996 $ ] in @xmath449 . this means that @xmath110 converges to @xmath991 $ ] in @xmath447 . ( 2)@xmath516\in \mathcal{n}^\psi(p)$ ] if and only if @xmath997 . since @xmath35 is asymptotically conservative , from ( 1 ) above , @xmath998 for any @xmath853 . the converse is also proven by the same argument . [ prop : rigidity_2 ] suppose that @xmath54 is not a torus with two holes . for @xmath946 , there is a homeomorphism @xmath999 of @xmath54 with the following property : for any @xmath1000 , @xmath435\in \mathcal{af}(\psi_{gm}(p))$ ] and @xmath853 , we have @xmath1001)$ ] . from the assumption and theorem [ thm : induced_automorphism ] , there is a homeomorphism @xmath999 of @xmath54 such that @xmath1002 . from theorem [ thm : associated_foliations_and_null_sets ] , if we take @xmath674\in \mathcal{af}(\psi_{gm}(q))$ ] , then @xmath1003).\ ] ] from ( 2 ) in proposition [ lem : rigidity_1 ] , for any @xmath346 , @xmath388 if and only if @xmath1004=[h_\omega(\alpha)]\in \mathcal{n}^\psi(q)$ ] . hence , we deduce that @xmath1005)\cap \mathcal{s}= \mathcal{n}^\psi(q)\cap \mathcal{s}=\mathcal{n}^\psi([f_\omega(g)])\cap \mathcal{s}\ ] ] where @xmath342 stands for a subset of @xmath449 in . therefore , the support of @xmath375 coincides with the support of @xmath1006 . in particular , any essential curve of @xmath378 is also that of @xmath1007 , and vice versa . as , we decompose @xmath187 as @xmath1008 let @xmath366 be the support of an arational component @xmath365 of @xmath187 . it is known that @xmath447 is metrizable ( cf . theorem 1.2 in @xcite ) . henceforth , we fix a distance @xmath1009 on @xmath447 . fix @xmath1010 . take a sequence @xmath1011 such that @xmath1012 and @xmath1013,[g_i])<1/n.\ ] ] since @xmath999 is a homeomorphism , @xmath1014 $ ] tends to @xmath1015 $ ] in @xmath356 ( and hence in @xmath447 ) as @xmath537 . by taking a subsequence , we may assume that @xmath1016,[f_\omega(g_i)])<1/n\ ] ] for all @xmath702 . let @xmath1017 be a sequence in @xmath55 converging to @xmath1018 $ ] in @xmath447 . since @xmath198 , by ( 1 ) of proposition [ lem : rigidity_1 ] , @xmath1019 converges to @xmath1014 $ ] in @xmath447 for all @xmath709 . by applying the diagonal argument and taking a subsequence if necessary , we can take @xmath1020 such that if we put @xmath1021 and @xmath217 , then @xmath1022 ) , d_\infty(\omega(z_n),[f_\omega(\alpha_n)])\}<2/n\ ] ] in @xmath447 . since @xmath999 is a homeomorphism of @xmath54 , @xmath1014 $ ] tends to @xmath1015 $ ] in @xmath356 and hence in @xmath447 . from , we have @xmath1023 ) ) \le d_\infty(\omega(z_n),f_\omega([\alpha_n ] ) ) + d_\infty(f_\omega([\alpha_n]),[f_\omega(g_i ) ] ) \to 0\ ] ] as @xmath537 . therefore @xmath262 converges to @xmath1015 $ ] . this means that @xmath1015\in \mathcal{a}(\omega\colon [ g_i])$ ] . since @xmath365 is an arational component of @xmath187 , @xmath1024)\subset \mathcal{n}^\psi([g_0])$ ] . therefore , by lemma [ lem : equivalent_1 ] , we conclude @xmath1025)= \mathcal{n}^\psi(q)\subset \mathcal{n}^\psi([f_\omega(g_i)])\ ] ] since @xmath853 and @xmath1015\in \mathcal{a}(\omega\colon [ g_i])$ ] . therefore , @xmath378 contains an arational component @xmath380 which is topologically equivalent to @xmath1026 . since the support of @xmath375 coincides with that of @xmath371 , the arational components of @xmath378 are contained in @xmath1027 . hence , the normal form of @xmath378 should be @xmath1028 where @xmath1029 and @xmath1030 . thus , @xmath375 is topologically equivalent to @xmath1006 . hence by corollary [ coro : topological_equivalence_null_set ] , we deduce @xmath1031)= \mathcal{n}^\psi([h])=\mathcal{n}^\psi(q),\ ] ] which is what we desired . [ prop : rigidity_3 ] for any @xmath946 , there is a unique isometry @xmath1032 on @xmath55 which is close to @xmath35 at infinity . we first suppose that @xmath54 is not a torus with two holes . take @xmath999 as in proposition [ prop : rigidity_2 ] . since @xmath999 is a homeomorphism of @xmath54 , @xmath999 induces an isometry @xmath1032 on @xmath55 . when @xmath54 is a closed surface of genus @xmath71 , there is an ambiguity of the choice of @xmath999 which is caused by the hyperelliptic involution . however , the isometry @xmath1032 is independent of the choice . let @xmath8 and @xmath1033 be convergent sequences at infinity satisfying that @xmath45 . from proposition [ prop : null_sets_and_accompany ] , there are @xmath187 , @xmath1034 , @xmath1035 such that @xmath1036 ) & = { \mathcal{acm}}({\bf x}^1)={\mathcal{acm}}({\bf x}^2)\\ \mathcal{n}^\psi([h_1 ] ) & = { \mathcal{acm}}(\omega({\bf x}^1 ) ) \\ \mathcal{n}^\psi([h_2 ] ) & = { \mathcal{acm}}(\xi_\omega({\bf x}^2)).\end{aligned}\ ] ] ( cf . ) . hence , our assertion follows from proposition [ prop : accumulation_points_and_accompany ] and the following claim : [ claim : gh_1h_2 ] it holds @xmath1037)=\mathcal{n}^\psi([f_\omega(g)])= \mathcal{n}^\psi([h_2]).\ ] ] let @xmath1038 and @xmath1039 . let @xmath1040 and fix @xmath801\in \mathcal{af}(\psi_{gm}(p))$ ] . from proposition [ prop : rigidity_2 ] , @xmath1041)=0 $ ] . since @xmath249 is taken arbitrarily in @xmath1042 , from the proof of proposition [ prop : null_sets_and_accompany ] , we have @xmath1043)=0 $ ] . hence @xmath1037 ) \subset \mathcal{n}^\psi([f_\omega(g)]).\ ] ] let @xmath1044)$ ] . for @xmath1045 , we take a subsequence @xmath93 of @xmath43 such that @xmath262 converges to @xmath273 . we may assume that @xmath93 converges to @xmath1046 . fix @xmath801\in \mathcal{af}(\psi_{gm}(p))$ ] . from proposition [ prop : rigidity_2 ] , we have @xmath1047)$ ] . from the construction of @xmath187 , @xmath807 is topologically equivalent to a subfoliation of @xmath187 . hence , @xmath1048)=\mathcal{n}^\psi(q)$ ] . since @xmath273 is taken arbitrarily from @xmath1045 , we can deduce that any sequence converging to @xmath881 is visually indistinguishable from @xmath104 ( cf . remark [ remark : accompanying_sequence ] ) . therefore , @xmath1049)$ ] and @xmath1050)=\mathcal{n}^\psi([f_\omega(g)]).\ ] ] since @xmath1032 is an isometry , @xmath1051 since @xmath1032 extends to @xmath447 homeomorphically and coincides with the action of @xmath999 on @xmath458 , we have @xmath1052)\cap \mathcal{pmf } & = { \mathcal{acm}}(\xi_\omega({\bf x}^2))\cap \mathcal{pmf } = \xi_\omega({\mathcal{acm}}({\bf x}^2))\cap \mathcal{pmf } \\ & = \xi_\omega(\mathcal{n}^\psi([g]))\cap \mathcal{pmf } = \mathcal{n}^\psi([f_\omega(g)])\cap \mathcal{pmf}.\end{aligned}\ ] ] this equality means that @xmath1053 . by corollary [ coro : topological_equivalence_null_set ] , we have @xmath1054 and @xmath1055)=\mathcal{n}^\psi([f_\omega(g)])$ ] . the case where @xmath54 is a torus with two holes follows from the fact that the teichmller space of @xmath54 is isometric to the teichmller space of a sphere with five holes . let @xmath790 with @xmath1056 . from proposition [ prop : rigidity_3 ] , there is an isometry @xmath1032 which is close to @xmath35 . since @xmath1057 , @xmath1032 is the identity mapping on @xmath55 , and hence @xmath35 is close to the identity . in this section , we consider the actions of rough homotheties on teichmller space . indeed , we shall prove theorem [ thm : teichmullerspace_homothety ] . as given in [ subsec : results ] , we recall a mapping @xmath1058 a @xmath63-_rough homothety _ if @xmath1059 for all @xmath1060 . suppose first that @xmath509 . we may assume that @xmath54 is not a torus with two holes . suppose to the contrary that there is a @xmath63-rough homothety @xmath35 with asymptotic quasi - inverse for some @xmath70 . notice from remark ( 2 ) in [ subsec : remarks_asymptotically_conservative ] that @xmath946 . take a homeomorphism @xmath999 on @xmath54 as proposition [ prop : rigidity_2 ] . let @xmath1061 . consider the projective classes @xmath516 $ ] and @xmath517 $ ] as points in @xmath449 . then , from , lemma [ lem : rigidity_1 ] and proposition [ prop : rigidity_2 ] , we have @xmath1062,[\beta])^{k } \le i_{x_0}([f_\omega(\alpha)],[f_\omega(\beta ) ] ) \le e^{d_0}i_{x_0}([\alpha],[\beta])^{k}\ ] ] where @xmath221 is a constant depending only on @xmath1063 and @xmath1064 . on the other hand , let @xmath1065 , where @xmath1032 is an isometry associated to @xmath35 taken as proposition [ prop : rigidity_3 ] . from the definition , @xmath1066 for @xmath358 . by the quasiconformal invariance of extremal length , we obtain @xmath1067,[\beta ] ) \le i_{x_0}([f_\omega(\alpha)],[f_\omega(\beta ) ] ) \le k_0 i_{x_0}([\alpha],[\beta ] ) \ ] ] since @xmath999 is a homeomorphism on @xmath54 and @xmath1068 . therefore , we deduce @xmath1069,[\beta])^{1-k } & \le k_0e^{d_0 } \label{eq : k - homothety_4 } \\ i_{x_0}([\alpha],[\beta])^{k-1 } & \le k_0e^{d_0 } \label{eq : k - homothety_4_2}\end{aligned}\ ] ] for any @xmath1061 with @xmath1070,[\beta])\ne 0 $ ] . since the left - hand sides of and are projectively invariant , when the projective classes @xmath516,[\beta]$ ] tend together to some projective measured foliation @xmath435\in \mathcal{pmf}$ ] with keeping satisfying @xmath1071 , the left - hand side in diverges if @xmath1072 , otherwise the left - hand side in diverges . in any case , we get a contradiction . suppose @xmath1073 . in this case , there exists an isometry @xmath1074 sending @xmath0 to the origin @xmath381 . furthermore , the gromov product @xmath1075 for @xmath1076 satisfies @xmath1077 , where @xmath1078 $ ] is the geodesic connecting between @xmath28 and @xmath29 ( cf . 2.33 in @xcite ) . suppose on the contrary that there is a @xmath63-rough homothety @xmath35 with @xmath70 . notice from theorem [ thm : acgroup_hyperbolic_group_boundary_continuous ] that @xmath35 extends to a self - homeomorphism on @xmath1079 . we may assume that @xmath1080 . from , for @xmath1076 , @xmath1081)-kd_{\mathbb{d}}(0,[x_1,x_2])|\le d_2\ ] ] for some constant @xmath1082 . therefore , for any @xmath1083 , we have @xmath1084 with positive constants @xmath1085 . since the lift of a self - homeomorphism on @xmath1079 to @xmath1086 is a monotone function , the extension of @xmath35 to @xmath1079 is differentiable almost everywhere on @xmath1079 . however , since the extension of @xmath35 satisfies , either @xmath35 or @xmath1087 is not differentiable any point on @xmath1079 if @xmath70 . this is a contradiction . the main result of this section is lemma [ lem : fills_curves ] . the estimates in the lemma looks similar to that in theorem 6.1 of @xcite . however , our advantage here is that we treat the extremal lengths of all non - trivial ( possibly peripheral ) curves of subsurfaces and give a constant @xmath1088 concretely ( cf . and ) . let @xmath1089 be a holomorphic quadratic differential on @xmath4 . the differential @xmath1090 defines a measured foliation on @xmath4 . we say that such a measured foliation the _ vertical foliation _ of @xmath1089 . the vertical foliation of @xmath1091 is called the _ horizontal foliation _ of @xmath1089 . by a _ step curve _ , we mean a geodesic polygon in @xmath4 the sides of which are horizontal and vertical arcs of @xmath1089 ( cf . figure [ fig : stepcurve ] ) . [ prop_strebel_stepcurve ] let @xmath1089 be a quadratic differential and @xmath400 the vertical foliation of @xmath1089 . let @xmath1092 be a simple closed step curve with the additional property that for any vertical side @xmath1093 of @xmath1092 the two neighboring horizontal sides @xmath1094 and @xmath1095 are on different sides of @xmath1093 ( there are no zeros of @xmath1089 on @xmath1092 ) . then , @xmath1096 where @xmath385 is the homotopy class containing @xmath1092 . it can be also observed that a step curve with the property stated in theorem [ prop_strebel_stepcurve ] is quasi - transversal . for instance , see the proof of proposition ii.6 or the curve ( 4 ) of figure 10 of expos 5 in @xcite . let @xmath668 be an essential subsurface of @xmath4 . denote by @xmath1097 a subset of @xmath342 consisting of curves which are non - peripheral in @xmath668 . let @xmath1098 be a subset of @xmath342 consisting of curves which can be deformed into @xmath668 . [ lem : fills_curves ] let @xmath668 be a connected , compact and essential subsurface of @xmath4 with negative euler characteristic . let @xmath1099 be a system of curves which fills up @xmath668 . then , for @xmath1100 , we have @xmath1101 where @xmath1102 and @xmath1103 depends only on the topological type @xmath1104 of @xmath4 and the number @xmath890 of the system @xmath1105 . in particular , we have @xmath1106 for all @xmath1107 . suppose first that @xmath385 is represented by a component of @xmath1108 . when @xmath385 is homotopic to a puncture of @xmath4 , @xmath1109 since @xmath4 contains an arbitrary wide annulus whose core is homotopic to @xmath385 . hence we have nothing to do ( in fact , we can set @xmath1110 ) . suppose that @xmath385 is not peripheral in @xmath4 . let @xmath1111 be a jenkins - strebel differential for @xmath385 on @xmath4 . let @xmath1112 be the characteristic annulus of @xmath1111 . we consider a compactification " @xmath1113 by attaching two copies of circles as its boundaries . the induced flat structure on @xmath1112 from @xmath1111 canonically extends to the compactification @xmath1113 and components of the boundary @xmath1114 are closed regular trajectories under this flat structure . there is a canonical surjection @xmath1115 ( the completion of @xmath4 at the punctures ) . namely , @xmath1116 is reconstructed by identifying disjoint vertical straight arcs in @xmath1114 along vertical saddle connections of @xmath1111 . ( in this sense , @xmath1117 is a quotient map ) . without any confusion , we may recognize the characteristic annulus @xmath1112 itself as a subset of @xmath4 . let @xmath1118 and @xmath1119 be the core trajectory in @xmath1112 and the geodesic representative of @xmath495 with respect to @xmath1111 respectively . since @xmath385 is parallel to @xmath1108 , by taking an isotopy , we may assume that @xmath1118 is a component of @xmath1108 . furthermore , since @xmath1120 , @xmath385 does not intersect any @xmath495 for all @xmath370 . hence , each @xmath1119 consists of vertical saddle connections . in other words , @xmath1119 is contained in the critical graph @xmath1121 of @xmath1111 in @xmath4 , which consists of vertical saddle connections of @xmath1111 . suppose @xmath1129 for some @xmath370 and @xmath1127 contains a vertical saddle connection @xmath1130 such that @xmath1131 for all @xmath370 . then , @xmath1130 intersects all @xmath1132 at most at endpoints ( critical points of @xmath1111 ) . let @xmath1133 . let @xmath1134 be a horizontal arc in @xmath1112 starting at @xmath1135 and terminating at a point of @xmath1136 . since the both side of @xmath1130 is in @xmath1112 , after @xmath1134 passes through @xmath1130 , @xmath1134 terminates at a point @xmath1137 . let @xmath1138 be a segment of @xmath1118 connecting @xmath1139 and @xmath782 ( cf . figure [ fig : a_gamma ] ) . set @xmath1140 . by definition , @xmath636 does not intersects any @xmath1119 and hence @xmath1141 for all @xmath370 . suppose first that @xmath1134 arrived @xmath782 from the different side from that where @xmath1134 departed at @xmath1139 ( cf . ( 1 ) of figure [ fig : a_gamma ] ) . then , we have @xmath1142 since the width of @xmath1112 is one and @xmath636 is a step curve with the property stated in proposition [ prop_strebel_stepcurve ] . hence @xmath636 is non - trivial and non - peripheral simple closed curve in @xmath4 . however , this contradicts that @xmath1105 fills up @xmath668 , since such a @xmath1143 contains homotopically non - trivial arc connecting @xmath1108 because @xmath385 is parallel to a component of @xmath1108 . suppose @xmath1134 arrived at @xmath782 from the same side as that where @xmath1134 departed ( cf . ( 2 ) of figure [ fig : a_gamma ] ) . we may also assume that @xmath1134 departs from @xmath1139 into @xmath668 . indeed , suppose we can not assume so . then , the component of @xmath1144 that lies on the same side as that of @xmath668 ( near @xmath1118 ) is covered by @xmath1145 , which contradicts what we assumed first . then , there is an open rectangle @xmath1146 in @xmath1112 such that @xmath636 and a segment in @xmath1122 surround @xmath1146 in @xmath1112 . from the assumption , we may assume that the closure of @xmath1147 , say @xmath23 , intersects some @xmath1119 . suppose that @xmath636 is trivial . then , @xmath23 is a disk in @xmath4 surrounded by @xmath636 , since @xmath1118 can be homotopic to the outside of @xmath23 . this means that @xmath1119 is contained in a disk @xmath23 because @xmath1119 does not intersect @xmath636 , which is a contradiction . by the same argument , we can see that @xmath636 is non - peripheral ( otherwise , @xmath1119 were peripheral ) . since @xmath1134 departs into @xmath668 at @xmath1139 and returns to @xmath1118 on the side where @xmath668 lies , after taking an isotopy if necessary , we can see that @xmath1134 contains a subsegment which is nontrivial in @xmath668 and connecting @xmath1108 , which contradicts again that @xmath1105 fills @xmath668 up . let us continue to prove lemma [ lem : fills_curves ] for peripheral @xmath1148 . we take @xmath1127 as in claim 1 . since both sides of every vertical saddle connection face @xmath1112 , @xmath1127 visits each vertical saddle connection at most twice . notice that the number of vertical saddle connections is at most @xmath1149 . since each vertical saddle connection in @xmath1127 is contained in some @xmath1119 , we have @xmath1150 since @xmath1119 is the geodesic representative of @xmath495 . since the width of @xmath1112 is one , from , we conclude @xmath1151 we next assume that @xmath385 is not parallel to any component of @xmath1108 . let @xmath1153 be components of @xmath1108 each of which is non - peripheral in @xmath4 . let @xmath566 and set @xmath1154 ( cf . it is possible that two curves @xmath1155 and @xmath1156 are homotopic in @xmath4 . in this case , we recognize @xmath1157 in . however , for the simplicity of the discussion , we shall assume that any two of @xmath1158 are not isotopic . the general case can be treated in a similar way . let @xmath1159 be the holomorphic quadratic differential on @xmath4 whose vertical foliation is @xmath1160 . since @xmath1161 in @xmath351 , @xmath1159 tends to @xmath1111 in @xmath1162 ( cf . see also theorem 21.3 in @xcite ) . let @xmath1163 and @xmath1164 denote the characteristic annuli of @xmath1159 for @xmath385 and @xmath393 , respectively . set @xmath1165 and @xmath1166 to be closed trajectories in homotopic to @xmath385 and @xmath393 , respectively . let @xmath1167 be the closure of the component of @xmath1168-neighborhood of the cores @xmath1166 , containing @xmath1163 . by definition , we may identify @xmath1167 with @xmath668 . let @xmath1169 be the geodesic representation of @xmath495 with respect to @xmath1159 . we fix an orientation on @xmath1165 . let @xmath220 be a component of @xmath1170 . let @xmath1171 be the set of points @xmath1172 such that the horizontal ray @xmath1173 departing at @xmath249 from the right of @xmath220 terminates at a curve in @xmath1174 before intersecting @xmath220 twice . let @xmath1175 be the set of @xmath1172 such that @xmath1173 terminates at a critical point of @xmath1159 . then , we claim let @xmath1178 . suppose @xmath1179 . since the completion @xmath1116 with respect to the punctures is closed , @xmath1173 is recurrent ( cf . 10 of chapter iv in @xcite ) . by the definition of @xmath1171 and @xmath1180 , @xmath1173 intersects @xmath220 at least twice before intersecting curves in @xmath1174 . hence , @xmath1173 contains a consecutive horizontal segments @xmath1134 and @xmath1181 such that each @xmath1182 intersects @xmath220 only at its endpoints , and does not intersect any curves in @xmath1174 . when one of the segments , say @xmath1134 , connects both sides of @xmath220 , @xmath220 contains a vertical segment @xmath1183 connecting endpoints of @xmath1182 , and two trajectories @xmath1134 and @xmath1183 make a closed curve @xmath242 on @xmath4 . since the two ends of @xmath1182 terminate at @xmath220 from different sides , the intersection number satisfies @xmath1184 and is greater than or equal to the width of @xmath1163 , by proposition [ prop_strebel_stepcurve ] . therefore @xmath242 is non - trivial and non - peripheral in @xmath4 . since @xmath1182 does not intersect @xmath1166 , @xmath242 is contained in @xmath1167 , where we have identified with @xmath668 . furthermore , @xmath242 is not peripheral in @xmath668 because @xmath242 has non - trivial intersection with @xmath385 . by definition , @xmath242 does not intersect all @xmath495 , which is a contradiction because @xmath1105 fills @xmath668 up . we assume that two ends of each @xmath1182 terminate at @xmath220 from the same side . in this case , we can also construct a simple closed step curve @xmath242 with the property stated in proposition [ prop_strebel_stepcurve ] from @xmath1134 , @xmath1181 and a subsegment of @xmath220 ( cf . figure [ fig : delta ] ) . : there are two cases . in the case ( 1 ) , the initial point of @xmath1134 and the terminal point of @xmath1181 are separated by the terminal point of @xmath1134 . the case ( 2 ) describes the other case.,height=188 ] this is a contradiction as above . thus we conclude that @xmath1176 . we show that @xmath1177 is open in @xmath220 . let @xmath1185 such that the horizontal ray @xmath1173 defined above does not terminate at critical points of @xmath1159 . by definition , the horizontal ray @xmath1173 terminate the interior of a straight arc contained in either @xmath1186 or @xmath1187 . hence , when @xmath1188 is in some small neighborhood of @xmath249 , @xmath1189 also terminates at such a straight arc , and hence @xmath1190 for all point @xmath873 in a small neighborhood of @xmath249 . let us return to the proof of case 2 of the lemma . let @xmath220 be a component of @xmath1191 . by definition , for @xmath1192 , @xmath1173 terminates at @xmath220 at most once before intersecting curves in @xmath1193 . since any horizontal ray @xmath1173 with @xmath1179 can terminate at a curve in @xmath1174 from at most two sides . hence for almost all point @xmath273 in a curve in @xmath1174 , there are at most @xmath1194 points in @xmath1171 such that the horizontal rays emanating there land at @xmath273 . from claim 2 , we get @xmath1195 where @xmath1196 means linear measure . since @xmath1186 is the geodesic representative of @xmath495 , the number of components of @xmath1170 is @xmath1197 therefore , we obtain @xmath1198 since @xmath1159 tends to @xmath1111 as @xmath1199 , for all @xmath1200 we find an @xmath566 such that @xmath1201 by corollary 21.2 in @xcite and the cauchy - schwarz inequality , where @xmath1202 from . notice that the number @xmath1203 of components of @xmath1108 satisfies @xmath1204 . indeed , we fix a hyperbolic metric on @xmath4 and realize @xmath668 as a convex hyperbolic subsurface of @xmath4 . let @xmath1205 and @xmath1206 be the genus and the number of punctures in @xmath668 . since @xmath1207 and @xmath668 is essential , by comparing to the hyperbolic area , we have @xmath1208 and hence @xmath1204 . j. alonso , t. brady , d. cooper , v. ferlini , m. lustig , m. mihalik , m. shapiro , h. short , notes on word hyperbolic groups , _ group theory from a geometrical viewpoint _ ( trieste , 1990 ) , 363 , world sci . publ . , river edge , nj , ( 1991 ) . a. douady , a. fathi , d. fried , f. laudenbach , v. ponaru , and m. shub , _ travaux de thurston sur les surfaces _ , sminaire orsay ( seconde dition ) . astrisque no . 66 - 67 , socit mathmatique de france , paris ( 1991 ) . n. ivanov , isometries of teichmller spaces from the point of view of mostow rigidity , _ topology , ergodic theory , real algebraic geometry _ ( eds . turaev , v. , vershik , a. ) , 131149 , amer . 2 , vol * 202 * , american mathematical society ( 2001 ) .

the giant galactic globular cluster @xmath3 centauri ( ngc 5139 , hereafter @xmath3 cen ) is not only the most massive globular cluster in the milky way , but its stars have also been shown to have a large spread in both age and metallicity ( see for instance , many contributions in van leeuwen et al . 2002 ) . in light of these discoveries , it has been proposed ( lee et al . 1999 , majewski et al . 2000 ) and now dynamically modeled ( tsujimoto & shigeyama 2003 ) that @xmath3 cen may be the nucleus of a disrupting dsph , similar to the galactic globular cluster m54 . the second largest globular cluster , m54 is currently believed to be the core of the sgr dsph ( da costa & armandroff 1995 , bassino & muzzio 1995 , sarajedini & layden 1995 , layden & sarajedini 2000 ) and lies at a nucleation point for the several different metallicity populations of sgr ( sarajedini & layden 1995 , majewski et al . 2003 ) . the recent observation of apparent tidal tails around @xmath3 cen by leon , meylan & combes ( 2000 , hereafter lmc00 ) bolsters the notion of this cluster being the remnant of a larger , now disrupted parent object , and , with @xmath3 cen and m54 as models , ree et al . ( 2002 ) , lee et al . ( 2001 , 2002 ) , and yoon et al . ( 2000 ) have proposed that other clusters with evidence of multiple populations , such as clusters with bimodal horizontal branches , may also be the nuclei of disrupting dsph satellite galaxies . two clusters proposed to be m54 analogues ( ree et al . 2002 ) are ngc 6388 and ngc 6441 , the third and fifth largest globular clusters in the milky way respectively , since both have observed characteristics suggesting internal metallicity spreads . moreover , high mean metallicity clusters like these ( see table 1 ) should have only a red stub of a horizontal branch ( hb ) , but ngc 6388 and ngc 6441 also have extended blue hb tails ( piotto et al . 1997 , hereafter p97 ) . a _ bimodal _ ( or extended ) metallicity distribution may explain both this curious hb morphology ( p97 ) as well as the unusually wide red giant branches ( rgb ) observed in these clusters . because it is difficult to understand how multiple generations of stars might be created in a stellar system of initial mass of order @xmath4 ( which is not massive enough to retain significant amounts of gas after an initial starburst ) , it is possible that the multi - population clusters are the remnants of _ formerly _ larger stellar systems , for example dsph galaxies , where multi - epoch star formation histories are commonly found . the verification of the existence of additional multiple - population clusters would increase the need to formulate a general process e.g. , the disruption of initially larger systems that leads to a more substantial _ class _ of these objects . these scientific questions are complicated by the presence of non - negligible differential reddening in the fields of some the most interesting examples of potential multi - modal clusters . in this paper , we concentrate on two effects of differential reddening relevant to the globular cluster evolution picture outlined above : the potential creation of false signatures of both tidal tails around , and of multiple metallicities within , globular clusters . these issues are explored in case studies of the three massive globular clusters @xmath3 cen , ngc 6388 and ngc 6441 . tidal tail searches around galactic satellites can be particularly vulnerable to differential reddening effects because search techniques typically rely on the detection of star count overdensities . lines of sight with a lower dust column density will naturally reveal more stars to a given magnitude limit . this may create an apparent overdensity of stars in regions with lower extinction that could be mistaken for , or overemphasize , density enhancements from tidal tails . in the case of the reported tidal tails around @xmath3 centauri by lmc00 , the contours of the detected tails curiously and closely correspond to contours of the prominent iras @xmath5 dust emission ( and therefore , presumably , optical extinction by dust ) in this field , as these authors have observed . because lmc00 did not correct for differential extinction in their optical starcounts ( s. leon , private communication ) , at minimum there must be some dust - shaping of the morphology of mapped @xmath3 cen tidal tails . we re - evaluate the possible impact of differential reddening on large field studies of @xmath3 cen by comparing a customized tidal tail search on raw 2mass data to one performed on 2mass data differentially dereddened according to the dust maps of schlegel et al . ( 1998 ) . the cause for the observed width of the rgb in ngc 6388 and ngc 6441 has been variously explained as differential reddening and/or the presence of multiple populations of different metallicity , with no clear consensus emerging . p97 originally concluded that differential reddening on scales greater than eight arcseconds was unlikely for ngc 6388 but possible for ngc 6441 , although only to a few hundredths of a magnitude . later evidence indicating the presence of strong differential reddening on small scales across both clusters was found by raimondo et al . ( 2002 ) using hst - wfpc2 small area surveys similar to those used by p97 . pritzl et al . ( 2001 , 2002 ) also find differential reddening in ngc 6441 and ngc 6388 using ctio @xmath6 and @xmath7 band photometry , although they note that in ngc 6388 the differential reddening probably covers a smaller magnitude range . if differential reddening were significant , the reddening vector could reproduce the sloping red hbs and spread the rgb stars perpendicular to the unreddened rgbs , inflating their apparent width . numerical experiments performed by raimondo et al . ( 2002 ) indicate that differential reddening of order @xmath8e@xmath9 0.1 would be substantial enough to cause an unreddened hb of similar metallicity ( e.g. , that of 47 tucanae ) to resemble a sloping red stub similar to that of ngc 6388 or ngc 6441 . nevertheless , raimondo et al . conclude that reddening is not the reason for the strange hb s since their sloping nature is present even in small area subsamples exhibiting a thin rgb , which suggests that a second effect is at work . pritzl et al . ( 2001 , 2002 ) and ree et al . ( 2002 ) argue that variable metallicity may adequately explain the observed features of the ngc 6388 and ngc 6441 cmds , and recent stellar evolutionary models ( e.g. , sweigart & catelan 1998 ) confirm that a metallicity spread can naturally create a sloping hb . specifically , ree et al . ( 2002 ) find that two distinct populations separated by 1.2 gyr and 0.15 dex in [ fe / h ] can reproduce both the observed wide rgbs and tilted hbs in the clusters . however , pritzl et al . ( 2001 , 2002 ) argue that the simple two - population model fails to reproduce the observed hb slope and that a more general metallicity _ spread _ is necessary . because selective extinction at the 2mass wavelengths @xmath0 ( 1.25 @xmath10 ) , @xmath1 ( 1.65 @xmath10 ) , & @xmath11 ( 2.17 @xmath10 , hereafter simply referred to as @xmath12 ) is roughly one tenth that in visual bands , the magnitude of differential reddening at these wavelengths is commensurately reduced . thus , we apply the 2mass point source catalog to test whether : ( 1 ) differential reddening affects interpretations of an _ infrared _ search for tidal tails around @xmath3 cen , and ( 2 ) the ngc 6388 and ngc 6441 rgbs are broadened at near infrared wavelengths where differential reddening will have less impact on the width of the cluster rgb s . @xmath0 , @xmath1 , and @xmath2 band data were drawn from the 2mass all - sky point - source working survey database as of july , 2002 . initial values of the mean field extinction , @xmath13 , and distance modulus , @xmath14 , for each cluster were obtained from the harris compilation as posted on the world wide web ( http://physun.physics.mcmaster.ca/globular.html ) on june 22 , 1999 ( see harris 1996 ) . these were converted to the required infrared extinction parameters using the standard extinction models of cardelli , clayton , and mathis ( 1989 ) . table 2 gives a complete list of the adopted extinction parameters . the lmc00 study of optical starcounts in the @xmath3 cen field makes no corrections for differential reddening because magnitude calibrations were not available for the majority of clusters in their study ( s. leon , private communication ) . as figures 1a and 1b illustrate , the proposed tidal tails of @xmath3 cen are from enhanced star counts whose pattern appears to follow regions of lower dust absorption on the iras @xmath5 map . this effect is particularly noticeable on the western edge of the cluster . lmc00 note and discuss the possibility that the star counts are affected by reddening , but determine that the tidal tails are of a sufficiently large s / n that differential reddening should not significantly affect their tidal tail search . by comparing starcount patterns obtained using raw 2mass data with those from 2mass data differentially dereddened according to the dust maps of schlegel et al . ( 1998 ) we intend to re - evaluate whether uncorrected differential extinction can produce a false tidal tail signal . the 2mass data has a @xmath15 completeness limit at @xmath16 mag in the direction of @xmath3 cen , yielding almost 500,000 stars within a @xmath17 box centered on the cluster . because a photometric search for diffuse tidal tails presents considerable challenges , not least of which is substantial field star contamination in this low latitude field , a two stage method is employed to minimize field star contamination and probe for starcount excesses as deeply into the surrounding field as possible : first , a coarse filtering mask in color - color - magnitude ( ccm ) space is used to extract only those stars whose ccm location is similar to that of stars observed at radii less than the cluster core radius . although this coarse mask does not take into account foreground / background contamination of the cluster core , it filters out a significant number of stars whose ccm locations are significantly different from those in the cluster . second , we then apply a signal - to - noise filter similar to that used by odenkirchen et al . ( 2001 ) to separate cluster stars from field stars and emphasize starcount overdensities in the potential tidal tails . in the first stage of filtering , a filtering mask is applied in ccm space to pre - select those ccm locations containing stars within 100 arcseconds of the center of the cluster ( approximately @xmath18 the core radius @xmath19 , see table 1 ) . we divide this irregularly shaped region in ( @xmath0 , ( @xmath20 ) , ( @xmath21 ) ) space into a @xmath22 grid in ccm space bounded by the limits of the ccm region ; this grid size is selected to achieve a balance between coarse binning and low number statistics . with the individual cells described by indices ( @xmath23 , @xmath24 , @xmath25 ) respectively , we define @xmath26 as the number of stars within 100 arcseconds of the cluster center that fall within each grid cell in ccm space . to obtain an estimate of the number of field stars included in @xmath26 , we represent the field star distribution , @xmath27 , by the stars within an annulus @xmath28 ( roughly three times the tidal radius of the cluster ) . we assume that the distribution of field stars is relatively even across the cluster and scale the field mask by the ratio @xmath29 of the areas over which the field and core regions were summed . the selection mask @xmath30 is defined as the ratio of the core mask to the sum of the core and scaled field masks . @xmath31 @xmath32 will have a value near one for regions of ccm space where the core sample @xmath26 is not substantially contaminated with field stars , and will decrease as contamination becomes more severe . the complete data set is filtered using this ccm selection mask : a star is rejected from the data set if the corresponding @xmath32 is less than some minimum value @xmath33 . this algorithm is a slightly more sophisticated means to eliminate stars unrelated to @xmath3 cen than simple preliminary cuts in the color - magnitude ( cm ) plane . we find that a value of @xmath34 allows us to select cluster stars over field stars most optimally and appears to preserve the cmd of the cluster center best . this preliminary filtering rejects roughly 90% of the stars in the original data set . in the second stage of filtering , we apply a modified version of grillmair et al.s ( 1995 ) method in the cm plane as described by odenkirchen et al . this second stage eliminates field stars more effectively than the first stage , but we find that the method does not perform well unless a coarser pre - selection of stars is performed first to reduce the data set to a size that the more computationally complex second - stage routine can handle . with some recycling of notation , we define a grid in @xmath35 space , with indices ( @xmath24 , @xmath25 ) respectively , and find the signal- to - noise ratio ( @xmath36 ) of probable cluster stars to field stars in each cm grid cell . we count the number of stars @xmath37 that fall within each grid cell and lie within @xmath38 of the cluster center ( chosen to correspond to the region in which lmc00 observe circular cluster surface density contours , see fig . 1a ) , and the number of stars @xmath39 that fall within the cm grid cell and lie between @xmath40 and @xmath41 arcminutes from the cluster center . the field star population is assumed to be uniform , and the expected number of field stars of a given ( @xmath24 , @xmath25 ) that fall within @xmath38 of the cluster core is subtracted from @xmath37 . with @xmath29 representing the ratio of the area of sky on which @xmath42 and @xmath43 respectively have been defined , the local s / n ratio of cluster to field stars for each grid cell ( @xmath24 , @xmath25 ) is , by standard poissonian statistics , @xmath44 it is possible to use @xmath45 as a mask for selection in the cm plane by eliminating all @xmath46 grid cells with @xmath45 less than some limiting value @xmath47 , which is chosen such that the number of stars in the tidal tail region ( as reported by lmc00 ) is a maximum relative to the field star population . the new matrix @xmath48 represents the total number of stars in each @xmath46 grid cell for the region and outer radius @xmath49 ) located directly north and south of the circle described by the inner radius . ] of the cluster in which tidal tails have been reported by lmc00 ( see fig . 1a ) , as shown by the dashed lines in fig . starting from @xmath50 ( where @xmath51 is the maximum value of all @xmath45 ) and iterating downward to @xmath52 , we calculate @xmath53 and @xmath54 for all ( @xmath24 , @xmath25 ) such that @xmath55 . hence , @xmath56 and @xmath57 respectively are the total number of field stars and cluster stars in lmc00 s tidal tail regions whose locations in @xmath58 space satisfy @xmath55 . the optimal value of @xmath47 is that which maximizes @xmath59 ; hence this final selection mask filters the data set for all stars which fall within cells in cm space such that @xmath60 . the positions of stars from this filtered set are then converted to a smooth density distribution by application of a two - dimensional gaussian kernel with an adaptive half maximum width ( e.g. , silverman 1986 ) for each star set to the distance of the 30@xmath61 nearest neighbor in the plane of the sky . each of these gaussians is evaluated at finely spaced grid points in the plane of the sky , and these values are summed to create an image of relative stellar densities at each location on the grid ( fig . 1c and 1d ) . we find that this @xmath36 optimization method worked efficiently to reduce the 2mass data set by a further 80 - 90% and brought out starcount overdensities around the cluster that have the appearance of tidal tails . focusing the definition of @xmath62 on the regions found by lmc00 to possess tidal tails is considerably more effective at bringing these apparent tails out of the background than defining @xmath62 more generally ; however , this particular approach obviously is not useful for a general survey for previously unknown tidal tail searches because it requires _ a priori _ knowledge of where tails may be found . we verified that the overdensities found by this method are not merely artifacts of the definition of @xmath62 in the region of the lmc00 expected tails by redefining @xmath62 in locations rotated around the cluster core by 90 degrees ( that is , protruding east - west as opposed to north - south ) . using this definition of @xmath62 , no starcount overdensities were detected around the cluster , which supports that the apparent tails found with the first @xmath62 definition had not been an artificial product of the processing method and that the method of sampling @xmath62 in the tail region is a crucial step in bringing those tails out of the background noise . in the analysis of the raw 2mass data described above a starcount excess to the north and south of the cluster is found . as fig . 1c shows , to within the limits imposed by the smoothing kernel , the location of the 2mass tails corresponds to that of the tails found by lmc00 . the same analysis has been repeated with 2mass data dereddened according to the extinction maps of schlegel et al . the schlegel et al . maps have a resolution of approximately 6.1 arcminutes , which corresponds to about 1500 cells of reddening data over a @xmath63 field of view , and show differential reddening of order @xmath64 mag over that field of view . in contrast to what is found for the raw 2mass data , the results of the analysis on the dereddened 2mass data do not reveal any elongation of the circular cluster contour lines or any overdensities resembling tidal tails ( fig . 2 plots radial surface density profiles for both original and dereddened 2mass data , and shows the surface density of the dereddened cluster to fall off smoothly with radius , like a king ( 1966 ) profile . in contrast , the surface density of stars in the non - dereddened cluster data appears to drop at @xmath65 ( @xmath66 60 arcmin ) corresponding to the low starcount regions at this radius to the east and west of the cluster and rise again at @xmath67 ( @xmath68 arcmin ) which corresponds to high starcount regions to the north and south of the cluster ( fig . 1c ) . given the fidelity with which the apparent infrared and optical tails follow the regions of lower reddening on the iras 100 micron map ( fig . 1b , particularly on the western and southern edges of the cluster ) , and the disappearance of the infrared tails in the dereddened cluster data , we speculate that by analogy the @xmath3 cen tidal tails previously detected with the optical starcounts , which are more affected by dust extinction , could have been artifacts of differential reddening . we must also point out , however , that our test of the 2mass database is not completely analagous to the lmc00 analysis : due to the @xmath16 mag completeness limit of 2mass we are unable to probe to the same effective depth as the lmc00 study . the differences in relative depth are demonstrated by the fact that while lmc00 observe reasonably circular contours out to a radius of about 45 arcminutes ( roughly the tidal radius , see table 1 ) , we only find circular contours out to about 40 arcminutes before we begin to observe distinct elongation of the contours into potential tidal tails ( fig . 1 ) . thus , the leon et al . counts reach fainter surface brightness limits than does 2mass and are sensitive to more diffuse , lower density tails than are we . nevertheless , we regard the 2mass test presented here as a cautionary example demonstrating that a proper accounting of differential reddening is an important step in mapping any tidal tails in the @xmath3 cen ( or any other heavily obscured ) field . to investigate the effects of differential reddening on the width of the rgb in the clusters ngc 6388 and ngc 6441 we use the uniform photometry of 2mass to look for spatial patterns in the distribution of red giant star colors within each cluster . 47 tucanae ( ngc 104 , hereafter 47 tuc ) is adopted as a control case for a cluster of similar mean metallicity but with little differential reddening ( average @xmath69 0.02 ) . we also compare our results to the @xmath3 cen data although this cluster has an overall [ fe / h ] much lower than the other clusters , it nonetheless provides a useful example of the effects of _ variable metallicity _ within a cluster over spatial scales with relatively little differential reddening on the scale of arcminutes ( @xmath70 mag , note that the previous section explored extinction variations in @xmath3 cen on the scale of _ degrees _ ) . analysis of rgb color _ spreads _ at different magnitudes in one cluster as well as differences between clusters is simplified if we remove the mean rgb color - magnitude relation for each cluster ( i.e. make the center of the rgb locus for each cluster appear vertical in the color - magnitude diagram ) . since the [ fe / h ] of the unreddened cluster 47 tuc is similar to that of ngc 6388 and ngc 6441 to within about 0.25 dex ( see table 1 ) , the shape , height and position of the rgbs for the three clusters in a reddening corrected , absolute color - magnitude space should be similar . hence , we use the well - defined 2mass cmd of 47 tuc to `` straighten '' the rgb of the clusters in our sample ( figure 3a ) . we convert apparent to absolute magnitudes using the distance moduli in table 2 and consider only that part of the rgb with absolute magnitude @xmath71 to minimize contamination from cluster horizontal branch ( hb ) as well as field stars . a legendre polynomial is fit to the 47 tuc rgb , with a rejection threshhold of 2@xmath72 iterated until convergence . the final polynomial is converted to power series coefficients up to third order , which are used to straighten the rgbs of all four clusters . we define an adjusted @xmath73 color given by : @xmath74 where @xmath75 values are given in table 2 , and the coefficients @xmath76 , @xmath77 , and @xmath78 are power series coefficients given in table 3 . similar equations are derived for the @xmath79 and @xmath80 cmds . using the adjusted values , histograms of colors readily show that the rgbs of ngc 6388 and ngc 6441 ( figure 4 ) are wider than those of 47 tuc and even that of @xmath3 cen , which is well established to have a metal spread from [ fe / h ] @xmath81 -2.2 to -0.7 ( vanture , wallerstein , & suntzeff 2002 ) . the latter result is somewhat surprising , given the fact that the metal - rich rgb of 47 tuc provides a poor approximation to the metal - poor rgb of @xmath3 cen and produces a clear tilt in the `` adjusted '' @xmath3 cen rgb ( fig . 4 ) . the fig . 4 histograms are divided into five @xmath82 colors bins in order to study the spatial distribution of rgb stars by color . the five color bins from the lowest value of @xmath82 ( most blue ) to the highest ( most red ) are referred to as bins a through e , respectively . because globular clusters are dynamically mixed on timescales that are short compared to chemical enrichment , there should be no significant difference in the spatial distribution of the stars in these five bins . for example , we show below ( section 4.2.1 ) that neither 47 tuc nor @xmath3 cen exhibit any such spatial variation . on the other hand , if differential extinction were significant , the distributions of stars by color would be uneven across the face of the cluster as a reflection of reddening variations . this is a test we now apply to ngc 6388 and ngc 6441 . various methods to divide the sample population into color bins were explored . the results are not very sensitive to which method is chosen . the following technique ensures similar treatment of each cluster : bin ranges are assigned to each cluster using a @xmath82 histogram of a subsample of the purest cluster stars , defined by an angular radius where the density of stars decreased to a limiting density given by : @xmath83 ( where the outer density is determined as the density of stars at the edge of the field of view ) . in all cases , this resulted in a sample of stars within a radius of between roughly one and three arcminutes , which is outside the core radius but well within the tidal radius of each cluster ( see table 1 ) . using a color histogram with binning width 0.02 mag , the central three bins are defined within the @xmath82 range from the central peak to the colors where counts decrease to @xmath84 of the central peak . bin c is centered at the peak of the histogram and has 1/3 of the width of b+c+d . the range of bin a is determined as the color range blue - ward of bin b up to where the counts first drop to one ( to eliminate extreme colored stars from the sample ) . likewise , bin e is defined as the range red - ward of bin d up to where the counts first drop to one ( fig . field star contamination ( or contamination by non - rgb stars in the cluster ) , not differential reddening , is assumed to be the reason for stars that lie very far from the fiducial rgb , and the described culling of the data set promises to retain only those stars that make up the bulk of the rgb and that could potentially have been scattered to their current locations in the cmd by differential reddening . these color bin definitions are applied to the full 300 arcsecond data set . similar binnings were created for @xmath85 and @xmath86 histograms , although only the @xmath82 and @xmath85 distributions were found to be useful in our analysis ( see below ) . before proceeding further , we apply one additional filter designed to preserve reddening effects and reduce noise from astrophysical and photometric scatter . this filter takes advantage of the fact that the degree of scatter due to reddening in one color should be correlated to that in another color . a legendre polynomial with an iterative @xmath88 rejection threshhold ( similar to that used earlier in section 4.1.1 ) was fit to the cluster @xmath82 vs. @xmath85 color - color diagram ( fig . 6 ) , and color - color cuts were performed to eliminate stars at greater than @xmath88 from the legendre fit . this independently derived fit for each cluster is very close to the fiducial reddening vector for standard extinction models with @xmath89 , lending further evidence to support the presence of differential reddening across each cluster . figs . 6 and 7 show color - color plots of @xmath82 versus @xmath85 and @xmath86 color , respectively , with the adopted outlier cuts shown . note that the dynamic range in color of @xmath86 is small compared to the reddening effect sought , and it is difficult to determine which stars appear redder than average as a result of differential reddening instead of photometric errors or astrophysical reasons . we therefore reject @xmath86 color from our analysis , since it does not have the discriminating power required to pick out the required distribution from the noise of points in figure 7 . spatial plots of the stars in each of the @xmath82 bins for 47 tuc , @xmath3 cen , ngc 6388 and ngc 6441 are shown in figures 8 - 11 respectively . as expected for a monometallic population with little reddening , 47 tuc shows no spatial segregation of the stars in the five bins . although giant stars in @xmath3 cen have been found to exhibit a spatial variation in metallicity ( jurcsik 1998 ) , this variation is on scales of roughly 20 arcminutes ; in contrast , the mixed metallicity population in our smaller field of view shows no spatial segregation by rgb color . both 47 tuc and @xmath3 cen have few stars in bins a and e relative to their total population . in contrast , ngc 6388 exhibits a pronounced abundance of redder stars ( bins d and e ) in the northern and northwestern regions of the cluster compared to southern regions ( fig . 10 ) , and the extreme bins a and e each contain significantly more stars relative to the total population of the cluster than these bins contained for 47 tuc and @xmath3 cen . comparison to the 47 tuc and @xmath3 cen templates for monometallic and multi - metallic clusters ( respectively ) having comparatively little differential reddening over small scales leads one to the conclusion that it is unlikely that the ngc 6388 color spread is due primarily to metallicity effects . rather , fig . 10 suggests that differential reddening on small scales is causing ( 1 ) a greater number of stars to appear in bins a and e , and ( 2 ) large spatial variations by observed stellar color . likewise , ngc 6441 ( fig . 11 ) shows an enhancement of redder stars ( bins d and e ) in the southeastern quadrant of the field of view , supporting the view that it too is subject to substantial differential reddening . using isochrone fitting in subfields , heitsch & richtler ( 1999 ) similarly find that the southeast region of ngc 6441 has higher @xmath13 than the northwestern region by about 0.15 mag . layden et al . ( 1999 ) concur with heitsch & richtler to within a few hundredths of a magnitude using rr lyrae stars and cmds in sectors of an annulus around the cluster center . however , in contrast to the distinct dichotomy in ngc 6388 ( redder stars in the northwest , bluer stars in the southeast ) , the spatial distribution of stars in bins a through c in ngc 6441 appears relatively constant ( with the possible exception of decreased counts in the southeastern quadrant of bin a ) , and the cluster presents the appearance of an inhomogeneously distributed redder population superposed on a centrally concentrated bluer population background . this effect may be an artifact of the severe contamination of the cluster by field stars since ngc 6441 is located in an extremely crowded field near the galactic center ( as vividly demonstrated by the washed out starcount map shown in figure 12 ) . indeed , heitsch & richtler ( 1999 ) note the strong contamination of their own cmd of ngc 6441 by field stars that cover the lower part of the cluster rgb . in this case , strong contamination by blue field stars may contribute to the apparent width of the cluster red giant branch . figure 12 shows 2mass starcount density ( panel a ) and mean stellar color ( panel b ) maps for all four of our clusters . in this figure , higher starcounts and redder mean colors appear whiter ( table 5 provides the numerical values for each greyscale ) . the field of view of these maps is more than 100 times larger than that used for the analyses in the rest of this section since this method only provides coarse resolution maps which , however , are useful for understanding qualitatively the global character and degree of patchiness of the dust obscuration around each cluster . for example , assuming that the average intrinsic star color in a given direction is relatively uniform over degree scales , the mean color maps can be interpreted as showing the general character of the reddening in the direction of each cluster . however , these maps have too coarse a greyscale to show color variations of order @xmath90 . 12 shows that the mean stellar color in the direction of ngc 6388 is not only patchier than that in the direction of 47 tuc or @xmath3 cen but that the number of redder stars appears to be greater to the north of the cluster than to the south , in agreement with the trend found above . distinct regions of higher reddening and lower reddening are demarcated by a boundary that passes close to the center of the cluster . differential reddening across this apparent dust boundary may produce the observed overabundance of redder rgb stars in the north - northwestern regions of the cluster compared to the south - southeastern regions . the ngc 6441 field mean color map also shows evidence of patchy extinction , although it is not obvious whether any of these patches could produce the observed reddening patterns near the cluster that are likely the case for ngc 6388 . however , the presence of large scale patchiness is sufficient to suggest the possibility of patchiness on the angular scales explored in section 4.2.1 . histograms of star colors within regions of greater reddening should be offset from those of subregions with lesser reddening . to study the color distributions in subsections of the cluster fields , which will have larger statistical noise , we apply a uniform width - normalized gaussian kernel ( e.g. , silverman 1986 ) to the @xmath82 color distribution of stars in ngc 6388 and ngc 6441 . figure 13 shows these smoothed distributions for the northwest and southeast quadrants of each cluster . the half maximum width of the kernel is the minimum value that produces a clearly defined central peak in the distribution for a given cluster . ngc 6388 required this smoothing width to be only 0.02 mag , while ngc 6441 required a wider gaussian of width 0.06 mag to smooth out secondary peaks adequately . the central peak of the distribution of the northwestern quadrant of ngc 6388 is offset from that of the southeastern quadrant by roughly @xmath91 0.061 mag , and the peaks of the northwestern and southeastern quadrants of ngc 6441 are offset by roughly @xmath92 0.072 mag . because the width of the kernel used for ngc 6441 is of order the offset between the peaks , the precision of the derived @xmath93 is less for ngc 6441 than for ngc 6388 . nevertheless , the @xmath93 results for both clusters are in agreement with the qualitative arguments for differential reddening presented earlier in sections 4.2.1 and 4.2.2 , and our value for the differential reddening across ngc 6441 is consistent with the more detailed optical extinction maps produced by layden et al . ( 1999 ) and heitsch & richtler ( 1999 ) to within 0.02 mag in @xmath75 . if such an offset exists between the color distribution of stars in different regions of each cluster , then the artificial width produced by the superposition of these offset distributions may contribute to the thicker rgb of the entire cluster . 14 simulates the effect of superimposing two offset color distributions by artificially reddening the northern half of the monometallic cluster 47 tucanae by 0.061 magnitudes in @xmath82 ; it is observed that while such a superposition can artificially inflate the width of the color distribution ( fig . 13 , lower panels ) , it is only by about half the amount required to re - create the width of the distributions observed in ngc 6388 and ngc 6441 . the inability of this simulation to entirely explain the rgb widths of these clusters may be due to the presence of a large reddening _ spread _ ( as opposed to the simplistic two value model used here ) , or it may be indicative of a contributing effect of differential metallicity . to test the significance of the differences in @xmath82 color distribution in different quadrants of each cluster , we applied the kolmogorov - smirnov ( ks ) test to compare spatial color distributions numerically . as a calibration of the method , we first compared the color distribution across the clusters 47 tuc and @xmath3 cen . as tables 6 and 7 show , the @xmath82 color distribution of stars in , for example , the northwestern quadrant of the clusters correlates with the distribution of stars in the southeastern quadrant with significance levels of about 0.5 and 0.25 for 47 tuc and @xmath3 cen respectively . these significance levels are consistent with a probability that stars within the two regions could have been randomly drawn from one parent distribution . tables 6 and 7 demonstrate that this relation significance is roughly similar for comparisons between all quadrants for both 47 tuc and @xmath3 cen . correlations between the southeastern and southwestern , and northeastern and northwestern quadrants of ngc 6388 have significance levels of roughly 0.9 and 0.5 respectively . in contrast , the ks test yields a probability of only 0.002 that the northwestern and southeastern quadrants share a similar color distribution ( table 8) . compare this to the variable metallicity cluster @xmath3 cen , which maintains a relatively high correlation significance of color distribution between all quadrants in spite of its metallicity spread . it appears unlikely that a metallicity spread is the dominant cause of the rgb width in ngc 6388 ( we expect mixing in clusters in any case ) . instead , the stars in these regions almost certainly come from statistically distinct color populations , pointing again to the presence of strong differential reddening . application of the ks test to the @xmath82 color distribution in the southeastern and northwestern quadrants of ngc 6441 yields a near zero likelihood of correlation ( table 9 ) . qualitatively , this is not surprising because there are very few stars from bins d and e in the northwestern quadrant while there are a considerably higher number of redder stars that lie in the southeastern quadrant . however , while these two quadrants are by orders of magnitude the least correlated regions of this globular cluster , correlations between all other combinations of quadrants are only of order 0.01 to 0.1 , which suggests ( as we have concluded in section 4.2.2 ) that reddening in ngc 6441 is patchy on smaller angular scales than in ngc 6388 . if differential reddening is producing the observed width in these clusters , the rgb histograms in sub - regions of each cluster should also be narrower than that of each cluster as a whole . using hst - wfpc2 visual band data for ngc 6388 and ngc 6441 , raimondo et al . ( 2002 ) find this to be the case for both clusters . in summary , it appears very probable that differential reddening is artificially inflating the apparent thickness of the rgb in these two clusters , although the thickness may not be fully explained by this effect alone . differential extinction complicates photometric studies of galactic stellar populations , and even nearby globular clusters with @xmath13 as low as about 0.1 mag may be visibly affected . in our search for tidal tails around the giant globular cluster @xmath3 cen we have found that differentially dereddening 2mass photometry largely eliminates apparent tidal tail candidates present in data that have not been corrected for extinction . given the similarity of the false 2mass tidal tail signature to the optical tails reported by leon et al , it would be beneficial to verify the diffuse optical starcount distribution in the @xmath3 cen field with deep visual photometry that accounts for the @xmath94 variations . we have also investigated the effects of differential reddening on the cmds of the clusters ngc 6388 and ngc 6441 , which exhibit abnormally large spreads in their rgbs . the analyses presented make use of spatial maps by stars in color bins , maps of mean stellar color , kernel - smoothed histograms of stellar color in sub - regions , and ks statistical analyses of the distribution of ( @xmath21 ) and ( @xmath20 ) colors across the clusters . these tests indicate that the northwestern quadrant of ngc 6388 is subject to about @xmath95 0.12 mag more reddening than the southeastern quadrant , and that the southeastern quadrant of ngc 6441 is more heavily reddened by about @xmath95 0.14 mag than the northwestern quadrant . these findings are consistent with previous results from heitsch & richtler ( 1999 ) and layden et al . ( 1999 ) . as numerical experiments performed here and by raimondo et al . ( 2002 ) have shown , this amount of differential reddening could produce anomalously wider red giant branches for these clusters ( in addition to the tilt observed in their horizontal branches , as demonstrated by raimondo et al . ) . raimondo et al . have shown that the rgb in sub - regions of these clusters is thin ( as expected if differential reddening were the cause of the width of the rgb across the entire cluster ) . athough we find that possible width inflation due to the superposition of thin rgbs from two _ quadrants _ subject to different amounts of differential extinction may not be sufficient to completely explain the observed width of the cluster rgbs , it may be possible that the rgb width could be explained more fully by a more realistic superposition of rgbs from smaller subregions . therefore , if not accounting for the entire rgb - widening effect , at minimum differential reddening may be exaggerating metallicity - induced rgb spreads and the actual spread in age or metalliticy within the clusters is likely not as large as has previously been suggested ( e.g. , pritzl et al . 2001 , 2002 ) . the number of known globular clusters showing clearly anomalous metallicity spreads remains small . with few bona fide members , this `` class '' of globular cluster can still comfortably be accomodated by origin pictures that are exceptional , e.g. , along the lines of the m54 - sgr paradigm . however , even in the obvious case of the multi - population @xmath3 cen , the m54 analogy may be less compelling than previously thought , and the picture that some globular clusters are the `` residue '' of dsph disruption more ambiguous . the authors would like to thank s. leon and g. meylan for helpful discussion regarding the effects of reddening on @xmath3 centauri . this publication makes use of data products from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center , funded by the national aeronautics and space administration and the national science foundation . bassino , l.p . and muzzio , j.c . 1995 , the observatory , 115 , 256 . cardelli , j.a . , clayton , g.c . and mathis , j.s . 1989 , , 345 , 245 . da costa , g.s . and armandroff , t.e . 1995 , , 109 , 2533 . grillmair , c.j . , freeman , k.c . , irwin , m. , & quinn , p.j . 1995 , , 109 , 2553 . harris , w.e . 1996 , , 112 , 1487 . heitsch , f. and richtler , t. 1999 , , 347 , 455 . jurcsik , j. 1998 , , 506l , 113 . king , i.r . 1966 , , 71 , 64 . layden , a.c . and sarajedini , a. 2000 , , 119 , 1760 . layden , a.c . , ritter , l.a . , welch , d.l . , and webb , t.m.a . 1999 , , 117 , 1313 . van leeuwen , f. , hughes , j.d . , & piotto , g. ( eds . ) 2002 , asp conf . 265 , ( san francisco : asp ) . lee , y.w . , rey , s.c . , ree , c.h . , joo , j.m . , sohn , y.j . , and yoon , s.j . 2002 , asp conf . 265 : omega centauri , a unique window into astrophysics , 305 . lee , y.w . , joo , j.m . , sohn , y.j . , rey , s.c . , lee , h.c . , and walker , a.r . 1999 , nature , 402 , 55l . leon , s. , meylan , g. , & combes , f. 2000 , , 359 , 907 . majewski , s.r . , patterson , r.j . , dinescu , d.i . , johnson , w.y . , ostheimer , j.c . , kunkel , w.e . , and palma , c. 2000 , proc . 35th liege international astrophysics colloq . , the galactic halo : > from globular clusters to field stars , ed . a. noels et al . ( belgium : institut dastrophysique et de geophysique ) , 619 . odenkirchen , m. , grebel , e.k . , rockosi , c.m . , dehnen , w. , ibata , r. , rix , h.w . , stolte , a. , wolf , c. , anderson , j.e . bahcall , n.a . , brinkmann , j. , csabai , i. , hennessy , g. , hindsley , r.b . , ivezi , z. , lupton , r.h . , munn , j.a . , pier , j.r . , stoughton , c. , and york , d.g . 2001 , , 548 , 165 . piotto , g. , sosin , c. , king , i.r . , djorgovski , s.g . , rich , r.m . , dorman , b. , renzini , a. , phinney , s. , and liebert , j. 1997 , in advances in stellar evolution , ed . rood & a. renzini ( cambridge : cambridge university press ) , 84 . pritzl , b.j . , smith , h.a . , catelan , m. , & sweigart , a.v . 2001 , , 122 , 2600 . pritzl , b.j . , smith , h.a . , catelan , m. , & sweigart , a.v . 2002 , , 124 , 949 . raimondo , g. , castellani , v. , cassisi , s. , brocato , e. and piotto , g. 2002 , , 569 , 975 . ree , c.h . , yoon , s.j . , rey , s.c . , and lee , y.w . 2002 , asp conf . 265 : omega centauri , a unique window into astrophysics , 101 . sarajedini , a. and layden , a.c . 1995 , , 109 , 1086 . schlegel , d. , finkbeiner , d. , & davis , m. 1998 , , 500 525 . silverman , b.w . 1986 , density estimation for statistics and data analysis ( london : chapman & hall ) . sweigart , a.v . , & catelan , m. 1998 , , 501 , l63 . trager , s.c . , king , i.r . , & djorgovski , s. 1995 , , 109 , 218 . tsujimoto , t. and shigeyama , t. 2003 , astro - ph/0302300 vanture , a.d . , wallerstein , g. , & suntzeff , n.b . 2002 , , 569 , 984 . yoon , s.j . , lee , c.h . , and lee , y.w . 2000 , american astronomical society meeting , 196 . cccccccc 47 tucanae & 0.124 & 13.37 & 13.26 & 0.04 & 0.021 & 0.011 & 0.009@xmath3 centauri & 0.299 & 13.92 & 13.63 & 0.12 & 0.060 & 0.033 & 0.027ngc 6388 & 1.240 & 16.54 & 15.44 & 0.40 & 0.209 & 0.109 & 0.090ngc 6441 & 1.395 & 16.62 & 15.38 & 0.45 & 0.234 & 0.123 & 0.102

in order to gain information about a quantum state of the radiation field one has to measure some observable . the measurement process unavoidably involves some kind of interaction , thus coupling the mode under examination to one or more other modes of the field . one has to admit that , in general , the measured observable is not defined on the sole hilbert space of the signal mode @xcite . on the contrary , it reflects properties of the global state , possibly entangled , coming from the interaction among the signal mode and the set of the probe modes . sometimes , it is possible to get rid of these probe modes , so that the statistics of the output is related only to the quantum statistics of the signal mode . this , as an example , is the case of balanced homodyne detectors @xcite . in that case , in fact , an appropriate rescaling of the output photocurrent allows to completely neglect the local oscillator in the definition of the measured observable , which simply results to be proportional to the signal field quadrature @xmath0 . however , in more general cases this procedure can not be pursued and the output statistics always remind us the way we are probing the signal under examination . this is a common feature of three relevant detection schemes in quantum optics , which provide different setup for jointly measuring a couple of photocurrents . they are the heterodyne @xcite , the eight - port homodyne @xcite and the recently introduced six - port homodyne detectors @xcite . it is a purpose of this paper to show that , although they involve very different ways of coupling signal to probe modes , they provide the same information on the signal under examination . actually , we will prove their full equivalence by demonstrating that their output photocurrents have the same operatorial form . more surprisingly , this remains true also when taking into account the inefficiency of photodetection , even though the latter occurs at very different stages in the three detection schemes . the paper will be organized as follows . in the next section we briefly review how to describe inefficient photodetection in terms of beam splitters and ideal detectors . in the three sections [ s : eig],[s : het ] and [ s : tri ] we will examine the three detection schemes respectively , in order to show that all of them jointly measure the real and the imaginary part of the complex photocurrent @xcite @xmath1 being @xmath2 the signal mode and @xmath3 a probe mode . the inefficiency of realistic photodetectors will be also taken into account , in order to show it does not affect the equivalence of the considered detection schemes . in section refs : pom we will derive the output statistics of considered schemes as members of the more general class of two - photocurrent devices . section [ s : out ] will close the paper with some concluding remarks . the final stage of any detection scheme is represented by photodetection , namely counting photon through their conversion to electric pulses . let us consider a light beam @xmath4 entering a phototube which converts to electric pulses a fraction @xmath5 of the incoming photons . by keeping open the detector window for a time interval @xmath6 , the probability @xmath7 of counting @xmath8 photons is given by @xcite @xmath9^m } \over{m ! } } \exp [ -\zeta\hat i ( t)t]\nr\right\ } \label{pc - gen}\;,\end{aligned}\ ] ] where @xmath10 denotes normal ordering of operator and @xmath11 is the beam intensity @xmath12 @xmath13 denotes the positive ( negative ) frequency part of the field . for a single - mode field excited in a stationary state eq . [ pc - gen ] can be rewritten as @xmath14 being @xmath15=1 $ ] the single mode field operator and @xmath16 the global quantum efficiency of the photodetectors . for unit quantum efficiency this coincides with the actual photon number distribution of the state under examination @xmath17 whereas , in the realistic case of non - unit quantum efficiency eq . ( [ pc - singlemode ] ) becomes a binomial convolution @xcite . in formula @xmath18 let us now consider the scheme in fig . [ f : eta ] . the signal mode @xmath2 is impinged in a beam splitter of transmissivity @xmath19 whose second port @xmath3 is placed in the vacuum . a perfect ( @xmath20 ) photodetection on the exiting mode reveals @xmath8 photons with a probability @xmath21 being @xmath22 the identity operator on the second input of the beam splitter and @xmath23 the evolution operator of the beam splitter . a straightforward calculation shows that @xmath24 eq . ( [ conv_n2 ] ) coincides with eq . ( [ conv_n ] ) for @xmath25 . this means that a photodetection process with efficiency @xmath26 is equivalent to a perfect photodetection process performed on the original signal mixed with vacuum by a beam splitter with a value of the transmissivity equal to the quantum efficiency @xcite . in the following we will adopt this equivalent scheme . eight - port homodyne detector is known for a long time for the joint determination of phase and amplitude of the field in the microwave domain . it was introduced in the optical domain by walker and carrol @xcite and successively analyzed by different authors @xcite . a schematic diagram of the experimental setup is reported in fig . [ f : eig ] . there are four balanced beam splitters whereas a @xmath27 phase shifter is inserted in one arm . the four input modes are denoted by @xmath28 , @xmath29 whereas the detected output modes are denoted by @xmath30 , @xmath29 . there are four identical photodetectors whose quantum efficiency is given by @xmath26 . the _ noise _ modes used to take into account inefficiency , according to the scheme of the previous section , are denoted by @xmath31 , @xmath32 . we consider @xmath33 as the signal mode , whereas @xmath34 is referred to be the idler of the device . the mode @xmath35 is unexcited , whereas @xmath36 is placed in a highly excited coherent state @xmath37 provided by an intense laser beam ( local oscillator ) . the detected photocurrents are @xmath38 , @xmath29 which form the eight - port homodyne observables @xmath39 the latter are derived by rescaling the difference photocurrent , each of them obtained in an homodyne scheme . for this reason eight - port homodyne is known also as double - homodyne detection . in eq . ( [ 8phot ] ) @xmath26 denotes the quantum efficiency of the photodetectors whereas @xmath40 is the intensity of the local oscillator . in order to obtain @xmath41 and @xmath42 in terms of the input modes we first note that the input - output mode transformation is necessarily linear , as only passive components are involved in the detection scheme @xcite thus , we can write @xmath43 where the transformation matrix can be computed starting from the corresponding transformations for the beam splitters and the phase shifter @xcite @xmath44 \label{matrix}\:.\end{aligned}\ ] ] eqs . ( [ linear ] ) and ( [ matrix ] ) together with the equivalent scheme for the inefficient detector leads to the following expression for the output modes @xmath45&+\sqrt{1-\eta}\ ; u_1 \nonumber \\ b_2 = & \sqrt{\eta}\;\left[a_1 + ia_2 -a_3-i a_4\right]&+\sqrt{1-\eta}\ ; u_2 \nonumber \\ b_3 = & \sqrt{\eta}\;\left[a_1- a_2 + ia_3- a_4\right]&+\sqrt{1-\eta}\ ; u_3 \nonumber \\ b_4 = & \sqrt{\eta}\;\left[a_1- ia_2 -a_3 + i a_4\right]&+\sqrt{1-\eta}\ ; u_4 \label{output}\:.\end{aligned}\ ] ] upon inserting eqs . ( [ output ] ) in eq . ( [ 8phot ] ) and considering the limit of highly excited local oscillator we obtain the eight - port photocurrents in terms of the input modes @xmath46 + o[\frac{1}{|z| } ] \nonumber \\ \fl \widehat { \cal z}_2&= & -\hat a_{1 } ( \pi/2 ) + \hat a_{2 } ( \pi/2 ) + \sqrt{\frac{1-\eta}{\eta } } \left [ \hat u_{3 } ( \pi/2 ) - \hat u_{4 } ( \pi/2 ) \right ] + o[\frac{1}{|z| } ] \label{zphot}\:.\end{aligned}\ ] ] in eq . ( [ zphot ] ) @xmath47 denotes a quadrature of the field . the complex photocurrent @xmath48 is given by @xmath49 for unit quantum efficiency , whereas for non unit quantum efficiency it becomes a gaussian convolution of eq . ( [ 8zed ] ) , we will consider this point in detail in [ s : pom ] . it is worth noticing here that the mode transformation defined by eqs . ( [ linear ] ) and ( [ matrix ] ) is distinctive for a canonical @xmath50-port linear coupler as defined in refs . it has been rigorously shown @xcite that a @xmath51-port linear coupler can always be realized in terms of a number of beam splitters and phase - shifters . however , this implementation is , in general , not unique . the interest of eight - port homodyne scheme lies in the fact it provides the minimal scheme for realizing a @xmath50-multiport @xcite . if one considers the multiport as a given _ black - box _ device the eight - port homodyne scheme can be depicted as in fig . [ f : rough ] . this will facilitate the comparison with the six - port homodyne detection presented in section [ s : tri ] . heterodyne detection scheme is known for a long time in radiophysics . it has been introduced in the domain of optics @xcite in order to describe the joint measurement of two conjugated quadratures of the field . the term heterodyne is used as the involved field modes are excited on different frequencies . in fig . [ f : het ] we show a schematic diagram of the detector . we denote by @xmath52 the signal field , whereas @xmath53 describes the local oscillator . the field @xmath54 accounts for the losses due to inefficient photodetection . the input signal is excited in a single mode ( say @xmath2 ) at the frequency @xmath55 , whereas also the local oscillator is excited in only one mode at the frequency @xmath56 . this local oscillator mode is placed in a strong coherent state @xmath37 by means of an intense laser beam . the beam splitter has a transmissivity given by @xmath19 , whereas the photodetectors shows quantum efficiency @xmath26 . the heterodyne output photocurrents are given by the real @xmath57 and the imaginary @xmath58 part of the complex photocurrent @xmath59 . the latter is obtained after the rescaling of the output photocurrent @xmath60 , which is measured at the intermediate frequency @xmath61 . by fourier transform of eq . ( [ intensity ] ) we have @xmath62 being @xmath63 positive and negative part of the output field . in terms of the input fields ( [ hetero1 ] ) can be written as @xmath64 \nonumber \\ & & \left[\sqrt{\eta\tau } \hat{\e}^{(+)}_s ( \omega ' ) + \sqrt{\eta ( 1-\tau)}\hat{\e}^{(+)}_{lo } ( \omega ' ) + \sqrt{1-\eta}\hat{\e}^{(+)}_l ( \omega ' ) \right ] \;.\label{hetero2}\end{aligned}\ ] ] heterodyne photocurrent is obtained by the following rescaling @xmath65 physically this definition corresponds to require a very intense local oscillator , which however is allowed only for a little mixing with the signal mode @xcite . in this limit only terms containing the local oscillator field @xmath66 at the frequency @xmath56 can survive in eq . ( [ hetero2 ] ) , so that we have @xmath67 where @xmath68 in writing eq . ( [ hetero5 ] ) we have substituted @xmath69 for the relevant modes involved . provided @xmath70 and @xmath71 to be noise modes placed in the vacuum the expression ( [ hetero5 ] ) for the heterodyne photocurrents leads to the identical output statistics of eight - port homodyne photocurrents , the mode @xmath72 playing the role of the idler of the device . the full equivalence of the two detection schemes has been thus proved . a linear , symmetric three - port optical coupler is a straightforward generalization of the customary lossless symmetric beam splitter . the three input modes @xmath73 , @xmath74 are combined to form 3 output modes @xmath75 , @xmath76 . in analogy to lossless beam splitters , which are described by unitary 2@xmath772 matrices @xcite , any lossless triple coupler is characterized by a unitary @xmath78 matrix @xcite . for the symmetric case we have the form @xmath79 where each matrix element @xmath80 represents the transmission amplitude from the @xmath81-th input port to the @xmath82-th output port , that is @xmath83 . such devices have already been implemented in single - mode optical fiber technology and commercial triple coupler have been available for some time @xcite . any triple coupler can be also implemented by discrete optical components using symmetric beam splitters and phase shifters only @xcite . as it has already mentioned in section [ s : eig ] , this is due to remarkable mathematical fact that that any unitary @xmath84-dimensional matrix can be factorized into a sequence of 2-dimensional transformations plus phase - shift @xcite . moreover , this decomposition is not , in general , unique . in fig . [ f : ttt ] a possible implementation of a triple coupler is schematically reported . experimental realizations of triple couplers has been reported for both cases , the passive elements case and the optical fiber one @xcite . let us now consider the measurement scheme of fig . [ f : thd ] . the three input modes are mixed by a triple coupler and the resulting output modes are subsequently detected by three identical photodetectors . the measured photocurrents are proportional to @xmath85 , @xmath86 given by @xmath87 after photodetection a fourier transform ( ft ) on the photocurrents is performed @xmath88 this procedure is a straightforward generalization of the customary two - mode balanced homodyning technique . in that case , in fact , the sum and the difference of the two output photocurrents are considered , which actually represent the fourier transform in a two - dimensional space . by means of the identity @xmath89 for the periodic ( modulus 3 ) kronecker delta @xmath90 , we obtain our final expressions for the fourier transformed photocurrents @xmath91 @xmath92 gives no relevant information as it is insensitive to the phase of the signal field , whereas @xmath93 and @xmath94 are hermitian conjugates of each other and contain the relevant information in their real and imaginary part . in the following let us assume @xmath33 is the signal mode and @xmath34 is fed by a highly excited coherent state @xmath95 representing the local oscillator . for large @xmath40 the output photocurrents are intense enough to be easily detected . they can be combined to give the reduced photocurrents @xmath96 \nonumber \\ \widehat { \cal z}_2 & = & \sqrt{3 } \frac{\hat { \cal i}_2 - \hat { \cal i}_3 } { 2 i|z|}=-\hat a_{1 } ( \pi/2 ) + \hat a_{3 } ( \pi/2 ) + o[\frac{1}{|z| } ] \label{thp}\;,\end{aligned}\ ] ] which we refer to as the _ triple homodyne photocurrents_. again the complex photocurrent @xmath97 has the form @xmath98 , being @xmath33 the signal mode and @xmath35 the idler of the device . when accounting for the non unit quantum efficiency @xmath26 of the photodetectors the output modes are written as @xmath99 so that the reduced photocurrents are now given by @xmath100 + o[\frac{1}{|z| } ] \nonumber \\ \fl \widehat { \cal z}_2= \sqrt{3 } \frac{\hat { \cal i}_2 - \hat { \cal i}_3 } { 2 i\eta |z|}= -\hat a_{1 } ( \pi/2 ) + \hat a_{3 } ( \pi/2 ) + \sqrt{\frac{1-\eta}{\eta } } \left [ \hat u_{1 } ( \pi/2 ) - \hat u_{3 } ( \pi/2 ) \right ] + o[\frac{1}{|z| } ] \label{6zphot}\:.\end{aligned}\ ] ] when , as it is the case , the modes @xmath101 are placed in the vacuum the six - port photocurrents in eq . ( [ 6zphot ] ) leads to the same statistics of the eight - port photocurrents in eq . ( [ zphot ] ) . indeed , they describe different devices leading to the same amount of information on the signal mode @xmath33 . some comments are , however , in order . by comparison of fig . [ f : rough ] and fig . [ f : thd ] it appears obvious that six - port homodyne is an optimized scheme relative to the eight - port one . one mode less is needed to reach the same amount of information , thus decreasing the possible sources of noise . the reason for this lies in the final stage of the two schemes . the fourier transform of the six - port photocurrents , in fact , is a more effective procedure relative to the double - homodyne final stage of the eight - port scheme . this is related to the noise suppression mechanism of homodyne detection . the latter , in fact , is generalized to the multi - mode case by the fourier transform rather than duplication of the original two - mode scheme . in this section we analyze with some detail the output statistics of an abstract two - photocurrent device . the latter is characterized by the joint measurement of the real @xmath57 and the imaginary @xmath58 part of the complex photocurrent @xmath102 when equipped with perfect photodetectors . on the other hand , in the realistic case of inefficient photodetection the photocurrent is given by @xmath103 in eqs . ( [ zed ] ) and ( [ zeta ] ) @xmath2 and @xmath3 denote single mode radiation field which can be excited in any quantum state . the operators @xmath70 and @xmath71 also denote single mode radiation field , however strictly placed in the vacuum states . they are used to simulate losses due inefficient photodetection , according to the equivalent detection scheme of section [ s : eta ] . in the following we will refer to the mode @xmath2 as the signal mode which is under examination , whereas the mode @xmath3 is in a known and fixed state , playing the role of the probe mode of the device . this is only for sake of convenience . it is obvious that the roles of the two modes can be interchanged and any argument can be reversed considering the mode @xmath3 as a signal mode . according to this scheme we denote the outcome probability density distribution by @xmath104 . the latter describes , in the complex plane , the state of the mode @xmath2 as probed by the mode @xmath3 . indeed , different choices for the probe mode lead to very different features in the probability distribution . each experimental random outcome @xmath105 from the joint measurement of @xmath57 and @xmath58 can be considered as a point @xmath106 in the complex plane . on the other hand , the output photocurrent @xmath59 is expressed as a sum of different contributions coming from different modes . therefore , it appears intuitively rather obvious that the resulting probability distribution will be given by a convolution . to be more specific let us start by considering the ideal case of unit quantum efficiency @xmath20 . we write the probability distribution @xmath104 as the fourier transform @xmath107 of the characteristic function @xmath108\right\}\label{xi}\:,\end{aligned}\ ] ] being @xmath109 the global density matrix describing both modes @xmath2 and @xmath3 . we consider probe mode to be independent on the signal mode , so that the input mode is factorized as @xmath110 upon substituting eqs . ( [ zed ] ) and ( [ rho ] ) in eq . ( [ xi ] ) we are able to write the characteristic function @xmath111 as a product @xmath112 being @xmath113 \right\}\ ] ] the displacement operator and @xmath114 the single mode characteristics function , the latter entering in the definition of the wigner function of a single mode radiation field @xcite @xmath115 we now insert eq . ( [ pxi ] ) in eq . ( [ ft ] ) . by means of eq . ( [ wdf ] ) and using the convolution theorem we arrive at the final result @xmath116 the symbol @xmath117 denoting convolution . ( [ con ] ) provides a justification for referring to the mode @xmath3 as the probe of the device . in fact , it acts as a filter on the signal wigner function . the latter is not a genuine probability distribution , as it can be negative when describing quantum interference . thus , it can not be directly sampled by experiments . however , the convolution ( [ con ] ) make it more regular , leading to @xmath104 which is a measurable distribution . phase space density as in eq . ( [ con ] ) have been already introduced by wodkiewicz @xcite to account for the effect of the measuring apparatus in a joint measurement of conjugated variables . originally , they have been termed phase - space _ propensities_. more recently , they also have been used in entropic description of quantum mechanical states @xcite . two - photocurrent devices appear as the natural setup to start from , in order to experimentally access such kind of phase - space distribution . wigner function , though it contains a complete description of the quantum state , can not be directly measured . the phase distributions @xmath104 are smoothed version of it , corresponding to the occurrence of additional noise of purely quantum origin . indeed , two - photocurrent devices provide the generalized joint measurement of position and momentum , thus unavoidably introducing additional noise by first principles @xcite . the crucial point to notice here is that they offer the remarkable possibility to manipulate this quantum noise . as it emerges from eq . ( [ con ] ) it can be redirected in the desidered region of the complex plane by suitable choice of the probe mode , according to the kind of information of which is of interest . in order to incorporate the effects of inefficient detection we first note that eq . ( [ zeta ] ) differs from eq . ( [ zed ] ) by two additional _ additive _ terms . therefore , further convolutions to the original wigner function are expected . indeed , the characteristic function of the whole device is now expressed as @xmath118 where characteristic function for the noise modes can be easily evaluated as @xmath119 \qquad j=1,2 \label{noisexi}\:.\end{aligned}\ ] ] upon inserting eqs . ( [ pxieta ] ) and ( [ noisexi ] ) in eq . ( [ ft ] ) we obtain our final result , namely the output probability distribution @xmath120 of a two - photocurrent device equipped with photodetectors of quantum efficiency @xmath26 @xmath121 \label{noisecon}\;,\end{aligned}\ ] ] being @xmath122 the probability distribution obtained for ideal photodetection and @xmath123 conclusions ----------- in this paper we have introduced the class of two - photocurrent devices . this kind of detectors are characterized by the fact that they jointly measure the real and the imaginary part of the complex photocurrent @xmath124 , where @xmath2 and @xmath3 describe two single mode of the radiation field . eight - port homodyne , heterodyne and six - port homodyne detectors belong to this general class , thus their fully equivalence in probing radiation field has been proved . it has also been proved that this equivalence still holds when the inefficiency of the photodetection process is taken into account . this is an interesting and unexpected result , as photodetection takes place at very different stages in the three detection schemes . the three schemes analyzed in this paper are equivalent from the point of view of provided information on the measured signal . nevertheless , they have different physical implementations which have to be compared . we pointed out the advantages of the six - port homodyne scheme in comparison with the customary eight - port one . actually , it provides the minimal scheme to access generalized phase space distribution @xcite . in a two - photocurrent device a generalized joint measurement of position and momentum is performed on the signal mode . this results in a smoothing of the signal wigner function to a measurable distribution , which represents the output probability distribution of the measurement . some additional noise is unavoidably introduced , according to the heisenberg principle for joint measurement . however , the filtering process has been shown to be a convolution with the wigner function of the probe mode . therefore , it is possible to manipulate and redirect the noise . a suitable choice of the probe mode enhances different features of the signal phase space distribution , according to the kind of desidered information . i would thank prof . harry paul for his kind hospitality in the group nichtklassiche strahlung of max - planck society and prof . mauro dariano for introducing me to the subject of phase - space measurements . i would also thank valentina de renzi , ole steuernagel , gabriel drobny , alexei chizhov and alfred wnsche for profitable discussions . this work has been partially supported by the university of milan by a scholarship for postgraduate studies in foreign countries . 99 busch p , lahti p j 1995 _ riv . nuovo cim . _ * 18 * 1 yuen h p , chan v w s 1983 _ opt . * 8 * 177 yuen h p , shapiro j h 1980 _ ieee trans . inf . theory _ * it-26 * 78 shapiro j h , wagner s s 1984 _ ieee j. quant . _ * qe-20 * 803 shapiro j h 1985 _ ieee j. quant . _ * qe-21 * 237 walker n g , carrol j e 1984 _ electron . lett . _ * 20 * 981 walker n g , carrol j e 1986 _ opt . _ * 18 * 355 walker n g 1987 _ j. mod . opt . _ * 34 * 15 lai y , haus h a 1989 _ quantum opt . _ * 1 * 99 paris m g a , chizhov a , steuernagel o 1996 _ opt . comm._*134 * 117 zucchetti a , vogel w , welsch d g 1996 _ phys . _ a*54 * 856 dariano g m , sacchi m f 1995 _ phys . * r4309 kelley p l , kleiner w h , 1964 _ phys . rev . _ * 136 * 316 mandel l , wolf e 1995 _ optical coherence and quantum optics _ , ( cambridge university press ) leonhardt u , paul h 1993 _ phys . _ a*48 * 4598 dariano g m , macchiavello c , paris m g a 1995 _ phys . _ a*198 * 286 noh j w , fougeres a , mandel l 1991 _ phys . rev . lett . _ * 67 * 1426 noh j w , fougeres a , mandel l 1992 _ phys . _ a*45 * 424 noh j w , fougeres a , mandel l 1992 _ phys . _ a*46 * 2840 bandilla a , ritze h h 1993 _ quantum opt . _ * 5 * 213 freyberger m , schleich w p 1993 _ phys . _ a*47 * r30 leonhardt u , paul h 1993 _ phys . rev . _ a*47 * 2460 dariano g m , paris m g a 1993 _ phys . _ a*48 * r4039 dariano g m , paris m g a 1994 _ phys . _ a*49 * 3022 freyberger m , heni m , schleich w p 1995 _ quantum opt . _ * 7 * 187 paris m g a 1996 _ phys . _ a*53 * 2658 paris m g a 1996 _ opt . _ * 124 * 277 stenholm s 1994 _ j. mod _ * 41 * 2483 stenholm s 1995 _ appl . _ b*60 * 243 trm p , stenholm s , jex i 1995 _ phys . rev . _ a*52 * 4853 jex i , stenholm s , zeilinger a 1995 _ opt . comm . _ * 117 * 95 reck m , zeilinger a , bernstein h j , bertani p 1994 _ phys . lett . _ * 73 * 58 trm p , jex i , stenholm s 1996 _ j. mod . opt . _ * 43 * 245 yuen h p 1982 _ phys . _ a*91 * 101 paris m g a 1996 _ phys . _ a*217 * 78 campos r a , saleh b e a , teich m c 1989 _ phys . _ a*40 * 1371 mattle k , michler m , weinfurther h , zeilinger a , zukowski m 1995 _ appl . _ b*60 * s111 weihs g , reck m , weinfurther h , zeilinger a 1996 _ opt . * 21 * 302 sheem s k 1981 _ j. appl . * 52 * 3865 hariharan p , sen d 1959 _ j. sci . _ _ 36 _ 70 zernike f 1950 _ j. opt . * 40 * 326 wigner e p 1932 _ phys . _ * 40 * 749 cahill k e , glauber r j 1969 _ phys . _ * 177 * 1857 cahill k e , glauber r j 1969 _ phys . rev . _ * 177 * 1882 wodkiewicz k 1984 _ phys . lett . _ * 52 * 1064 wodkiewicz k 1986 _ phys . _ a*115 * 304 wodkiewicz k 1988 _ phys . _ a*129 * 1 buzek v , keitel c h , knight p l 1995 phys . a*51 * 2575 buzek v , keitel c h , knight p l 1995 phys . rev . a*51 * 2594 she c y , heffner h 1966 _ phys . rev . _ * 152 * 1103 arthurs e , goodman m s 1988 _ phys _ * 60 * 2447 paris m g a , steuernagel o 1996 , in preparation . ( 10,10)(-5,-5)(-2,-4.5)(0,1)0.6(-2,-3.5)(0,1)7.5 ( -4,2)(1,0)2(-4.5,-2)(1,0)0.6 ( -1.2,1.2)(-1,1)1.65(1.2,-1.2)(1,-1)1.65 ( -1.2,-1.2)(-1,-1)1.65(1.2,1.2)(1,1)1.65 ( -3.5,-2)(1,0)7.5(-4,-4.5)(1,0)2 ( -4.5,-4)(0,1)2(2,-4)(0,1)3.5 ( 2,.5)(0,1)3(2,3.9)(0,1)0.6 ( -4,2)(1,0)7.5(3.9,2)(1,0)0.6 ( 2,4.5)(1,0)2(4.5,2)(0,1)2 ( 2,3.5)(1,0).4(2,3.5)(-1,0).4 ( -2,-3.5)(1,0).4(-2,-3.5)(-1,0).4 ( -3.5,-2)(0,1).4(-3.5,-2)(0,-1).4 ( 3.5,2)(0,1).4(3.5,2)(0,-1).4 ( -4.5,-4.5)(2,0)(4.5,4.5 ) ( 2,3.5)(-2,-3.5 ) ( -3.5,-2)(3.5,2 ) ( -2,-1.3)(0,-1)0(-2,-3)(0,-1)0 ( 1,-2)(-1,0)0(2.7,-2)(-1,0)0 ( -1.3,-2)(-1,0)0(-3,-2)(-1,0)0 ( -2,1)(0,-1)0(-2,2.7)(0,-1)0 ( 2,1.3)(0,1)0(2,3)(0,1)0 ( 1.3,2)(1,0)0(3,2)(1,0)0 ( -1,2)(1,0)0(-2.7,2)(1,0)0 ( 2,-1)(0,1)0(2,-2.7)(0,1)0 ( 4.25,4.35 ) ( -4.75,-4.65 ) ( 1.7,-.1 ) ( 4.2,-2.15)@xmath35(-4.6,2)@xmath33 ( -1.5,-3.9)@xmath125(-4.1,-1.3)@xmath126 ( 2.5,3.7)@xmath127(3.9,1.2)@xmath128 ( 1.7,-4.5)@xmath36(-2.35,4.35)@xmath34 ( 15.00,20.00)(25.00,20.00)(15.00,30.00)(25.00,30.00 ) ( 15.00,40.00)(25.00,40.00)(15.00,50.00)(25.00,50.00 ) ( 14.00,50.00)(14.00,40.00 ) ( 14.00,30.00)(14.00,20.00 ) ( 52.00,50.00)(65.00,50.00)(65.00,40.00)(52.00,40.00 ) ( 52.00,30.00)(65.00,30.00)(65.00,20.00)(52.00,20.00 ) ( 65.00,47.00)(65.00,53.00)(65.00,43.00)(65.00,37.00 ) ( 65.00,33.00)(65.00,27.00)(65.00,23.00)(65.00,17.00 ) ( 65,20)(65,30 ) ( 65,40)(65,50 ) ( 68.00,50.00)(75.00,50.00)(75.00,40.00)(68.00,40.00 ) ( 68.00,30.00)(75.00,30.00)(75.00,20.00)(68.00,20.00 ) ( 89.00,45.00)(100.00,45.00)(81.00,23.00)(75.00,20.00 ) ( 85.00,45.00)(85.00,25.00 ) ( 100.00,45.00)(100.00,25.00 ) ( 100.00,25.00)(89.00,25.00)(75.00,50.00)(81.00,47.00 ) ( 75.00,40.00)(81.00,43.00)(75.00,30.00)(81.00,27.00 ) ( 9.,53.)(0,0)[cc]@xmath33(9.,43.)(0,0)[cc]@xmath34 ( 9.,33.)(0,0)[cc]@xmath35(9.,23.)(0,0)[cc]@xmath36 ( 57,53.)(0,0)[cc]@xmath129(57,43.)(0,0)[cc]@xmath130 ( 57,33.)(0,0)[cc]@xmath131(57,23.)(0,0)[cc]@xmath132 ( 72,53.)(0,0)[cc]@xmath128 ( 72,43.)(0,0)[cc]@xmath127 ( 72,33.)(0,0)[cc]@xmath126 ( 72,23.)(0,0)[cc]@xmath125 ( 105.00,45.00)(0,0)[lc]@xmath133 ( 105.00,25.00)(0,0)[lc]@xmath134 ( 85,45)(0,0)[cc]**(85,25)(0,0)[cc]* * ( 14,7)(-4.5,-3.5)(-2,0)(1,0)6.5(-2,-0.2)(1,0 ) ( -.8,-.2)(1,0)0(0,-2)(0,1)4 ( -.3,-2)(0,1)(.3,-2)(0,1 ) ( -.3,-.8)(0,1)0(.3,-.8)(0,1)0 ( -1,-1)(1,1)2(-.8,0)(1,0)0 ( 0,-.8)(0,1)0(1,0)(1,0)0 ( 0,1)(0,1)0(4.5,0 ) ( 4.5,-.5)(0,1)1(5,0)(1,0).5 ( 5.5,0)(1,1).8(5.5,0)(1,-1).8 ( 6.5,1)(6.5,-1 ) ( 6.7,1)(1,0)1.8(6.7,-1)(1,0)1.8 ( 8.5,1)(1,0)0(8.5,-1)(1,0)0 ( -3.,.4)@xmath135 ( -4.49,-.7)@xmath136 ( -.75,-3)@xmath137 ( 3,-1)(0,1)(3,0)(0,1)0 ( 3,-1.3)@xmath138 ( 2.8,0)(0,1)(2.8,0)(0,-1)0 ( 2,1.3)@xmath139 ( 6,1.5)@xmath140(6,-2)@xmath141 ( 8.65,1)@xmath142(8.65,-1)@xmath143 ( 143.33,53.00)(25.00,25.00 ) ( 50.33,50.00)(55.33,50.00)(50.33,40.00)(55.33,40.00 ) ( 50.33,30.00)(55.33,30.00)(55.33,30.00)(65.33,40.00 ) ( 55.33,40.00)(65.33,30.00)(65.33,40.00)(70.33,40.00 ) ( 65.33,30.00)(70.33,30.00)(55.33,50.00)(70.33,50.00 ) ( 78.33,30.00)(80.33,30.00)(80.33,30.00)(80.33,30.00 ) ( 80.33,30.00)(85.33,30.00)(80.33,40.00)(85.33,40.00 ) ( 80.33,50.00)(85.33,50.00)(80.33,50.00)(70.33,40.00 ) ( 70.33,50.00)(80.33,40.00)(70.33,30.00)(72.33,30.00 ) ( 85.33,50.00)(87.33,50.00)(85.33,40.00)(95.33,40.00 ) ( 85.33,30.00)(95.33,30.00)(93.33,50.00)(95.33,50.00 ) ( 95.33,50.00)(100.33,50.00)(95.33,40.00)(100.33,40.00 ) ( 95.33,30.00)(100.33,30.00)(100.33,40.00)(110.33,50.00 ) ( 100.33,50.00)(110.33,40.00)(110.33,50.00)(120.33,50.00 ) ( 110.33,40.00)(120.33,40.00)(100.33,30.00)(120.33,30.00 ) ( 120.33,30.00)(130.33,40.00)(120.33,40.00)(130.33,30.00 ) ( 130.33,30.00)(140.33,30.00)(130.33,40.00)(140.33,40.00 ) ( 120.33,50.00)(140.33,50.00)(90.33,50.00 ) ( 75.33,30.00)(50.33,50.00 ) ( 50.33,40.00)(50.33,30.00 ) ( 140.33,50.00)(140.33,40.00 ) ( 140.33,30.00 ) ( 57.33,34.00 ) ( 75.66,30.00)(0,0)[cc]@xmath144 ( 90.66,50.00)(0,0)[cc]@xmath145 ( 145,50.00)(0,0)[cc]@xmath146 ( 145,40.00)(0,0)[cc]@xmath147 ( 145,30.00)(0,0)[cc]@xmath148 ( 46.33,50.00)(0,0)[cc]@xmath149 ( 46.33,40.00)(0,0)[cc]@xmath150 ( 46.33,30.00)(0,0)[cc]@xmath151 ( 67.00,36.00)(0,0)[cc]bs ( 82.00,46.00)(0,0)[cc]bs ( 112.00,46.00)(0,0)[cc]bs ( 132.00,36.00)(0,0)[cc]bs ( 9.34,36.33)(1,0)10.00(64.67,36.00)(5,6)9.00 ( 73.67,46.66)(1,0)10.00(83.67,46.66)(5,-6)8.67 ( 92.34,36.33)(-4,-5)8.33(84.00,26.00)(-1,0)10.33 ( 73.67,26.00)(-5,6)8.33(83.67,46.66)(4,3)5.67 ( 89.34,51.00)(1,0)15.33(92.00,36.00)(1,0)12.33 ( 84.00,26.00)(5,-4)5.33(89.34,21.66)(1,0)15.33 ( 73.34,26.00)(-1,-1)4.33(9.34,51.33 ) ( 9.34,36.33)(9.34,21.99 ) ( 9.34,36.33)(9.34,51.33 ) ( 105.00,51.00)(105.00,36.00 ) ( 105.00,21.66)(73.34,46.66 ) ( 84.00,46.66)(92.34,36.00 ) ( 84.34,26.33)(65.34,36.33 ) ( 73.34,26.00 ) ( 78.00,36.00)(0,0)[cc]ft ( 6.67,53.83)(0,0)[cc]@xmath149 ( 6.67,38.83)(0,0)[cc]@xmath150 ( 6.67,24.44)(0,0)[cc]@xmath151 ( 37.67,53.83)(0,0)[cc]@xmath146 ( 37.67,38.83)(0,0)[cc]@xmath147 ( 37.67,23.83)(0,0)[cc]@xmath148 ( 57.67,54.33)(0,0)[cc]@xmath152 ( 57.67,39.33)(0,0)[cc]@xmath153 ( 57.67,24.99)(0,0)[cc]@xmath154 ( 107.67,54.33)(0,0)[cc]@xmath155 ( 107.67,39.33)(0,0)[cc]@xmath156 ( 107.67,24.99)(0,0)[cc]@xmath157 ( 68.67,51.00)(1,-1)4.67(9.34,51.33)(1,0)10.33 ( 9.34,22.00)(1,0)10.00(44.67,36.33)(-1,0)12.00 ( 41.34,36.33)(1,0)6.30 ( 47.67,36.33 ) ( 47.67,22.00 ) ( 47.67,51.08 ) ( 47.67,47.58)(0,1)7.00(47.67,32.83)(0,1)7.00 ( 47.67,18.50)(0,1)7.00(51.34,36.33)(1,0)14.00 ( 68.67,51.00)(-1,0)17.67(69.00,21.66)(-1,0)17.89 ( 51.67,21.66)(0,1)0.00(47.67,51.00)(-1,0)13.67 ( 47.67,21.66)(-1,0)13.00(45.67,51.00 ) ( 45.67,21.66)(45.67,36.33 ) ( 53.67,51.00)(53.67,36.33 ) ( 53.67,21.66 )

a wide range of soft materials exhibit slow relaxations strongly reminiscent of the glassy dynamics observed in hard condensed matter . these materials include concentrated or strongly interacting colloidal suspensions , emulsions , surfactant systems , polymeric or colloidal gels , and foams @xcite . most of these materials are out of thermodynamical equilibrium : for systems far from equilibrium , a continuous evolution of the mechanical and dynamical properties is observed , in analogy with the slowing down of the dynamics in aging hard glasses . several different terms are used in the literature to designate out - of - equilibrium soft materials , depending on the context and the authors views : gels , glasses , and jammed @xcite systems are but a few examples . here , we will term them quite generally `` soft glassy systems '' @xcite or simply soft glasses . soft glasses are non - ergodic systems , since they are able to explore only a restricted portion of the total phase space . experimentally , however , determining whether or not a given system is non - ergodic may prove to be a difficult task , since only a limited range of length and time scales can be probed . in the following we will therefore use the term `` soft glassy system '' with a wider connotation , including also materials that are ergodic but still exhibit very slow relaxations . soft glassy systems attract a wide interest . on the one hand , they are ubiquitous in industrial applications , for example in the food , cosmetic , paint , pharmaceutical , and oil recovery industry . for most applications , understanding and controlling their dynamical and rheological properties , as well as their long term evolution and stability is of fundamental importance . on the other hand , soft glasses are much studied at a more fundamental level . for example , systems such as colloidal hard spheres are taken as model systems for investigating the general behavior of glasses , their relevant time and length scales being more easily accessible than in molecular glasses . moreover , soft systems allow one to control in great detail and to significantly vary the interactions between their constituents . therefore , it is possible to investigate how different interactions influence the dynamical behavior of soft glassy materials and the generality of the behavior observed for model systems can be tested . until very recently , the experimental characterization of the dynamics was limited to relatively short time scales ( e.g. up to a few hundred seconds in light scattering experiments ) and only the average dynamics were accessible . theoretically , very few quantitative models were available , and their applicability was restricted to a few model systems ( e.g. the mode coupling theory , mct @xcite , applied to colloidal suspensions interacting via a hard sphere potential ) . recent advances include progress both on the conceptual and experimental side . the concept of jamming @xcite was proposed to describe in a unified way albeit somehow qualitatively the ergodic to non - ergodic transition in many systems , including molecular glasses , granular materials , and colloids . on the other hand , the mct has been extended to include concentrated systems with attractive interactions ( for a review see reference @xcite ) and attempts have been made to extend its validity to more diluted attractive systems @xcite , thus bridging the gap between colloidal gels and glasses . thus , both the jamming picture and the mct attempt to give a unified vision of the ergodic to non - ergodic transition , the scope of the former being wider and the latter providing more quantitative predictions . finally , dynamical heterogeneity has emerged as a key feature of the slow dynamics of glassy systems @xcite and many theories have been developed based on or incorporating dynamical heterogeneity . on the experimental side , recent advances include both the design of new systems with interactions that can be fine - tuned and the development of new methods and techniques to measure very slow and heterogeneous relaxations in out - of - equilibrium systems . experimentally , controlling the transition to non - ergodicity ( possibly in a reversible way ) is often quite difficult . as a consequence , systems for which a transition from a fluid to a solid state can be reversibly obtained are particularly interesting . they include thermoresponsive systems such as thermoreversible sticky hard spheres @xcite or emulsions @xcite , swellable microgel particles @xcite , star polymers @xcite , and surfactant systems @xcite . the most popular techniques for probing the dynamics of soft systems are probably optical microscopy and dynamic light scattering : both of them have benefited of major improvements in the last years . time - resolved confocal microscopy , together with digital image processing , allows the trajectories of thousands of fluorescently dyed particles to be measured simultaneously @xcite , thus providing detailed information on the dynamics at a microscopic level . motion on very long time scales ( up to several days ) has been measured by bleaching a cubic volume of the sample and by measuring the fluorescence recovery due to the diffusion of the unbleached particles @xcite . in light scattering experiments , information on the sample dynamics are obtained by measuring the time autocorrelation function of the scattered intensity , @xmath0 . in the single scattering regime @xcite , the dynamic structure factor ( or intermediate scattering function ) @xmath1 is proportional to @xmath2 , where the intensity autocorrelation function is measured at a scattering vector @xmath3 and the proportionality factor depends on the setup detection . the advent of high brilliance coherent x - ray sources has made possible x - photon correlation spectroscopy ( xpcs ) @xcite , which extends dynamic light scattering experiments to much higher values of @xmath4 . for highly multiply scattering samples , the diffusing wave spectroscopy ( dws ) @xcite formalism allows the particle mean squared displacement to be extracted from @xmath0 , provided that the dynamics be homogeneous . until recently , the applicability of light scattering and xpcs techniques to soft glasses has been severely limited by the need to average the measured intensity autocorrelation function over a duration much longer than the system relaxation time . this limitation has been overcame by the so - called multispeckle technique @xcite , where a multielement detector is used ( typically a ccd camera ) and the intensity autocorrelation function is averaged not only over time , but also over distinct speckles corresponding to different elements of the detector . this scheme allows one to drastically reduce the duration of a measurement , thus making dynamic scattering techniques suitable for systems with very slow or non - stationary dynamics . a conceptually similar approach is followed in the so - called interleaved method , where a point detector is used and the sample is slowly rotated in order to illuminate sequentially the detector with different speckles @xcite ; for each speckle the intensity correlation function is calculated for time delays multiple of the period of rotation of the sample and @xmath5 is finally obtained by averaging over all speckles . a recent implementation of this scheme , more simple and efficient , is described in reference @xcite . to conclude this brief overview of recent advances in scattering methods , we remark that most scattering experiments probe the average dynamics and hence are not sensitive to dynamic heterogeneities . information on temporal heterogeneity of the dynamics can be extracted from higher order correlation functions @xcite or using the time resolved correlation ( trc ) method @xcite , as it will be discussed in some detail in section [ sec : hetero ] . in this paper we will review the experimental work of the last few years on soft glassy systems , with an emphasis on microscopy , light scattering and , to a less extent , rheology experiments . references will be made also to some of the numerical and theoretical investigations related to experiments , as well as to experiments on granular materials that share intriguing analogies with soft glasses . the literature is very abundant , reflecting the growing interest in this field . we have chosen to focus on four areas that seem to us particularly interesting and promising . deals with the observation of two types of colloidal glasses ( repulsive and attractive ) , and the success of the mct in modelling the slow dynamics of these systems . systems very far from equilibrium , where stress relaxation appears to play a major role in the slow dynamics are reviewed in section [ sec : outofequilibrium ] . section [ sec : perturb ] discusses the effect of a perturbation on a glassy system , both in the linear and in the non - linear regime . finally , dynamical heterogeneities in the slow dynamics are addressed in section [ sec : hetero ] . hard spheres have been investigated as a model system for the glass transition for many years . experimentally , colloidal particles such as poly(methyl methacrylate ) ( pmma ) spheres sterically stabilized and suspended in an organic solvent have been shown to behave as hard spheres , forming a glass phase when the volume fraction , @xmath6 , becomes larger than @xmath7 @xcite . on the other hand , the mode - coupling theory ( mct ) introduced by gtze @xcite provides detailed predictions on the dynamics of hard spheres when approaching the glass transition . for a supercooled liquid of hard spheres and below @xmath8 . although in some experiments centrifugation is used to realize such a quench , in most cases supercooled samples are prepared by shear melting . ] the mct predicts a two - step decay of the dynamic structure factor @xmath1 . the initial relaxation ( @xmath9 relaxation ) is due to the rattling of the particles inside the cage formed by their neighbors . the @xmath9 relaxation is followed by a plateau and a second relaxation ( @xmath10 relaxation ) , due to cage - escape processes . at the glass transition , the ideal mct predicts that the characteristic time of the @xmath10 relaxation diverges and thus that the plateau extends to all times , indicating a complete dynamical arrest . note however that on very long time scales ergodicity may be restored by thermally activated processes ( hopping ) , and attempts have been done to include such processes in the mct @xcite . good agreement has been found between the mct and dynamic light scattering measurements of the dynamic structure factor of hard sphere colloidal suspensions ( for a review on experimental tests of the mct , see @xcite ) , although the critical packing fraction @xmath11 predicted by the mct is smaller than the experimentally measured glass transition volume fraction @xmath12 . to overcome this discrepancy , the @xmath13 volume fractions @xmath14 and @xmath15 are often used when comparing experiments and theoretical or numerical work @xcite . in hard spheres , the dramatic slowing down of the dynamics when approaching @xmath8 is due to the cage effect : the motion of any given particle is increasingly hindered by its neighbors as the particles are packed more tightly . the glass transition is therefore driven by the repulsive ( excluded volume ) interaction between the spheres , and the arrested phase thus formed is termed _ repulsive _ glass . recent theoretical @xcite and experimental @xcite work has shown that the addition of short - ranged , low - energy attractive interactions can lead , surprisingly , to the melting of such a repulsive glass . this can be understood in the framework of the cage effect : particle bonding due to attractions results in the increase of the available free volume , thus loosening and eventually opening the cage . if the strength of the attraction is increased further , however , a new arrested phase is formed , because the bonds are sufficiently long - lived to effectively confine the particles . this arrested phase is referred to as an @xmath16 glass , although the term gel is also found in the literature . for a short general introduction to the topic of attractive @xmath17 repulsive glass , see @xcite . a brief and clear review of the experimental work on this topic is given in @xcite , while reference @xcite discusses recent theoretical advances . in the following , we will briefly recall the main results of mct and then focus on the experimental work . theoretical phase diagram for hard spheres of radius @xmath18 with a square well attractive interaction of width @xmath19 and depth @xmath20 ( data from reference @xcite ) . the different curves are labelled by the relative well width , @xmath21 . note the reentrant shape of the lines separating the fluid region to the glassy regions for @xmath22 . for @xmath23 a line separating the attractive and the repulsive glass phases ends in the @xmath24 singular point . the @xmath25 singular point ( circled cross ) is estimated to be located on the @xmath26 line ( not shown for clarity ) . ] shows the theoretical phase diagram for particles interacting via a hard - core repulsion ( @xmath27 for @xmath28 , where @xmath29 is the inter - particle potential and @xmath18 is the particle radius ) and a square well attraction ( @xmath30 for @xmath31 ) @xcite . the different curves are labelled by the relative width of the attractive well , @xmath32 . although the exact location of the lines depends on the detailed shape of the potential and the chosen approximated expression for the static structure factor , the features shown in are quite general . the lower part of the phase diagram corresponds to strong attractions ( @xmath33 ) and is occupied by the attractive glass . the upper - right region is occupied by the repulsive glass , while the upper - left part corresponds to the fluid phase . note that at small enough @xmath34 ( typically less than about 0.1 ) the phase diagram has a reentrant shape : there exists a range of volume fractions for which , at constant @xmath6 , a sample may be in the fluid phase or in either the attractive or the repulsive glass phase , depending on the strength of the attractions . for very short - ranged attractions ( @xmath35 less than about 0.04 ) , at high @xmath6 the attractive and repulsive glass regions are contiguous , separated by a line ending in the so - called @xmath24 singular point ( open square with plus in ) . when @xmath34 increases , the line separating the attractive and the repulsive glasses progressively shrinks , until the line and the @xmath24 singularity vanish and the so - called @xmath25 singularity is observed ( open circle with cross in ) . in the vicinity of the @xmath24 and @xmath25 singular points , as well as at the intersection of the attractive and repulsive glass lines , the system is predicted to relax with a very broad distribution of time scales , resulting in a logarithmic decay of the dynamic structure factor @xmath1 and a power law behavior of the particle mean squared displacement @xcite . simulations performed by using either a square - well @xcite or the asakura - osawa potential @xcite , have confirmed the predictions of the mct . experimentally , the more extensive tests of the two glasses scenario have been performed by three independent groups . pham and coworkers @xcite have used nearly monodisperse , sterically stabilized pmma particles to which polystyrene , a non - adsorbing polymer , was added to induce a short - ranged attraction via the depletion effect , leading to an attraction well described by the asakura - osawa potential ( for a review on this much studied experimental system , see @xcite ) . the strength of the attraction can be controlled by changing the amount of the polymer , while the range is of the order of the radius of gyration of the polymer . in references @xcite the phase diagram of this system is established and both the structure and the dynamics of the glass phases are studied in detail by means of light scattering . the so called two - color technique @xcite is used to suppress multiple scattering , and a recently introduced `` echo '' scheme @xcite is used to obtain properly ensemble - averaged measurements even when dealing with non - ergodic samples . as expected from the mct , at fixed @xmath6 the non - ergodicity or debye - waller factor @xmath36 for the attractive glass is found to be higher than that for the repulsive glass , because the motion of clustered particles is more restrained than that of repulsive spheres , which can rattle in the cage formed by their neighbors . accordingly , the liquid - order peak in the static structure factor , @xmath37 , shifts to larger @xmath4 vectors when going from the repulsive to the attractive glass . additionally , @xmath37 is less sharp for attractive glasses , since they are locally more disordered than repulsive glasses . in the vicinity of the @xmath24 point , a logarithmic decay of @xmath1 is observed over more than three decades in time . beyond the @xmath24 point , where the mct predicts that the attractive and repulsive glasses should merge into a single arrested phase , the non - ergodicity factor becomes independent of attraction strength , suggesting that indeed the two glasses phases are indistinguishable . however , quite intriguingly , the short - time behavior of @xmath1 remains different , a feature also reported in reference @xcite . similar detailed investigations , leading to comparable results , are presented by eckert and bartsch and by chen and coworkers . eckert and bartsch study a concentrated suspension of microgel particles , to which linear polystyrene is added to induce an attraction via the depletion effect @xcite . the microgel particles are bidisperse so as to suppress crystallization . this two - component system is mapped onto an effective ( polydisperse ) one - component system of slightly soft spheres ( the repulsive barrier is modelled by an inverse power potential @xmath38 ) , and is studied by static and dynamic light scattering . the multispeckle technique @xcite is used to access the long time dynamics and for studying non - ergodic samples . chen and coworkers @xcite investigate by small angle neutron scattering ( sans ) and dynamic light scattering ( dls ) a concentrated aqueous solution of l64 , a triblock copolymer of the pluronics family . l64 is composed of two end sections of polyethylene oxide ( peo ) and a central section of polypropylene oxide ( ppo ) . both ppo and peo become increasingly hydrophobic when the temperature @xmath39 is raised , but ppo does so at a faster rate . as a consequence , when @xmath39 is increased micelles are formed with a ppo core and a peo corona . at high enough temperature , the peo corona becomes sticky because of hydrophobic interactions , thus providing an attraction between the micelles . the experimental control parameters for this system are @xmath39 and the block copolymer concentration , @xmath40 . for comparison with the mct phase diagram , the @xmath39-@xmath40 experimental phase diagram is mapped onto a @xmath6-@xmath20 diagram by fitting the static structure factor obtained by sans to that predicted for adhesive hard spheres . note that the dls data are taken at a @xmath4 vector much smaller than that corresponding to the peak of @xmath37 , in contrast to the work of pham @xmath41 @xmath42 and eckart and bartsch . several features of the two glasses scenario predicted by the mct have been observed also in other systems . references @xcite report a rheology , sans and dynamic light scattering investigation of a solution of soft spheres , made of diblock copolymer micelles . an attraction can be induced by synthesizing diblock polymers with a controlled amount of stickers ( ethyl acrylate units ) . two distinct glass phases and a reentrant behavior are observed . the rheology experiments show that the elastic modulus @xmath43 of the attractive glass is larger than that of the repulsive glass , a feature predicted by the mct . note , however , that for these soft objects the difference in @xmath43 for attractive and repulsive glasses ( about a factor of 2 ) is much less marked than for attractive hard spheres ( more than a decade , see @xcite ) . the melting of a glass of soft spheres ( star polymers ) and a reentrant gelation upon the addition of a linear polymer has been observed by stiakakis and coworkers @xcite . although the phenomenology is strongly reminiscent of that observed in other systems with depletion interactions , the authors point out that the melting is due to the decrease of the range of the soft repulsion between stars upon addition of the linear polymer , while the gelation results from bridging flocculation induced by the long polymers chains . finally , the existence of a reentrant glass transition in laponite suspensions has been speculated , but not yet proved experimentally @xcite . qualitatively , the experiments mentioned above strongly support the mct picture of attractive and repulsive glassy states and show that this scenario is robust with respect to changes in the exact shape of the potential . a quantitative comparison is however much more difficult . indeed , even for the experimental system closer to the theoretical models , i.e. the hard spheres with depletion interactions studied by pham and coworkers , it is difficult to exactly locate a sample on the theoretical phase diagram , because of uncertainties in the parameters describing the potential and in the relative volume fraction @xmath14 of a sample @xcite . a comparison of the experimental and theoretical phase diagrams is given in reference @xcite , showing that the mct somehow underestimates the extent of the fluid pocket in between the attractive and repulsive glass regions . in view of the success of the mct in modelling suspensions of attractive particles at high volume fraction , attempts have been made to extend it to lower @xmath6 , in order to describe as a glass transition the `` weak gelation '' observed in attractive systems where particle bonds are non - permanent . in the `` cluster mode - coupling theory '' of kroy and coworkers @xcite , the mct is applied to clusters of particles , rather than to the particles themselves . this theory predicts the existence of a cluster phase that may freeze if the effective cluster volume fraction is large enough . recent simulations @xcite and experiments @xcite are consistent with this picture . a more detailed discussion of the delicate interplay between aggregation , dynamical arrest and phase separation that arises in relatively low @xmath6 suspensions is given in @xcite . for a discussion of colloidal gelation in connection to the glass transition , see also the review by trappe and sandkhler @xcite , which focuses on the elastic properties of these materials . in spite of the large amount of theoretical , numerical and experimental work of the last years , several questions are still open . we mention here two topics that are likely to focus the interest of researchers in the next future and are connected to the subject of this review : dynamical heterogeneities and aging . simulations on high @xmath6 attractive glasses suggest that , although the mct correctly captures the average dynamics , the particle motion is very heterogeneous @xcite , a feature also observed at intermediate volume fractions ( @xmath44 ) @xcite . new experiments are needed to characterize dynamical heterogeneity in attractive glasses and the new approaches described in may prove useful . no systematic experimental investigation of the aging behavior in attractive @xmath17 repulsive glasses is available yet , although preliminary dls data are presented in reference @xcite . they suggest a different aging in attractive and repulsive glasses : for the former , the dynamics keep slowing down for up to 10 days , while no significant aging effects beyond 2 days are reported for the latter . the destabilization of the attractive glass with time , due to activated dynamics ( bond breaking ) , has also been investigated in a series of numerical works @xcite . the slow dynamics of out - of - equilibrium soft materials have been investigated experimentally for a wide variety of systems , ranging from model systems of hard spheres to more complicated materials . quite generally , these experiments demonstrate that the dynamics at very long time scales are not completely frozen : many soft materials in the glassy phase have been shown to exhibit ultra - slow relaxations . after reviewing these investigations , we will describe the aging behavior that is very often associated with the slow dynamics . finally , internal stress relaxation will be discussed as a possible mechanism for slow dynamics and aging in many soft glassy materials very far from equilibrium . as discussed in , the ideal mct predicts the divergence of the @xmath10 relaxation time at the glass transition : therefore for @xmath45 the dynamic structure factor should exhibit a plateau after the initial decay ( @xmath9 relaxation ) . the pioneering dynamic light scattering experiments on colloidal hard spheres glasses of van megen and coworkers @xcite hinted however at a second decay of @xmath1 , after the plateau . unfortunately , the accessible time delays were too short to properly characterize this slow relaxation . qualitatively similar results are obtained by pham and collaborators @xcite . recent experiments confirm indeed that a hard sphere glass is not a fully arrested state and cage - escape plays a decisive role even in the glass phase . using a confocal scanning microscope , simeonova and kegel @xcite follow in real - time the fluorescence recovery after photo - bleaching of a small volume within a concentrated suspensions of hard spheres and derive from their measurements the mean squared displacement ( msd ) , @xmath46 , of the colloids . in the glass phase , they find a subdiffusive behavior @xmath47 , which extends over seven decades in time . the largest displacements are of the order of the particle size . this result is intriguing : it is fundamentally different from the ideal mct predictions and is moreover different from the light scattering results of van megen _ et al . _ , who observe an intermediate plateau of the msd , before detecting a further increase of @xmath46 at very large @xmath48 . interestingly , thanks to their extremely long experiments , simeonova and kegel have recently demonstrated that the subdiffusive power law growth of the msd corresponds actually to a transient regime ( lasting up to @xmath49 brownian times , or about @xmath50 days ! ) , after which a plateau is eventually reached @xcite . apart from these few reports , experiments probing the slow dynamics of hard spheres systems deep in the glass phase are still scarce . other systems , however , provide an opportunity to investigate in details the dynamical state of soft materials far from the fluid - to - solid transition . indeed , for several soft materials , the characteristic times are intrinsically smaller than that of hard - spheres systems , and a quench to a state far from equilibrium may be realized more easily . these systems include fractal colloidal gels , dilute suspension of colloidal charged platelets , concentrated suspensions ( or pastes ) of colloidal particles stabilized both by electrostatic and steric repulsions , and compact arrangements of soft deformable particles ( multilamellar vesicles or polyelectrolyte microgels ) , the volume fraction of the latter being essentially @xmath51 . thanks to the multispeckle dynamic light scattering method @xcite , the dynamics have been probed for time delays as long as tens of hours . a remarkable result of these experiments is that , for several soft glassy materials , the intensity autocorrelation function @xmath52 relaxes completely to zero at very long @xmath48 , a behavior indicative of structural changes over the length scale probed by the experiment . these findings have been obtained by dynamic light scattering both in the multiple scattering ( diffusing wave spectroscopy , dws @xcite ) and in the single scattering regime ( dls ) @xcite , as well as by x - photon correlation spectroscopy ( xpcs ) @xcite . diffusing wave spectroscopy measurements are performed on very turbid samples , or on transparent systems to which probe particles are added to increase the turbidity . using dws , a slow relaxation well separated from the fast dynamics by a plateau in @xmath52 has been measured for systems as varied as dilute suspensions of charged clay platelets ( laponite ) forming a `` wigner glass '' @xcite , colloidal pastes @xcite , and concentrated colloidal gels @xcite . the quantitative interpretation of the decay of @xmath52 is not straightforward : the standard dws formalism allows the msd to be calculated from @xmath0 , but it requires the motion of all scatterers to be spatially and temporally homogeneous @xcite , a condition that is likely to be violated in many glassy systems ( see ) . it is also worth noting that in these experiments the length scale probed is at most of the order of a few tens of nanometers , much smaller than the typical size of the sample constituents . qualitatively similar features have been observed in single scattering experiments on a variety of systems , including fractal colloidal gels with strong and irreversible @xcite or weak , thermoreversible @xcite bonds between the particles , compact arrangements of soft elastic spheres @xcite or of emulsion droplets @xcite , polycrystals made of block copolymer micelles @xcite , and suspensions of laponite @xcite . note that dls probes length scales much larger than those explored in a dws experiment , typically up to the size of the sample constituents or more . the full decay of the dynamic structure factor extracted from dls experiments is therefore particulary significative , since it is indicative of slow rearrangements that eventually change the system configuration on length scales comparable to or larger than the size of its constituents . an additional valuable feature of dls is the possibility to easily measure how the dynamics depend on length scale . this is accomplished by changing the @xmath4 vector at which the experiment is performed . several different @xmath4-dependencies of the characteristic time @xmath53 for the slow relaxation have been measured . abou and coworkers @xcite have found @xmath54 for a suspension of clay particles ( laponite ) , as expected for a diffusive process , such as cage - escape mechanisms . however , less intuitive scaling laws have also obtained by several groups on various samples . solomon and coworkers @xcite reported @xmath55 for a colloidal gel of adhesive particles , while bellour and coworkers have found for the same material as abou _ et al . _ but in a different aging regime , see also subsection [ subsec : aging] @xmath56 . these results remain largely unexplained . an intriguing @xmath57 scaling has been measured for a wide variety of materials , including both attractive systems ( fractal colloidal gels @xcite ) and repulsive systems ( compact arrangements of soft elastic spheres @xcite or of emulsion droplets @xcite , a polycrystal made of copolymer micelles @xcite , and laponite in xpcs experiments that probe the high @xmath4 regime @xcite ) . this scaling suggests that the slow dynamics be due to a `` ballistic '' motion of the particles , in the sense that the average particle displacement grows linearly with time ( this ultra - slow regime should not be confused with the ballistic motion of colloids observed on very short time scales , before the onset of brownian motion @xcite ) . a very unusual shape of the dynamic structure factor accompanies always this scaling : @xmath1 decays as a `` compressed exponential '' , i.e. @xmath58 $ ] , with a `` compressing '' exponent larger than one , @xmath59 . illustrates both the @xmath57 scaling and the compressed exponential shape of @xmath60 for the colloidal fractal gel studied in @xcite . note that the same compressed exponential shape was observed in reference @xcite , although with a somehow smaller value of @xmath61 . a ) : dynamic structure factor measured simultaneously at various scattering vectors for a fractal gel made of strongly attractive polystyrene particles , with @xmath62 . the curves are labelled by the magnitude of the scattering vector in @xmath63 . b ) : the sama data as in a ) are plotted as @xmath64 $ ] @xmath17 @xmath65 . all data collapse onto a straight line , thus demonstrating that @xmath66 $ ] , with @xmath67 . c ) : double logarithmic plot of the @xmath4 dependence of the relaxation time @xmath53 . the line has a slope of @xmath68 , suggesting a @xmath57 scaling . to avoid overcrowding the figure , not all the curves that were fitted to obtain the data shown in c ) are plotted in a ) and b ) . data taken from reference @xcite . ] because the relaxation is faster than exponential and due to the @xmath57 scaling , all models based on cage - escape processes must be ruled out . instead , we have proposed @xcite the dynamics to be due to randomly distributed internal stress sources acting on the sample , whose response is that of an elastic solid . simple scaling arguments lead then to the observed peculiar shape of the dynamic structure factor @xcite . a more refined model based on the microscopic description of local rearrangement events has been developed for the case of a colloidal gel @xcite . experiments testing the notion of internal stress as the driving force for the slow dynamics are reviewed in sub . finally , we note that there are strong indications that in this type of systems the dynamics is heterogeneous . the unusual dynamics reported above may therefore be due to a series of discrete rearrangement events , as it will be discussed in more details in . the properties of out - of - equilibrium systems continuously change with time as the system slowly evolves towards its equilibrium configuration . as a consequence , correlation functions , e.g. measured by dynamic light scattering , and response functions , e.g. stress or strain relaxation in rheological experiments , depend not only on time delay , @xmath48 , as for time - translation invariant systems , but also on the waiting time or sample age , @xmath69 . typically , the dynamics becomes progressively slower as the sample ages and the system spends more and more time in the metastable states it visits , the energy of such states becoming increasingly lower ( see references @xcite for an introduction to aging phenomena ) . very generally , age is defined as the time elapsed since the sample was quenched from a fluid to an out - of - equilibrium solid state . for hard condensed matter systems ( e.g. structural , polymeric or spin glasses ) , this is usually realized by means of a temperature quench . by contrast , soft materials are very often ( albeit not always ) initialized by applying a strong mechanical shear ; hence age is defined as the time interval between the end of the applied shear and the beginning of the measurement . aggregation phenomena leading to a percolated network may also provide a route to an out - of - equilibrium state : in this case , @xmath70 is taken as the percolation ( or gelation ) time . the first experimental reports of aging phenomena in soft condensed matter are light scattering experiments performed on hard spheres colloidal glass @xcite . although the accessible time window in these experiments was too short to investigate in details the aging dynamics , the data unambiguously show a slowing down of the dynamics as the sample ages , when the volume fraction of the sample is larger than @xmath8 . by contrast , no dependence of the dynamic structure factor with sample age can be detected when @xmath6 is smaller than @xmath8 . hence , the results of van megen and coworkers nicely demonstrate the dramatic dynamical differences between a sample in the supercooled fluid state and one in the glass state , in spite of the minute differences in sample composition . more recently , simeonova and kegel @xcite , using the fluorescent recovery after photobleaching technique , have found clear evidence of aging behavior in hard sphere glasses , since they measure a particle msd whose growth slows down with increasing @xmath69 . similar results have been obtained by courtland and weeks @xcite , who used confocal microscopy to follow in real - time the three - dimensional motion of individual particles in a colloidal glass . they find that the msd initially grows linearly with time , then tends to plateau , and finally exhibits a slight upturn at very large time ( similarly to what observed for the cage escape process in supercooled fluids ) . for glasses , the time scale of the upturn increases with @xmath69 . they also find that the dynamics is both spatially and temporally heterogeneous , although characterizing dynamical heterogeneities is more difficult than in supercooled samples , because the particle motion is very restrained and the statistics is poorer ( more details on dynamical heterogeneities can be found in ) . at all ages the most mobile particles form clusters ; however no obvious evolution of the cluster size or morphology with age is observed . to conclude the overview of experiments on the aging of hard spheres systems , we recall the preliminary dynamic light scattering results in reference @xcite , which compare the aging dynamics of a repulsive hard sphere colloidal glass to those of an attractive glass ( see ) and suggest different behaviors : the repulsive glass seems to stop aging after about one day , while the dynamics of the attractive glasses keep on slowing down after @xmath71 days . aging behavior has been most extensively investigated in soft materials which greatly depart from a hard spheres model system . these materials include colloidal pastes made of concentrated silica @xcite or polystyrene particles @xcite ( both stabilized by a combination of electrostatic and steric repulsions ) , diluted @xcite or more concentrated @xcite fractal colloidal gels of polystyrene beads , diluted suspensions of charged clay platelets ( laponite ) @xcite ( to which polystyrene beads may be added to perform dws experiments @xcite ) , compact arrangements ( @xmath72 ) of surfactant multilamellar vesicles @xcite , polyelectrolyte microgel particles @xcite , or polycrystals of copolymeric micelles @xcite . for these systems , rheological tests are very often employed to probe aging phenomena , in addition to light scattering experiments . rheological experiments , e.g. measurements of the stress relaxation following a step strain in the linear regime , are often compared to available theories , one of the most successful being the soft glassy rheology ( sgr ) model of sollich and coworkers @xcite . this model , derived from the trap model of bouchaud @xcite accounts for the rheology of out - of - equilibrium materials . in brief , in the trap model non - interacting particles evolve through a hopping mechanism in an energy landscape with wells of depth @xmath73 . the distribution of well depths is fixed and the evolution of @xmath74 , the probability for a particle to be in a trap of depth @xmath73 at time @xmath48 , is governed by thermally activated hopping . this model leads to aging phenomena when the average trapping time diverges . in the sgr model , the material is divided into elements which yield above a critical yield strain @xmath75 , thus relaxing stress . yield events ( i.e. rearrangements of the particles ) are seen as hops out of the trap , and yield energy is identified as the trap depth @xmath73 . the activation barrier is @xmath76 where @xmath77 is the maximum elastic energy before yielding , @xmath78 is the local shear strain of an element , and @xmath79 is an elastic constant . the link between the microscopic parameters and the macroscopic strain applied to the sample is straightforward : in between rearrangements , the local strain follows the macroscopically imposed strain @xmath80 . in reference @xcite , fielding , sollich and cates study the role of aging in the rheology of soft glassy materials . they describe several rheological tests suitable to investigate the ( linear and non - linear ) rheology of aging soft materials , whose features are qualitatively similar to experimental results . experimentally , aging behavior has been observed in dws @xcite , single light scattering @xcite , and linear rheology @xcite experiments . in all cases , the characteristic time @xmath53 of the correlation functions ( for light scattering experiments ) , of the stress relaxation ( in a step strain experiment ) , or of the strain evolution ( in a creep experiment ) increases continuously with sample age . however , very different aging laws @xmath81 have been found for the various materials investigated . in many cases a power law @xmath82 is observed , but the aging exponent @xmath83 may significantly vary . indeed , values of about @xmath51 @xcite , smaller than @xmath51 ( @xmath84 ) @xcite , and larger than @xmath51 ( @xmath85 and @xmath86 in references @xcite and @xcite , respectively ) have all been observed , reflecting in the language of glasses full aging , sub - aging and hyper - aging , respectively . a peculiar , very fast aging regime has been observed both in fractal colloidal gels @xcite and in laponite samples @xcite , for which @xmath53 is found to grow exponentially with sample age . this fast aging regime is eventually followed by a slower growth of @xmath53 , corresponding to full aging . note that the work by bellour _ et al . _ @xcite , who studied the aging of laponite over a wide interval of @xmath69 and observed both regimes , rationalizes the apparently conflicting findings of references @xcite and @xcite , which were probing only the exponential and the full aging regime , respectively . the experimental work reviewed in this section clearly shows that , albeit aging behavior is a ubiquitous feature of out - of - equilibrium soft materials , the detailed evolution of the dynamics with @xmath69 greatly depends on the particular system that is investigated . the large spectrum of aging behaviors measured experimentally hints at differences in the microscopic mechanisms at play in the various samples and stands as a challenge for more detailed theories . in subsec . [ subsec : slow ] , we have mentioned that a large variety of soft disordered materials were found to exhibit an unusual slow dynamics , with a compressed exponential shape for the dynamic structure factor ( @xmath87 $ ] , with @xmath88 and @xmath57 ) , indicative of ballistic motion @xcite . we have argued that these peculiar features could be rationalized with simple arguments , based on the concept that the dynamics be due to the relaxation of internal stresses . several experimental observations hint at the key role played by internal stresses in leading to the final relaxation of @xmath1 . cipelletti @xmath41 @xmath42 and manley @xmath41 @xmath42 @xcite have observed particle syneresis in colloidal gels made of attractive particles . when the gel is strongly anchored to the container walls ( as it is usually the case ) , tensile stress builds up in the sample as a result of the decrease of the interparticle distance , due to syneresis . further evidence for the role of internal stress is provided by experiments on concentrated emulsions , whose dynamics was initialized by centrifugation @xcite . the dynamics is systematically faster in the direction of the centrifugation acceleration , along which most of the internal stress has been presumably built in . for laponite suspensions , bandyopadhyay and coworkers argue that internal stresses are generated by the increase of the interparticle repulsion , due to the dissociation of ions at the surface of the particle , as revealed by the increase of the conductivity of the suspension with sample age @xcite . finally , similar dynamics were observed for a micellar polycrystal @xcite and compact arrangements of multilamellar vesicles ( mlvs ) @xcite , two systems for which the fluid - to - solid transition is induced by a temperature jump . for these samples , we expect internal stress to be built up due to the rapid growth of randomly oriented crystallites or mlvs , respectively . interestingly , linear rheology measurements on the mlvs @xcite show that the characteristic time of the mechanical response of this material follows exactly the same aging law as the characteristic time of the final decay of @xmath1 . since rheology probes the response to an external stress , while the final relaxation of @xmath1 is ascribed to internal stress , this concordance gives further support to the concept of stress relaxation as a key ingredient in determining the dynamics of disordered systems and their evolution with sample age . dense arrangements of mlvs @xcite provide ideal samples to test in great details the notion of internal stress as a driving force for the slow dynamics . as mentioned above , the fluid - to - solid transition of this material can be driven by a temperature jump . this has to be contrasted with most soft glassy systems , where the fluid - to - solid transition is obtained upon cessation of a large applied shear , which certainly influences the initial configuration of internal stress . for the mlvs , the physical origin of the internal stress is the elastic energy stored in the deformation of the vesicles , which are expected to depart from a spherical , elastically relaxed shape , due to their rapid and disordered growth at the fluid - to - solid transition . on the other hand , the linear elastic modulus @xmath43 is equal to the density of elastic energy stored by the material when it is deformed in the linear regime @xcite . therefore , both the internal stress and @xmath43 share the same microscopic origin , i.e. the deformation of the mlvs . it follows that the slow dynamics should depend crucially on @xmath43 , if indeed they are driven by the relaxation of internal stress . in order to test this conjecture , we have recently measured the aging dynamics of mlvs samples for which @xmath43 was varied over more than one order of magnitude @xcite . we find that the slow dynamics is faster for systems with a higher elastic modulus , in agreement with the hypothesis that the higher @xmath43 the larger the internal stress . remarkably , the slow dynamics and the aging can be entirely described by the evolution of an effective viscosity , @xmath89 , defined as the characteristic time measured in a stress relaxation rheology test times @xmath43 . the concept of effective viscosity is found to be robust , since at all time @xmath89 is independent of @xmath43 , of elastic perturbations , and of the rate at which the sample is driven from the fluid to the solid state . a simple model that links @xmath89 to the internal stress created at the fluid - to - solid transition is proposed . in this model , the ballistic motion of the mlvs results from a balance between a driving force , associated to the local internal stress acting on a region containing several mlvs , and a viscous drag . ( note that this model is similar to the one proposed in reference @xcite to describe the fast dynamics associated with the rattling of soft microgel particles within the cage formed by their neighbors . ) in this picture , aging results from a weakening of the driving force , due to the progressive relaxation of internal stress . indeed , we find that @xmath43 slowly decreases with @xmath69 , as expected if the measured elastic modulus is the sum of a ( constant ) `` equilibrium '' elastic modulus ( corresponding to an ideal , totally relaxed configuration of the sample , where all mlvs have a spherical shape ) and the internal stress , which decreases with @xmath69 . similar arguments should apply also to other soft systems , such as concentrated emulsions , whose elasticity , similarly to that of the mlvs samples , depends on the deformation of their constituents . by contrast , a very different behavior is observed for hard particles , for which the elasticity results essentially from excluded volume interactions ( entropic origin ) . _ study the age - dependent rheology of a colloidal paste made of silica particles @xcite . they find that the elastic modulus of the paste increases logarithmically with sample age , defined as the time elapsed since the system has been fluidified by a strong mechanical shear , while the elastic modulus of a mlv sample decreases . the authors suggest a link between the aging that they observe in stress relaxation experiments and the spontaneous increase of the elastic modulus . a phenomenological model which incorporates as main ingredients spontaneous aging and mechanical rejuvenation @xcite is able to reproduced the essential features of the various rheological tests performed experimentally , in particular the stress relaxation experiments , although in the model the elastic modulus of the sample is fixed . the same phenomenological model gives also results in good agreement with start - up flow experiments @xcite , where the time evolution of the stress is measured as a constant shear rate is imposed to the sample . they find both experimentally and theoretically that the stress overshoots before the system actually starts to flow . the amplitude of the overshoot increases with sample age , thus suggesting that the older the sample the larger the stress that has to be stored before flow can occur . this suggests that the system strengthens with age , in contrast with the decrease of @xmath43 observed for the mlvs . interestingly , a similar link between slow dynamics , aging behavior and stress relaxation has been proposed to explain the dynamics of a gently vibrated granular pile @xcite . kabla and debrgeas measure by multispeckle dws the two - time intensity autocorrelation function of the light scattered by the pile , @xmath90 . the sample is vibrated by applying `` taps '' , whose amplitude is small enough not to induce macroscopic compaction , but large enough to trigger irreversible particle displacements on a microscopic scale . the characteristic decay time of @xmath90 increases roughly linearly with @xmath69 , in surprising analogy with many glassy systems . by invoking arguments similar to those of bouchaud s trap model @xcite , the authors explain this slowing down as the result of a slow evolution of the ( gravitationally - induced ) stress distribution in the pile . weak grain contacts are progressively replaced by stronger contacts , leading to the observed aging of the dynamics and the strengthening of the pile . the literature reviewed in this section indicates that internal stress and its time evolution are quite general ingredients in attempts to explain the slow dynamics and the aging of many glassy systems . however , internal stress may play very different roles , either acting as a driving force for the dynamics ( as proposed , e.g. , for the mlvs ) , or evolving in response to other processes ( e.g. thermal activation or applied vibrations ) . the experiments reviewed here show that the role of internal stress , as well as the evolution of the elastic response during aging , may depend also on the way the samples are initialized and on the microscopic origin of the elasticity . indeed , the internal stress can not be the same when a system is fluidized by applying a mechanical shear or , on the contrary , without perturbing it mechanically . similarly , different aging behaviors of the stress distribution and the elasticity are to be expected in samples where the elasticity is due to the deformation or the bending of the individual constituents , or where it derives from excluded volume interactions . a great amount of experimental , numerical and theoretical work has been devoted to investigating the response of glassy soft materials to an external perturbation , usually a mechanical one . there are several reasons justifying such a broad interest . on one hand , these materials are ubiquitous in industrial applications , where their mechanical properties are of primary importance . on the other hand , applying a ( large ) mechanical perturbation is a way to modify the dynamical state and the aging behavior of a soft glass : in this case the dynamics may depend on the mechanical history , much as the dynamics of a molecular or spin glass depends on its thermal history . this analogy lies on the similar role of temperature and , e.g. , strain in driving the fluid - to - solid transition , as proposed by liu and nagel @xcite . finally , there is an intense debate on the relationship between response functions ( in the linear regime ) and correlation functions in glassy systems and the break down of the fluctuation dissipation theorem ( fdt ) . in this section , we will first review experiments probing the response of a system to a strong perturbation ( non - linear regime ) , and then discuss experimental work addressing the possible violation of the fdt . habdas _ et al . _ have performed original experiments @xcite where they measure the average velocity and the velocity fluctuations of a magnetic bead submitted to a force , @xmath91 , and immersed in a dense colloidal suspension ( in the supercooled fluid state ) . as also observed in simulations @xcite , they find that the average velocity varies as a power law with @xmath91 . they also find a threshold force @xmath92 for motion to be observed , which initially increases with the colloid volume fraction , but eventually saturates , suggesting that @xmath92 does not diverge at the glass transition . moreover , they do not observe an increase of velocity fluctuations when approaching the glass transition , as one might have expected if the probe environment became more heterogeneous . they suggest that @xmath92 be related to the strength of the cage and that the existence of @xmath92 hints at local jamming even if the colloidal suspension is globally in a liquid phase . two interesting extensions of this work would be _ i ) _ to perform the same type of measurements in the glass phase and _ ii ) _ to look at the rearrangements of the particles due to the forced motion of the magnetic probe . we note that experiments similar to _ ii ) _ have been conducted in a two - dimensional granular material where one follows the motion of all grains in response to the forced motion of one grain @xcite . note that in the experiment by habdas _ the motion of the magnetic probe is much larger than the brownian motion of the surrounding particles . as a consequence , this experiment probes at a microscopic level the non - linear rheology of the suspensions . macroscopic non - linear rheology experiments have been performed on a variety of systems . although it has been known for several years , and has been widely used in experiments , that the dynamics of soft glassy materials can be initialized by submitting them to a strong shear , recent experiments have investigated in more details the effects of shear on the slow dynamics and the aging of these materials . one important issue is to understand how the slow and age - dependent evolution of the rheological properties correlates with the mechanical history of the materials . a first answer is provided by cloitre _ @xcite , who have studied a paste made of soft microgel colloidal particles . they measure the strain recovery and the creep of a sample ( initially fluidified by applying a stress larger than the yield stress , @xmath93 ) submitted to a `` probe '' stress @xmath94 of variable amplitude . they find that the time evolution of the strain is age - dependent but that all the experimental data collapse on a master curve when the time is normalized by @xmath95 . the aging exponent @xmath83 decreases continuously from @xmath51 ( full aging ) to @xmath96 ( no aging ) as @xmath94 increases ( @xmath97 for @xmath98 ) . they propose that @xmath83 can be used to quantify the partial mechanically - induced rejuvenation of the sample . light scattering has also been used extensively to investigate the interplay between slow dynamics and shear rejuvenation . using the recently introduced light scattering echo technique @xcite , petekidis and coworkers @xcite have looked at the effect of an oscillatory shear strain on the slow relaxation of a colloidal glass of hard spheres , finding support for a speedup of the slow dynamics due to shear . bonn and coworkers @xcite have measured by dws the characteristic relaxation time @xmath53 of a glassy suspension of laponite particles , to which a shear of rate @xmath99 has been applied . similarly to the results of reference @xcite , they find that @xmath53 decreases after imposing a shear ; moreover @xmath53 is smaller for larger @xmath99 . the observed shear rejuvenation is in agreement with theoretical predictions @xcite . more quantitatively , ozon _ et al . _ @xcite have shown using clay particles ( smectite ) that partial rejuvenation depends only on the amplitude of the applied strain and not on its frequency nor its duration . they find that the relative decrease of the relaxation time due to a sinusoidal shear varies exponentially with strain amplitude , @xmath80 , implying that the mechanical energy input is proportional to @xmath80 . this result is in contradiction with the soft glassy rheology model of sollich _ et al . _ @xcite , which assumes a @xmath100 dependence , resulting from a local elastic response . this discrepancy is only but an example of how subtle and counterintuitive the mechanisms for shear rejuvenation may be . by investigating in detail the dynamics of a colloidal paste subjected to an oscillatory mechanical strain of variable duration , amplitude and frequency , viasnoff and lequeux @xcite have indeed demonstrated that the characteristic relaxation time of the mechanically perturbed sample ( as measured by dws ) may be either smaller or larger than that of an unperturbed sample . the former case corresponds to rejuvenation , while the latter has been termed `` overaging '' . an illustration of the rejuvenation and overaging effects is shown in figure [ fig : overaging ] . note that overaging is reminiscent to observations of the dynamics of a foam by cohen - addad and hhler @xcite , who have found that the bubble dynamics is strongly slowed down after applying a transient shear . viasnoff and lequeux interpret their findings using bouchaud s trap model for glassy dynamics @xcite ( see subsec . [ subsec : aging ] for a short description ) , assuming that shear is strictly equivalent to temperature . they calculate the probability for a particle to be in a trap of depth @xmath73 a time @xmath48 after a temperature jump of small amplitude . they find that the @xmath39-jump ( correspondent to the shear perturbation in their experiments ) modifies the lifetime distribution : at large times , the deep traps may be overpopulated compared to the unperturbed case , while populations of the shallow traps are equal in both cases . this change of the distribution of life times for the traps leads to an average relaxation time which is longer for the perturbed sample , and hence to overaging . physically , overaging demonstrates that a moderate shear could help the system to find a more stable configuration , by allowing it to explore more rapidly a larger portion of the energy landscape . ( left ) strain history and ( right ) normalized relaxation time @xmath101 ( ratio of the relaxation time of the perturbed sample over relaxation time of the unperturbed sample ) as a function of sample age . rejuvenation corresponds to @xmath102 and overaging corresponds to @xmath103 ( gray region ) . curves are labelled by strain amplitudes . overaging occurs for low strain amplitude . ( adapted from reference @xcite ) ] shear rejuvenation may lead to very peculiar rheological responses . coussot , bonn and coworkers have indeed observed for several systems @xcite a critical stress above which the viscosity @xmath104 continuously decreases with time and below which @xmath104 increases until flow is totally arrested . the authors interpret the viscosity bifurcation as resulting from a competition between aging and rejuvenation and propose a simple phenomenological model that captures the main experimental observations . an alternative way to perturb a system have been recently studied by narita _ the authors have measured by dws the slow dynamics of concentrated colloidal suspensions upon drying . they consider that in this experimental configuration a uniaxial compressive stress is generated in the sample . they attempt to map their previous findings of mechanically - induced rejuvenation and overaging on their experimental results . implicitly , they assume that the effects on the internal stress in the aging dynamics of the sample are similar to those of an external strain . detailed experiments are clearly needed to address this issue . finally , recent experiments of simeonova and kegel @xcite demonstrate that soft glasses can be considerably perturbed also by gravitational stress . by changing the solvent in which the colloids are suspended , the density difference between the colloids and the solvent , @xmath105 , is varied . two systems with gravitational lengths @xmath106 equal respectively to @xmath107 and @xmath71 @xmath83 m ( @xmath108 , with @xmath109 the boltzmann constant and @xmath18 the radius of the colloids ) are used . they find that gravity accelerates significantly the aging of the colloidal glass . thus , a direct parallel with the overaging effect can be drawn : gravity , similarly to a moderate shear , can help the system to find a more stable configuration . on the other hand , gravity prevents a full exploration of the configuration space , ultimately suppressing crystallization . indeed , space experiments have shown that hard sphere suspensions that are in a glass phase on earth do crystallize in microgravity conditions @xcite . how gravity couples to the particle rearrangements on a microscopic scale remains an open issue . for systems at equilibrium , the response to an external perturbation is related to the correlation function of the observable to which the perturbation field is conjugated . let us consider an observable @xmath110 and its normalized time autocorrelation function @xmath111 . if the system is perturbed by a field @xmath106 conjugated to @xmath110 , the response function is @xmath112 . at equilibrium , the fluctuation - dissipation theorem ( fdt ) relates the time derivative of the correlation function with the response : @xmath113 where @xmath109 is the boltzmann constant and @xmath39 is the system temperature . two common examples of the fdt are the nyquist formula that relates the voltage noise across a resistor to the electrical resistance and the stokes - einstein relation , which relates the diffusion of a particle in a solvent to the solvent viscosity . in deriving the fdt , time - translation invariance is required : as this assumption is not fulfilled for out - of - equilibrium systems , for them the fdt does not hold . for out - of - equilibrium systems , it has been proposed @xcite that the fdt could be generalized by introducing an effective temperature , @xmath114 . the time derivative of the correlation function and the response function are then related through @xmath115 . note that @xmath116 is expected to depend on the observation time scale and on the sample age and to be higher than the temperature @xmath39 of a bath with which the out - of - equilibrium material is in thermal contact . in the past , violations of the fdt have been experimentally observed for structural glasses @xcite , spin glasses @xcite , and polymer glasses @xcite ; moreover , an effective temperature has been introduced to describe granular materials @xcite . by contrast , experiments testing the fdt in out - of - equilibrium soft materials are still very rare . using available data from the literature , bonn and kegel @xcite examine the generalized stokes - einstein relation , @xmath117 $ ] , for hard sphere suspensions . here @xmath118 and @xmath18 are the frequency - dependent diffusion coefficient and the radius of the particles , respectively , and @xmath119 is the frequency - dependent viscosity of the suspension . they derive @xmath118 from dynamic light scattering experiments and @xmath119 from rheology measurements of either the high shear rate viscosity of the suspension , when considering the short - time diffusion coefficient , or the complex modulus in the linear regime , when considering the long - time diffusion coefficient . they find that , in a wide range of volume fractions ( between @xmath120 and @xmath121 ) , the fdt is obeyed ( @xmath122 ) for the short time diffusion coefficient , i.e. , at the larger volume fractions , for the particle dynamics within the cage . by contrast , for supercooled suspensions the fdt is strongly violated at longer times . the departure from fdt is quantified by @xmath123 , where @xmath124 and @xmath125 are reduced parameters given in units of the diffusion coefficient of the particle at infinite dilution and of the solvent viscosity , respectively . they deduce that @xmath126 can be as large as @xmath127 at low frequencies and for a volume fraction @xmath128 ; @xmath129 is found to decrease from @xmath130 to @xmath131 as @xmath6 increases from @xmath132 to @xmath133 . as expected theoretically , they find that @xmath134 at high frequency and that at low frequency @xmath116 is larger than @xmath39 and decreases continuously with @xmath135 following a power law . finally , we point out that , although the violation of the fdt is clearly evidenced in this work , the exact values of the effective temperatures should be taken with extreme caution since they are derived from experiments carried out with different particles . abou and gallet @xcite have measured the effective temperature of a colloidal glass of clay particles ( laponite ) , using a modified stokes - einstein relation ( the detailed theoretical formalism for the diffusion of a particle in an aging medium can be found in references @xcite ) . they seed the colloidal glass with micron - sized particle probes that can be tracked using light microscopy , and use these probes as a thermometer by measuring both the mean squared displacement and the mobility of the beads . from these measurements , they extract an effective temperature @xmath116 at a fixed frequency of @xmath51 hz . they find a non - monotonic variation of @xmath116 with sample age : @xmath116 is equal to the bath temperature @xmath39 for a young sample , it increases up to @xmath136 at intermediate age ( @xmath137 min ) and then decreases when the sample further ages . the authors relate the non - monotonic variation of @xmath116 with sample age to the evolution of the characteristic time for the relaxation of the colloidal glass as measured by dynamic light scattering , and argue that @xmath138 when the characteristic frequencies of the slow modes measured in dls are comparable to the frequency at which @xmath114 is measured . measuring the frequency dependence of the effective temperature would be needed to confirm this interpretation . the same experimental system has been investigated by the group of ciliberto @xcite . they probe the electrical and rheological properties of the material during the formation of the soft glass . for the dielectric experiments , the clay solution is used as a conductive liquid between two electrodes . the set - up allows the frequency dependence of both the electrical impedance and the voltage noise to be measured , from which an effective temperature is derived , by means of a generalized nyquist formula . they find that @xmath114 is a decreasing function of frequency and reaches the bath temperature at high frequency . moreover , @xmath116 decreases as the sample ages , while the fdt is strongly violated for young samples : @xmath116 can be larger than @xmath139 k at low frequency ( @xmath51 hz ) and small age @xcite . they show @xcite that the origin of the large violation is the highly intermittent dynamics characterized by large fluctuations of the voltage noise ( see for more details on the dynamic heterogeneities ) . we note that , in this line of thought , crisanti and ritort @xcite have shown numerically how an effective temperature can be extracted from the probability distribution function of intermittent events . because of the strong fluctuations of the noise in the experiments of ciliberto and coworkers , the value of @xmath116 extracted from the voltage data is extremely sensitive to rare events and to the duration over which data are collected . this may explain the discrepancy between references @xcite and @xcite concerning the numerical value of the effective temperature obtained by dielectric measurements ( about one order of magnitude ) . bellon and ciliberto @xcite have also tested the fdt for the rheological properties of the same system . to that end , they have built a novel `` zero - applied stress '' rheometer @xcite , by which the ( very small ) strain induced by thermal fluctuations can be measured . although strong aging properties are measured in the rheological responses of the material ( as measured also for the dielectric properties ) , no violation of the fdt could be detected . the authors propose several possible explanations for the discrepancy between dielectric and rheological measurements . in particular , they point out that the strong fluctuations of the voltage noise could result from the dissolution of ions in the solution . ion dissolution has been very recently confirmed by bandyopadhyay and coworkers @xcite who argue that the dissociation of the ions at the surface of the particles leads to the increase of the interparticle repulsion and is at the origin of the slow dynamics and aging of laponite samples . because the increase of the interparticle repulsion may lead to a rearrangement event only when the internal stress thus generated exceeds the ( local ) yield stress , fluctuations in the mechanical properties are expected to be much less strong than fluctuations of the dielectric properties , which are more sensitive to the release of charges . on the other hand , the rheology measurements of the group of ciliberto are in principle similar ( albeit not identical ) to those of abou and gallet , since both probe the mechanical properties of the material . abou and gallet measure a weak violation of the fdt ( @xmath140 is at most equal to @xmath141 at a frequency of @xmath51 hz ) , while the measurements of ciliberto s group do not detect any violation . however , it should be pointed out that the rheology experiments of ciliberto and coworkers may lack the sensitivity required to measure violations as small as those reported by abou and gallet . whether these experimental results are indeed conflicting remains therefore an open question . to conclude , we remark that experimental works are still scarce , owing to the difficulty of measuring simultaneously the response function and the fluctuation in experiments on soft materials . laponite clay suspensions have been mostly investigated and the results obtained so far by two independent groups and using different techniques display both qualitative and quantitative discrepancies . on the other hand , simulations works on a binary lennard - jones mixture @xcite have shown that the effective temperature is independent on the chosen observable , while it has been shown theoretically @xcite that the effective temperature does depend on the observable in the glass phase of bouchaud s trap model . more theoretical , numerical , and experimental investigations are needed to rationalize these contrasting findings . dynamical heterogeneity is now recognized as a fundamental feature of the slow dynamics of supercooled fluids and glasses in hard condensed matter , thanks to the large body of experimental , numerical and theoretical work carried out in the last decade ( a general review on dynamical heterogeneity can be found , e.g. , in reference @xcite ) . most early observations of heterogeneous dynamics focussed on temporal heterogeneity : the coexistence of different relaxation times was identified as the source of the non - exponential relaxations observed in glass formers . spatial heterogeneity was often invoked as the most plausible physical origin of this coexistence ( see @xcite and references therein ) . subsequent experimental and numerical work has shown that indeed the dynamics of glass formers is spatially heterogeneous , and spatial heterogeneity has been related to the cooperative nature of the slow dynamics ( for reviews on spatial heterogeneity that focus on experimental and numerical work , see for example references @xcite and @xcite , respectively ) . cooperativity plays a central role in many recent theories , where the glass transition is explained as a dynamical ( as opposed to thermodynamic ) transition driven by the divergence of the size of regions that undergo cooperative rearrangements . following this approach , analogies have been drawn with critical phenomena , the static correlation length of the latter being replaced by a suitable dynamical correlation length @xcite . it should be noted that most theories are developed in the framework of spin models or the so - called dynamically facilitated ( or kinetically constrained ) models @xcite , where the motion of on - lattice particles depends on the number of occupied neighboring sites . making quantitative connections between the results for these systems and molecular glass formers or colloidal systems may be therefore difficult . however , we note that recent numerical and theoretical work has shown that the concept of the glass transition being a dynamical critical phenomenon can be successfully applied also to lennard - jones glass formers @xcite and in the framework of the mode coupling theory @xcite . experimentally , soft materials provide a unique opportunity to study in great detail temporal and spatial heterogeneity in supercooled fluids and glasses , because the relevant length and time scales are more easily accessible than for hard condensed matter systems . in this section , we will review recent experiments that probe dynamical heterogeneity in a variety of systems , ranging from model hard sphere suspensions to more complicated glassy samples , such as colloidal gels and concentrated surfactant phases . most experiments are performed using time - resolved confocal scanning microscopy @xcite or recently introduced light scattering methods that allow temporal heterogeneities to be measured @xcite , as it will be discussed in the following subsections . optical microscopy and digital imaging processing allow one to follow simultaneously the individual trajectories of a large number of particles ( up to thousands ) , tracking their position to an accuracy of a few tens of nanometers @xcite . three - dimensional motion can be studied thanks to time - resolved , laser - scanned confocal microscopy . once the particle trajectories are known , a wide range of statistical quantities can be calculated in order to detect , characterize , and quantify dynamical heterogeneity . direct comparison with simulation work is possible , making optical microscopy a powerful experimental technique for investigating slow dynamics in glassy soft materials ( for a recent review of video microscopy applied to colloidal suspensions , see reference @xcite ) . early optical microscopy measurements of dynamical heterogeneities were performed by kasper and coworkers , who studied the self diffusion of tracer particles in a concentrated hard sphere fluid @xcite . the tracer particles have a core - shell structure : the outer layer is identical in composition to the host particles , while the inner core has a large optical contrast with the fluid and the host particles . the trajectories of the tracer particles are followed in a thin slice of a three - dimensional sample , allowing the mean squared displacement ( msd ) and the self part of the van hove correlation function , @xmath142 , to be calculated . @xmath142 represents the ( density of ) probability that a particle moves a distance @xmath143 in a time step @xmath48 . for a diffusive process , @xmath142 is a gaussian distribution . kasper @xmath41 @xmath42 find that in concentrated suspensions @xmath142 departs from a gaussian behavior , the deviations being increasingly marked as @xmath6 approaches @xmath8 . these deviations are due to a small but significant fraction of displacements @xmath144 larger than expected . deviations from gaussian behavior are quantified by the so - called non - gaussian parameter @xmath145 , defined as the fourth moment of @xmath146 , properly normalized ( @xmath147 for a gaussian distribution ) . similar experiments are reported by kegel and van blaaderen @xcite , who use confocal microscopy to study in two dimensions the dynamics of concentrated suspensions of nearly density- and refractive index - matched colloids behaving as hard spheres . in contrast to the experiment of kasper @xmath41 @xmath42 , all particles are tracked ( as opposed to tracer particles only ) , improving dramatically the statistics . for concentrated suspensions in the fluid phase , @xmath142 is reasonably well described by the sum of two gaussian distributions , corresponding to two distinct populations of `` fast '' and `` slow '' particles . this result clearly demonstrate the heterogenous nature of the dynamics , in agreement with several simulation works ( see , e.g. , reference @xcite ) . weeks and coworkers performed the first three - dimensional optical microscopy investigation of the slow dynamics of supercooled colloidal suspensions @xcite . they use pmma particles stained with a fluorescent dye and suspended in an organic solvent that nearly matches both the density and the refractive index of the particles . note that under the reported experimental conditions the particles are slightly charged @xcite . they find a non - gaussian @xmath146 , with a non - gaussian parameter that has a peaked shape as a function of the time step @xmath48 . the peak position corresponds to the characteristic time of the @xmath10 relaxation , and its height increases when @xmath6 approaches @xmath8 , indicating increasingly heterogeneous behavior . in the glass phase , no clear peak is observed . in analogy with previous simulation work @xcite , weeks and coworkers study the spatial arrangement of the most mobile particles ( the mobility is measured for time intervals comparable to the @xmath10 relaxation time ) . they find that these particles form clusters that have a fractal morphology ( fractal dimension @xmath148 ) and whose size , for supercooled suspensions , increases with increasing @xmath6 , up to a radius of gyration of about 10 particle radii . for glasses , defining the most mobile particles is more difficult , since no @xmath10 relaxation is observed and the cluster size depends sensitively on the way the mobility is calculated @xcite . when the particle trajectories are slightly averaged over time to remove the contribution of brownian motion , the clusters observed in glasses are similar to those observed for supercooled samples . interestingly , no change in the cluster size is observed for glasses during aging , thus ruling out the increase of the correlation length of the dynamics as a possible origin for the slowing down of the dynamics @xcite . in glassy systems , the relationship between dynamical heterogeneity and local structure is a long - standing open question . various experiments suggest that particle mobility is related to the degree of local disorder and packing , although it should be stressed that very small changes in the local structure are associated to huge variations in the dynamics . in the same three - dimensional experiment discussed above , weeks and weitz have shown that the clusters of the most mobile particles occupy regions with lower local density and higher disorder @xcite . two - dimensional systems allow a more direct investigation of the relationship between local structure and dynamical heterogeneity . cui and coworkers have studied the dynamics of a single layer of concentrated monodisperse particles confined in a quasi - two - dimensional cell @xcite ( note that this system does not exhibit a glass phase , but rather crystallizes at high enough volume fraction ) . at relatively high volume fractions and intermediate time scales , the dynamics is very heterogeneous , with two distinct populations of particles . the fast particles move in a string - like fashion strongly reminiscent of that observed in simulations of supercooled fluids @xcite along channels formed by the disordered boundaries between regions of slowly - moving , quasi - ordered particles . on very long time scales , infrequent , large displacements are associated with transient regions of lower density created by density fluctuations . a very elegant realization of a two - dimensional colloidal glass former has been recently reported by knig and coworkers @xcite . superparamagnetic colloidal particles are confined at the water - air interface of a hanging liquid drop ; the interactions between particles can be fine - tuned by applying a magnetic field , thereby fixing the `` temperature '' of the system @xcite . mixtures of particles of two different sizes are used , in order to prevent crystallization . the structure is analyzed in terms of locally ordered patterns of small and large particles that maximize the packing density @xcite . dynamical heterogeneities appear to be associated with the regions where the local order is frustrated and the local packing is looser @xcite . although in light scattering experiments no microscopic information on the individual particle trajectories are available , this technique still presents some attractive features compared to optical microscopy . the scattering volume is typically larger than the sample volume imaged by microscopy , thus allowing a better statistics to be achieved . a large experimental volume may be particularly important when the dynamics is spatially correlated : for example , for samples near @xmath8 , the size of the clusters of fast - moving particles measured in the experiments by weeks @xmath41 @xmath42 @xcite was comparable to the full field of view , thus making difficult their precise characterization . moreover , particles used in light scattering experiments are typically smaller than those used for imaging : as a consequence , the time scales of the slow dynamics are not as prohibitively long as for larger particles . in addition , small particles are less influenced by external fields such as gravity . finally , experiments can be performed both in the single and in the strongly multiple scattering limits , thus extending the possible choice of systems ( by contrast , optical microscopy requires nearly index - matched suspensions , except for two - dimensional systems ) . traditional light scattering techniques , however , do not provide direct information on dynamical heterogeneity , because of space and time averaging . in fact , the intensity correlation function @xmath149 has to be averaged over extended periods of time in order to achieve an acceptable accuracy ( typically up to four orders of magnitudes longer than the largest relaxation time in the system ) , and the detector collects light scattered by the whole illuminated sample . indirect measurements of dynamical heterogeneities are still possible : for example , the non - gaussian parameter @xmath145 that quantifies deviations from diffusive behavior can be measured ( see e.g. reference @xcite ) . however , it should be noted that a non - zero value of @xmath145 could be due either to the coexistence of fast and slow populations of particles , as discussed in the previous subsection , or to the same non - diffusive behavior shared by all particles . traditional light scattering experiments lack the capability of discriminating between these two contrasting scenarios . higher - order intensity correlation functions contain more information on temporal heterogeneities of the dynamics @xcite . in particular , lemieux and durian have studied the dynamics of the upper layer of grains in a heap upon which grains are steadily poured at a flow rate @xmath150 @xcite . they measure the 4-th order intensity correlation function , @xmath151 , where @xmath152 is the multiply scattered intensity measured by a point - like detector and the average is over time @xmath48 and a time lag by @xmath53 . this is different from the notation used in previous sections of this paper and in most numerical and theoretical works where @xmath48 and @xmath153 are used , respectively . ] . they show that , contrary to @xmath5 , @xmath154 allows intermittent , avalanche - like processes ( occurring at low @xmath150 ) to be distinguished from continuous dynamics ( occurring at high @xmath150 ) . moreover , both the statistics of the avalanches and the motion of grains within a single avalanche can be obtained by analyzing @xmath154 . the technique proposed by lemieux and durian still requires to average the ( higher - order ) intensity correlation functions over time . this poses a problem when studying very slow or non - stationary processes , as it is often the case for soft glasses . cipelletti and coworkers have recently proposed an alternative approach , termed time resolved correlation ( trc ) @xcite , where one takes full advantage of the multispeckle method . in a trc experiment , a ccd camera is used to record a time series of pictures of the speckle pattern of the light scattered by the sample ( both single scattering and dws measurements are possible ) . the dynamics is quantified via @xmath155 , the instantaneous degree of correlation between pairs of speckle patterns recorded at time @xmath48 and @xmath156 : @xmath157 . here , @xmath158 is the intensity measured by the @xmath159-th pixel and averages are over all ccd pixels . because any change in the sample configuration results in a change in the speckle pattern , @xmath155 quantifies the overlap between sample configurations separated by a time lag @xmath53 , as a function of @xmath48 . the time average of @xmath155 yields the intensity correlation function @xmath160 usually measured in light scattering , while the raw @xmath155 is analogous to the two - time correlation function studied numerically or experimentally for non - stationary ( e.g. aging ) systems . note however that in most simulations and experiments the two - time correlation function is averaged over a short time window or over different realizations of the system , in order to reduce its `` noise '' . by contrast , the essence of the trc method is to extract useful information from the fluctuations of @xmath161 . time resolved correlation measurements for a granular material gently tapped . the dynamics of this athermal system is due to the taps ; therefore both the time , @xmath48 , and the time delay , @xmath53 , are expressed in number of taps . inset : @xmath155 as a function of @xmath48 for a fixed time delay @xmath162 . note the rare , large drops of the trc signal , indicative of intermittent dynamics . main plot : for the same system , @xmath155 is plotted as a function of time delay , for fixed @xmath163 . the step - like relaxation of @xmath161 is due to large , rare rearrangement events similar to those that yield the downward spikes in the inset . adapted from reference @xcite . ] in order to investigate the temporal heterogeneity of the dynamics , it is useful to plot @xmath155 as a function of time @xmath48 , for a fixed lag @xmath53 . for temporally homogeneous dynamics , one expects @xmath155 to be constant ( except for small fluctuations due to measurement noise ) , as verified on dilute suspensions of brownian particles @xcite . on the contrary , a large drop of @xmath161 at time @xmath48 would be indicative of a sudden rearrangement event occurring between @xmath48 and @xmath156 and leading to a significant change of the sample configuration . large drops of @xmath161 have been indeed observed in trc measurements on a variety of systems , including colloidal fractal gels and concentrated surfactant phases @xcite , flocculated concentrated colloidal suspensions @xcite , and granular materials @xcite . as an example , the inset of shows trc data for the granular material studied in references @xcite : large drops of the degree of correlation are clearly visible , thus demonstrating the intermittent nature of the slow dynamics . these sudden rearrangements are also visible in the two - time correlation function , provided that no average is performed , as discussed above . the main plot of illustrates this point by showing @xmath155 as a function of @xmath53 for a fixed @xmath48 , for the same system as in the inset . the rearrangement events result in a discontinuous , step - like relaxation of the correlation function ( note that a similar behavior was observed in simulations on a molecular glass former , when the two - time correlation function was not averaged over different system realizations @xcite ) . by contrast , the @xmath164 intensity correlation function @xmath149 is often indistinguishable from that of a system with homogeneous dynamics . this demonstrates that great care should be used in applying standard equilibrium methods to extract from @xmath5 quantities such as the particles msd , because these methods usually assume the dynamics to be homogeneous . in particular , the preliminary experiments reported in reference @xcite on the same systems for which the `` ballistic '' motion associated with internal stress relaxation was observed ( see sections [ subsec : slow ] and [ subsec : internalstress ] ) raise the issue of how to reconcile intermittent dynamics with ballistic motion . a simple explanation would be to assume that the dynamics is due to a series of individual rearrangements and that the particle motion in a given region of the sample resulting from distinct rearrangements is highly correlated ( uncorrelated motion would lead to a diffusive - like behavior ) . the model proposed by bouchaud and pitard for the slow dynamics of colloidal fractal gels is indeed based on this idea @xcite . an important point in the trc experiments reported above is that the technique allows one to probe directly temporal heterogeneity , not spatial heterogeneity , because each ccd pixel receives light issued from the whole scattering volume . however , the very existence of large temporal fluctuations of @xmath155 demonstrates albeit indirectly that the dynamics is spatially correlated over large distances . more precisely , in order to observe intermittent dynamics the number @xmath165 of dynamically independent regions contained in the scattering volume has to be limited , because for @xmath166 many rearrangement events would occur in the sample at any given time , leading to small , gaussian fluctuations of @xmath155 . similar arguments have been invoked in recent simulation and theoretical works to explain non - gaussian fluctuations in systems with extended spatial or temporal correlations ( see for example references @xcite ) . in particular , the probability density function ( pdf ) of @xmath161 at fixed @xmath53 is in many cases strongly reminiscent of the `` universal '' gumbel distribution @xcite or of a generalized gumbel - like pdf @xcite . these distributions are characterized by an asymmetric shape , with an ( asymptotically ) exponential tail . we note however that the detailed shape of the pdf of @xmath161 depends sensitively on the lag @xmath53 @xcite , a feature also found in theoretical work @xcite . in the trc experiments discussed above , individual rearrangement events can be identified and the probability distribution of the time interval , @xmath167 , between such events can be calculated . for the flocculated suspension studied in reference @xcite sarcia and hbraud find a power law distribution @xmath168 . interestingly , this behavior agrees with theoretical predictions derived in the framework of the trap model for glassy systems , in the regime where a limited number of dynamically independent regions are observed @xcite . for other experimental systems , individual rearrangement events may not be distinguishable , because the drop of @xmath161 associated with one single event may be negligible . in this case , a measurable loss of correlation is always due to the cumulative effect of many events , even for the smallest delays @xmath53 accessible experimentally . deviations from temporally homogeneous dynamics are still detectable , however . an example is provided by the dynamics of a shaving cream foam @xcite : @xmath155 is found to exhibit large fluctuations on a time scale much longer than the average relaxation time of the intensity autocorrelation function . as a consequence , the two - time correlation function decays smoothly at any time ( no step - like relaxation such as that in b is observed ) , but its relaxation time slowly fluctuates with time @xmath48 . these fluctuations are quantified by introducing the variance of @xmath161 , defined by @xmath169 . @xmath170 is the analogous for trc experiments of the generalized dynamical susceptibility @xmath171 introduced in theoretical and numerical works on spin systems , hard spheres , and molecular glass formers @xcite . note that @xmath171 is proportional to the volume integral of the so - called four - point density correlation function , which compares the change of the local configuration around the position @xmath172 during a time interval @xmath53 to the corresponding quantity for position @xmath173 @xcite . because @xmath171 is related to the spatial correlations of the dynamics , this parameter provides a quantitative link between temporal and spatial dynamical heterogeneity . for the foam , @xmath170 has a peaked shape , the largest fluctuations of the dynamics being observed on a time scale comparable to @xmath174 , the average relaxation time of the system . this feature is strongly reminiscent of the behavior of the systems investigated numerically and theoretically in references @xcite . in simulations of glass formers , the height of the peak of @xmath171 increases when approaching the glass transition @xcite . this behavior has been interpreted as due to the increase of the size of dynamically correlated regions . the foam and the ising spin models studied in reference @xcite provide a means to test this hypothesis on systems for which there is a natural characteristic length that increases with time , due to coarsening ( this length is the bubble size and the size of parallel spin domains for the foam and the ising models , respectively ) . indeed , for both systems the same dynamic scaling of fluctuations with domain size is observed : @xmath175 , where @xmath174 is the average relaxation time of the correlation function and @xmath176 is the number of bubbles ( or spin domains ) contained in the system . similar trc experiments have been performed on a very polydisperse colloidal paste by ballesta and coworkers @xcite . they find that @xmath170 has a peaked shape , similarly to the systems reported above . moreover , the height @xmath177 of the peak evolves in a very surprising way : initially @xmath177 increases with increasing volume fraction @xmath6 , similarly to the growth of the peak in supercooled molecular systems upon cooling @xcite . in contrast with molecular systems , however , @xmath177 reaches a maximum value at intermediate volume fractions and is drastically reduced at the highest @xmath6 experimentally achievable @xcite . more experiments on different systems will be needed to test the generality of this behavior . in particular , model systems such as hard spheres suspensions are an attracting choice , since detailed predictions on the behavior of @xmath170 in the framework of the mct are now available @xcite . dynamical heterogeneity has been measured also by other techniques and in different systems . strongly intermittent behavior has been observed in the dielectric signal for a laponite glass @xcite , as already mentioned in section [ subsec : fdt ] . as a result , the pdf of the voltage signal measured in these experiments exhibits a non - gaussian behavior , with roughly exponential tails similar to those reported in reference @xcite . the microscopic origin of these events is still unclear . attempts to measure intermittent behavior in the rheological response of the same laponite system using the `` zero - shear '' rheometer @xcite mentioned in were , to date , not successful . interestingly , heterogeneous dynamics has been reported for a variety of granular systems , using different techniques . intermittency has been measured by trc in granular systems gently vibrated @xcite . these experiments show that the evolution of the system configuration is not continuous , but rather occurs through a series of discrete rearrangement events , similar to those reported in references @xcite for colloidal systems . on a more microscopic level , marty and dauchot @xcite and pouliquen and coworkers @xcite have tracked the motion of individual `` grains '' in two - dimensional and three - dimensional granular systems submitted to a cyclic shear . they find that the grain motion is strongly reminiscent of that of colloidal glasses : a cage effect is observed and deviations from a gaussian behavior in the pdf of the grain displacement are reported , similarly to the results of the experiments discussed in . collectively , these analogies support and extend to a microscopic level the unifying picture underlying the concept of jamming proposed by liu and nagel @xcite . as a final remark , we note that intriguing analogies exist also between the heterogeneity of the spontaneous dynamics of soft glasses and the behavior of many complex fluids under shear . indeed , for concentrated colloidal suspensions and surfactant systems the macroscopic rheological response to an applied constant strain or stress has been shown to exhibit large temporal fluctuations @xcite . on the other hand , simulations and experiments have shown that the stress relaxation and the flow of foams , ultrasoft glasses and concentrated surfactant systems are both temporally and spatially heterogeneous ( see e.g. references @xcite ) . although this heterogeneous behavior is usually associated to the existence of different microstructures ( e.g. disordered / ordered onion phases in @xcite ) , simulations of a simple model of an ( athermal ) generic yield stress fluid suggest that heterogeneity may exist even in structurally homogeneous fluids @xcite . in these simulations , the flow of a yield stress fluid sheared at a constant rate is shown to be due to discrete rearrangement events , whose characteristic size diverges for a vanishing shear rate . the basic ingredients of this model ( localized plastic events due to a microscopic yield stress and elastic propagation of the local stress relaxation ) are certainly relevant for many soft glasses . in particular , close analogies may be drawn between slowly sheared yield stress systems and elastic glassy materials such as colloidal gels where internal stresses are progressively built up and induce particle rearrangements . more work will be needed to explore in depth these analogies and to fully assess their relevance and generality . in this paper , we have reviewed the experimental work of the last few years on the slow dynamics in soft matter . in particular , research in four areas has been discussed : the existence of two different glass states ( attractive and repulsive ) , the dynamics and the aging of systems far from equilibrium , the effect of an external perturbation on glassy materials , and dynamical heterogeneity . as it was pointed out in many occasions in the preceding sections , these topics are closely related and indeed the same systems have been often studied in the context of two or more of these areas . at the end of each section we have listed what are , in our opinion , the most relevant questions that are still open : here , we limit ourselves to some brief final considerations . a great effort has been directed to developing unified approaches that may account for the similarities in the slow dynamics of many soft materials and for the fascinating analogies with hard condensed matter glasses and granular media . the most successful theory is probably the mct : initially , it was restricted to hard sphere suspensions , but it has now been extended to include attractive systems at moderate to high volume fractions . however , it should be noted that the deep physical reasons of its success are still somehow unclear ; for example , dynamical heterogeneity , now recognized as a key feature of the slow dynamics of most glassy systems , is not included in standard formulations of the mct . very recent theoretical work shows that quantitative predictions on dynamical heterogeneity can be made in the framework of the mct : experiments that test these predictions would certainly shed new light on the limits to the validity of the mct . an alternative unifying approach is the jamming scenario , which is however a conceptual tool rather than a fully - developed quantitative theory . its appeal resides in the large variety of systems whose fluid - to - solid transition may be rationalized in a unified picture , ranging from colloidal suspensions to molecular glasses and granular materials . the analogies in the local dynamics between glass formers and granular media that have been rapidly reviewed in this paper ( cage effect , dynamical heterogeneity ) provide additional support to the jamming scenario . another intriguing similarity hinted to by recent work is the increase of heterogeneity in unperturbed glassy systems when temperature or packing fraction approaches a critical value , which is paralleled by the growth of spatial and temporal heterogeneity in yield - stress fluids when the shear rate vanishes . this analogy is particularly suggestive in the framework of jamming , because of the similar role that is attributed to temperature ( or interparticle potential and volume fraction ) and stress in driving the fluid - to - solid transition . while the mct and the jamming scenario provide some guidance in describing systems that approach the non - ergodicity transition from the fluid phase , our understanding of materials deeply quenched in an out - of - equilibrium phase is less advanced . although some common features are observed ( dynamical heterogeneity and a general trend for the dynamics to slow down with sample age being two of the most prominent ) , a large palette of different behaviors is observed , which remains largely unexplained ( see , e.g. , the various aging regimes discussed in this review ) . the concept of effective temperature may prove useful to describe quantitatively out - of - equilibrium soft materials and their aging ; however , current experimental determinations of @xmath178 are still too scarce and yield contradictory results . in particular , simultaneous measurements of the effective temperature for various observables and in a wider variety of systems will be necessary . finally , we observe that stress relaxation appears to be an important ingredient of the slow dynamics of many soft glasses ; in spite of that , it has been generally neglected in theoretical approaches , with the exception of the model for colloidal gels cited in section [ subsec : slow ] . recent advances in the investigation of slow dynamics in glassy soft matter have been made possible by close interactions between theory , simulation and experiments . on the experimental side , new methods and techniques have been developed to extend measurements to out - of - equilibrium and very slowly relaxing systems . a great effort is currently done to obtain spatially- and time - resolved information on the dynamics , and to identify the most insightful quantities to be extracted from the raw data in order to characterize heterogeneous behavior . future advances will likely include the combined ( and possibly simultaneous ) use of these methods , in order to achieve a more complete understanding of the various physical mechanisms that drive the relaxation of glassy soft matter , as well as of their interplay . conceptually , the focus will be in identifying and explaining the `` universal '' features of the slow dynamics in vastly different systems . we thank the numerous colleagues that shared with us their results before publication . e. zaccarelli , v. viasnoff , and a. kabla kindly provided us with their data for some of the figures of this review : we thank them warmly . we are indebted to many people for very useful and stimulating discussions ; in particular we wish to thank l. berthier , e. pitard , and v. trappe . we are grateful to present and past coworkers in our group : p. ballesta , a. duri , and s. mazoyer . financial support for our research on slow dynamics was provided by the french ministre pour la recherche ( aci jc2076 ) , rgion languedoc - roussillon , cnrs ( pics 2410 and projet mi - lourd `` dynamiques lentes et vieillissement de matriaux '' ) , cnes ( grants no . 02/4800000063 and 03/4800000123 ) , and the european community through the `` softcomp '' network of excellence and the marie curie research and training network `` arrested matter''(grant no . mrtn - ct-2003 - 504712 ) . l. c. thanks the institut universitaire de france for supporting his research . see for example _ soft and fragile matter_. edited by cates m and evans m ( bristol : institute of physics publishing 2000 ) liu a j and nagel s d 1998 jamming is not just cool anymore _ nature _ * 396 * 21 sollich p , lequeux f , hbraud p and cates m e 1997 rheology of soft glassy materials _ phys . _ * 78 * 2020 gtze w and sjogren l 1992 relaxation processes in supercooled liquids _ reports on progress in physics _ * 55 * 241 dawson k a 2002 the glass paradigm for colloidal glasses , gels , and other arrested states driven by attractive interactions _ curr . opin . colloid interface sci . _ * 7 * 218 kroy k , cates m e and poon w c k 2004 cluster mode - coupling approach to weak gelation in attractive colloids _ phys . * 92 * 148302 richert r 2002 heterogeneous dynamics in liquids : fluctuations in space and time _ j. phys . : condens * 14 * r703 stieger m , pedersen j s , lindner p and richtering w 2004 are thermoresponsive microgel model systems for concentrated colloidal suspensions ? a rheology and small - angle neutron scattering study _ langmuir _ * 20 * 7283 chen w r , chen s h and mallamace f 2002 small - angle neutron scattering study of the temperature - dependent attractive interaction in dense l64 copolymer micellar solutions and its relation to kinetic glass transition _ phys . e _ * 66 * 021403 habdas p and weeks e r 2002 video microscopy of colloidal suspensions and colloidal crystals _ curr . colloid interface sci . _ * 7 * 196 simeonova n b and kegel w k 2003 real - space recovery after photo - bleaching of concentrated suspensions of hard colloidal spheres _ faraday discuss . _ * 123 * 27 berne b j and pecora r 1976 _ dynamic light scattering _ ( new - york : john wiley & sons , inc . ) diat o , narayanan t , abernathy d l and grubel g 1998 small angle x - ray scattering from dynamic processes _ curr . colloid interface sci . _ * 3 * 305 weitz d a and pine d j 1993 _ diffusing - wave spectroscopy _ in _ dynamic light scattering _ ed brown w ( oxford : clarendon press ) wong a p y and wiltzius p 1993 dynamic light - scattering with a ccd camera _ rev . instrum . _ * 64 * 2547 bartsch e , frenz v , baschnagel j , schartl w and sillescu h 1997 the glass transition dynamics of polymer micronetwork colloids . a mode coupling analysis _ j. chem . phys . _ * 106 * 3743 cipelletti l and weitz d a 1999 ultralow - angle dynamic light scattering with a charge coupled device camera based multispeckle , multitau correlator _ rev . instrum . _ * 70 * 3214 viasnoff v , lequeux f and pine d j 2002 multispeckle diffusing - wave spectroscopy : a tool to study slow relaxation and time - dependent dynamics _ rev . instrum . _ * 73 * 2336 muller j and palberg t 1996 probing slow fluctuations in nonergodic systems : interleaved sampling technique _ prog . colloid polym . sci . _ * 100 * 121 pham k n , egelhaaf s u , moussaid a and pusey p n 2004 ensemble - averaging in dynamic light scattering by an echo technique _ rev . instrum . _ * 75 * 2419 lemieux p a and durian d j 1999 investigating non - gaussian scattering processes by using n - th order intensity correlation functions _ j. opt a _ * 16 * 1651 lemieux p a and durian d j 2001 quasi - elastic light scattering for intermittent dynamics _ app . _ * 40 * 3984 cipelletti l , bissig h , trappe v , ballesta p and mazoyer s 2003 time - resolved correlation : a new tool for studying temporally heterogeneous dynamics _ j. phys . : condens . matter _ * 15 * s257 pusey p n and van megen w 1986 phase - behavior of concentrated suspensions of nearly hard colloidal spheres _ nature _ * 320 * 340 pusey p n and van megen w 1987 observation of a glass - transition in suspensions of spherical colloidal particles _ phys . lett . _ * 59 * 2083 gtze w 1999 recent tests of the mode - coupling theory for glassy dynamics _ j. phys . : condens . matter _ * 11 * a1 van megen w and underwood s m 1994 glass - transition in colloidal hard spheres - measurement and mode - coupling - theory analysis of the coherent intermediate scattering function _ phys . e _ * 49 * 4206 bergenholtz j and fuchs m 1999 nonergodicity transitions in colloidal suspensions with attractive interactions _ e _ * 59 * 5706 fabbian l , gtze w , sciortino f , tartaglia p and thiery f 1999 ideal glass - glass transitions and logarithmic decay of correlations in a simple system _ phys . * 59 * r1347 dawson k , foffi g , fuchs m , gtze w , sciortino f , sperl m , tartaglia p , voigtmann t and zaccarelli e 2001 higher - order glass - transition singularities in colloidal systems with attractive interactions _ phys . rev . e _ * 63 * 011401 eckert t and bartsch e 2002 re - entrant glass transition in a colloid - polymer mixture with depletion attractions _ phys . lett . _ * 89 * 125701 pham k n et al . 2002 multiple glassy states in a simple model system _ science _ * 296 * 104 chen s h , chen w r and mallamace f 2003 the glass - to - glass transition and its end point in a copolymer micellar system _ science _ * 300 * 619 bhatia s r and mourchid 2002 a gelation of micellar block polyelectrolytes : evidence of glassy behavior in an attractive system _ langmuir _ * 18 * 6469 sciortino f 2002 disordered materials - one liquid , two glasses _ nat . * 1 * 145 poon w c k 2004 colloidal glasses _ mrs bulletin _ * 29 * 96 gtze w and sperl m 2002 logarithmic relaxation in glass - forming systems _ phys . rev . e _ * 66 * 011405 sperl m 2003 logarithmic relaxation in a colloidal system _ phys . e _ * 68 * 031405 sperl m 2004 dynamics in colloidal liquids near a crossing of glass- and gel - transition lines _ phys e _ * 69 * 011401 foffi g , dawson k a , buldyrev s v , sciortino f , zaccarelli e and tartaglia p 2002 evidence for an unusual dynamical - arrest scenario in short - ranged colloidal systems _ phys . e _ * 65 * 050802 zaccarelli e , foffi g , dawson k a , buldyrev s v , sciortino f and tartaglia p 2002 confirmation of anomalous dynamical arrest in attractive colloids : a molecular dynamics study _ phys . e _ * 66 * 041402 sciortino f , tartaglia p and zaccarelli e 2003 evidence of a higher - order singularity in dense short - ranged attractive colloids _ phys . lett . _ * 91 * 268301 puertas a m , fuchs m and cates m e 2002 comparative simulation study of colloidal gels and glasses _ phys . lett . _ * 88 * 098301 puertas a m , fuchs m and cates m e 2003 simulation study of nonergodicity transitions : gelation in colloidal systems with short - range attractions _ phys . e _ * 67 * 031406 cates m e , fuchs m , kroy k , poon w c k and puertas a m 2004 theory and simulation of gelation , arrest and yielding in attracting colloids _ j. phys . : condens . * 16 * s4861 puertas a m , fuchs m and cates m e 2004 dynamical heterogeneities close to a colloidal gel _ phys . _ * 121 * 2813 pham k n , egelhaaf s u , pusey p n and poon w c k 2004 glasses in hard spheres with short - range attraction _ phys * 69 * 011503 poon w c k 2002 the physics of a model colloid - polymer mixture _ j. phys . : condens . matter _ * 14 * r859 segre p n , van megen w , pusey p n , schatzel k and peters w 1995 2-color dynamic light - scattering _ j. mod . opt . _ * 42 * 1929 chen w r , mallamace f , glinka c j , fratini e and chen s h 2003 neutron- and light - scattering studies of the liquid - to - glass and glass - to - glass transitions in dense copolymer micellar solutions _ phys . * 68 * 041402 eckert t and bartsch e 2003 the effect of free polymer on the interactions and the glass transition dynamics of microgel colloids _ faraday discuss . _ * 123 * 51 grandjean j and mourchid a 2004 re - entrant glass transition and logarithmic decay in a jammed micellar system . rheology and dynamics investigation _ europhys . lett . _ * 65 * 712 bergenholtz j , poon w c k and fuchs m 2003 gelation in model colloid - polymer mixtures _ langmuir _ * 19 * 4493 stiakakis e , vlassopoulos d , likos c n , roovers j and meier g 2002 polymer - mediated melting in ultrasoft colloidal gels _ phys . rev . _ * 89 * 208302 tanaka h , meunier j and bonn d 2004 nonergodic states of charged colloidal suspensions : repulsive and attractive glasses and gels _ phys . e _ * 69 * 031404 del gado e , fierro a , de arcangelis l and coniglio a 2003 a unifying model for chemical and colloidal _ europhys . lett . _ * 63 * 1 del gado e , fierro a , de arcangelis l and coniglio a 2004 slow dynamics in gelation phenomena : from chemical gels to colloidal glasses _ phys . * 69 * 051103 segre p n , prasad v , schofield a b and weitz d a 2001 glasslike kinetic arrest at the colloidal - gelation transition _ phys . lett . _ * 86 * 6042 trappe v and sandkhler p 2004 colloidal gels - low - density disordered solid - like states _ curr . opin . colloid interface sci . _ * 8 * 494 foffi g , zaccarelli e , buldyrev s , sciortino f and tartaglia p 2004 aging in short - ranged attractive colloids : a numerical study _ j. chem . phys . _ * 120 * 8824 zaccarelli e , foffi g , sciortino f and tartaglia p 2003 activated bond - breaking processes preempt the observation of a sharp glass - glass transition in dense short - ranged attractive colloids _ phys . lett . _ * 91 * 108301 saika - voivod i , zaccarelli e , sciortino f , buldyrev s v and tartaglia p 2004 effect of bond lifetime on the dynamics of a short - range attractive colloidal system _ phys . e _ * 70 * 041401 van megen w , mortensen t c , williams s r and mller j 1998 measurement of the self - intermediate scattering function of suspensions of hard spherical particles near the glass transition _ phys . rev . e _ * 58 * 6073 mortensen t c and van megen w 1999 dynamic light scattering measurements of single particle motions in hard - sphere suspensions near the glass transition _ slow dynamics in complex systems : eighth tohwa university symposium_. edited by tokuyama m and oppenheim i. american institute of physics 3 bandyopadhyay r , liang d , yardimci h , sessoms d a , borthwick m a , mochrie s g j , harden j l and leheny r l 2004 evolution of particle - scale dynamics in an aging clay suspension _ phys . lett . _ * 93 * 228302 ediger m d 2000 spatially heterogeneous dynamics in supercooled liquids _ annu . * 51 * 99 glotzer s c 2000 spatially heterogeneous dynamics in liquids : insight from simulation _ j. non - cryst . solids _ * 274 * 342 garrahan j p and chandler d 2003 coarse - grained microscopic model of glass formers _ proc u. s. a. _ * 100 * 9710 whitelam s , berthier l and garrahan j p 2004 dynamic criticality in glass - forming liquids _ phys . lett . _ * 92 * 185705 chamon c , charbonneau p , cugliandolo l f , reichman d r and sellitto m 2004 out - of - equilibrium dynamical fluctuations in glassy systems _ j. chem . phys . _ * 121 * 10120 toninelli c , biroli g and fisher d s 2004 spatial structures and dynamics of kinetically constrained models of glasses _ phys . rev . lett . _ * 92 * 185504 biroli g and bouchaud j - p 2004 diverging length scale and upper critical dimension in the mode - coupling theory of the glass transition _ europhysics letters _ * 67 * 21 ritort f and sollich p 2003 glassy dynamics of kinetically constrained models _ adv . phys . _ * 52 * 219 crocker j c and grier d g 1996 methods of digital video microscopy for colloidal studies _ j. coll . interface sci . _ * * 179**298 kasper a , bartsch e and sillescu h 1998 self - diffusion in concentrated colloid suspensions studied by digital video microscopy of core - shell tracer particles _ langmuir _ * 14 * 5004 kegel w k and van blaaderen 2000 a direct observation of dynamical heterogeneities in colloidal hard - sphere suspensions _ science _ * 287 * 290 kob w , donati c , plimpton s j , poole p h and glotzer s c 1997 dynamical heterogeneities in a supercooled lennard - jones liquid _ phys . rev . lett . _ * 79 * 2827 weeks e r , crocker j c , levitt a c , schofield a and weitz d a 2000 three - dimensional direct imaging of structural relaxation near the colloidal glass transition _ science _ * 287 * 627 yethiraj a and van blaaderen a 2003 a colloidal model system with an interaction tunable from hard sphere to soft and dipolar _ nature _ * 421 * 513 donati c , douglas j f , kob w , plimpton s j , poole p h and glotzer s c 1998 stringlike cooperative motion in a supercooled liquid _ phys . lett . _ * 80 * 2338 weeks e r and weitz d a 2002 properties of cage rearrangements observed near the colloidal glass transition _ phys . lett . _ * 89 * 095704 cui b x , lin b h and rice s a 2001 dynamical heterogeneity in a dense quasi - two - dimensional colloidal liquid _ j. chem . phys . _ * 114 * 9142 knig h , zahn k and maret g 2004 glass transition in a two - dimensional system of magnetic colloids _ slow dynamics in complex systems _ aip conference proceedings * 708 * 40 knig h , hund r , zahn k and maret g 2004 experimental realization of a colloidal glass former in 2d _ submitted _ zahn k , mendezalcaraz j m and maret g 1997 hydrodynamic interactions may enhance the self - diffusion of colloidal particles _ phys . lett . _ * 79 * 175 knig h 2004 local particle rearrangements in a two - dimensional binary colloidal glass former _ slow dynamics in complex systems _ aip conference proceedings * 708 * 76 knig h 2004 elementary triangles in a two - dimensional ( 2d ) binary colloidal glass former _ submitted _ lemieux p a and durian d j 2000 from avalanches to fluid flow : a continuous picture of grain dynamics down a heap _ phys . lett . _ * 85 * 4273 duri a , ballesta p , cipelletti l , bissig h and trappe v 2004 fluctuations and noise in time - resolved light scattering experiments : measuring temporally heterogeneous dynamics _ to appear in fluctuation and noise letters _ and _ preprint cond - mat/0410150 _ sarcia r and hbraud p 2004 using diffusing - wave spectroscopy to study intermittent dynamics _ proceedings of the 2004 photon correlation and scattering conference _ 60 . ( also at http://gltrs.grc.nasa.gov/reports/2004/cp-2004-213207.pdf ) kabla a 2003 dsordre et plasticit dans les milieux diviss : mousses et matriaux granulaires phd thesis , universit paris vii . ( also at http://tel.ccsd.cnrs.fr/documents/archives0/00/00/36/26/ ) caballero g , lindner a , ovarlez g , reydellet g , lanuza j and clment e 2004 experiments in randomly agitated granular assemblies close to the jamming transition _ preprint cond - mat/0403604 _ kob w and barrat j l 2000 fluctuations , response and aging dynamics in a simple glass - forming liquid out of equilibrium _ eur . j. b _ * 13 * 319 pitard e 2004 finite - size effects and intermittency in a simple aging system _ preprint cond - mat/0409323 _ mayer p , bissig h , berthier l , cipelletti l , garrahan j p , sollich p and trappe v 2004 heterogeneous dynamics of coarsening systems _ phys * 93 * 115701 bramwell s t , holdsworth p c w and pinton j f 1998 universality of rare fluctuations in turbulence and critical phenomena _ nature _ * 396 * 552 bramwell s t et al . 2000 universal fluctuations in correlated systems _ lett . _ * 84 * 3744 clusel m , fortin j - y and holdsworth p c w 2004 criterion for universality - class - independent critical fluctuations : example of the two - dimensional ising model _ phys . e _ * 70 * 046112 castillo h e , chamon c , cugliandolo l f and kennett m p 2002 heterogeneous aging in spin glasses _ phys . lett . _ * 88 * 237201 sibani p , jensen h j intermittency , aging and record fluctuations _ preprint cond - mat/0403212 _ bissig h 2004 dynamics of two evolving systems : coarsening foam and attractive colloidal particles phd thesis , university of fribourg . ( also at http://www.unifr.ch/physics/mm/scm_publications.php?type=theses ) donati c , franz s , glotzer s c , parisi g 1999 theory of non - linear susceptibility and correlation length in glasses and liquids _ preprint cond - mat/9905433 _ laevic n , starr f w , schroder t b and glotzer s c 2003 spatially heterogeneous dynamics investigated via a time - dependent four - point density correlation function _ j. chem . phys . _ * 119 * 7372 ballesta p , ligoure c and cipelletti l 2004 temporal heterogeneity of the slow dynamics of a colloidal paste _ slow dynamics in complex systems _ aip conference proceedings * 708 * 68 ballesta p , duri a , ligoure c and cipelletti l , unpublished . marty g and dauchot o 2004 subdiffusion and cage effect in a sheared granular material _ preprint cond - mat/0407017 _ pouliquen o , belzons m and nicolas m 2003 fluctuating particle motion during shear induced granular compaction_phys . lett . _ * 91 * 014301 lootens d , van damme h and hbraud p 2003 giant stress fluctuations at the jamming transition _ phys . rev . lett . _ * 90 * 178301

constraints on dark matter self - scattering cross - section can be obtained from astrophysical observations . upper limits on the cross - section per unit mass , @xmath6 have been computed , most famously , from the bullet cluster @xcite which is a plane of the sky collision between two galaxy clusters . the distribution of mass was measured using lensing , and that of the gas using observations of its x - ray emission . from the consistency of the mass - to - light ratios of the sub - cluster and the main cluster , which implies that very little dark matter was lost due to self - interaction during the merger , @xcite derive an upper limit on the cross - section , as @xmath7 @xmath3 g@xmath4 . based on the same argument , but using @xmath8-body simulations to model the collision , @xcite derive @xmath9 @xmath3 g@xmath4 . the authors note that because the separation on the sky between the centroid of the visible galaxies and dark matter is consistent with zero , the dark matter s interaction cross - section is @xmath10 @xmath3 g@xmath4 . relaxed clusters offer a different route to the upper limit on @xmath6 . random scattering between dark matter particles will tend to make the dense cluster core more spherical . @xcite used an elliptical core of ms2137 - 23 to place an upper limit of @xmath11 @xmath3 g@xmath4 . the value relies on the assumption that the measured ellipticity comes solely from the cluster core , instead of projection of mass from outside the core , and that there has been no recent merger activity in the cluster . this is the most stringent upper limit on @xmath6 from astrophysical observations . clusters also present a potential opportunity to measure a lower limit on @xmath6 ( c.f . @xcite ) consider galaxies in the central regions of a cluster . these galaxies , usually ellipticals , consist of a compact stellar component and a more extended dark matter halo . as galaxies orbit the cluster centre , they move through the ambient dark matter halo of the cluster . the galaxies stellar matter and dark matter interact with the ambient dark matter differently . the motion of stars is subject to the gravitational field of the cluster and is unaffected by individual dark matter particles . the galaxies dark matter particles , on the other hand , can interact and scatter off the cluster s dark matter , thereby retarding the forward motion of galaxy s dark matter , making it lag behind the stellar component of the galaxy . if the scattering cross - section and the density of dark matter are large enough , this drag force will eventually result in a spatial separation between the stellar and dark matter components of the galaxy , which might be detectable . one can quantify the above scenario and estimate the separation as a function of relevant parameters . assume that the mass of the core of the cluster interior to the centre of the galaxy is @xmath12 , the dark matter and the stellar masses of the galaxy are @xmath13 and @xmath14 , respectively , and their average distance from the core are , @xmath15 . let the cross - sectional area of the galaxy s dark matter halo be @xmath16 . the force of gravity between the cluster core and the two components of the galaxy are , respectively , @xmath17 @xmath18.\label{newtdm}\ ] ] in the second equation gravity is partly counteracted , or screened by the self - interaction and scattering between the core s and galaxy s dark matter particles . the quantity @xmath19 is the covering factor , or the ` optical depth ' , and is assumed to be small . over time , the difference in the two accelerations amounts to a displacement , or separation @xmath20 if ( i ) the galaxy orbit were exactly radial , ( ii ) the cluster s dark matter space density scaled as @xmath21 , so @xmath22 were independent of radius at all epochs , ( iii ) the spatial distribution of stars and dark matter within the galaxy were point - like , and ( iv ) the expression in the square brackets in eq . [ newtdm ] were an appropriate way to quantify the retarding force , then the displacement would be exactly @xmath23 . of these , assumption ( iv ) is the least objectionable and will lead to a relatively small error . assumption ( iii ) , if it is not valid and dark matter halos are actually extended and overlapping , will tend to underestimate the force in equation ( [ newtdm ] ) and hence the separation @xmath24 . then a larger cross - section would be required to produce a given displacement . hence a point - like mass distributions is conservative , if a lower limit on @xmath6 is being sought . assumptions ( i ) and ( ii ) , and the @xmath25 dependence of @xmath24 in eq . [ disp ] combined summarize the dynamical evolution of the cluster and the galaxy s orbit , and are not necessarily independent of each other . for a typical relaxed cluster the central density profile is not too far from @xmath21 , however , it is not expected to be exactly that , and would have most likely varied over time . consequently , the shape of the galaxy s orbit and how it evolved in the past is important , and would constitute the largest source of uncertainty in @xmath24 . equation [ disp ] can be rewritten as @xmath26 thus , for typical values of masses and distances , and assuming that the self - scattering cross - section is at most a few decades below the most stringent astrophysically obtained upper limit , the separation between the dark and visible components of galaxies will be or order of kpc , and hence potentially observable . our order - of - magnitude estimate can be compared to the results of @xmath8-body simulations carried out by @xcite , who modeled the bullet cluster . for the specific geometry and parameters of that cluster , and assuming @xmath27 @xmath3 gm@xmath4 , the separation between the stellar and dark matter in the sub - cluster can be read off from their figure 5 , as @xmath28 kpc . this is smaller than the separation we get for the parameters assumed in eq . [ dispunits ] . the main difference between the two estimates is due to the nature of the encounters in the two cases . in the bullet cluster , the two merging clusters made one passage past each other , so that the sub - cluster spent little time in the dense core region of the larger cluster . for the bullet cluster , the current separation of the sub - components is 720 kpc . a merger velocity of 4700 km s@xmath4 ( following * ? ? ? * ) gives an encounter time of @xmath29 yrs . the velocity may well be significantly lower @xcite . but @xcite report that velocity of @xmath30 2800 km s@xmath4 would not greatly change their results . in the scenario we outlined above , the galaxies have been orbiting the inner cluster core for about a hubble time . also , @xmath31 kpc is a more appropriate value for the bullet cluster . taking these in to account , eq . [ dispunits ] gives @xmath32 kpc , comparable to their estimate of @xmath28 kpc . thus , the bullet cluster itself is a ( limited ) test of eq . [ dispunits ] . this comparison with the bullet cluster demonstrates that the longer duration of the encounter and smaller distances in our scenario give rise to a larger separation , more likely to be detected . abell 3827 is rather unique ; it is a nearby example of a multiple galactic cannibalism in progress , where the massive central ellipticals are closely surrounded by lensed images . the cluster s dynamical state , its proximity to us , and the fortuitous lensing configuration make it a prime candidate for detecting any displacement between the dark and visible matter due to non - zero dark matter scattering cross - section . in this paper we use strongly lensed images to reconstruct the mass distribution in the core of abell 3827 . as we will show , the core contains a galaxy whose visible and total mass are apparently separated by a few kpc . one possible interpretation of this result is the scenario described in the previous section . the cluster s lensing features were recently discovered by @xcite , whose findings we summarize below . abell 3827 ( @xmath33 ) appears to be in the last stages of merger @xcite . the outer cluster looks smooth and relaxed , and there are no significant secondary galaxy concentrations . the core of the cluster contains five ellipticals , called @xmath34 , within @xmath35 kpc of the centre , which appear to be in the process of spiraling inwards to form a single central galaxy . these ellipticals have probably undergone several orbits around the centre and have been tidally stripped , as evidenced by the extended @xmath36 diffuse light around the core . @xcite estimate the velocity dispersion of the cluster to be @xmath37 km s@xmath4 , making abell 3827 a rather massive cluster . the relative radial velocities of @xmath38 are within 400 km s@xmath4 of each other . the velocity of @xmath39 is about @xmath40 km s@xmath4 compared to the systemic velocity . @xmath41 is probably not a part of the cluster core because its velocity is about @xmath42 km s@xmath4 and its magnitude is @xmath43 fainter compared to that of the other four ellipticals . the cluster core lenses ( at least ) two background sources ; the main source , at @xmath44 , is extended and consists of at least three knots , @xmath45 . all three appear to be lensed into quads , but in one quad only three images are identified , and in another , only two . the second source , at @xmath46 , forms a single arc ; a counter arc may be lost in in the glare of the ellipticals . based on their spatial distribution , the images of the quads can be labelled by their time order , even though time delays have not been measured ( cf . accordingly , we will write @xmath47 to mean the @xmath48th arriving image of the @xmath49th knot within source @xmath50 . the images to the north - east of the cluster s centre of light are the 1st arriving ones ; the two sets of images in the west are ( clockwise ) the 3rd and 2nd arriving , and finally the images to the south - east are 4th arriving . the 2nd and 3rd arriving images are close together and form a nearly merging pairs across the critical curve . one can loosely divide the lens modeling methods into two categories , depending on the prior information they use . the fixed - form methods assume that mass follows light , and parametrize this relation with a handful of scalings . free - form methods use different types of priors ; most notably they do not require that mass trace light . the authors of the discovery paper , @xcite carried out a lensing mass reconstruction using lenstool , a fixed - form method @xcite . the reconstruction reproduces the positions of only 5 of the 10 images well . the mean scatter between observed and recovered images is less than @xmath51 , while the astrometric uncertainty is @xmath52 . overall , the mass model has @xmath53 per degree of freedom , indicating that there is additional information in the astrometry . this is not surprising ; the dynamical state of the cluster core is probably unsettled , so it is quite possible that light does not follow mass . this , combined with fact that abell 3827 has a relatively large number of lensed images ten makes a free - form reconstruction a logical choice . we use a free - form reconstruction method , pixelens @xcite . the reconstruction is done in a circular window , which is divided into many equal - sized square pixels . the mass within each pixel is uniformly distributed but each pixel s mass value is allowed to vary independently of others . pixelens takes advantage of the fact that the lensing equation is linear in the mass pixels and in the coordinates of the source , and so it is relatively straightforward to come up with solutions that reproduce image positions exactly . in fact , since the unknowns outnumber the knowns ( coordinates of the lensed images ) by a large factor , an infinity of solutions are possible . pixelens generates a user - specified number of individual models , according to a weak and adjustable prior of which the ensemble average is considered in this paper . the prior and the model - sampling technique are motivated and explained in detail in the above papers , so here we just remark on two points that are important for the modelling . first , along with the image positions , a notional centre must be given . the default behaviour of pixelens is to require the local density everywhere to point at most @xmath54 away from the centre . in this work we increase the allowed angle ( called dgcone for `` density - gradient cone '' ) somewhat , typically to @xmath55 this allows more substructure , as appears to be demanded by the data . second , to account for the mass exterior to the modeling window , pixelens can add external shear of any magnitude . the shear axis can be within @xmath54 of a user - specified shear direction , denoted by shear . we typically set shear 60 on the basis of the axis of the diffuse light on arcminute scale around the cluster core which permits any shear orientation from @xmath56 to @xmath57 . the version of the lens - reconstruction software used is included as an online supplement . other examples of free - form techniques for cluster lenses are described in @xcite and @xcite . to start with , we create a reference , or fiducial reconstruction . it uses all the 10 images identified by @xcite . as the centre we use the centre of light ( also the centre of their figure 2 frame ) , which is about @xmath58 west of n2 . at the redshift of the cluster , @xmath59 , the critical surface mass density for sources at infinity is @xmath60 gm/@xmath3 , and @xmath61 corresponds to 1.93 kpc , for @xmath62 , @xmath63 , and @xmath64 . the diameter of the reconstruction window was set to 29 pixels , with the scale of @xmath65/pixel , so the angular and physical size of the window are about @xmath66 and 77 kpc , respectively . the fiducial pixelens reconstruction is shown in figure [ kmass ] ; the mass density contours are spaced by @xmath67 . the map is an average of 500 individual models . ( the mass map is not sensitive to the number of individual models in the ensemble as long as it is greater than @xmath68 , the pixel size within the range @xmath69-@xmath58 per pixel , or the orientation of the reconstruction window with respect to the cardinal directions . ) figure [ massenc ] is a plot of mass enclosed as a function of radius . our fiducial reconstruction is indistinguishable from the average of 18 reconstructions discussed in sections [ effp ] and [ effs ] , and shown as the red thick line . the thin dotted lines are all the 18 models , and the errorbars represent their rms dispersion . the two blue solid flanking lines enclose 90% of the 500 individual mass models of the fiducial reconstruction . note that at large radii , i.e. outside the images , the enclosed mass begins to level off since the image positions do not require any mass at those radii . however , the 90% errorbars are fully consistent with the density profile continuing on with the same slope as at smaller radii . figure [ karriv ] shows the arrival time contours of the @xmath70 and @xmath71 knots of source @xmath50 . from the contours , we see that the model indicates extra images appearing next to the first arriving image of @xmath70 . these may be spurious , or may actually be present , depending on the local mass substructure . they do not , however , affect the larger - scale features of the mass maps , which are the focus of this paper . the plot of arrival - time contours also provides a simple model for the ring - like extended image ( cf . * ) which in this case arises from the extended source @xmath50 . detailed comparison would , however , be very difficult because the ring is faint and in many places overlaid with the light from the ellipticals . we have not attempted to use the arc @xmath72 as a lensing constraint , since it has no identified counter image . in principle it is possible to extract some strong - lensing information from this arc . if pointlike features could be identified , they could be put in as multiple - image constraints in the usual way . another possibility would be to assume two fictitious image positions along the arc , which in models would have the effect of forcing a critical curve to pass through the arc . the mass contours in figure [ kmass ] show deviations from circular symmetry , which are due to mass inhomogeneities in the lens . we isolate these by subtracting a smooth component , as follows . figure [ profile ] shows pixel mass as a function of the distance from the centre of the fiducial reconstruction . as already mentioned when discussing figure [ massenc ] , the lack of lensing constraints beyond about 20 kpc leads pixelens to put very little mass into those pixels ; so the tapering off of points in figure [ profile ] is an artifact . the average projected density profile slope where the images constrain the mass well , i.e. around @xmath73 , is @xmath74 , and is representative of most mass maps in this paper . this is shown as the thick straight line in this log - log plot ; the other two lines have slopes of @xmath75 and @xmath76 , and represent the reasonable range of possible power - law slopes . to isolate the mass inhomogeneities we subtract circularly symmetric density profile of slope @xmath77 from the reconstructed mass maps ; a change of @xmath78 in the assumed smooth profile slope would not have made a significant difference . the normalization of the subtracted profile is adjusted such that the contours of substructure , or the excess mass density left after the subtraction of the smooth cluster profile , contain 0.5 , 1 , 2 , 4 and 8 % of the total mass within the full @xmath66 window . this normalization is calculated separately for each mass map and each of the five contour levels . figure [ fiducial ] shows the five contour levels of the mass substructure , as red solid lines ; thicker lines represent contours of smaller enclosed mass . images are shown as black solid dots , and labeled at the periphery of the figure . the arc @xmath79 has no detected counter images . image @xmath80 was not identified by @xcite , but looking at their figure 2 , the present authors feel @xmath80 is the location of that image . it is not used in the fiducial reconstruction but is used in some other ones . galaxies are denoted by green squares and labeled as in @xcite ; @xmath41 is a smaller square because the galaxy is probably background and less massive . figure [ fiducial ] shows that the mass substructure does not trace the visible galaxies , an indication that the cluster core is not in equilibrium , but is dynamically disturbed . there is one dominant mass clump , to the north east ( upper left ) of galaxy @xmath81 , but not centered on @xmath81 . this clump is the main subject of the paper , and we will refer to it as the ne-@xmath81 clump . a smaller clump is between galaxies @xmath82 , and seems to be avoiding @xmath83 . the following sections will demonstrate that even though the details of the reconstructions differ , the main features of this fiducial mass map , and most importantly the existence and location of ne-@xmath81 clump , and the lack of mass associated with @xmath83 , are robust to changes in priors , centre of the reconstruction window , and , to some degree , the subsets of images used in reconstruction . figure [ lots3 ] shows maps of substructure mass contours for nine reconstructions . the image set used is the same as in the fiducial case , but the priors are different . the assumed values of dgcone , and shear are shown in each panel . since the cluster core is most likely not in equilibrium , we tried three different lens centres . the fiducial one is called @xmath84 , and was identified by @xcite as the cluster s centre of light ; it sits between @xmath83 and @xmath85 . centre @xmath86 is shifted by @xmath87 from @xmath84 and is close to @xmath85 . centre @xmath88 is shifted by @xmath89 with respect to @xmath84 ; it coincides with the location of galaxy @xmath83 , the central - most elliptical in the core . the three choices for the lens centre are used in the reconstructions of the three rows of figure [ lots3 ] , respectively . their location is marked with a cross . ( we remark again that the lens centre in pixelens is needed because the density gradients , used for dgcone , are calculated with respect to the lens centre . the other utility of the lens centre will be put to use in section [ cent ] . ) figure [ lots3 ] shows that all reconstructions , with the possible exception of the middle panel of the third row , require a mass lump to the ne of galaxy @xmath81 . the size and position of the lump vary between panels , but in none of these is the lump centered on @xmath81 ; it is always to the upper left of that galaxy . the other common feature is that @xmath83 has relatively little excess mass associated with it . the nearby mass clump sometimes encompasses @xmath85 or @xmath39 , but not @xmath83 . figure [ lots4 ] shows the effect of leaving out some of the lensed images from the reconstruction . the images plotted in each panel are the only ones used in the corresponding reconstruction . the lens centre is always @xmath84 , and the pixelens priors are shown in each panel . most changes to the image set makes little difference to the global features of the recovered mass maps , and specifically the ne-@xmath81 mass clump . since this mass clump is of primary interest for us , we examine its dependence on the images more closely . from this figure , and other reconstructions not shown here , we know that the single most important image that determines the location of the clump near @xmath81 is image @xmath90 . in the lower right panel @xmath90 was not used , and the mass clump moved to a different , lower , or more southerly , location . in that reconstruction , both the mass distribution and the image configuration have an approximate bilateral symmetry , with the axis inclined by @xmath91 to the @xmath92-axis . the location of @xmath90 predicted by such a bilaterally symmetric mass distribution would have been very close to the observed @xmath93 . however , the observed @xmath90 is about @xmath94 clockwise from @xmath93 . to make that happen , one needs a large mass clump roughly where pixelens puts the ne-@xmath81 clump . this can be understood in terms of the arrival time surface . @xmath90 is a saddle point image . a sufficiently large mass just outside of the image circle will create a local bump in the arrival time surface and a saddle point will be created between it and the centre of the lens . in section [ expl ] we will present additional arguments to show why @xmath90 requires a large mass to exist beyond the location of galaxy @xmath81 . one might be tempted to say that @xmath90 has been misidentified , and is actually much closer to @xmath93 . if that is assumed , the reconstructed mass and observed galaxy positions bear no correspondence to each other , as shown in the lower right panel of figure [ lots4 ] . furthermore , the imaging data as presented in @xcite shows a definite , though faint smudge at the location of @xmath90 . we conclude that @xmath90 is real , and hence the true mass distribution is close to what our fiducial reconstruction looks like . the mass substructure , i.e. the excess left after subtracting the smooth cluster profile , of all eighteen reconstructions of figures [ lots3 ] and [ lots4 ] are superimposed in figure [ superpose ] . for clarity , we plot only the contours enclosing 1% of the total mass in every reconstruction . the figure highlights the features common to most reconstructions : the massive ne-@xmath81 mass clump , and the secondary clump around galaxies @xmath85 and @xmath39 , which avoids @xmath83 . the main goal of the paper is to estimate the lower limit on the dark matter cross - section using eq . [ dispunits ] as applied to galaxy @xmath81 and the ne-@xmath81 mass clump . in this section we estimate the parameters associated with the ne-@xmath81 clump , namely , its mass , @xmath13 , size , @xmath95 , and distance from @xmath81 , @xmath24 , and give a measure of dispersion in these properties between reconstructions . figure [ superpose ] gives a visual impression of the systematic uncertainties in our reconstructions . because the shape and location of the ne-@xmath81 clump is not exactly the same in various reconstructions , there is no unique way of quantifying its average properties and estimating the dispersion in these properties between the various mass maps . we chose to use an iso - density mass substructure contour to define the clump , and the corresponding mass excess within that contour as the clump s mass , its center of mass as the center of the clump , and the enclosed area , @xmath50 , to define the size of the clump , i.e. @xmath96 . first , we need to decide what to take as the boundary of the clump . in most reconstructions , including the fiducial one , the mass substructure contours containing 0.5 and 1% of the total mass are disjointed , and enclose separate mass clumps , while the contours containing 4 and 8% of the total mass form one contiguous region . this suggests that @xmath97 of the total recovered mass is in the individual clumps , therefore we take the closed 1% contour located to the ne of the @xmath81 galaxy as defining the clump . two of the reconstructions , the bottom middle in figure [ lots3 ] and the bottom right in figure [ lots4 ] have no excess mass at 1% level in the vicinity of @xmath81 , so we exclude them . with these choices , the average and rms dispersion in the mass of the ne-@xmath81 clump based on the 16 maps is @xmath98 , its distance from the centre of the elliptical @xmath81 is @xmath99 , and its size @xmath100 . our fiducial reconstruction gives @xmath101 , @xmath102 , and @xmath103 , all of which are within the rms of the 16 maps , so when estimating @xmath6 in section [ dmx ] we will use the fiducial values . we also need to estimate @xmath104 , the clumps distance from the cluster centre . equation [ disp ] assumes that the centre of the elliptical galaxy and the centre of its displaced dark matter halo are equidistant from the cluster centre , i.e. @xmath105 since the displacement @xmath24 is small . therefore we take @xmath106 , same for all reconstructions . the uncertainty in @xmath104 comes from the uncertainty in the location of the cluster centre . in section [ cent ] below we show that the latter is of order @xmath107 , which translated into a fractional error , which is smaller than that in the other two relevant lengths , @xmath24 or @xmath95 . while the errors in the estimated properties of the ne-@xmath81 mass clump are relatively small , @xmath108 , we stress that these are not the main sources of uncertainty in estimating the parameters of the dark matter particle cross - section , @xmath6 . the dominant sources of uncertainty are the unknown dynamics of the cluster , the evolution in cluster core s density profile slope , and the galaxy orbit shape , especially that of elliptical @xmath81 ( see section [ intro ] ) , and can not be quantified using lensing data . to properly address these one would need to carry out a series of n - body simulations with a range of initial conditions . based on their observations , @xcite quote image astrometric uncertainty of @xmath52 . since pixelens reproduces image positions exactly , we need to check if astrometric errors could have affected our results . here we redo the fiducial reconstruction but randomly shift the observed images within @xmath109 of their true positions . five realizations are shown in figure [ fake ] . to reduce run - time , we used 300 models to create an ensemble , instead of the usual 500 . since all five reconstructions look very similar and have small rms dispersions ( @xmath110 and @xmath111 ) , smaller than the ones presented in the section [ props ] we conclude that astrometric errors have not affected our results . in addition to the ne-@xmath81 clump , the other notable feature of the reconstructions is that regardless of the input model parameters ( see figures [ fiducial]-[fake ] ) there is very little mass excess associated with @xmath83 , the centrally located elliptical galaxy . this suggests that the peak of the mass distribution is not very close to @xmath83 . here we ask which one of our three trial centres , @xmath84 , @xmath86 , or @xmath88 comes closest to the true cluster centre , defined as the location of the largest mass concentration . we use the following procedure . a pixelens pixel can assume any mass value provided it agrees with the image positions and priors , like dgcone and shear . if there is no smoothing , the resulting maps usually show a lot of fluctuations between neighbouring pixels . to avoid that , we impose a nearest neighbour smoothing constraint by requiring that a pixel can not have a mass value larger than twice the average of its eight neighbours . the only exception to this rule is the pixel at the centre of the reconstruction window , which has no imposed upper bound . it is reasonable to suppose that the closer the reconstruction centre is to the real centre , the more mass the central pixel will contain . from our total ( not substructure ) mass maps we calculate the fraction of total mass contained in the central pixel . for the centres @xmath84 , @xmath86 and @xmath88 we calculate that percentage of mass as @xmath112 , @xmath113 , and @xmath114 respectively . these values are averages and rms for the three sets of three maps shown in figure [ lots3 ] . recall that @xmath88 is centered on galaxy @xmath83 , whereas @xmath84 and @xmath86 correspond to blank spots between @xmath83 and @xmath85 . since the central pixel of @xmath88 contains the least mass , and has the smallest dispersion of the three cases , it supports our finding in the earlier sections that there is very little excess mass associated with the optical light of elliptical @xmath83 . of the three , centre @xmath86 is closest to the largest concentration of mass in the cluster . finding the centre of the cluster is not the main goal of the paper , however this section demonstrates that the uncertainty in the location of the centre is of order of @xmath107 , and hence smaller than fractional uncertainties in the other length measurements entering eq . [ dispunits ] , i.e. @xmath24 and @xmath95 , therefore when estimating @xmath104 it is reasonable to use @xmath84 as the centre of the cluster , as we have done in section [ props ] . to further test the robustness of the ne-@xmath81 mass clump we carry out a reconstruction where pixelens is allowed to add point masses at specified pixel locations . in other words , we allow some pixels to override the nearest neighbour constraint described in section [ cent ] . we chose pixels containing the observed galaxies @xmath81-@xmath39 , and use pixelens command ptmass x y area1 area2 , where x and y is the location of the point mass , and area1 area2 are the lower and upper limits on the area enclosed by its einstein ring . we used @xmath115 and @xmath0 arcsec@xmath116 as representative values for mass associated with individual galaxies . figure [ ptmass ] shows the results . the contour levels of the mass substructure are the same as before , at 0.5 , 1 , 2 , 4 , and 8% of total . there is a substantial amount of mass associated with individual galaxies , including @xmath81 , but a large mass to the upper left of @xmath81 is still required , as shown by the 1% substructure mass contour . in fact , the ne-@xmath81 clump , without the @xmath81 point mass , roughly contains @xmath117 , similar to what we find for the fiducial reconstruction , in section [ dmx ] . also consistent with earlier findings , is that compared to other ellipticals , @xmath83 has the least amount of mass associated with it . in this section we create a series of four synthetic mass distributions that illustrate why an excess mass lump is needed where pixelens places it , to the north east of @xmath81 . these synthetic maps have mass distributions similar enough to the real cluster to make the comparison meaningful . however the images produced by them do not attempt to exactly reproduce the observed images . in figure [ fourmassc ] the mass map in the upper left is roughly what abell 3827 would look like if mass followed light , in other words , mass clumps are placed at the observed locations of galaxies @xmath81-@xmath39 , indicated by the green squares . the blue contour has the critical surface mass density for @xmath118 . the black solid dots are the locations of the observed images of source @xmath70 . the four extended islands of red points mark the locations of images produced by this synthetic mass configuration . these images were selected such that the second arriving images cluster around the observed @xmath119 . the other images were unconstrained . the locations of the 1st , 2nd and 3rd arriving images can be relatively well reproduced by the mass - follows - light scenario . however , the 4th arriving images are consistently far from the observed @xmath90 . this is the basic reason why mass - follows - light models do not work . what does one have to do to ( approximately ) reproduce @xmath90 ? in the upper right panel of figure [ fourmassc ] we move the mass clump associated with @xmath81 to the left ( east ) . the 4th arriving images are hardly affected . next , in the lower left panel we keep the mass clump at its original position but increase its mass by @xmath120 . the configuration of images changes but the location of @xmath90 is still not reproduced . finally , in the lower right panel we move the mass clump to the left _ and _ increase its mass by @xmath120 . now the bulk of the 4th arriving images move to where the observed location of @xmath90 . this is basically the mass distribution that pixelens finds , with its massive ne-@xmath81 clump . a further test of the veracity of pixelens s reconstructions is to use the images created by the synthetic mass distributions of the previous section as input for pixelens . we use the mass distributions of the upper left and lower right panels of figure [ fourmassc ] . for each of these we take two quads from the red dots in figure [ fourmassc ] , for a total of eight images to be used as lensing constraints . the two panels of figure [ pxmass ] show the synthetic mass distributions as the smooth contours , while the jagged contours are the pixelens reconstructions . the black mass contours are at 0.25 , 0.5 , 0.75 and 2 of the critical surface mass density for @xmath118 , while the blue contour is at the critical density . the magenta empty circles are the images used in the reconstruction . the first of the two synthetic mass models ( left panel ) is approximately elliptical and does not have a mass lump to the left of the centre . pixelens recovers the mass map quite well ; most importantly it does not create a fictitious massive clump near the 4th arriving image . the second synthetic mass map ( right panel ) does have a mass clump , and pixelens recovers it well . the reconstructions of the preceding sections and the tests with synthetic lenses show that the two main features of the lensing mass reconstructions are robust . ( 1 ) the ne-@xmath81 mass clump is present consistently in most reconstructions , and it is never centered on galaxy @xmath81 . ( 2 ) the second , less massive clump is near galaxies @xmath85 and @xmath39 , but avoids @xmath83 , so that there is no excess mass associated with that central elliptical . we interpret ( 1 ) to mean that the visible and the dark components of @xmath81 are separated , and we hypothesize that the cause of the separation is the scenario described in section [ intro ] , namely that the scattering between the galaxy s and the cluster s dark matter particles induced the galaxy s halo to lag behind its stellar component . other interpretations of the observed separation are discussed in section [ other ] . conclusion ( 2 ) also speaks to the dynamically disturbed nature of the cluster core , though there is not enough information to speculate what happened to the halo of @xmath83 and how the mass clump is related to the nearby @xmath85 and @xmath39 . we now concentrate on ( 1 ) , and estimate @xmath6 from eq . [ dispunits ] and the parameters based on the reconstructions presented in sections [ effp ] and [ effs ] and summarized in section [ props ] . we do not consider the rms dispersions quoted in that section here , because the main source of uncertainty is not in the measured properties of the ne-@xmath81 clump , but in the dynamics of the cluster . for a conservative lower limit on @xmath6 we want to use the smallest linear distances , @xmath104 , @xmath95 , and @xmath24 consistent with the reconstruction , so we use the sky projected quantities , and make no attempt to estimate the corresponding three - dimensional values . to obtain @xmath12 we measure the projected mass within @xmath104 from figure [ massenc ] and halve it to account for the core being a sphere instead of a long tube ; @xmath121 . one major source of uncertainty is related to the quantity @xmath22 entering eq . [ dispunits ] . first , if @xmath22 depends on radius , i.e. distance from cluster center , then the unknown shape and ellipticity of @xmath81 s orbit and hence how it ` samples ' @xmath22 becomes an issue . second , the radial dependence of @xmath22 may well be time dependent . from our lensing reconstruction we can measure how this quantity scaled with radius at the present epoch . the projected @xmath122 , which implies that the space density @xmath123 , and mass enclosed @xmath124 , so @xmath125 . this is a relatively weak radial dependence , but it does not tell us much about the temporal dependence . in future work , these uncertainties can be constrained through n - body simulations that sample possible merger histories and have the present day configuration resembling abell 3827 . the other large source of uncertainty is also related to cluster dynamics , and it is @xmath5 , the duration of the encounter , because it can not be measured directly from the reconstruction . a lower limit on @xmath5 would be one dynamical time at the present radius if @xmath81 , roughly @xmath126yr . the upper limit is the age of the universe , @xmath127 yrs . the most likely @xmath5 would probably correspond to several dynamical times within the extended core region of the cluster , or @xmath128 yr . putting all the parameters together in to eq . [ dispunits ] we express the lower limit on the dark matter self - interaction cross - section as @xmath129 while this value is safely below the astrophysically measured upper limit on self - scattering , it is interesting to compare it to the estimates of the self - annihilation cross - section for dark matter particles , because in most particle models the latter are expected to be larger than the scattering cross - sections . to facilitate the comparison we express our result in terms of cross - section per particle of mass @xmath130 , and attach a subscript to indicate that this refers to self - scattering , @xmath131 if the dark matter particle is a thermal relic , its self - annihilation cross - section can be expressed in terms of dark matter density and typical velocity @xcite , @xmath132 a different constraint comes from considering the observed abundance of annihilation by - products . for neutrinos , @xcite get an upper bound which is more stringent than provided by other experimental and observational considerations , @xmath133 for a range of particle masses between @xmath134 and @xmath135 gev . while our result is well above ( i.e. inconsistent with ) the thermal relic estimate , @xmath136 , for particle masses above @xmath134 mev , it is comparable to @xmath137 , especially if particle mass is somewhat below @xmath138 gev . here we offer other interpretations for the separation between the ne-@xmath81 mass clump and the elliptical galaxy @xmath81 . \(i ) the ne-@xmath81 mass clump could belong to the line of sight structures , and not the cluster core , or galaxy @xmath81 . in that case the alignment of ne-@xmath81 clump with the cluster ellipticals is not expected . however , there are no additional light concentrations in the direction of the ne-@xmath81 clump , so if this mass is external to the cluster core , it would still need to have a very high mass - to - light ratio . \(ii ) the ne-@xmath81 mass clump could be primarily due to the baryonic gas associated with galaxy @xmath81 , which has been separated from its parent galaxy by ram pressure of the cluster gas . however , in the cluster s core one would expect the gas of individual ellipticals to have been stripped and dispersed on short time - scales , and in the case of a late stage merger , as in abell 3827 , long before the galaxies of the merging clusters reach the new common cluster core . observations of other massive clusters do not show x - ray emitting gas associated with individual galaxies in the core . examination of an archival chandra image of abell 3827 also shows no discernible substructure ( proposal i d 08800836 ) . ( iii ) because the visible and dark components of @xmath81 have different spatial extents , the gravitational field gradient of the cluster could have resulted in different forces acting on the two components , which could lead to the observed separation . however , it is hard to imagine how tidal forces would lead to such a large offset . another possibility is that the stellar component of @xmath81 is wobbling around the centre of its dark matter s potential well . separation due to this effect have been observed , but only for the central cluster galaxies . @xmath81 is not the central elliptical ; @xmath81 and ne-@xmath81 clump are both about @xmath0 kpc from cluster s largest mass concentration ( section [ cent ] ) . \(iv ) finally , because the lensed images are faint , extended and close to bright ellipticals , the identification and location of images could be partially wrong . since image positions are the most important input for pixelens mass reconstruction , misidentified images could lead to wrong recovered mass maps . additional observations would help ; for now we note that random astrometric uncertainties , of order @xmath139 , in the image positions do not affect the main features of the reconstruction ( section [ uncert ] ) . in this paper we argue that if the dark matter self - scattering cross - section is only a few orders of magnitude below its most stringent astrophysically determined upper limit , @xmath140 @xmath3 g@xmath4 , then it should be detectable in the cores of massive galaxy clusters . the dark matter particles of individual galaxies that have been orbiting the cluster centre for about a hubble time will be scattering off the dark matter of the cluster , resulting in a drag force , not experienced by the visible stellar component of the galaxies . over time this drag will amount to a spatial separation between the dark and visible components of the galaxy , whose magnitude is estimated by eq . [ dispunits ] . we suggest that at least one elliptical in the core of abell 3827 is an example of such a scenario . our free - form lensing reconstructions show that there a massive dark clump , the ne-@xmath81 clump , located about @xmath141 , or @xmath1 kpc from the centre of light of @xmath81 . we performed many mass reconstructions to test the robustness of the ne-@xmath81 clump , and quantified the uncertainty in its properties . a series of synthetic mass models , presented in section [ tests ] , further support the existence of the clump and its separation from the visible component of @xmath81 . we interpret this clump as the dark matter halo of @xmath81 which is lagging behind its stellar counterpart . if this interpretation is correct , we estimate the lower limit on the cross - section as given in eq . [ estim ] . the largest source of systematic uncertainty in this equation relates to the unknown dynamics of the cluster and central galaxies , the shape and inclination of the orbit of @xmath81 , and the true three dimensional distances and separations within the core . in section [ other ] we note that other interpretations of the observed separation can not be ruled out at this point . future observations as well as numerical simulations should be able to differentiate between the various scenarios , and also quantify the uncertainties associated with the dynamical state of the cluster . using purely collisionless simulations . the stellar and dark matter components of superimposed , mass distributions of different several orbits around the cluster centre the two components wobble there is no need for non - zero @xmath6 , section [ intro ] , check if an observable separation can develop over a hubble time . as already mentioned , abell 3827 is special in having very favorable parameters for the detection of the dark matter - light offset . first , a3827 is a very massive cluster , hosting several massive ellipticals in the very core . @xcite note that within a similar radius , abell 3827 is slightly more massive than abell 1689 @xcite , a cluster with over 100 strong lensing features . they also note that the central galaxy of abell 3827 is perhaps the most massive cd galaxy in the local universe . second , the fortuitous location of the background source a produces strong lensing features that are optimal in configuration and distance for mass reconstruction in the most relevant regions ; the low redshift of abell 3827 and source a ensure that the images are close to the cluster centre , @xmath142 . these factors make it unlikely that there are many other clusters out there where the offset would be detectable . llrw would like to thank henry lee for discussing with her the images of this unusual object prior to publication , keith olive for discussions about particle dark matter , and larry rudnick for help with chandra data archive . beacom , j. f. , bell , n. f. & mack , g. d. 2007 , phys.rev.lett . , 99 , 231301 bertone , g. , hooper , d. & silk , j. 2005 , phys.rept . , 405 , 279 carrasco , e. r. , gomez , p. l. , verdugo , t. , lee , h. , diaz , r. , bergmann , m. , turner , j. e. h. , miller , b. w. & west , m. j. 2010 , apj , 715 , l160 clowe , d. , bradac , m. , gonzalez , a. h. , markevitch , m. , randall , s. w. , jones , c. , zaritsky , d 2006 , apj , 648 , l109 coe , d. , fuselier , e. , bentez , n. , broadhurst , t . , frye , b. , ford , h. , 2008 , apj , 681 , 814

soft x - ray transients ( sxts ) are close binary systems which undergo large amplitude outbursts with a recurrence time of about @xmath1 years ( see e.g. white 1994 for a review ) . in most cases sxts disappear from the x - ray sky between outbursts leaving only the companion star observable as a dim optical source . most sxts are low mass x - ray binaries ( lmxbs ) , that is , binary systems in which a neutron star or a black hole accretes matter lost by a roche - lobe filling low - mass stellar companion . a minority of sxts show x - ray bursts during decline from outburst ( e.g. koyama et al . 1981 ) and must contain neutron stars . in several other sxts , observations of the companion have allowed measurement of a mass function which is larger than the maximum mass of a neutron star ( orosz et al . 1994 ; casares & charles 1994 ; orosz et al . 1995 ; casares , charles & marsh 1995 ; bailyn et al . these are certain to be black holes . for the remaining sxts there is no direct evidence on whether they have neutron stars or black holes , but it is believed that the majority of them contain accreting black holes ( see tanaka & lewin 1995 and van paradijs & mcclintock 1995 ) . most of the outburst energy in an sxt is emitted in rather soft ( 1 to few kev ) x - rays . it has been claimed that the spectra of black hole sxts ( bsxts ) are characterised in outburst by an ` ultra soft ' ( white et al . 1984 ) component , and that this is not seen in neutron star sxts . however , in a few cases involving good black hole candidates ( e.g. v404 cyg ) such a component has not been observed ( harmon et al . at least one black hole sxt was observed to emit very high energy photons , namely nova muscae 1991 which had an e@xmath2@xmath3 annihilation line ( goldwurm et al . 1993 ) . on the other hand some neutron star sxts have also been observed to emit above 100 kev ( barret et al . it is therefore not clear if there are any spectral characteristics that would allow , in general , to distinguish between neutron star and black hole transients . the binary periods of sxts generally lie in the range between 5 hours and 6 days . ( the ` ultra - soft ' transient system 4u 1630 - 47 , [ parmar , angelini & white 1995 ] could be an exception if its period is confirmed to be 601.7@xmath4 days . ) sxts are therefore lmxbs with main sequence dwarf companions or , at longer periods , ( sub)giant companions ( king 1993 ) . as in the case of their close cousins , the dwarf novae , which have central white dwarfs instead of neutron stars or black holes , two kinds of models have been proposed to explain the outbursts of sxts . hameury , king & lasota ( 1986 ) proposed a model in which outburts are due entirely to a mass transfer instability in the x - ray illuminated regions of the mass - losing companion . it was shown , however , by gontikakis & hameury ( 1993 ) that the existence of sxts with orbital periods under @xmath5 h is a problem since the model is unable to produce the observed short ( @xmath6 s ) timescales ( see also lasota 1995 ) . the dwarf nova disk instability model ( dim ) ( see e.g. cannizzo 1993 ) has been extended to the case of bsxts by huang & wheeler ( 1989 ) and mineshige & wheeler ( 1989 ) . below we discuss problems and difficulties that this model encounters when confronted with observations . in addition to the above two proposals , one should also consider models in which the two mechanisms are at work in the same source . for example , the disk instabilty could be triggered by an enhanced mass transfer ( emt ) . such mixed " or `` hybrid '' models have been proposed for the superoutbursts of su uma type dwarf novae ( duschl & livio 1989 ; smak 1995,1996 ; lasota , hameury & hur 1995 ) and suggested as a possible explanation of sxt outbursts ( lasota 1996 ) . in this paper we analyse how various versions of the disk instabilty model could be applied to outbursts of bsxts . in section 2 we discuss the difficulties of the dwarf nova dim when it is applied to bsxts . section 3 gives a short review of the advection - dominated flow model of quiescent bsxts ( narayan , mcclintock & yi 1996 ) . in section 4.1 we show that the ` hybrid ' emt - triggered dim could apply to bsxts . in section 4.2 we show that the outer cold disk in bsxts could be globally unstable and so bsxt outbursts could be triggered by a thermal disk instability , as proposed independently by mineshige ( 1995 ) . we discuss problems and difficulties in section 5 and present conclusions in section 6 . the dim is based on the fact that accretion disk equilibria are unstable at effective temperatures in the range @xmath7 k , depending on the radial distance from the center . it is conventional to discuss the instability in the @xmath8 plane , where @xmath9 is the surface density of the disk and @xmath10 is the effective temperature . at a given radius , disk equilibria trace a characteristic ` @xmath11 ' shaped curve on this plane , as indicated by the example shown in figure 1 ( based on [ hameury , hur & lasota 1995 ] ) . the middle branch of the s ( shown by a dotted line ) represents thermally and viscously unstable equilibria and can not represent physically accessible states . therefore , locally , an accretion flow has to choose between the two stable branches of the s ( solid lines ) , either the upper ` hot ' branch or the lower ` cold ' branch . the upper branch exists only above a minimum surface density @xmath12 and minimum mass accretion rate @xmath13 , where the value of @xmath10 corresponds to @xmath12 , while the lower branch exists only below a certain @xmath14 and @xmath15 . note that the limiting surface densities and accretion rates vary with the radius @xmath16 . if the steady mass transfer rate @xmath17 supplied by the secondary star corresponds to conditions in the unstable middle branch , i.e. if @xmath18 , then the disk can not maintain a steady state configuration and is forced into a local ` limit cycle ' behaviour , alternating between the cold and hot branches . a dwarf nova outburst or sxt outburst would then correspond to the transition from the cold to the hot state . observations indicate that dwarf nova and sxt outbursts are global events where the entire disk makes the transition . in order to produce such a global outburst , it is required that the pre - outburst quiescent disk be entirely on the lower branch at all radii @xmath16 , i.e. we must have @xmath19 at all @xmath16 . = 8.2 cm in general , the critical density of the cold branch can be represented as ( smak 1993 ) : @xmath20 where @xmath21 is the mass of the central object , @xmath0 is the shakura - sunyaev viscosity parameter of the thin disk , and the index @xmath22 so that @xmath14 is an increasing function of @xmath16 . the @xmath23 exponent is usually negative , @xmath24 . in a disk around a black hole in a binary system the radius may vary by 5 to 6 orders of magnitude so that eq . ( [ e2.1 ] ) would imply very low densities in the quiescent inner disk . for example , introducing numerical values in eq . ( [ e2.1 ] ) according to the formula given by smak ( 1993 ) , we obtain @xmath25 where @xmath26 is the scaled radius in units of the schwarzschild radius @xmath27 , and the mass is scaled according to @xmath28 . according to the dim , the quiescent disk in a dwarf nova or sxt system is not in global equilibrium . instead , the local accretion rate increases with distance from the center ( see eq . ( [ e2.5 ] ) and the effective temperature is constant with radius ( see eq . ( [ e5.2 ] ) . this last property is confirmed by observations ( e.g. wood et al . 1986 ; 1989 ) . assuming that outbursts begin near the inner disk edge and taking @xmath29 , smak ( 1993 ) obtains for the recurrence time @xmath30 where @xmath31 is the radius of the inner edge of the disk and @xmath32 is the kinematic viscosity coefficient . in most bsxts the recurrence time is longer than @xmath5 yr . therefore , we obtain from eq . ( [ e2.3 ] ) a condition on the viscosity coefficient : @xmath33 where @xmath34 is the recurrence time in units of @xmath35 s. combining eq . ( [ e2.2 ] ) and eq . ( [ e2.4 ] ) , we then obtain the following inequality which has to be satisfied by the accretion rate at the inner edge of a quiescent disk if the dim is correct ( lasota 1995 ) : @xmath36 the dim of mineshige & wheeler ( 1989 ; mw ) satisfies this inequality . ( [ e2.5 ] ) provides an observational test of the dim . several bsxts have been observed by ginga ( mineshige 1992 ) and rosat ( see verbunt 1995 ) , and two systems , a0620 - 00 and v404 cyg , have been detected at levels corresponding to accretion rates ( assuming an efficiency of 0.1 , see below ) of at least @xmath37 g s@xmath38 for a0620 - 00 and @xmath39 g s@xmath38 for v404 cyg ( mineshige 1992 ; mcclintock , horne & remillard 1995 ; verbunt 1995 ) . for reasonable values of @xmath40 , these accretion rates are several orders of magnitude larger than the values implied by eq . ( [ e2.5 ] ) for @xmath41 ( inner edge at the last stable orbit as in the dwarf nova type model ) . therefore , the detections of x - ray radiation from quiescent sxts contradict the disk instability model , at least in the version proposed by mw ( mineshige 1992 ) . we note that in the case of a062000 , the prototypical bsxt , the optical luminosity suggests a mass accretion rate of @xmath42 g s@xmath38 in the outer part of the accretion disc ( mcclintock et al . 1995 ) , at @xmath43 . the observation of x - ray emission in quiescent sxts at low luminosities poses an insurmountable obstacle to models in which the accretion disk is of the shakura sunyaev type with @xmath41 . on the one hand , the requirement of eq . ( [ e2.5 ] ) ) can be reconciled with the deduced mass accretion rate of @xmath44 only if we assume a very low viscosity in the inner disk , e.g. @xmath45 which is almost equivalent to having no viscosity at all . here we have assumed @xmath46 . according to the dim ( see e.g. cannizzo et al . 1995 ) @xmath23 should be of the form @xmath47 , where @xmath48 is the disk semi - thickness . cannizzo et al . ( 1995 ) find @xmath49 , @xmath50 and the values of @xmath23 they obtain are well in excess of those required by eq . ( [ e2.5 ] ) confirming the failure of the dim to account for the observed properties of quiescent sxts . on the other hand , the x rays have to be produced by a process connected to viscosity , and if we assume a standard thin disk with @xmath51 as implied by the x - ray observations , the effective temperature will be so low that the disk will radiate hardly any x - rays at all . in other words , the required value of @xmath23 in the dwarf nova type model is unreasonable , and even if we ignore this problem , it is impossible to fit both the low luminosity and the temperature of the radiation simultaneously . the advection - dominated model of narayan , mcclinotck & yi ( 1996 ) ( hereafter nmy ) presents a solution to this problem . the fundamental property of advection - dominated accretion flows ( adafs ) is that the efficiency with which energy is radiated is very low , so that most of the heat released by viscous friction ( or other processes ) is advected into the central black hole and only an almost infinitesimal amount is radiated away . models of adafs were first constructed for optically thick flows with angular momentum ( begelman 1979 ; begelman & meier 1982 ; abramowicz et al . 1986 , 1988 ) . the models correspond to very high ( super - eddington ) accretion rates and have not yet found an application to real astrophysical objects . as pointed out by narayan and popham ( 1993 ) , advection may play an important role also in the case of optically thin flows which are notoriously non - efficient radiators . in this case , the viscously heated gas flows into the black hole on a shorter timescale than its cooling time and therefore a large fraction of the energy is advected rather than radiated . models of optically thin advection - dominated flows were constructed in a series of papers by narayan & yi ( 1994 , 1995a , b ) and abramowicz et al ( 1995 ) , chen ( 1995 ) and chen et al . ( 1995 ) . the narayan & yi ( 1995b ) version of the adaf model deals with a two - temperature plasma which is radiatively cooled by bremsstrahlung , synchrotron emission and comptonization . this model was first successfully applied to the galactic center source sgr a@xmath52 ( narayan , yi & mahadevan 1995 ) , and soon after applied to quiescent sxts ( nmy ) . recently the narayan & yi ( 1995b ) model was applied to the dwarf active galactic nucleus ngc 4258 ( lasota et al . 1996 ) . the accretion flow in the nmy model consists of two parts : outside some radius @xmath53 the accreting gas takes the form of a standard , cold , geometrically thin , keplerian disk . however , inside @xmath53 , the accretion proceeds in the form of an adaf . here , the gas becomes extremely hot , with the ions reaching temperatures @xmath54 k and the electrons going up to @xmath55 k. the very high ion temperature causes the gas to be very thick in the vertical direction ; indeed , the flow is essentially quasi - spherical rather than disk - like , though it is still partially supported by rotation . for simplicity , nmy assumed that the accretion rate is constant throughout the whole flow . in the nmy model , the optical and uv radiation is produced by the outer thin disk , while the inner advection - dominated flow , because of its very high temperature , produces all the hard radiation in x - rays and soft @xmath56-rays . even though the same @xmath57 flows through the outer and inner zones , the model quite naturally explains the low luminosity of the x - rays as a consequence of the extremely low radiative efficiency of the adaf . the viscous energy dissipation in the adaf is significantly larger than that in the outer thin disk since most of the gravitational potential energy is released in the adaf . however , virtually all of this energy is advected into the black hole and only a very tiny fraction is radiated . this effect is so strong that the luminosity of the adaf is actually less than that of the outer disk . nmy used their model to fit the spectra of three quiescent bsxts : a6020 - 00 , v 404 cyg and x - ray nova muscae 1991 . the fits to the observed optical uv to x - ray spectra of the three sxts are very good . the models depend basically on two parameters , @xmath58 ( where @xmath23 refers to the viscosity parameter in the adaf and is distinct from @xmath0 of the outer thin disk ) and @xmath53 . the other parameters are either given directly by observations ( for example the black hole mass ) or the fits turn out to be insensitive to their values ( for example the ratio of the gas to the total pressure ) . the presence of an inner hot advection - dominated flow removes the contradiction between the observed x - ray emission and the properties of a shakura - sunyaev type disk . there remains the question of why bsxts go into outbursts . could the mechanism be related to the hot inner flow ? abramowicz et al ( 1995 ) and narayan & yi ( 1995b ) showed that there is a maximum accretion rate above which there are no optically thin advection - dominated solutions . in the narayan & yi model the critical accretion rate is @xmath59 for @xmath60 and @xmath61 for @xmath62 . figure 2 , taken from narayan ( 1995 ) and based on chen et al . ( 1995 ) , represents the structure of all possible accretion solutions for the case of a 10 @xmath63 black hole . label 1 refers to a region where the only allowed solution is an advection - dominated one , label 2 refers to a radiatively - cooled zone , and label 3 corresponds to a region where both types of solutions are allowed . in region 4 there are no stable solutions at all . one could speculate that if the accretion rate at which matter is brought to the hot adaf corresponded to region 4 , the system would be forced into a thermal runaway and a limit cycle ( see e.g. chen et al . there is , however , a problem with this scenario . nmy found that they needed @xmath64 in the advection - dominated inner zone of their models . for such @xmath23 , zone 4 exists only for mass accretion rates above @xmath65 . but the mass transfer rates in quiescent sxts estimated by nmy correspond to @xmath66 . for such low accretion rates an adaf solution is always present . furthermore , adafs are thermally stable by construction ( abramowicz et al . 1995 , narayan & yi 1995b , kato et al . 1995 ) the hot inner flows in quiescent sxts are therefore stable and we can not expect to find there the cause of outbursts . thus the outburst mechanism in sxts must be associated with the outer cold disk which could be , or become thermally unstable . below we will investigate the mechanisms that could cause an outburst in the outer thin disk . in this and the next sections we will study the stability properties of the outer ` standard ' thin disk , which extends from the transition radius @xmath67 to the outer radius @xmath68 . as explained in section 2 , for the disk to be globally unstable , the mass transfer rate @xmath17 must satisfy at some radius the inequality @xmath69 where @xmath15 and @xmath70 are defined in section 2 . the range of unstable mass transfer rates depends on the size of the disk . as an example figure 3 shows surface density profiles of equilibrium accretion disks around a @xmath71 black hole , where the disk has been truncated at @xmath67 in accordance with the nmy model . the accretion rate @xmath57 is constant for each curve . the unstable part of the curve is represented by a dotted line , the cold stable configurations correspond to the solid line . models by hameury et al . ( 1995 ) were used in this calculation with a thin disk viscosity parameter @xmath72 . note that this is smaller than the value @xmath64 which nmy found was necessary for a consistent model of the advection - dominated region . there is no contradiction in this since there is no reason for @xmath23 to be the same in the advection dominated and standard regions ; one should rather expect @xmath23 to be lower in the outer thin disk compared to the inner hot flow . a globally stable , cold accretion disk can not be represented by a curve that crosses the @xmath73 line . assuming @xmath74 at @xmath68 , we can have a situation where with decreasing @xmath75 , we reach the condition @xmath76 at some radius @xmath77 . the disk can be in cold equilibrium only for @xmath78 . continuing to @xmath79 , the segment of the curve between @xmath14 and @xmath12 is unstable while the curve to the left of @xmath12 represents hot equilibria . for each mass transfer rate @xmath17 , there exists therefore a critical radius @xmath80 outside of which the disk is globally stable in the cold state . from figure 3 we can see that the global state of the disk will depend on the exact value of @xmath53 relative to @xmath80 . for example , for @xmath81 , if the outer disk is truncated at @xmath82 we have a globally stable disk whereas if @xmath83 the disk is globally unstable . for the moment we consider @xmath53 to be a free parameter that is fixed by best ` fit ' to the spectrum of the quiescent sxt in combination with other parameters such as the black hole mass , @xmath23 etc . ( see nmy ) . physically , of course , the value of @xmath53 should be determined by the process that leads to the formation of the adaf , such as coronal evaporation ( see e.g. meyer & meyer hofmeister 1994 ) or heating by diffusive energy transport ( honma 1995 ) . the theories of these processes have not yet developed to the point where they can make robust predictions of @xmath53 . lasota ( 1995 ) studied a similar problem in the context of dwarf nova outbursts . some su uma type dwarf novae show only very rare and very long superoutbursts and no ( or almost no ) ` normal ' outbursts . the best known system in this class is wz sge . such dn systems are the closest relatives of bsxts . the dim applied to dwarf novae requires some modifications before it can be applied to the whole class of observed events . in particular , the so called uv delay problem and the x ray and uv emissions from quiescent dns require a ` hole ' in the inner disk regions ( see e.g. meyer & meyer hofmeister 1994 ) . this ` hole ' can be produced by magnetic disruption if the white dwarf is ( weakly ) magnetized ( livio & pringle 1992 ) or it can be the result of evaporation ( meyer & meyer hofmeister 1994 ) . lasota ( 1995 ) show that if the inner regions of accretion disks in quiescent dwarf nova systems are removed , the remaining disk is globally stable for mass transfer rates @xmath84 g s@xmath38 . this implies that ( super)outbursts in such systems have to be triggered by an enhanced mass transfer from the companion . they suggest that the lack of normal outbursts in wz sge results because of its low mass transfer rate : there are no outbursts because the disk is stable . a superoutburst would be triggered by an emt which would put the disk into a globally unstable state ; in other words , superoutbursts would be due to a disk instability generated by an increased mass transfer . observations show that the mass transfer increases prior to and during the superoutburst so that such a hybrid mechanism could be at work in su uma s ( smak 1995 , 1996 ) . the alternative model ( see osaki 1996 for a review ) in which superoutbursts are due to a pure disk instablity , the so called tidal thermal instability , is yet to be confirmed by the observational properties of su uma ( smak 1991 ) and requires the viscosity to be extremely low in the case of wz sge . if @xmath53 in ( some ) bsxts is such that in quiescence the outer thin disk is globally stable the same reasoning would apply also to these systems and ( some ) sxt outbursts could be due to an emt triggered disk instability . if , for a given mass transfer rate , the inner radius of the outer thin accretion disk @xmath53 is smaller than the critical radius @xmath85 , the disk will be globally unstable and will undergo dn type outbursts . we show in this section that we obtain reasonable recurrence times without any unusual assumption about the value of the viscosity parameter @xmath0 . in order to estimate properties of the outbursts we closely follow the steps used by smak ( 1993 ) in the analysis of wz sge . as discussed in section 2 . , the disk outburst begins when the density exceeds somewhere the critical value @xmath14 . the type of resulting outburst depends on where in the disk this happens . for a constant @xmath17 , the type of outburst depends on two characteristic times : the time @xmath86 it takes the matter accumulating at the outer disk to build up a surface density larger than @xmath14 , and the viscous time @xmath87 which measures the time it takes matter to diffuse inward and cross the @xmath14 barrier somewhere nearer to the inner disk boundary . if @xmath88 one obtains a so - called outside - in " ( ` type a ' ) outburst beginning at the outer disk edge ; in the opposite case one has an inside - out " ( type b ) outburst ( smak 1984 ) . below , we will consider inside - out " outbursts . as discussed by smak ( 1993 ) these outbursts have longer recurrence times ( longer length of the outburst cycle ) than the `` outside in '' outbursts and give also upper limits on @xmath0 . one should note here that although there are cases where it is claimed that bsxt outbursts began in the inner regions ( chen et al . 1993 ) , the observational evidence is far from being conclusive . one should note however that the present discussion concerns the _ outer _ thin disk so that even an inside - out " outburst in the outer disk would most probably be an outside - in " outburst from the point of view of the inner , hot accretion flow . rescaling eq . ( [ e2.5 ] ) to the characteristic parameters of the outer disk we get @xmath89 as before , @xmath0 denotes the viscosity parameter in the thin disk . if we replace @xmath57 by the luminosity of the thin disk : @xmath90 the recurrence time of the instability can be written as @xmath91 eq . ( [ e4.5 ] ) reveals the severe problem faced by the dwarf - nova type disk instability model of sxts and how the problem is resolved in the nmy two - zone model . if we set the inner edge of the thin disk to be @xmath92 , corresponding to the last stable orbit , then for the observed x - ray luminosity of @xmath93 , eq . ( [ e4.5 ] ) shows that we need an absurdly small @xmath94 . on the other hand , if we set @xmath95 as found by nmy from their spectral fits , then we obtain the right recurrence timescale with @xmath96 , consistent with the values of @xmath0 required by the ` standard ' dim for dwarf novae ( livio & spruit 1991 ) . however , our estimated value of @xmath0 is very sensitive to the values of @xmath53 and @xmath97 , so that the values @xmath98 , @xmath53 and @xmath97 can not be much different from those assumed in eq . ( [ e4.5 ] ) for @xmath0 to be not much smaller than @xmath99 . in the case of a0620 - 00 the optical luminosity is @xmath100 ( mcclintock et al . 1995 ) so that if one wished to keep @xmath101 , eq . ( [ e4.5 ] ) requires @xmath102 . at @xmath103 this optical luminosity corresponds to an accretion rate @xmath104 . 3 shows however that for @xmath102 the outer disk is globally stable so that eq . ( [ e4.5 ] ) does not apply because to go into outburst the system requires an emt . on the other hand , if we allowed the value of @xmath0 to be as low as @xmath105 , which is required for the pure dim to work in the case of wz sge ( smak 1993 ) , the transition radius could be @xmath106 depending on the value of the black hole mass . we thus conclude that bsxt outbursts can be triggered by the ` normal ' dwarf nova type instability in the outer cold , keplerian disk . the critical factor which allows us to obtain good agreement on the recurrence time with a `` reasonable '' value of @xmath0 is the truncation of the thin disk at a large radius @xmath53 . since such values of @xmath102 are also required by the emt model , this could be viewed as an independent confirmation of the truncated disk model of nmy . another feature of the nmy model is that it provides a natural explanation for the spectral evolution seen in sxts during outburst ( narayan 1996 ) . when the outer disk goes into outburst and the mass accretion rate increases suddenly , the first response of the system will be for the @xmath57 in the inner advection - dominated zone also to increase proportionately . since the luminosity of the inner zone varies roughly as @xmath107 , the x - ray luminosity will go up enormously . at the same time , since the electron temperature is very high , the spectrum will remain hard and there will be emission out to a few @xmath108 kev . model calculations indicate that the photon spectral index is expected to be @xmath109 ( narayan 1996 ) when the luminosity reaches @xmath110 . meanwhile , the increased @xmath57 will probably cause the outer thin disk to diffuse inward on a viscous time . therefore , after some delay , the outer thin disk will move into the advection - dominated zone , perhaps extending all the way into the black hole . when this happens , we will have a very luminous soft x - ray source with a spectrum consistent with that of a shakura - sunyaev disk . a0620 - 00 and nova muscae 1991 appear to have made this transition , whereas v404 cyg remained a hard source throughout the outburst . perhaps the thin disk did not penetrate all the way to the center in the latter case . as the outburst dies down , we expect these stages to occur in reverse . first , the thin disk will shrink back , leaving behind a strong hard x - ray source . later , as @xmath57 reduces , the system will revert to the quiescent state where there will be a very low x - ray luminosity and most of the viscous energy will be advected into the black hole . the stages described above , which appear reasonable under the advection - dominated model , match quite well with observations of several sxts in outburst . the nmy model gives a satisfactory description of the quiescent state of bsxts , and as we have shown in this paper also provides a reasonable explanation of the recurrence times of the outbursts and the spectral evolution during outburst . all of these were difficult to explain with the previous model where the thin disk was assumed to extend all the way down to the black hole . there is , however , an inconsistency in the nmy model . the spectral ` fits ' are made under the assumption that the outer disk is stationary , which means in practice , that a temperature profile @xmath111 is used to calculate the spectrum . for the parameters required for an acceptable spectral fit , the assumed values of @xmath112 , @xmath23 and @xmath53 are such that @xmath113 so that a stationary disk is globally unstable . such a disk would therefore be subject to the dwarf - nova type instability , which means that in the quiescent state the accretion rate would not be constant with radius . this implies a self - inconsistency in the nmy model . a further problem arises from the @xmath10 required to fit the optical spectra . taking for the effective temperature ( cf . frank et al . 1992 ) @xmath114^{1/4 } , \label{e5.1}\ ] ] and replacing @xmath57 by the expression given in eq . ( [ e4.3 ] ) we see that @xmath115 which shows that the effective temperature in a quiescent dwarf nova disk is practically independent of the radius . for @xmath116 and a recurrence time @xmath117 60 days , typical parameters of a dwarf nova , eq . ( [ e5.2 ] ) gives ( for @xmath118 ) @xmath119 , very close to the observed values ( wood et al . 1986 ) . the problem arises when we apply the same formula to quiescent sxts . for sxt parameters , the effective temperature turns out to be @xmath120 k which is too low to agree with observations . both in a0620 - 00 and v404 cyg , the optical and uv data imply significantly higher temperatures . it should be emphasized that this problem is not peculiar to the nmy model , but will be faced by any sxt model which invokes a dwarf nova type non - steady quiescent accretion disk . a possible explanation for the higher disk temperature could be that the outer disk is optically thin . the strong h@xmath23 line in the spectra of these disks suggests that a substantial part of the emission comes from optically thin gas . it is then natural for the color temperature to be higher than the effective temperature . finally , one should note that the failure of the dim to account for the properties of quiescent sxts has been also noticed by cannizzo et al . ( 1995 ) and kim et al . ( 1996 ) who suggested ` non - standard ' inner disk structures that do not invoke advection . the detailed mechanism by which bsxts go into outburst will depend on the values of @xmath57 , @xmath23 and @xmath121 . these parameters , or rather the ratio @xmath58 and @xmath121 , are the free parameters which nmy used to fit their models to observed spectra of quiescent sxts . the physical conditions in the outer disk are very similar to those that are encountered in accretion disks of su uma systems . for this reason also the problems with the origin of the outbursts ( or rather superoutbursts ) are the same . at the low accretion rates that are required by observations of bsxts in quiescensce , the accretion disk may be globally stable for @xmath122 . in this case only an increase in the mass transfer from the secondary can bring the disk into an unstable state and trigger the outburst . to make the model complete one should find why the secondary increases its transfer rate on timescales typical of sxts . one should note however that variations of the transfer rate on various timescales _ are observed _ in many close binary systems . when @xmath57 , @xmath23 and @xmath121 correspond to a globally unstable outer disk configuration , a dn type instability will necessarily occur . the effect of the outburst on the inner hot adaf should be similar to that of a emt triggered outburst since both types of outbursts are basically due to the same thermal instability . the main result of this paper is that the disk instability model works much better for sxts if the thin disk is truncated at a large transition radius @xmath53 as in the nmy model rather than having the disk extending down to @xmath123 as in the model by mineshige & wheeler ( 1989 ) . the region inside the radius of truncation would then be filled with a very hot , optically thin , ultra - low radiative efficiency , advection - dominated flow . nmy showed that such a model gives a good fit to the spectra of quiescent sxts , while narayan ( 1996 ) suggested that it may be consistent with the spectral evolution observed in sxts during outburst . in this paper , we show that the recurrence times of sxt outbursts are satisfactorily predicted with a reasonable value of the viscosity parameter @xmath96 in the outer disk ( or @xmath124 in the worst case ) . in contrast , in the dwarf nova type model , one requires an implausibly low value @xmath94 . one interesting point to bear in mind is that it is possible to have binaries where @xmath57 , @xmath23 and @xmath53 correspond to a globally stable outer disk in a cold low state , and where in addition the secondary star for whatever reason does not have any tendency to exhibit a large amplitude emt instability . in such a system , we would have a black hole binary which is permanently in the quiescent state , with most of the accretion energy being swallowed by the black hole . the system would be extremely dim in x - rays and quite under - luminous even in the optical and would be very hard to detect . in principle , the galaxy could have large numbers of such nearly silent black holes , which would be the binary equivalent of the underluminous galactic nuclei discussed by fabian & rees ( 1995 ) . we are grateful to jean - marc hur for providing us with figures 1 and 3 . we thank jean - marie hameury for discussion and comments and craig wheeler for helpful critical remarks . rn was supported in part by nasa grant nag 5 - 2837 ( to the center for astrophysics ) and nsf grant phy 9407194 ( to the institute for theoretical physics ) . iy acknowledges support from suam foundation . abramowicz m. a. , chen , x. , kato , s. , lasota , j. p. , regev , o. , 1995 , , l37 abramowicz , m. a. , czerny , b. , lasota , j. p. , szuszkiewicz , e. , 1988 , , 646 abramowicz , m. , lasota , j. p. , xu , c. , 1986 , variability and instabilities of accretion disks , in : g. swarup , g. , kapahi , v.k . 119 , quasars . reidel , dordrecht , p. 376 barret , d. et al . , 1993 , , 241 bailyn , c.d . , orosz , j.a . , mcclintock , j.e . , remillard , r.a . , 1995 , nature , submitted begelman , m. c. , 1979 , , 237 begelman , m. c. , meier , d. l. , 1982 , , 873 cannizzo , j.k . , 1993 , the limit cycle instability in dwarf nova accretion disks , in : wheeler , j. craig ( ed . ) accretion disks in compact stellar systems , world scientific publishing , singapore , p.6 cannizzo , j.k . , chen , w. , livio , m. , 1995 , , 880 casares , j. , charles , p.a . , 1994 , , l5 casares , j. , charles , p.a . , marsh , t.r . , 1995 , , l45 chen , w. , livio , m. , gehrels , n. , 1993 , , l5 chen , x. , 1995 , , 641 chen , x. , abramowicz , m. a. , lasota , j. p. , narayan , r. , yi , i. , 1995 , , l61 duschl , w.j , livio , m. , 1989 , 209 183 fabian , a. c. , rees , m. j. , 1995 , , l55 frank , j. , king , a. r. , & raine , d. , 1992 , accretion power in astrophysics , cup , cambridge goldwurm , a. , 1993 , , 293 gontikakis , c. , hameury , j .- , 1993 , 271 , 118 hameury , j .- m . , hur , j .- m . , lasota , j .- , 1996 , in preparation hameury , j .- m . , king , a.r . & lasota , j .- , 1986 , 162 , 71 harmon , b.a . , 1994 , batse observations of transient hard x - ray sources , in : fichtel , c. , gehrels , n. , norris , j.p . the second compton symposium , aip , new york , p. 210 honma , f. , 1995 , pasj , in press kim , s .- w , wheeler , j.c . , mineshige , s. , 1996 , submitted king , a.r . , 1993 , , l5 koyama , k. , et al . , 1981 , , l27 lasota , j .- , 1995 , mechanisms for dwarf nova outbursts and soft x - ray transients , in : van den heuvel e.p.j , van paradijs j. ( eds ) compact stars in binaries . proceedings of the iau symposium 165 , kluwer , p. 43 lasota , j .- , 1996 , models of soft x - ray transients and dwarf novae , in : wood j.h . cataclysmic variables and related objects . proceedings of iau coll . 158 , in press lasota , j .- p . , hameury , j .- m . , hur , j .- m . , 1995 , 302 , l29 lasota , j .- p . , abramowicz , m.a . , chen , x.m . , krolik , j.h . , narayan , r. , yi.i . , 1996 , in press livio m. , pringle j.e . , 1992 , , 23p mcclintock , j.e . , horne , k. , remillard , r.a . , 1995 , , 358 meyer , f. , meyer - hofmeister , e. , 1994 , 288 , 175 mineshige , s. , 1995 , submitted mineshige , s. , , 1992 , , 117 mineshige , s. , wheeler , j.c . , 1989 , , 241 narayan , r. , mcclintock , j. e. , & yi , i. , 1996 , , 821 ( nmy ) narayan , r. , popham , r. , 1993 , nature , 362 , 820 narayan , r. , yi , i. , 1994 , , l13 . 1995a , , 231 . 1995b , , 710 narayan , r. , yi , i. , mahadevan r. , 1995 , nature 374 , 623 orosz , j.a . , bailyn , c.d . , remillard , r.a . , mcclintock , j.e . foltz , c.b . , 1994 , , 848 orosz , j.a . , bailyn , c.d . , mcclintock , j.e . , remillard , r.a . , 1995 , submitted osaki , y. 1996 , in : iau colloquium 158 , cataclysmic variables and related objects , ed . j. wood & a. evans ( dordrecht : kluwer ) , in press parmar , a.n . , angelini , l. & white , n.e . 1995 452 , l129 smak , j.i . , 1984 , , 161 . , 1991 , , 269 . , 1993 , , 101 . , 1995 , , 355 . , 1996 , dwarf nova outbursts and superoutbursts . in : wood j.h . cataclysmic variables and related objects . proceedings of iau coll . 158 , in press tanaka , k. & lewin , w.h.g . 1995 black - hole binaries . in : lewin , w.h.g . , van paradijs , j. , van den heuvel , e.p.j . x - ray binaries cup , cambridge , p. 121 van paradijs , j. & mcclintock , j.e . 1995 , optical and ultraviolet observations of x - ray binaries . in : lewin , w.h.g . , van paradijs , j. , van den heuvel , e.p.j . x - ray binaries cup , cambridge , p.58 verbunt , f. 1995 in : van den heuvel e.p.j , van paradijs j.(eds ) compact stars in binaries . proceedings of the iau symposium 165 , kluwer , in press verbunt , f.1994 285 , 903 white , n.e . 1994 , the galactic distribution of black holes in x - ray binaries , in : aip conf . 308 , the evolution of x - ray binaries , ed . holt & c.s . day ( new york : aip ) , 53 white , n.e . , kaluzienski , j.l . & white , j.h . 1984 , the spectra of x - ray transients , in : woosley , s.w . , ( ed . ) high energy transients in astrophysics , aip conf . 115 , aip , new york , p. 31 wood , j. 1986 , , 629 wood , j. , horne , k. , berriman , g. , & wade , r. 1989 , 974

tidal theories have gained great attention in the last decade due to the large number of discovered close - in planets which orbit their parent stars with periods of a few days . according to classical theories ( e.g. darwin 1880 ; kaula 1964 ; mignard 1979 ) , the long - term tidal evolution of these planets results in orbital decay and circularization , on time - scales which depend on the physical properties of the interacting systems ( hut 1981 ; dobbs - dixon , lin & mardling 2004 ; ferraz - mello , rodrguez & hussmann 2008 ; jackson , barnes & greenberg 2009 ; rodrguez & ferraz - mello 2010 ) . the rotation of short - period planets is also affected by tidal interactions . it is nowadays accepted that the spin rate of planets soon after their formation was much higher than their orbital mean motion ( e.g. agnor , canup & levison 1999 ; kokubo & ida 2007 ) . however , the cumulative tidal effects causes the planet to spin down , reaching its stationary value in time - scales much shorter than those needed to circularize the planet orbit . this evolution ultimately results in the synchronous rotation ( the spin state in which the orbital and rotation periods coincide ) , which occurs when the planet orbit is circularized ( hut 1981 ; ferraz - mello et al . 2008 ) . for non - circular orbits , the synchronous rotation only occurs if an additional torque acts to counteract the tidal torque . the additional gravitational torque may be generated by planets with a rocky - like composition having a permanent equatorial deformation or frozen - in figure ( see goldreich & peale 1966 ; greenberg & weidenschilling 1984 ) . these torques , in contrast to the tidal ones , are conservative . in the context of the two - body problem , the gravitational interaction of the star with an asymmetric planet drive the planet rotation into different regimes of motion , including oscillations around exact spin - orbit resonances ( goldreich & peale 1966 ) . when dissipative tidal effects are taken into account , the oscillations are damped and the planet rotation is trapped into a resonance . indeed , the rotational motion of some solar system bodies can be explained by the combined effects of the tidal and asymmetric torques ( goldreich & peale 1966 ; greenberg & weidenschilling 1984 ; correia & laskar 2009 ) . mercury is an example of a rocky planet with a rotation state currently evolving in the 3/2 spin - orbit resonance ( pettengill & dyce 1965 ) . several models of the secular spin evolution of mercury have been developed , including i ) the chaotic evolution of its orbit ( correia & laskar 2004 ) ; ii ) effects of the core - mantle friction ( correia 2006 ; correia & laskar 2009 ) ; iii ) the evolution starting with retrograde and subsequent synchronous motion combined with impact events ( correia & laskar , 2012 ; wieczorek et al . 2012 ) ; iv ) the consideration of non - linear tidal theories ( makarov 2012 ) . however , despite the large amount of previous works , the question on how the current resonant state of mercury was achieved still remains under discussion . concerning the rotation of earth - like close - in extrasolar planets , a near synchronous motion is usually assumed in view of tidal evolution . however , correia , levrard & laskar ( 2008 ) have shown that when an additional torque is considered , the equilibrium rotation of tidally evolved planets can assume rotational states which depart from synchronism . moreover , as the capture probability in a non - synchronous spin - orbit resonance requires a non - circular orbit ( goldreich & peale 1966 ; celletti & chierchia 2008 ) , the rotation of several recently discovered super - earth planets in eccentric orbits can be currently evolving under a resonant motion , provided the planet figure can support a permanent equatorial deformation . as the eccentricity is tidally damped , several resonant trappings would become unstable and the planet rotation ultimately reaches synchronization for almost circular orbits . however , the locking in a non - synchronous resonance with the very short orbital period can protect certain planet s longitudes from solar radiation , providing more favorable conditions with respect to habitability as compared to non - resonant rotation ( dobrovolskis 2007 ) . moreover , heller , leconte & barnes ( 2011 ) studied the equilibrium rotation of super - earths and the implications on their habitability . they found that , under assumption of an evolution driven by tidal torques , the equilibrium value of the rotation period of gj 581 * d * is probably about two times smaller than the orbital period . in a recent work , callegari & rodrguez ( 2012 ) have analysed the rotation of extrasolar planets with orbital periods less than 33 days and masses below the mass of neptune . depending on one parameter related to the equatorial ellipticity and also on the orbital eccentricity , they identified regions of chaotic rotation in the phase space of some particular planets like 55 cnc**e * * , gj 1214**b * * , gliese 876**d * * , among others . the present investigation aims at two main aspects : i ) how the orbital evolution of the planet , due to tidal interaction with the central star , is altered when the asymmetric torque is considered ; and ii ) to analyse the occurrence of captures in spin - orbit resonant motions under combined effects of both torques for the planets kepler-10**b * * , 55 cnc**e * * and gj 3634**b**. these planets are examples of hot super - earths , which are close - in planets having a few earth - masses with a rocky - like composition . we are interested in the past behavior in the history of the rotation and the subsequent evolution to synchronization . we also discuss the possibility of a current evolution of the rotation of these planets into a super - synchronous resonance state . moreover , we investigate the implications on the orbital evolution due to such temporary resonant trappings . despite the continuously growing number of discovered planets , their rotation state and equilibrium figures remain unknown . in this work , we perform several numerical simulations of the evolution of the systems , assuming several values of unknown parameters , such as equatorial ellipticity and initial rotation . the results allow us to understand the past history of these systems , their current orbital configuration and final rotational states . it should be noted that the majority of previous works concerning spin - orbit interactions with tidal dissipation have been addressed in the context of a fixed keplerian orbit , with constant values of semi - major axis and eccentricity ( e.g. , celletti , froeschl & lega 2007 ; celletti & chierchia 2008 ) . this assumption is adequate for solar system bodies , where tidal interactions are weak and , consequently , variations of orbital elements are small . on the other hand , the existence of extrasolar planets with orbital periods of a few days indicates that their current orbital configurations would be the consequence of strong tidal interaction with the host star . in this case , the keplerian approximation is no longer valid . thus , we analyse spin - orbit interactions as the orbit evolves due to tidal effects . solving the exact equations of motion of the problem , we show that the forces arising from the potential of a asymmetric planet also contributes to the orbital variation , accelerating the orbital decay and the eccentricity damping , but only in the cases in which the rotation is locked in resonance . the spin - orbit problem can also be modeled by the use of averaged equations of motion , where only terms with secular and resonant arguments are retained ( e.g. celletti & chierchia 2008 ; correia & laskar 2009 ) . through this model , we compute the averaged variations of semi - major axis and eccentricity due to the combination of asymmetric and tidal forces . we also give an explicit expression for the critical value of the equatorial prolateness , for which a given spin - orbit resonance would become unstable . the study of the particular cases of 2/1 , 3/2 and 1/1 spin - orbit resonances indicates a good agreement with the results of the numerical simulations of the exact equations . this paper is structured as follows . in section [ model ] , we present the exact equations of motion of the spin - orbit problem . section [ num ] shows the results of the numerical simulations , illustrating the evolutions of kepler-10**b * * , gj 3634**b * * and cnc 55**e**. we show that , assuming a nonzero value of the equatorial ellipticity , different temporary resonant trappings of the rotation can occur , with implications in the orbital decay and circularization . an analytical approach in the context of the averaged equations is shown in section [ mean ] , explaining the results obtained in the numerical simulations . finally , section [ discussion ] is devoted to discussion and conclusions . with the reference axis @xmath0 , which is assumed coincident with the orbital pericenter direction . the angle between the long axis and the star - planet direction is given by @xmath1 , where @xmath2 is the true anomaly of the planet . ] we consider a planet orbiting a central star with a period of a few days . hereafter , we refer to @xmath3 and @xmath4 as the planet and star masses , respectively ; @xmath5 and @xmath6 as the planet radius and the instantaneous star - planet distance , respectively . the elements @xmath7 stand for the semi - major axis , eccentricity and mean - motion of the astrocentric orbit of the planet , respectively . we suppose that the rotating planet body can be approximated by a homogeneous triaxial ellipsoid , with the equatorial axes @xmath8 and @xmath9 and the rotation axis @xmath10 . we assume that @xmath11 and @xmath10 is normal to the orbital plane ( i.e. , no obliquity is considered ; planar problem ) . in addition , we assume that the equatorial asymmetry ( @xmath12 ) corresponds to a frozen - in body or permanent deformation of the planet . the assumption above can be supported under consideration of a rocky composition of the planet body . this is the case of super - earth planets , which are the subject of the present investigation . the gravitational potential of the planet at the distance @xmath6 from its centre is given by ( e.g. beutler 2005 , p. 102 ) @xmath13 where @xmath14 is the gravitational constant and the angles , which has not been included in the present paper ( see ferraz - mello et al . @xmath2 and @xmath15 are defined in figure [ elips ] . note that @xmath16 is the angular rotation velocity of the planet . the parameter @xmath17 describes the equatorial ellipticity of the planet s gravity field and is related to the principal momenta of inertia @xmath18 through @xmath19 , where @xmath20 is the structure constant given by @xmath21 and @xmath22 is the polar moment of inertia . we neglect the contribution of terms arising from the oblateness of the rotating body , that is , @xmath23 and higher order terms . in fact , in the planar problem , the gravitational potential corresponding to the @xmath23term depends only on the distance @xmath6 . in this case , the potential generates a radial force with zero torque that does not affect the rotation evolution . at the distance @xmath6 from the centre of the planet , the gravitational field @xmath24 gives rise to a force on a point mass @xmath4 which is given by adopted in beutler ( 2005 ) , the corresponding force is computed as the positive gradient of the potential . ] @xmath25 where @xmath26 and @xmath27 are unit vectors with origin in the star ( see fig . [ elips ] ) . the force @xmath28 alters the orbital component of the angular momentum , whereas the reaction force governs the evolution of the spin through the creation of the torque on the planet given by @xmath29 where @xmath30 and @xmath31 . thus , @xmath32 note that , for @xmath33 and @xmath19 , the above equation agrees with the classical expression for the torque ( e.g. danby 1962 ; goldreich & peale 1966 ) . the effects of the dynamical tides are raised by the gravitational action of the star on the inelastically deformed rotating planet . due to the internal viscosity of the body , there exists a delay @xmath34 ( time lag ) between the star perturbation and the maximum deformation . the relationship between @xmath34 and tidal frequencies ( which are linear combinations of @xmath35 and @xmath36 ) depends on the internal rheology , which is poorly known even for solar system bodies . for sake of simplicity , in this work , we adopt a linear model with constant @xmath34 ( darwin 1879 ; mignard 1979 ; hut 1981 ; rodrguez et al . 2011a ) ; for a review of other tidal models , the reader is referred to efroimsky & williams ( 2009 ) and ferraz - mello ( 2012 ) . according to mignard ( 1979 ) , the tidal force and torque are given by @xmath37,\ ] ] and @xmath38,\ ] ] where @xmath39 and @xmath40 is the second degree love number . in this paper , we only consider the tides raised by the star on the planet , neglecting the tides raised on the star by the planet . indeed , the strength of stellar tide is proportional to the planet mass and can be safely neglected in the case of super - earths ( see rodrguez & ferraz - mello 2010 ) . note however that , after circularization of the planet orbit , stellar tides may be the only source of orbital decay . in the reference frame centered in the star ( see figure [ elips ] ) , the exact equations of motion of the planet can be now written as @xmath41 the perturbing forces @xmath28 and @xmath42 alter the keplerian orbit of the planet which is defined by the first term of the first equation , while the corresponding torques in the second equation produce variations of the planet rotation . note that , in the planar case , this last equation can be written in scalar form , because both torques are normal to the orbital plane . in this section we apply the analytical model described in the previous section , in order to investigate the motion of three known super - earths planets , namely kepler-10**b * * , 55 cnc**e * * and gj 3634**b**. for this task , the exact equations of motion ( [ mov1 ] ) were integrated numerically using the ra15 code ( everhart 1985 ) . we stress that no scaling parameter , often used to accelerate the tidal evolution , was used in our simulations ( see , for instance , tittemore & wisdom ( 1988 ) , for the validity of the usage of the scaling parameter ) . table 1 shows the physical parameters and current orbital elements of the systems considered in this paper . [ tab1 ] .physical parameters and current orbital elements of the analysed systems ( @xmath43batalha et al . 2011 ; @xmath44gillon et al . 2012 ; @xmath45bonfils et al . 2011 ) . the first two planets were discovered through the transit technique , allowing us a good estimation of their radii . @xmath46 and @xmath47 stand for mass and radius of the earth , respectively . [ cols="^,^,^,^,^,^",options="header " , ] numerical integrations require a setting of the initial configurations of the systems . in the case of problems including dissipative processes , the choice is complicated , mainly , due to the inadequacy of the simulations of the past evolution of dissipative systems through integrations backward in time ( see details in michtchenko & rodrguez 2011 ) . however , as shown in that paper , the final state of a migrating system is independent of its initial position in the phase space , defined by three values of parameters : the mass and semi - major axes ratios and the total angular momentum . as a consequence , to assess the orbit configuration in the past , we have to set solely the starting semi - major axis value . the initial eccentricity should satisfy the conservation of the total angular momentum of the system . the total angular momentum of the system is composed of the orbital component written as @xmath48 , and the rotational component given by @xmath49 . using kepler s third law , it is easy to show that the ratio of rotational and orbital components is of the order of @xmath50 ( see rodrguez et al . 2011a ) ; since @xmath51 , the contribution of rotation to the total angular momentum can be neglected . therefore , considering that the orbital angular momentum is nearly conserved during tidal migration , we obtain that @xmath52 where the subscripts ini " and current " stand for initial and current orbital elements , respectively . in this way , choosing an initial value for the semi - major axis and knowing the current values of @xmath53 and @xmath54 , equation ( [ am ] ) allows us to constrain the initial value of the eccentricity ( see next section and specially figure [ am - fig ] ) . it should be emphasized that the value obtained in such a way must be considered as an approximation of the initial value of the eccentricity , since the planets may belong to systems with two or more planets . the value of @xmath34 is determined by the relationship between the tidal frequency @xmath55 and the quality factor @xmath56 . in the case of a linear approximation , we have @xmath57 , where @xmath58const . it is known that , in the fourier spectrum of the tidal potential , the most important frequencies ( that is , those involved in terms of order zero and order one in the eccentricity ) are @xmath59 , @xmath60 and @xmath61 . the first one is associated to the term which is independent of the eccentricity , while the two last frequencies correspond to terms which are proportional to @xmath54 ( darwin , 1879 ; ferraz - mello et al . 2008 ) . using the frequency @xmath62 and a modified version of the quality factor written in terms of the love number @xmath40 as @xmath63 , we obtain @xmath64 actually , the spin rotation of exoplanets is not available from observations , thus we choose its initial value arbitrarily but considering a super - synchronous motion ( @xmath65 ) . finally , according to callegari & rodrguez ( 2012 ) , the equilibrium shape attained by the planet due to the gravitational interaction with the star , results in a prolateness , or equatorial ellipticity , which can be computed as @xmath66 . knowing that @xmath19 we can obtain the magnitude of @xmath17 for the planets with estimated radius . for the planets kepler-10**b * * and 55 cnc**e * * ( see section [ k10 ] and [ 55cnc ] ) , we obtain @xmath67 and @xmath68 , respectively , assuming @xmath69 . for sake of comparison with some rocky solar system tidally evolved bodies , we note that @xmath70 in the case of mercury ( smith et al . 2012 ) , @xmath71 for titan ( iess et al . 2010 ) and @xmath72 for io ( schubert et al . 2004 ) . the integration of the exact equations of motion usually requires a long computation time . for this reason , we decided to work with a small set of initial spin - orbit configurations and physical parameters . a statistical analysis taking into account many different initial conditions could give us a more accurate global picture of the past spin - orbit evolution of the planets . as pointed out by correia & laskar ( 2009 ) , such statistical study is more adequate in the context of the averaged equations of the spin - orbit problem ( see section [ mean ] ) . ( solid black curve ) . the dotted line corresponds to the case @xmath73 , where the trappings 2/1 and 3/2 occur . after escaping from the 3/2 resonant motion , the rotation follows the sts towards a synchronization between the rotation and orbital motions . ] kepler-10**b * * is a very close - in planet with a period of @xmath74 0.84 days , orbiting an old ( @xmath75 12 gyr ) sun - like star ( batalha et al . the physical parameters and the current orbital elements of the planet are summarized in table 1 . the estimated mass and radius values ( @xmath76 , @xmath77 ) suggest a rocky composition of the planet , which can be considered as a super - earth ( valencia 2011 ) . the transit technique measurements do not allow a reliable estimation of the current orbital eccentricity , which was arbitrarily fixed at zero in this work . this choice is supported by our numerical simulations of the whole system , which includes also the exterior companion , kepler-10 * c*. the results have shown that the long - term evolution , due to the coupled effect of tides and mutual interaction , could result in the complete circularization of both planetary orbits ( rodrguez et al . , in preparation ) . we adopt the initial semi - major axis at @xmath78 au , which provides @xmath79 through equation ( [ am ] ) . in addition , we set @xmath80 and @xmath81 . for the tidal parameter , we assume @xmath82 ( a value usually adopted for rocky planets ) and , solving equation ( [ k2dt ] ) , we obtain @xmath83 1.1 min . in section [ sett ] we have estimated that , for kepler-10**b * * on its current orbit , @xmath67 . it should be emphasized that @xmath17 would be smaller if the frozen - in shape of the planet was attained at larger distances than @xmath84 . hence , in order to investigate different possible scenarios , we consider three different values of @xmath17 , namely , @xmath85 , @xmath86 and @xmath87 . for kepler-10**b * * depending on the value of @xmath17 . different evolutions are observed , including captures in 2/1 , 3/2 and 1/1 spin - orbit resonances ( sors ) . the solid black curve corresponds to the evolution for @xmath88 ( i.e. , when only the tidal torque exist ) and coincides with the stationary tidal solution ( sts , see text for discussion ) of the rotation . the captures in the synchronous motion are detailed on the small panel . ] figure [ stat - k10 ] shows the variation of @xmath89 as a function of the eccentricity of kepler-10**b**. the curve with segments line corresponds to the stationary tidal solution ( hereafter sts ) of the rotation ( @xmath90 ) , obtained when only the tidal torque is considered . the sts is a function which depends only on the planet eccentricity , and is given by @xmath91 ( see equations ( 42 ) or ( 43 ) of hut ( 1981 ) ; see also section [ mean ] for more details ) . the stationary rotation evolves with the tidally damped eccentricity and the system reaches synchronization ( @xmath92 ) only for circular orbits . note that for small values of @xmath54 , there follows that @xmath93 ; for this reason , the sts is also called pseudo - synchronous solution ( but this is valid only for @xmath94 ) . the solid black curve in figure [ stat - k10 ] corresponds to the result of the numerical simulation performed with @xmath88 ( only tidal forces are considered ) . in this case , the stationary rotation is attained almost instantaneously ( when compared to the time of circularization ) . note that both curves are in good agreement , at least for @xmath95 ; hence , in this eccentricity interval , we can refer to the solution obtained with @xmath88 as sts . we also note that the curve for @xmath88 reaches the sts for @xmath89 close to two ; however , this is incidental and unrelated with the 2/1 trapping . figure . [ rot - k10 ] shows the time variation of @xmath89 . for the three considered values of @xmath96 , the rotation rapidly encounters the value @xmath97 and thus the planet rotation is captured in the 2/1 spin - orbit resonance ( hereafter sor ) . on one hand , for @xmath98 and @xmath99 , the rotation escapes from the 2/1 trapping approximately at 80 kyr , when the rotation drops to the value @xmath100 , implying capture into 1/1 sor and synchronous motion . however , it is worth noting that , for @xmath98 , the synchronism is attained in a form of jump , whereas for @xmath99 the rotation evolves without capture for a very short time before the 1/1 sor ( see small panel in fig . [ rot - k10 ] ) . we have also plotted the result for @xmath88 , showing that synchronization is attained around 400 kyr . it is interesting to note that the introduction of the equatorial asymmetry of the planet reduces the time to reach synchronization ; for instance , for @xmath98 , it is reduced by a factor of 5 . on the other hand , when @xmath73 , the 2/1 sor is destabilized at around 48 kyr giving rise to the trapping in the 3/2 sor ( @xmath101 ) . thus , the permanence of the rotation in the 2/1 sor is temporary and its duration clearly depends on the @xmath17 value . the rotation evolves in a mercury - like state for some time , when a new instability occurs , after which an evolution free of trapping starts . note that during the period of non - capture , the rotation returns to the stationary value , as can be seen in the dotted curve of fig . [ stat - k10 ] when @xmath102 . finally , the synchronization is reached at around 270 kyr . and @xmath54 depend on the specific capture in which the rotation is trapped in . as different values of @xmath17 lead to several captures , different scenarios are expected for the orbital evolution , despite the final values of the elements being the same for all @xmath17 ( see next figure ) . ] shows the final evolution . the orbital decay continues until circularization is achieved . the current semi - major axis of kepler-10**b * * coincides with the one obtained in the simulation . ] figures [ ae - zoom - k10][ae - k10 ] show the time evolution of semi - major axis and eccentricity of kepler-10**b**. we note that , when the rotation evolves under capture in resonance ( that is , for @xmath96 ) , the rates of variations of the elements are larger than the one corresponding to the pure tidal case ( @xmath88 ) , with the difference increasing as the eccentricity is damped ( when the comparison is done for a given sor ) . moreover , for the same value of @xmath17 , the rate of orbital decay and circularization seems to depend on the specific sor in which the planet rotation is trapped in . for example , in the case of @xmath73 , the evolutions of @xmath53 and @xmath54 show two different regimes as the rotation evolves first in the 2/1 and then in 3/2 trappings ( see section [ mean ] ) . we note that the time necessary to get an orbit with @xmath103 is much greater for @xmath104 . in fact , when the rotation evolves without capture , the time - scale for circularization increases , according to figures [ ae - zoom - k10 ] and [ ae - k10 ] . it is worth noting that , in the case of @xmath96 , the synchronous motion can be attained even for @xmath105 . for instance , in the case of @xmath99 , the system is captured in 1/1 sor at roughly 120 kyr , when @xmath106 . this fact confirms the classical result that , for eccentric orbits , the synchronous motion is only possible if there exists an additional torque which counteracts the tidal torque ( see goldreich & peale 1966 ; ferraz - mello et al . 2008 and references therein ) . we also note that after circularization , the 1/1 sor is the only possible resonant motion . figure [ am - fig ] shows the evolution of the system in the plane of orbital elements @xmath53 and @xmath54 . the solid line shows the result of numerical simulation in the case of @xmath98 . the dashed line corresponds to equation ( [ am ] ) , which is independent on the value of @xmath17 . note as both curves are in very good agreement , in such a way confirming our assumption about the conservation of the orbital angular momentum ( section [ sett ] ) . . the solution of equation ( [ am ] ) is shown by the dashed line and is compared to the result of the numerical simulation for the case @xmath98 , giving a good agreement ( the curves for the other two @xmath17 values also agree with the solution of equation ( [ am ] ) , but are not shown here for sake of simplicity ) . the starting position of the planet is indicated by the star symbol . ] gj 3634 b is a super - earth planet with mass @xmath107 orbiting a m2.5 dwarf each 2.65 days with an eccentricity of @xmath108 ( bonfils et al . since there is no transit detection for this planet , its radius is unknown . however , valencia , sasselov & oconnell ( 2007 ) have shown that there is a maximum radius that a rocky terrestrial planet may achieve for a given mass . a larger radius would imply a large amount of water and even the possibility of a massive volatile envelope . indeed , for a planet with mass @xmath109 the maximum terrestrial radius would be 11,600 km ( @xmath110 ) , which result in a mean density of @xmath111 . hence , we assume for gj 3634 b that @xmath112 to obtain @xmath113 . the adopted initial orbital elements are @xmath114 au and @xmath115 , while @xmath116 , @xmath81 and @xmath82 ( @xmath117 min ) . at the current value of the semi - major axis the estimated value of @xmath17 is @xmath118 ( section [ sett ] ) ; in this work , we perform the analysis using two values of @xmath17 : @xmath86 and @xmath87 . as a function of @xmath54 ( left panel ) and the time evolution of @xmath53 ( right panel ) for the planet gj 3634**b**. depending on @xmath17 , the planet rotation may be currently evolving in either 2/1 or 3/2 sor . note that the time - scale of orbital decay is drastically reduced provided the planet can support a permanent deformation resulting in a nonzero @xmath17 value . the vertical and horizontal dotted lines indicates the current values of @xmath54 and @xmath53 , respectively , title="fig : " ] as a function of @xmath54 ( left panel ) and the time evolution of @xmath53 ( right panel ) for the planet gj 3634**b**. depending on @xmath17 , the planet rotation may be currently evolving in either 2/1 or 3/2 sor . note that the time - scale of orbital decay is drastically reduced provided the planet can support a permanent deformation resulting in a nonzero @xmath17 value . the vertical and horizontal dotted lines indicates the current values of @xmath54 and @xmath53 , respectively , title="fig : " ] figure [ rot - exce - gj3634 ] shows the variation of @xmath89 as a function of @xmath54 ( left panel ) and the time variation of @xmath53 ( right panel ) . for @xmath99 , the rotation is rapidly trapped in the 2/1 sor , remaining in such resonant configuration during the most time of the tidal eccentricity damping . only at low eccentricities , close to 0.032 , the system leaves the 2/1 sor dropping directly into the synchronous motion . at variance , for @xmath73 , the rotation quickly reaches the stationary ( non - resonant ) state given by the sts curve ( @xmath88 ) in figure [ rot - exce - gj3634 ] ( left ) . the system evolves along the sts curve up to close approach to the 3/2 sor . then , the system is captured and remains in this resonance until dropping in the synchronized state , at @xmath119 . in both cases , the final configuration of the planet s orbit is characterized by circularization and synchronization , as in the case of kepler-10**b**. however , the current eccentricity of the gj 3634**b * * is different from zero ( its location at 0.08 is shown by the dotted vertical line in the left panel of figure [ rot - exce - gj3634 ] ) . note that the current eccentricity is reached when the rotation still evolves in the 2/1 or 3/2 sor ( depending on @xmath17 ) . this fact suggests that gj 3634**b * * could be presently involved in one of such resonant trappings , in particular , it could be found in a mercury - like rotational state ( @xmath120 ) . however , it should be kept in mind , that this conclusion depends on the choice of specific values of @xmath17 and @xmath121 . indeed , the results of an additional simulation performed with @xmath122 , have shown that the trapping of the system in the 3/2 sor is destabilized at @xmath123 . since @xmath124 , the planet rotation would be currently evolving without capture ( along the sts ) for such a small @xmath17 value . the semi - major axis evolution ( figure [ rot - exce - gj3634 ] , right panel ) shows that the tidal decay is faster when the rotation of the planet is trapped in the 2/1 and 3/2 spin - orbit resonant motions . the fact that the current eccentricity of the planet orbit is different from zero suggests that , in the absence of other perturbations , the orbital decay when @xmath92 or @xmath125 ( see equation ( [ a-11-tot2 ] ) ) . ] of the planet is presently in progress and , consequently , the final semi - major axis will be smaller than the current one . 55 cnc**e * * is a hot super - earth planet , with the mass @xmath126 and the radius @xmath127 , orbiting a sun - like star each 0.74 days with @xmath128 ( winn et al . 2011 ; gillon et al . the averaged density of the planet is comparable to that of earth , suggesting a rock - iron composition supplemented by a significant mass of water , gas , or other light elements ( @xmath129 ) indicate that the possibility of 55 cnc**e * * to be a mini - neptune instead of a super - earth should not be ruled out . ] ( winn et al . 2011 ) . for numerical simulations , we adopt @xmath130 au , @xmath131 , @xmath80 and @xmath81 . the tidal history of the system indicates that the magnitude of @xmath121 must be higher than the commonly adopted values for rocky planets ( ferraz - mello 2012 ) . in this work , we use @xmath132 ; this value allows us to obtain @xmath133 s. we perform numerical simulations for two different values of @xmath17 : @xmath85 and @xmath86 . for the purpose of further comparison with the results reported in callegari & rodrguez ( 2012 ) , we also consider the values @xmath134 and @xmath135 . figure [ rot - exce-55cnc ] shows the evolution of rotation as a function of the planet s eccentricity . the captures inside the 5/2 , 2/1 , 3/2 , 4/3 and 1/1 resonances can be observed . for @xmath134 , the planet rotation evolves under capture in the 3/2 sor almost up to the orbital circularization , with @xmath89 roughly varying between 1.46 and 1.54 . as a function of @xmath54 ( in logarithmic scale ) , for the planet 55 cnc**e**. we have used @xmath98 in addition to the value adopted in callegari & rodrguez ( 2012 ) @xmath134 . also values ten times smaller were taken into account for sake of completeness . the vertical dotted line indicates the current value of @xmath54 , namely , 0.057 . the trappings in the rsos 5/2 , 2/1 , 3/2 , 4/3 and 1/1 appear during the evolutions . note the large amplitude oscillation of @xmath89 inside the 3/2 and 4/3 cases with @xmath17 of the order of @xmath85 . depending on @xmath17 , 55 cnc**e * * can be an example of a planet currently evolving in the 5/2 , 2/1 or 3/2 sor . ] together with the sts curve . as for previously analysed planets , the orbital decay is faster when the planet rotation is trapped in a resonant motion . the current value of @xmath53 is obtained for all values of @xmath17 . ] in a recent work , callegari & rodrguez ( 2012 ) have shown that , for @xmath128 and @xmath136 ( or @xmath134 ) , the phase space of the rotation ( @xmath137 ) of 55 cnc**e * * , between the main spin - orbit resonances , is dominated by a large region of chaotic motion . this region increases as larger values of the eccentricity are considered . capture in spin - orbit resonance can work as a mechanism to prevent a chaotic rotation , provided that the dissipation is large enough in order to ensure the corresponding trapping . according to our simulations ( see figure [ rot - exce-55cnc ] ) , destabilizations of different resonances occur as the eccentricity is tidally damped ( see also next section ) , giving rise to new temporary trappings with a subsequent evolution to synchronism . in this way , a chaotic behavior of the rotation can be avoided by a continuous evolution under resonant motion . it is important to note that , at least for the cases shown in figure [ rot - exce-55cnc ] , there is no interval of non - resonant motion between two successive trappings , as in the case of kepler-10**b * * ( see figure [ rot - k10 ] for @xmath99 and @xmath87 ) . an additional simulation done with @xmath138 ( not shown in figures ) , has resulted in a 2/1 trapping until @xmath139 , indicating that the planet rotation may be currently evolving in resonant motion , for such a small value of @xmath17 . figure [ elem-55cnc ] shows the time variation of semi - major axis . it is worth noting that the orbital decay results in a more rapid evolution towards @xmath84 when the planet rotation is trapped in a sor . also , the rate of variation depends on the specific resonance . from figure [ elem-55cnc ] we see that the evolution when @xmath140 , ( capture in 2/1 ) is almost two times faster than the one for @xmath134 ( 3/2 capture ) . we will return to this discussion in the next section . in order to better understand the results previously obtained through numerical simulations of the exact equations of motion , in this section , we develop an analytical approach in the context of the averaged equations of the spin - orbit problem . the equation of motion for the spin in equation ( [ mov1 ] ) can not be solved analytically . however , because we are interested in the evolution inside spin - orbit resonances , we will consider planetary spin rates having values such that @xmath141 , where @xmath142 is a half - integer . our investigation is restricted to the cases @xmath143 , @xmath144 and @xmath145 but can be easily extended to any other sor . we introduce the `` resonant '' angle @xmath146 , where @xmath147 is the mean anomaly of the planet orbit and @xmath15 was defined in section [ model ] . note that the angle @xmath148 gives the orientation of the long axis whenever the planet passes through the pericenter of its orbit ( i.e. , @xmath149 when @xmath150 , see figure [ elips ] ) . it is easy to show that , inside the spin - orbit resonance given by the parameter @xmath142 , the resonant angle is a slowly varying variable when compared to the orbital period of the planet . indeed , we can write that @xmath151 , where @xmath152 and , consequently , @xmath153 . now , replacing @xmath15 by @xmath154 in equation ( [ t22 ] ) and averaging over the orbital period , we obtain the classical result for the averaged torque on the planet produced by the permanent equatorial deformation ( see goldreich 1966 ; goldreich & peale 1966 ) : @xmath155 where @xmath156 are power series of @xmath54 with the leading term of order @xmath157 ( see appendix a for a literal expansion up to seventh order ) . the averaged tidal torque on the planet is given by @xmath158 where @xmath159 and @xmath160 is a suitable function , whose expansion in powers of @xmath54 is shown in appendix a ( hut 1981 ) . at resonance , the equilibrium solution of the rotation is given by the condition @xmath161 and , resolving for @xmath162 , we obtain @xmath163 if we know the maximum value that @xmath148 can reach without disturbing the trapping in the @xmath142sor , we may obtain the minimum value of @xmath17 necessary for capture in the same resonant motion . replacing @xmath143 in equation ( [ j22-gral ] ) and expanding the ratio @xmath164 in powers of @xmath54 , up to order 6 , we obtain @xmath165 we note that , for @xmath166 , the above expression is the same obtained in ferraz - mello et al . ( 2008 ) , for synchronous asymmetric planets . should be done , where @xmath167 is the phase lag of the tidal wave whose frequency is @xmath168 . ] the averaged variation of the orbital elements can be obtained using the euler - gauss equations ( see beutler 2005 , p. 230 ) , which are written in terms of the radial and transverse components of the perturbing force . the result for the semi - major axis is @xmath169 replacing @xmath162 by the corresponding expression given in equation ( [ j22 - 11 ] ) , we obtain @xmath170 in the context of the linear tidal model , the averaged variation of the semi - major axis due to tides is given by equation ( 9 ) of hut ( 1981 ) , valid for any values of @xmath142 and @xmath54 . the expansion up to @xmath171 reads @xmath172}+{\cal o}(e^8).\ ] ] replacing @xmath143 in the above equation we obtain @xmath173 the total averaged variation of the semi - major axis is then given as @xmath174 and thus @xmath175 at the same order of approximation , the above equation coincides with the result of previous works ( see ferraz - mello et al . ( 2008 ) and references therein ) . when only the tidal torque on the planet is taken into account , the non - resonant sts is given by equation ( [ sync ] ) . in this case , the synchronism is possible only for circular orbits . if instead of @xmath143 , we replace @xmath176 in equation ( [ a - tide - gral ] ) , we obtain @xmath177 we define @xmath178 as the ratio between averaged semi - major axis variation in the @xmath142sor ( @xmath179 ) and in the tidal stationary solution of the rotation ( @xmath180 ) , that is @xmath181 comparing equations ( [ a-11-tot2 ] ) and ( [ a-11-tide - stat ] ) , we note that @xmath182 , indicating that the rate of semi - major axis decay is stronger when the rotation is locked in the 1/1 sor , compared to the case when it evolves following the sts . this result is in agreement with wisdom ( 2008 ) , who has shown that the rate of energy dissipation in a body with synchronous rotation is larger than in a asymptotic non - synchronous rotation . we repeat the procedure described in the previous section to obtain the averaged variation of the planet eccentricity . the resulting expressions are : @xmath183 @xmath184 @xmath185 @xmath186 similarly to the semi - major axis variation case , @xmath187 , where @xmath188 is the analogous for the eccentricity of equation ( [ ra ] ) . we analyse in this section the spin - orbit behavior when the planet evolves in the 3/2 and 2/1 sors , employing the same method described in the previous section has no term independent of @xmath54 and thus equation ( [ j22-gral ] ) can not be expanded around @xmath189 . ] . we have @xmath190 and @xmath191 as a function of orbital eccentricity , for @xmath145 , 3/2 and 1 . in the cases of @xmath192 the ratio of rates of variation of @xmath53 in the sor and in the sts increases for small @xmath54 , showing that the orbital decay is enhanced ( when compared with the evolution in the sts ) whenever the planet rotation evolves under resonance with the orbital motion . ] figure [ razao ] shows a plot of @xmath178 for the three analysed resonances as a function of the eccentricity . on one hand , we note that @xmath193 and @xmath194 increase for small eccentricity , indicating that the major effect on the orbital evolution due to the resonance captures occurs as @xmath54 decreases , in agreement with the results of the numerical simulations ( see figures [ ae - zoom - k10 ] , [ rot - exce - gj3634 ] and [ elem-55cnc ] ) . on the other hand , for large @xmath54 , @xmath195 is small ( mainly in the case @xmath144 ) as also shown in figures [ ae - zoom - k10 ] and [ rot - exce - gj3634 ] , where there is a small difference between the evolutions with and without @xmath17 ( i.e , @xmath196 ) . when the synchronous motion is analysed , we note in figure [ razao ] that @xmath197 is an increasing function of @xmath54 , indicating that the orbital decay when the rotation is trapped in the 1/1 sor is large for large @xmath54 . however , as simulations show , synchronism is attained for very low values of @xmath54 , making the orbital decay almost negligible . the results above are also valid for @xmath198 . it is interesting to note that @xmath178 and @xmath198 are only functions of the eccentricity and do not depend on any physical parameter . it is worthwhile mentioning that @xmath199 averages to zero unless the condition @xmath141 is assumed ( whit @xmath142 integer ) , that is , the averaged orbital evolution due to the triaxial torque is expected only for resonant cases . as it has been noted , equation ( [ j22-gral ] ) can be used to obtain the minimum value of @xmath17 necessary for the stability of the @xmath142sor . calling @xmath200 the one value given by the right hand of equation ( [ j22-gral ] ) , the stability condition for a resonant spin state requires that @xmath201 . the above condition is frequently found in literature as strength criterion of the resonance ( see murray & dermott 1999 , p. 205 ) . figure [ c22 ] shows the functional dependence of the absolute value of @xmath200 with the eccentricity , obtained for kepler-10**b**. the dashed horizontal lines represent three values of @xmath17 used in the numerical simulations shown in section [ k10 ] , namely , @xmath202 and @xmath87 . the case @xmath145 , shown by the blue curve , indicates that @xmath203 for @xmath204 , @xmath205 for @xmath206 and @xmath207 for @xmath208 . a good agreement is obtained doing the comparison with the results of the numerical simulations for kepler-10**b * * , shown in figure [ rot - exce - k10 ] . note as the 2/1 sor destabilizes when @xmath209 , @xmath206 and @xmath208 . the same can be observed for the 3/2 sor , in which the rotation escapes from the trapping when @xmath123 if @xmath73 , in concordance we use @xmath210 , taking also into account that truncated series in @xmath54 were used for the same figure . ] with the result of figure [ c22 ] ( orange curve ) . since for roughly @xmath211 the condition @xmath201 is satisfied more easily for @xmath143 as compared to @xmath192 , there is no problem in understanding the occurrence of synchronous motion for very small eccentricity , as can be also verified in figure [ rot - exce - k10 ] . it is worth noting that instead to compute a critical @xmath17 value , the condition for stationary solutions @xmath212 also enables us to obtain a critical value of @xmath54 for a given fixed @xmath17 . the resonant motion would destabilize whenever the eccentricity becomes smaller than the critical value for each sor . this approach is more convenient in those cases for which the observations allow an estimation of @xmath17 ( see correia & laskar ( 2004 ) for the study of the rotation of mercury ) . given by equation ( [ j22-gral ] ) and calculated with parameters of the planet kepler -10**b**. the calculation of @xmath200 can be used as a guide to investigate the stability of a given sor as a function of eccentricity ( see text for discussion ) . note that in the case of @xmath143 we have @xmath213 for @xmath214 ( see equation ( [ j22 - 11 ] ) ) , indicating that the synchronism is the only possible resonant motion for circular orbits ( see goldreich & peale 1966 ; celletti & chierchia 2008 ; callegari & rodrguez 2012 ) . ] as a function of the orbital eccentricity for kepler-10**b * * , resulting from the analysis of numerical simulations of the exact equations of motion ( see section [ k10-res ] ) . if a comparison with the results of figure [ c22 ] is done , we obtain a good agreement for the values of @xmath54 for which each sor destabilizes ( see text for details ) . ] the numerical exploration of the planet rotation which is subject to the conservative triaxial gravitational torque of the star shows different behaviors , including the oscillation around the equilibrium points of spin - orbit resonances ( see callegari & rodrguez 2012 and references therein ) . when dissipative effects produced by tidal torques are included , the rotation of close - in bodies can be captured in a resonant motion ( goldreich 1966 ; goldreich & peale 1966 ; celletti & chierchia 2008 ) . the capture in a specific resonance depends on the eccentricity @xmath54 and on parameters related to tides and equatorial ellipticity ( @xmath56 and @xmath17 , respectively ) . hence , since @xmath54 is tidally damped during the inward migration , each specific capture should become unstable and the rotation can achieve another resonant state , which , at the same time , should result in a temporary trapping . when the orbit is circularized , the final evolution results in the synchronization between the orbital and rotational periods ( i.e , the 1/1 spin - orbit resonance ) . in this paper , we analysed the coupled spin - orbit evolution of kepler-10**b * * , gj 3634**b * * and 55 cnc**e * * short - period super - earth planets . the numerical simulation of the exact equations of motion have shown that , depending on @xmath17 , the rotation displays a number of resonance trappings for which @xmath141 , including the cases @xmath215 , 2 , 3/2 , 4/3 and 1 . moreover , the results indicate that gj 3634**b * * can be an example of an exoplanet with a rotation state similar to mercury s rotation ( @xmath144 ) , provided the body figure can support a permanent equatorial asymmetry as small as @xmath73 . 55 cnc**e * * can currently evolve under 5/2 or 3/2 trappings if @xmath216 and under 2/1 if @xmath140 or even if @xmath17 is as small as @xmath217 . the rotation of 55 cnc**e * * may have a chaotic behavior for @xmath216 ( callegari & rodrguez 2012 ) . however , we have shown that a scenario is possible in which the rotation evolves to synchronism through successive trappings in sor . indeed , the eccentricity damping due to tides provides the destabilization mechanism required to escape from one given capture and subsequently evolve to lower order resonances ( see correia & laskar 2004 , 2009 ) . we have shown that different spin - states corresponding to various resonance captures can occur in extrasolar systems . although the 1/1 spin - orbit coupling is the final state of evolution , it is not reached in every case as present configuration . for current non - circular orbits , trappings in the 3/2 sor or other resonance configurations can be sustained over long times . the latter would depend on the orbital eccentricity , the planet s @xmath17 value and its dissipation factor . given that these parameters are not well - determined for terrestrial extrasolar planets , we note that the 1/1 coupling may not be the current rotation state of some planets . this would have consequences on the stellar insulation and therefore on the evolution of the atmosphere and surface of the planet . when only the tidal torque is taken into account ( @xmath88 ) , the rotation evolves to the stationary tidal solution ( sts ) . we have shown that under assumption of @xmath96 , the evolution along spin - orbit resonance ( sor ) leads to higher rates of orbital decay and circularization than the corresponding ones when the rotation evolves in the sts . moreover , for a given @xmath17 value , the variation rates of the elements depend on the specific trapping in which the rotation is trapped in , and increases with @xmath142 ( see figure [ ae - zoom - k10 ] ) . we have also calculated the averaged rates of variations of semi - major axis and eccentricity including the gravitational and tidal torques . moreover , through the analytical calculation of the critical value of @xmath17 it is possible to predict the values of the eccentricity for which a given sor would become unstable . the comparison with the direct numerical simulations have shown a good concordance for the analysed planets . the importance of the @xmath17 value on the rotational history of the planet is illustrated in the example of kepler-10**b**. it is worth noting that the planet rotation of kepler-10**b * * can evolve under the influence of the 2/1 sor for longer time - scales ( i.e. , small @xmath54 ) provided the planet figure can support a permanent equatorial deformation characterized by @xmath216 . for smaller @xmath17 , the 2/1 resonant state would become unstable for relatively large eccentricity ( figures [ c22 ] and [ rot - exce - k10 ] ) . we stress that the results of the numerical simulations provide different possible spin - orbit evolutions depending on the considered initial values of elements and rotation . the @xmath56 value and the specific tidal model would also play an important role in determining the rotation history of the planet , since the temporary trapping in sor depends on the rate of eccentricity damping , which is @xmath56-dependent . in fact , in a recent paper , makarov ( 2012 ) has applied the model by efroimsky ( 2012 ) and successfully explained the current resonant state of mercury within the considered model . our results concerning the association of the critical @xmath17 value with the capture probability are not new in the context of averaged equations of the spin - orbit problem ( see for instance correia & laskar 2004 , 2009 ) . however , in this work , we are able to compare the results of the averaged equations with those resulting from the numerical simulations of the exact equations of motion . a model of planet interior including the effect of viscous friction at the core - mantle boundary would provide quantitatively different results ( see peale 2005 , and correia & laskar 2009 , for the case of mercury ) . according to previous works the chances of resonant capture increase if the latter effect is taken into account ( goldreich & peale 1967 ; peale & boss 1977 ) . however , in order to highlight the contribution of @xmath17 to the orbital evolution , in this paper we adopt a simple model neglecting a possible layered structure of the planet . in the present study we have assumed constant @xmath17 values . however , depending on the adjustment to hydrostatic equilibrium @xmath17 may be itself time - dependent . the time - scale for adjustment to the completely relaxed state depends on material parameters , e.g. , rigidity and viscosity which are highly temperature dependent . furthermore , it will depend on the spin - orbit coupling the planet is trapped in . for the application to terrestrial planets given here the constant , i.e. frozen , @xmath17 configuration is justified because adjustment of solid rocky planets to equilibrium usually occurs on geological time - scales . therefore , the states described here do not necessarily represent hydrostatic configurations . in this work we have considered a model of one interacting planet , despite the fact that kepler-10**b * * and 55 cnc**e * * belongs to multiple - planet systems with exterior ( more massive ) companions . previous works have shown that the mutual secular interaction with the companion may affect the eccentricity evolution of the tidally evolved planet , increasing the circularization time - scale ( mardling 2007 ; rodrguez et al . 2011a ; rodrguez et al . moreover , the planetary perturbations can produce large excursions of the innermost planet eccentricity , providing a mechanism for increasing or decreasing the capture probability of the spin - orbit resonances ( correia & laskar 2004 , 2009 ) . however , a numerical simulation of 55 cnc system including the exterior companion 55 cnc**b * * has shown little differences with the single planet case . indeed , the evolution of the inner orbit eccentricity is almost not affected by the mutual interaction between the planets and thus the rotation evolution of 55 cnc**e * * can be safely analysed within the one planet model . an investigation about the rotation of asymmetric close - in super - earths including an exterior companion will be addressed in a future investigation . the authors acknowledge the support of this project by fapesp ( 2009/16900 - 5 ( a.r ) ; 2006/58000 - 2 ( n.c ) ) and cnpq . the authors also gratefully acknowledge the support of the computation centre of the university of so paulo ( lcca - usp ) . we are grateful to the reviewer for his / her valuable comments . we show here the explicit expressions for the functions @xmath156 and @xmath218 . the exact form of @xmath156 is given by

[ sec : man ] according to lagrange , the greatest and most fortunate of mortals had been isaac newton ( 16421727 ) , because it is only possible to discover once the system of the world.@xcite newtonian mechanics is no longer regarded as the last word on the workings of the universe , but it seems likely that as long as there continue to be physics students they shall begin by learning newton s laws of motion.@xcite newton s scientific reputation has long been secure , but the details of his personal and intellectual development have undergone much posthumous scrutiny and second - guessing , from which no clear portrait has emerged . according to theoretical physicist martin gutzwiller , `` although many documents concerning newton s life and work have been discussed at great length [ ... ] , he remains a lonely and mysterious figure with achievements to his credit that have no equal in the history of science.''@xcite the iconoclastic mathematician augustus de morgan and other 19th - century scientists fought the british scientific establishment s hero - worship of newton by dwelling upon and sometimes exaggerating the least attractive aspects of his personality.@xcite stephen hawking s vivid diatribe on newton s `` deviousness and vitriol , '' in his bestselling _ brief history of time_,@xcite belongs to that tradition . even such a great admirer of newton s creative genius as the soviet mathematician vladimir arnold could not discuss newton s career without some disparaging commentary on his character and beliefs.@xcite historian richard s. westfall , who dedicated more than twenty years to his authoritative biography of newton,@xcite later expressed a personal loathing for his subject.@xcite to be sure , the picture of newton that can be gleaned today from the conflicting accounts of those who knew him and from his multifarious personal papers is neither transparently coherent nor conventionally seductive.@xcite computer scientist ernest davis has characterized newton as `` a genius '' and also `` a very strange man , with a very different viewpoint , who lived in a very different world , and who died almost 300 years ago.''@xcite newton was high - strung , wary and secretive in his dealings with others , and intolerant of criticism , but there are few instances in which he can be convicted of bad faith . intellectual rivalries , contests , intrigues , and priority disputes were particularly frequent and acrimonious at a time when scientific publication was in its infancy and savants still depended on aristocratic patronage.@xcite in his notorious and unedifying disputes with hooke,@xcite flamsteed,@xcite and leibniz,@xcite newton did have legitimate grievances . those controversies generated more heat than others of the period primarily because of the greater significance of the scientific work involved.@xcite in his lifetime , newton s `` natural philosophy '' withstood many attacks , most of them misguided and some of them malicious.@xcite in the fraught intellectual climate of the period , those controversies threatened to spill into religious and political disputes that could have cost newton his livelihood and reputation.@xcite newton s guardedness in publishing and explaining his discoveries , as well as his prickly defense of his honor when faced with criticism , do not seem quite so peculiar or paranoid in their historical context . had he not enjoyed restoration england s relative freedom of thought and , from an early age , the security of his cambridge professorship , it is hard to see how his revolutionary work could have been accomplished . in evaluating newton s character , one should also bear in mind that during the last thirty years of his life he served as chief officer of the royal mint , earning nearly unanimous praise for his diligence and honesty . according to mint official and historian sir john craig , newton was not only personally conscientious but also `` a good judge and handler of men , '' endowed with `` magnetism which in many engendered an extraordinary regard and respect.''@xcite in the estimation of the eminent economist lord keynes , newton was `` one of the greatest and most efficient of our civil servants.''@xcite newton s efforts to improve the coinage contributed significantly to the nation s finances.@xcite after his appointment to the mint had made him a wealthy man , he became known to his extended family and to others for his charitable giving.@xcite [ sec : occult ] keynes had acquired many of newton s private manuscripts.@xcite he was therefore aware of the great scientist s interests in both alchemy and theology , leading him to declare in 1946 that newton had not been `` the first of the age of reason '' but rather `` the last of the magicians.''@xcite scholars have since debated whether newton s alchemical and religious pursuits are properly characterized as `` magical '' and how they relate to his work in physics and mathematics . the mature newton showed no interest in astrology,@xcite communication with spirits , ritual magic , or other of the common occult pursuits of the day.@xcite the editor of newton s mathematical papers , d. t. whiteside , could find `` no hint of any belief by him in number - mysticism,''@xcite while recent scholarship has documented newton s hostility to neo - platonism , gnosticism , kabbalism and other esoteric traditions influential in the renaissance.@xcite newton s private writings do evince a keen involvement with alchemy and its arcane imagery , but at the time there was no clear boundary between alchemy as mysticism and alchemy as early modern chemistry . moreover , the otherworldliness of alchemy as practiced in newton s day may have been exaggerated by 19th - century occultists and those influenced by their interpretations.@xcite newton s alchemical interests , which he shared with eminent contemporaries such as robert boyle and john locke , undoubtedly influenced his atomism.@xcite on the other hand , the notion that alchemy contributed to his conception of gravity as an action at a distance may be , according to the most recent scholarship , `` something of a canard.''@xcite leibniz , who rejected the newtonian theory of gravity , originated the claim that newton conceived of gravity as an `` occult quality , '' an aspersion that newton strenuously denied.@xcite explaining that he did not `` feign hypotheses '' about the mechanism by which gravity acted across empty space , newton presented his theory as a mathematical description of gravity s observable effects , an attitude wholly compatible with modern scientific principles.@xcite according to a tradition collected by voltaire , when the elderly newton was asked how he had discovered his celebrated law of universal gravitation , he replied : `` by thinking on it continually.''@xcite [ sec : religion ] [ t ] newton was a devout but unorthodox christian who privately rejected the doctrines of the holy trinity@xcite and the immortality of the soul as unscriptural and idolatrous.@xcite he dedicated a considerable effort to establishing a chronology of human civilization , combining his belief in the wisdom of the ancients and in the historicity of the bible with innovative uses of astronomy and other quantitative techniques.@xcite a failure from the standpoint of our current understanding of history , that work does shed light on newton s attitudes towards the various branches of learning . philosopher richard popkin tried to make theology the key to newton s intellectual development , provocatively asking why `` one of the greatest anti - trinitarian theologians of the 17th century '' took `` time off to write works on natural science , like the _ principia mathematica_?''@xcite but the fact that , in due course , newton s mathematics and physics were decisively embraced by an international community of scholars , the vast majority of whom were either ignorant of his religious convictions or actively hostile to them , suggests that newton s theology can contribute little to elucidating the _ content _ of his science.@xcite even newton s belief that divine intervention is needed to keep the planets on their regular orbits as theologically charged a claim about nature as he ever published@xcite was based on the mathematical laws of gravity that he had abstracted from astronomical observations , and the problem that he thereby raised , the dynamical stability of the solar system , has continued to occupy physicists and mathematicians to this day.@xcite in my view , if there is something to learn today from newton s religion , and more broadly from his life beyond his immortal work in the exact sciences , it is a lesson close to max weber s `` elective affinities '' between the worldview associated with certain historical strands of protestant christianity , and the scientific and industrial revolutions.@xcite according to historian stephen snobelen , `` in his biblicism , piety and morality , newton was a puritan through and through.''@xcite the central paradox of newton s life is close to what weber underlined in his analysis of the growth of capitalism : that the puritans , whose convictions seem so remote from our perspective , contributed decisively to creating the modern world.@xcite in the context of the radical english puritanism of the 17th century of the religion of oliver cromwell , john milton , and the pilgrim fathers the man isaac newton gains some intelligibility.@xcite puritanism and its legacy have long been contentious in english - language historiography , as reflected , for instance , in the widely diverging evaluations of oliver cromwell s rule.@xcite newton s own ( doctrinally heterodox ) puritanism may account , in part , for the antipathy of some of his biographers . historian frank manuel saw the prosecution of counterfeiters during newton s tenure at the mint as an opportunity for the great man to `` hurt and kill without doing violence to his scrupulous puritan conscience , '' adding that `` the blood of the coiners and clippers nourished him.''@xcite rupert hall dismissed this `` sadistic vampire '' portrait as `` blood - tub victorian melodrama rather than biography.''@xcite [ sec : knight ] [ t ] on 16 april 1705 , in a ceremony conducted at trinity college , cambridge , queen anne elevated isaac newton to the rank of knight bachelor.@xcite at the time , that honor was usually granted to military officers and senior figures in the national and local governments , as well as to rich merchants and others with political connections.@xcite newton s knighthood was probably a royal favor to his political patron , lord halifax . halifax , a prominent figure in the whig political party , hoped thereby to promote newton s candidacy in the upcoming parliamentary election , but newton came last at the polls and never again sought elective office.@xcite the lawyer james montagu ( lord halifax s brother and a future attorney general ) and the academic john ellis ( master of gonville and caius college and vice - chancellor of the university of cambridge ) were knighted along with newton . more than a few `` new men '' of obscure ancestry , like ellis , figure among the knights bachelor created by queen anne.@xcite for his part , sir isaac soon sought to establish his status as a gentleman.@xcite in november he submitted to the college of heralds an affidavit detailing his ancestry and claiming sir john newton , bart.,@xcite a rich landowner from culverthorpe , in lincolnshire , as a third cousin . this was supported by a declaration signed by sir john.@xcite that genealogy might seem suspicious , since isaac newton s father had been a relatively prosperous but illiterate yeoman farmer.@xcite it was not uncommon , in that day and age , for a wealthy man to pay an accommodating officer for the right to bear a coat of arms , or to some other hereditary privilege , based on a fictional pedigree . in fact , the second sir john newton , bart . ( father of the cousin who supported isaac newton s affidavit ) had inherited his baronetcy from a sir john newton to whom he was unrelated . the first sir john , a native of gloucestershire , was childless . he agreed to pass on his baronetcy to his rich namesake and putative cousin , resident in lincolnshire , in exchange for help settling his financial obligations . but archival research has established that the genealogy in isaac newton s 1705 affidavit is genuine and that he prepared it from the available records with his habitual scrupulousness.@xcite according to accounts collected by sir david brewster , later in his life newton claimed in private conversation that his male - line ancestry might have been scottish.@xcite yet it seems quite unlikely that newton could have seriously entertained that idea , which contradicts the well - documented genealogy that he had submitted to the college of heralds . moreover , it is clear from their mutual correspondence that isaac newton consistently regarded sir john newton as head of their common clan . sir john s heir , the rich and politically connected sir michael newton , would serve as chief mourner at isaac newton s funeral in 1727.@xcite [ sec : blazon ] [ t ] isaac newton s personal coat of arms is shown in fig . [ fig : cambridge ] . this escutcheon is carved on a stone tablet above the front door of woolsthorpe manor , the home in woolsthorpe - by - colsterworth , near grantham , lincolnshire , where newton was born and where he made some of his greatest intellectual breakthroughs after returning there when the university of cambridge was closed due to the great plague of 166566 . that tablet , shown in fig . [ fig : woolsthorpe ] , was installed in 1798 by edmund turnor , whose father had purchased woolsthorpe manor in 1733.@xcite in the language of heraldry , fig . [ fig : cambridge ] is described ( `` blazoned '' ) as `` sable , two shinbones in saltire , the dexter surmounted of the sinister argent . '' this requires some explanation for the uninitiated : `` sable '' means black and refers to the background color . a `` saltire '' is an x - shaped cross . `` argent '' means white and identifies to the color of the bones . the `` sinister '' bone ( which runs from the upper right to the lower left ) surmounts the `` dexter '' ( running from the upper left to the lower right ) . the last detail is interesting . the reader may care to verify that the crossbones of fig . [ fig : cambridge ] are arranged like the blades of a pair of left - handed scissors . even with symmetrical handles , ordinary ( i.e. , right - handed ) scissors can not be turned so that they will be congruent with fig . [ fig : cambridge ] : they will always look like the crossbones in fig . [ fig : barrs ] . the design of fig . [ fig : cambridge ] is therefore unequivocally left - handed , or , to use the latin term , _ sinister_. the blazon to which isaac newton was entitled by virtue of the genealogy in the 1705 affidavit did not specify which bone should lie on top.@xcite the arms of the newton baronets show the dexter bone surmounting the sinister ( see fig . [ fig : barrs]).@xcite on the other hand , samuel newton ( 16281718 ) , mayor of cambridge , who was unrelated to sir isaac but with whom sir isaac must have become acquainted during his years at the university , displayed his own coat of arms with the sinister bone surmounting the dexter.@xcite the coat of arms of isaac barrow , newton s mentor and predecessor as lucasian professor of mathematics , also features a sinister saltire , in his case formed by two crossed swords . an ornate carving of it , by the eminent artist grinling gibbons , graces the end of a bookshelf in trinity college s wren library , built while barrow was master and newton a fellow of that college.@xcite [ sec : symbolism ] like other aspects of his life , newton s heraldry resists easy interpretation . newton did not create the crossbone device , which is ancestral . on the other hand , one might attribute some meaning to his adoption of such a stark design ( which only a couple of decades later would become indelibly associated with the jolly roger by the pseudonymous captain charles johnson s _ general history of pyrates_@xcite ) , rather than seeking a new grant of arms from the college of heralds ( which he might well have obtained , given his eminence ) . newton s efforts to uncover factual accounts of history and natural science concealed in ancient sources , such as the bible , greek myths , or the _ corpus hermeticum _ , were surely related to the `` emblematic worldview '' of other renaissance intellectuals who sought wisdom by deciphering and correlating the symbolisms that they believed lay hidden in all things . but , as matt goldish has underlined , `` newton s relationship to symbols was not mystical , like that of the rosicrucians or other hermetics , but rather mathematical : he regarded them as a puzzle for which available clues provide a logical solution.''@xcite in his study of ancient sources , newton always sought precise , rational meanings that he thought had been obfuscated after the corruption of mankind s first religion.@xcite the alchemical literature of his day , with its deliberate encoding of chemical recipes in riddling allegories,@xcite must have encouraged that approach . scholars who have taken seriously the characterization of newton as `` last of the magicians '' have tended to downplay this _ disenchanting _ tendency , as well as his consistent rejection of metaphysical speculation divorced from rigorous experimental knowledge.@xcite with this in mind , three non - exclusive possibilities for newton s interpretation of his own escutcheon appear plausible : as a conventional mark of his social rank and family connections , as a stark _ memento mori _ like the skulls and hourglasses of puritan gravestones,@xcite or as a mathematical symbol connected to what would much later be called _ chirality_. though wholly conjectural , the last possibility is intriguing in light of other intellectual developments . [ sec : chirality ] an object is said to be chiral if it can not be superimposed on its mirror image . the term , coined by lord kelvin in 1893,@xcite is taken from the greek word for hand ( @xmath0 ) , the human hands being familiar instances of chiral objects : the left and right hands are mirror images of each other , and a left - hand glove will not fit over a right hand , or vice - versa . some 19th - century mathematical scientists , including clerk maxwell and willard gibbs , referred to the geometrical transformation that replaces a figure with its mirror image as a `` perversion,''@xcite a usage that is now largely obsolete but persists in some scientific contexts.@xcite awareness of the incongruence of mirror images is much older . modern chemists often credit ren descartes ( 15961650 ) with this insight , based on the remark attributed to him that `` any man who , upon looking down at his bare feet , does not laugh , has either no sense of symmetry or no sense of humor.''@xcite i have been unable to find this quote in descartes s published works or correspondence , but descartes did have a sophisticated understanding of what would later be termed chirality , for he famously explained magnetism as a flow of tiny particles shaped as left - handed or right - handed corkscrews , whose mutual action could therefore produce either attraction of repulsion.@xcite newton rejected such _ ad hoc _ , unverifiable hypotheses as `` fictions.''@xcite but when it came to the double refraction ( now called birefringence ) exhibited by crystals of iceland spar , newton proposed an interpretation in terms of the `` sides '' of light.@xcite ( these `` sides , '' akin to the linear polarizations of modern physical optics , are not chiral . ) did descartes s interest in the incongruence of mirror images influence newton ? though i have found no direct evidence of this , the possibility is not far - fetched . newton was both descartes s greatest disciple and his greatest critic . in the words of mathematician and science historian clifford truesdell , `` if the steely - splendid preface to the _ principia _ is not just a bolt from the blue ( and where in science as such bolts to be found ? ) , who else than descartes can be its grandsire?''@xcite [ sec : rightful ] the young newton first came to the attention of the `` republic of letters '' as creator of the first practical reflecting telescope.@xcite an issue that might have caught newton s attention is how a telescope can alter an image s orientation . galileo s old refracting telescope leaves orientation unchanged , but kepler s improved refractor rotates it by 180@xmath1 . the image can be made upright by reflection ( a desirable feature when observing objects on land ) , at the cost of reversing left and right . newton s telescope used a pair of mirrors , one curved , the other flat . each mirror reverses chirality , but the net effect is that the image is rotated while chirality is preserved . in his later years , newton had further occasion to consider the incongruence of mirror images when he sought to estimate the date of the expedition of the argonauts from what he took to be a contemporaneous description of the positions of the equinoxes and solstices with respect to the constellations . in the course of that work , newton had to correct for inconsistencies in the chiralities of constellations as drawn in stellar atlases . he referred to the corrected stellar maps as being `` rightly delineated , '' confusing some of his critics.@xcite the reason for those inconsistencies is that constellations may be depicted as they are actually seen from the earth or as they would appear on the surface of a celestial globe ( i.e. , from an imaginary vantage point outside the celestial sphere).@xcite [ sec : philosophy ] [ t ] in 1768 , immanuel kant considered chirality ( which he called the problem of `` incongruent counterparts '' ) in the context of the philosophical debate on whether the concept of space is reducible to the relations between concrete objects.@xcite this was a question that dated back to the disputes between newton and leibniz.@xcite kant argued that the spatial relations between the fingers of a given hand are the same whether it be a left or a right hand and that the fact that the two hands are distinguishable must therefore be explained by appeal to an external geometrical space in which the hands exist . according to kant , this supported newton s belief in the existence of an absolute space , which leibniz had denied.@xcite chirality figures also in ludwig wittgenstein s _ tractatus logico - philosophicus _ of 1921 . wittgenstein pointed out that a right - hand glove could be worn over a left hand if the three - dimensional glove could be turned around in a fourth spatial dimension.@xcite ( an equivalent observation had been made by mathematician august mbius in 1827.@xcite ) wittgenstein s comment is characteristically gnomic , but he seems to have concluded from this that chirality is tied to the space in which the object is embedded ( and therefore to the ways in which it is allowed to move ) rather than being a property of the object itself.@xcite [ sec : chemistry ] [ t ] the young louis pasteur , who had just completed his doctorate in chemistry and physics , was faced with a puzzle . he knew that @xmath2-tartaric acid , extracted from fermented grape juice , is optically active ( meaning that it rotates the plane of the polarization of light passing through it ) . factory - made `` paratartaric acid '' ( now called racemic tartaric acid ) has the same elemental composition as @xmath2-tartaric acid , yet crystallizes differently and is optically inactive . in 1848 , pasteur discovered that the artificially synthesized acid could make two distinct crystals , shown in fig . [ fig : tartrate ] , which are mirror images of each other . by carefully picking out crystals of one form and dissolving them , he produced a substance identical to the natural @xmath2-tartaric acid.@xcite pasteur s work established that racemic tartaric acid is a mixture of two substances whose molecules are mirror images of each other ( `` enantiomers '' ) , and that only one of the two forms occurs in nature . a decade afterwards , his experiments on tartrate fermentation led pasteur to conclude that chirality is an essential factor in molecular recognition within biological systems.@xcite pasteur s brilliant elucidation of molecular chirality would play a key role in the growth of biochemistry . writing a century later , the great crystallographer and molecular biologist j. d. bernal described it as pasteur s `` first and in some ways his greatest scientific discovery.''@xcite pasteur saw deep significance in the homochirality of biomolecules , writing that `` life as manifested to us is a consequence of the dissymmetry of the universe.''@xcite why life on earth evolved to use only one enantiomer of such organic substances as amino acids and sugars remains mysterious.@xcite [ sec : particle ] in 1956 , theoretical physicists t. d. lee and c. n. yang proposed that the weak nuclear interaction , unlike the other forces of nature , could distinguish between mirror images . soon afterwards , c. s. wu confirmed this experimentally when she found that in radioactive decays of cobalt-60 nuclei the electron produced is more likely to be emitted in a direction opposite to the nucleus s intrinsic angular momentum ( `` spin '' ) , making the decay process left - handed.@xcite a measurement of the spin of a massless particle along the direction of its linear momentum has only two possible outcomes , which are mirror images of each other and are therefore called chiralities . for the photon ( the massless particle of light ) , these correspond to its two circular polarizations.@xcite elementary particles that do have mass ( such as the electron ) are now regarded as mixtures of the two chiralities.@xcite the weak interaction acts exclusively on the left - handed component of particles and on the right - handed component of antiparticles . the standard model ( sm ) of high energy physics is therefore formulated in terms of chiral quantum fields.@xcite in 1964 , cronin and fitch reported experimental evidence that there is also a slight asymmetry between the interactions of left - handed particles and right - handed antiparticles . in 1973 , kobayashi and maskawa showed how this effect , known as `` charge - parity '' ( cp ) violation , was possible within the theoretical context of the sm . the amount of cp violation in the sm is , however , far too small to explain why the universe contains as much matter as it does , but no antimatter . how cp violation in the early universe can have been large enough to account for the matter content that we see today is one of the great unsolved problems of both cosmology and particle physics.@xcite [ sec : armiger ] [ t ] like other wealthy armigers of the period , sir isaac had his coat of arms painted on the doors of his carriage . a french commentator mistook it for a `` death s head '' and interpreted it as evidence that newton whose christian convictions were problematic for the free - thinking _ philosophes _ who embraced his science had `` entirely taken to religion '' in his old age.@xcite ( laplace and others would later pursue the idea that newton s religiosity was a symptom of senility or derangement , perhaps subsequent to his nervous breakdown of 1693.@xcite ) members of other english families surnamed newton also bore crossbones on their coats of arms.@xcite that heraldic device appears to be medieval and might have been intended originally as a symbol of warlike prowess . several armigerous newtons ( including the newton baronets ) displayed , as a crest above the shield with the crossbones , the image of an `` eastern prince '' ( or in some cases a `` naked man '' ) kneeling on his left knee and surrendering his sword.@xcite according to a family tradition , sir ancel gorney , an ancestor of the newtons , fought along king richard i in the third crusade and captured a muslim prince at ascalon.@xcite in 1722 , william stukeley , a personal friend of sir isaac and a noted antiquarian , engraved a drawing of the church where newton was baptized ( see fig . [ fig : stukeley ] ) , and later included it in his manuscript biography of newton.@xcite that illustration incorporates , as a sort of caption , an imaginary monument with a heraldic decoration and a latin inscription . the kneeling figure in the crest suggests that stukeley based his decoration on the heraldry of the newton baronets.@xcite the coat of arms appears as well on the portrait shown in fig . [ fig : seeman ] , made in the final years of newton s life . the use of the heraldic device suggests that this particular painting closely patterned after an earlier one by enoch seeman was made for the newton family.@xcite note that it depicts the dexter chirality for the crossbones , contrary to what is shown in figs . [ fig : cambridge ] and [ fig : woolsthorpe ] . the dexter chirality is also used in the arms carved by john woodward in 175556 above the stalls in the chapel of trinity college.@xcite [ sec : sinister ] the fictional protagonist of neal stephenson s 2003 novel _ quicksilver _ visits newton at woolsthorpe manor during the great plague and anachronistically notes the crossbones of fig . [ fig : woolsthorpe ] , a device whose `` awfulness '' embarrasses him as an englishman.@xcite in fact , despite living in an age when heraldry was highly prized as a mark of identity and status , newton left a tenuous heraldic trail . i have been unable to confirm that he personally chose , or even used , the sinister chirality illustrated in fig . [ fig : cambridge].@xcite as detailed above , the relevant documents from the college of heralds specify no chirality , while surviving instances in which a coat of arms is associated with isaac newton generally show the dexter chirality of the newton baronets , with the significant exception of the tablet pictured in fig . [ fig : woolsthorpe ] . the sinister chirality was associated with the newtons of cambridge , whom isaac newton must have known but to whom he was unrelated . the genteel turnor family of stoke rochford assiduously cultivated the memory of their illustrious countryman . figure [ fig : roubillac ] shows their copy of the bust of newton by sculptor louis franois roubillac , whose original resides in the trinity college library . antiquarian edmund turnor ( 17541829),@xcite grandson of the man of the same name who purchased woolsthorpe manor in 1733 , recorded only the dexter chirality as connected with sir isaac s family.@xcite but his father ( another edmund ) may have had reason to install such a permanent and conspicuous depiction of the sinister chirality at the most significant of all locales associated with newton s life.@xcite in heraldry , a diagonal band running from the upper left - hand to the lower right - hand corner is called a `` bend . '' the mirror image , with the band running from the upper right - hand to the lower left - hand corner , is identified as a `` bend sinister . '' the reason for this terminology is that if the coat of arms were drawn on the front of a battle shield , the top of the bend sinister would be near the knight s left shoulder . was newton aware that in the english heraldry of his day a bend sinister often marked the arms of a bastard child?@xcite [ sec : wonderchild ] newton s father had died three months before his birth . when his mother remarried , the three - year - old newton was left in the care of his maternal grandparents . treated almost as an orphan , newton had a lonely and unhappy childhood , in which some biographers have seen a source of his adult neuroses . in keynes s view , `` isaac newton , a posthumous [ son ] born with no father on christmas day,@xcite 1642 , was the last wonderchild to whom the magi could do sincere and appropriate homage.''@xcite it is easy to miss newton s personal dimensions or to mistake them altogether . like everyone else , he had a genealogy , both familial and intellectual . he was a creature of his own time and place , his work a part of what einstein called `` the doubtful striving and suffering of his generation.''@xcite yet he was also a prime mover in what may be the grandest cultural shift in history . he left no biological descendants , but he is an ancestor to modern humankind . neither the standards of his day nor those of ours are quite apt for giving his full measure . and neither the hero - worship of his first biographers nor the various revisionisms of subsequent scholars have produced a compelling picture of how such a man was possible.@xcite on his deathbed , newton refused the last rites of the church of england . he died on the morning of 20 march 1727 . a week later his body lay in state in the jerusalem chamber at london s westminster abbey . on 4 april the lord chancellor and five other noblemen carried the remains to their resting place beneath the abbey s nave , under a gravestone marked _ hic depositum est quod mortale fuit isaaci newtoni _ ( `` here lies what was mortal of isaac newton '' ) . i thank neer asherie and joseph gal for correspondence on pasteur s work , for bringing ref . to my attention , and for suggestions on improving this manuscript . jed buchwald and mordechai feingold pointed out refs . and . daniel garber and kurt smith assisted the search for the source of the quotation attributed to descartes in ref . . rob iliffe kindly provided advance copy of the relevant chapters in ref . . andrew gray and bernard juby gave expert heraldic advice ( ref . ) . i also thank jos cayasso for producing figs . [ fig : cambridge ] and [ fig : barrs ] , tom hayes for help procuring titles from the harvard libraries , jordy geerlings for correspondence on hatzfeld s anti - newtonianism ( ref . ) , don campbell for discussions about heraldic marks of bastardy ( ref . ) , as well as ari belenkiy and vanessa lpez for general feedback . finally , thanks are due to the national portrait gallery s rights and images department for permission to use fig . [ fig : seeman ] , and to ken redmore of the society for lincolnshire history and archaeology for permission to use fig . [ fig : roubillac ] . j. delambre , `` notice sur la vie et les ouvrages de m. le comte j .- lagrange , '' in http://archive.org/details/uvresdelagrange22natigoog[_uvres de lagrange _ ] , vol . i , ( paris : gauthier - villars , 1867 ) , p. xx . r. s. westfall , _ never at rest _ , ( cambridge : cambridge u. p. , 1980 ) . this remains the foremost scholarly effort to document and interpret newton s career , but at several points it betrays an imperfect grasp of newton s aims , methods , and results . compare , e.g. , westfall s dismissive treatment of newton s lunar theory ( pp . 540548 ) with refs . . there is now a vast secondary literature on newton , much of it by the `` newton industry '' that developed in the 1960s after scholars began in earnest to explore his private manuscripts . this literature is riven by disagreements on newton s personality and motives , on the significance of his interests in alchemy and religion , and on their relation to his achievements in the exact sciences . even the technical interpretation of some aspects of newton s published work in mathematics and physics has proved controversial . see r. s. westfall , `` the changing world of the newtonian industry , '' j. hist . ideas http://dx.doi.org/10.2307/2708719[*37*(1 ) , 175184 ] ( 1976 ) ; `` technical newton , '' isis http://dx.doi.org/10.1086/357657[*87*(4 ) , 701706 ] ( 1996 ) ; m. mazzotti , `` the two newtons and beyond , '' brit . j. hist . http://dx.doi.org/10.1017/s0007087406008764[*40*(1 ) , 105111 ] ( 2007 ) ; m. feingold , `` honor thy newton , '' early sci . http://dx.doi.org/10.1163/157338207x166416[*12*(2 ) , 223229 ] ( 2007 ) . johann bernoulli , who as a leibnizian partisan did much to sow discord with the newtonians , later resorted to a false pretense in a priority dispute against his own son daniel . see c. truesdell , `` the hydrodynamics of daniel and john bernoulli ( 17271740 ) , '' in _ leonhardi euleri opera omnia _ , ser . 2 , vol . 12 , ( lausanne : orell fssli turici , 1954 ) , pp . xxiii xxxviii ; `` experience , theory , and experiment , '' in _ an idiot s fugitive essays on science _ , ( new york : springer - verlag , 1984 [ 1955 ] ) , pp according to truesdell , the elder bernoulli lived in `` a world of challenges , enmities , secret methods , and anagrams '' ( p. 11 ) . this was newton s world also , but newton detested it and for much of his life would rather have allowed his discoveries to pass unnoticed than engage in the baroque polemics needed to defend them . after newton published his groundbreaking work on color in 1672 , hooke s hasty and dismissive response opened a frustrating controversy that helped drive newton away from physics and from publication for over a decade ( see ref . , ch . 7 ) . the animosity between hooke and newton was later aggravated by hooke s outspoken insistence that newton s _ principia _ failed to credit him with the original insight that the elliptical orbits of planets may be explained by an inverse - square law for gravity . hooke lacked the knowledge of calculus necessary to establish this rigorously , but his 1679 correspondence with newton may have helped to turn newton s attention back towards the subject and to clarify newton s thinking on orbits , by then already quite advanced . see m. nauenberg , `` robert hooke s seminal contribution to orbital dynamics , '' phys . perspect . http://dx.doi.org/10.1007/s00016-004-0226-y[*7*(1 ) , 434 ] ( 2005 ) . for a sympathetic summary of newton s position with regard to flamsteed , the first astronomer royal , see m. p. silverman , review of d. h. clark s and s. p. h. clark s _ newton s tyranny _ , am . j. phys . http://dx.doi.org/10.1119/1.1561274[*71*(5 ) , 507508 ] ( 2003 ) . the notion that the mature newton exercised tyrannical control of the british scientific establishment is undermined by flamsteed s ultimate success in keeping halley and newton from dictating the terms for the publishing of his observations . even within the royal society that he presided , newton never enjoyed unanimous support , even though his international reputation far exceeded any other member s ( see ref . , pp . 696697 ) . for a full account of the feud between newton and leibniz over the history of calculus , see a. r. hall , _ philosophers at war _ , ( cambridge : cambridge u. p. , 1980 ) . according to hall , `` leibniz was by no means a modest man . he was really different from newton in this respect '' ( p. 48 ) . see also d. bertoloni meli , _ equivalence and priority : newton versus leibniz _ , ( oxford : clarendon press , 1993 ) ; n. guicciardini , `` against leibniz , '' in _ isaac newton on mathematical certainty and method _ , ( cambridge , ma : mit press , 2009 ) , pp . 329384 . even the amiable christiaan huygens , second only to newton among the mathematical scientists of the period , was embroiled in `` perpetual controversy '' with members of the royal society , including hooke . see m. feingold , `` huygens and the royal society , '' _ de zeventiende eeuw _ http://www.dbnl.org/tekst/_zev001199601_01/_zev001199601_01_0004.php[*12*(1 ) , 2236 ] ( 1996 ) . for an extreme instance of anti - newtonianism see j. c. f. de hatzfeld , http://archive.org/details/caselearnedrepr00hatgoog[_the case of the learned _ ] , ( london : t. churchill , 1724 ) , a screed published during newton s lifetime and dedicated to king george i , which accuses newton of `` vulgar errors '' and of overturning `` both natural and revealed religion . '' it is based on a confused reading of earlier leibnizian critiques , including the conviction that a `` dr . orfireus '' had overturned newtonian mechanics by finding the secret of perpetual motion.@xcite hatzfeld s polemical career is discussed in j. geerlings , `` anti - newtonianism and radical enlightenment , '' in _ newton and the netherlands _ , eds . e. jorink and a. maas , ( amsterdam : amsterdam u. p. and leiden u. p. , 2012 ) , pp . the broader history of anti - newtonianism up to the mid 18th century is surveyed in j. i. israel , _ enlightenment contested _ , ( oxford : oxford u. p. , 2006 ) , ch . 8 , but the treatment of some of the relevant scientific questions is not altogether satisfactory . newton rejected the reality of ghosts and demons , characterizing `` enchanters , magicians , sorcerers , necromancers , and witches '' as `` deceivers and cheats . '' see s. d. snobelen , `` lust , pride , and ambition : isaac newton and the devil , '' in _ newton and newtonianism _ , eds . j. e. force and s. hutton , ( dordrecht : kluwer , 2004 ) , http://dx.doi.org/10.1007/1-4020-2238-7_8[pp . 155181 ] . l. m. principe , `` reflections on newton s alchemy in light of the new historiography of alchemy , '' in _ newton and newtonianism _ , eds . j. e. force and s. hutton , ( dordrecht : kluwer , 2004 ) , pp . http://dx.doi.org/10.1007/1-4020-2238-7_10[205219 ] ; _ the secrets of alchemy _ , ( chicago : u. of chicago p. , 2012 ) . in my opinion , however , principe overstates the continuity between what were later distinguished as `` alchemy '' and `` chemistry . '' metallic transmutation , an ongoing project in the west for fifteen hundred years , lost nearly all credibility in the course of the 18th century . the only plausible explanation of this that i can see is that the new culture of open scientific debate , which had only begun to grow in newton s lifetime , soon revealed that belief in such transmutation had been sustained by a mix of secrecy , confusion , and fraud . w. r. newman , _ atoms and alchemy _ , ( chicago : u. of chicago p. , 2006 ) ; `` newton s early optical theory and its debt to chymistry , '' in _ lumire et vision dans les sciences et dans les arts _ , eds . d. jacquart and m. hochmann , ( geneva : droz , 2010 ) , pp . 283307 . w. r. newman , `` the significance of newton s alchemical research , '' in _ new dictionary of scientific biography _ , vol . 1 , n. koertge , ( detroit : charles scriber s sons , 2008 ) , pp . see also pp . 2729 in i. b. cohen and g. e. smith , `` introduction , '' in _ the cambridge companion to newton _ i. b. cohen and g. e. smith , ( cambridge : cambridge u. p. , 2002 ) , pp . 132 . commentary by eminent 20th - century physicists that stresses the modernity of newton s approach includes a. einstein , `` the mechanics of newton and their influence on the development of theoretical physics , '' in _ einstein s essays in science _ , ( mineola , ny : dover , 2009 [ 1934 ] ) , pp . 2839 ; r. p. feynman , _ the character of physical law _ , ( cambridge , ma : mit press , 1965 ) , ch . 1 ; s. chandrasekhar , _ newton s _ principia _ for the common reader _ , ( oxford : clarendon press , 1995 ) , chs . 2 , 19 . for a brief and insightful overview of newton s anti - trinitarianism , see m. wiles , _ archetypal heresy _ , ( new york and oxford : oxford u. p. , 1996 ) , pp . newton s conclusion that the dogma of the trinity reflected a corruption of the early christians simple faith by the hellenistic taste for metaphysics is consistent with the interpretation later articulated by gibbon , the great rationalist historian : e. gibbon , _ the history of the decline and fall of the roman empire _ ii , ( new york : modern library , 2003 [ 1781 ] ) , ch . xxi . on newton s heterodox religious views and the extent to which he concealed them , see s. d. snobelen , `` isaac newton , heretic : the strategies of a nicodemite , '' brit . j. hist . http://dx.doi.org/10.1017/s0007087499003751[*32*(4 ) , 381419 ] ( 1999 ) . r. h. popkin , `` newton s biblical theology and his theological physics , '' in _ newton s scientific and philosophical legacy _ p. b. scheurer and g. debrock , ( dordrecht : kluwer , 1988 ) , http://dx.doi.org/10.1007/978-94-009-2809-1_4[pp . 8197 ] . classical mechanics as it is taught today derives largely from the efforts of euler , lagrange , and other continental scientists to deploy leibniz s analytical techniques in order to facilitate and generalize the application of newton s physical principles . this development following as it did the bitter and ideologically charged disputes among cartesians , leibnizians , and newtonians seems to me inexplicable except by its success in bringing a wide range of natural phenomena into a coherent and predictive mathematical account : i.e , by its ability to fulfill its explicit objective . i. newton , `` query 31 , '' in _ opticks _ , 4th ed . , ( new york : dover , 1979 [ 1730 ] ) , pp . 376406 . the relevant part of his query is quoted and annotated in h. g. alexander ( ed . ) , _ the leibniz - clarke correspondence _ , ( manchester : manchester u. p. , 1956 ) , pp . 175183 . s. tremaine , `` is the solar system stable ? '' _ the institute newsletter _ , summer 2011 , ( princeton : institute for advanced study ) , pp . 1 , 6 . http://www.ias.edu/about/publications/ias-letter/articles/2011-summer/solar-system-tremaine newton s life offers a nearly ideal illustration of the `` worldly asceticism '' with which weber credited the puritans and some other protestant sects ; cf . m. weber , _ the protestant ethic and the spirit of capitalism _ , ( london : routledge , 2001 [ 1930 ] ) , ch . 4 ; r. iliffe , `` self - experiment , sex , and the care of the self , '' lecture delivered at the center for science , technology , medicine & society of the u. of california , berkeley , 11 apr . 2013 http://www.youtube.com/watch?v=r8arrhktxvc ; ref . . newton s anti - trinitarianism might not be an insurmountable obstacle to locating his worldview within the sphere of puritanism . milton , a key figure in cromwell s regime , arrived at anti - trinitarian views decades before newton : m. dzelzainis , `` milton and antitrinitarianism , '' in _ milton and toleration _ , eds . s. achinstein and e. sauer , ( oxford : oxford u. p. , 2010 ) , http://dx.doi.org/10.1093/acprof:oso/9780199295937.003.0009[pp . 171185 ] . in new england , the orthodox calvinism of the pilgrim fathers would evolve into an increasingly liberal unitarianism : c. wright , _ the beginnings of unitarianism in america _ , ( hamden , ct : archon press , 1976 [ 1955 ] ) . newton is frequently identified as the first scientist knighted by the british monarch ( see , e.g. , ref . ) . yet francis bacon , an early advocate of experimental science , had been knighted in 1603 . christopher wren , sometime savilian professor of astronomy at oxford , was knighted in 1673 . bacon held high political offices and wren was an eminent architect , so that their knighthoods probably had little to do with their scientific pursuits . but newton s knighthood was probably not intended as recognition of his science either . le neve , http://archive.org/details/lenevespedigrees00lene[_le neve s pedigrees of the knights _ ] _ made by king charles ii , king james ii , king william iii and queen mary , king william alone , and queen anne _ , ed . g. w. marshall , ( london : mitchell & hughes , 1873 ) . `` bart . '' is the traditional abbreviation used after the name of a baronet . baronetcies are the only hereditary honors granted by the british crown that are not titles of nobility . both knights and baronets are styled `` sir , '' but knighthood is not inheritable . c. w. foster , `` sir isaac newton s family , '' reports and papers of the architectural and archological societies of the county of lincoln , county of york , archdeaconries of northampton and oakham , and county of leicester * 39*(1 ) , 162 ( 1928 ) . a yeoman owned the land that he farmed . according to the prevailing conventions , he ranked above laborers , artisans , and husbandmen , who owned little or no land of their own , and below gentlemen , who owned enough land to live on the rent thereof , without having to work at all . d. brewster , https://archive.org/details/memoirsoflifewri02brewuoft[_memoirs of the life _ ] , _ writings , and discoveries of sir isaac newton _ , 2nd ed . ii , ( edinburgh : edmonston and douglas , 1860 ) , pp . this report is wholly based on hearsay . brewster ( known as a scientist for his pioneering work on optical polarization ) was newton s first serious biographer , but some of his contemporaries and most subsequent historians have criticized his treatment of newton as `` hero - worship '' ( see ref . ) . it is natural that brewster , a proud scot , should have been eager to claim newton as a countryman . t. wotton , `` newton of barrs - court , gloucestershire , '' in http://openlibrary.org/books/ol24432884m/the_english_baronetage[_the english baronetage _ ] , vol . iii , part i , ( london : tho . wotton , 1741 ) , pp . 145149 . j. guillim , _ a display of heraldry _ , 6th ed . , ed . g. mackenzie , ( london : for r. and j. bonwicke and r. wilkin , in st . paul s church - yard , and j. walthoe and tho . ward , in the temple , 1724 ) , http://books.google.co.cr/books?id=4fwzaqaamaaj&pg=p142[p . 142 ] ; s. newton , _ the diary of samuel newton _ , ed . j. e. foster , ( cambridge : cambridge antiquarian soc . , 1890 ) , http://archive.org/stream/diaryofsamuelnew00newtrich#page/vi/mode/2up[p . vi ] . c. johnson , ( identified in this edition as daniel defoe ) , _ a general history of the pyrates _ , ed . m. schonhorn , ( mineola , ny : dover , 1999 [ 1724 ] ) , pp . 68 , 226 , 234 , 352 ; d. cordingly , _ under the black flag _ , ( new york : random house , 1996 ) , pp . xix xx , 114119 . see also http://commons.wikimedia.org/wiki/file:every,henry.jpg ( accessed 28 jul . 2013 ) a. r. hall , _ isaac newton : adventurer in thought _ , ( cambridge : cambridge u. p. , 1992 ) , pp . 198200 , 344348 . rob iliffe refers to newton s approach in these matters as a `` counter - allegorical hermeneutic.''@xcite w. r. newman and l. m. principe , `` the chymical laboratory notebooks of george starkey , '' in _ reworking the bench _ , eds . f. l. holmes , j. renn and h .- j . rheinberger , ( dordrecht : kluwer , 2003 ) , pp . d. levitin , `` scholarly and scholastic contexts for newton s general scholium , '' talk delivered at the conference _ isaac newton s general scholium to the _ principia : _ science , religion and metaphysics _ , u. of king s college , halifax , 2426 oct . http://isaacnewton.ca/general-scholium-symposium/ on the interpretation given to a similar heraldic device by john newton ( 17251807 ) , the abolitionist clergyman who authored the popular hymn `` amazing grace , '' see r. southey , https://archive.org/stream/selectionsfroml02wartgoog#page/n439/mode/2up[letter to john rickman dated 31 jan . 1836 ] , in _ selections from the letters of robert southey _ , vol . iv , ed . j. w. warter , ( london : longman , brown , green , longmans & roberts , 1856 ) , pp . 432433 ; notes and queries ( ser . 5 ) http://books.google.co.cr/books?id=7mieaaaayaaj&pg=pa445[*8*(206 ) , 445 ] ( 1877 ) . see , e.g. , d. c. walker ( ed . ) , _ origins of optical activity in nature _ , ( amsterdam and new york : elsevier , 1979 ) , p. vii ; c. c. gaither and a. e. cavazos - gaither ( eds . ) , _ gaither s dictionary of scientific quotations _ , 2nd ed . , ( new york : springer , 2012 ) , p. 2455 . i. newton , queries 25 , 26 , and 29 , in _ opticks _ , 4th ed . , ( new york : dover , 1979 [ 1730 ] ) , pp . 354361 , 370374 . for a critical appraisal of newton s work on birefringence , see d. brewster , https://archive.org/details/memoirsoflifewri01brewuoft[_memoirs of the life _ ] , _ writings , and discoveries of sir isaac newton _ , 2nd ed . i , ( edinburgh : edmonston and douglas , 1860 ) , pp . 187190 . here brewster s commentary is particularly interesting , as it concerns a scientific subject on which he was a leading expert and original contributor . for an overview of newton s notions about the microscopic structure of matter , see a. e. shapiro , `` newton s optics and atomism , '' in _ the cambridge companion to newton _ , i. b. cohen and g. e. smith , ( cambridge : cambridge u. p. , 2002 ) , pp . 227255 . a similar confusion accounts for the north - south reversal of all constellations except orion on the zodiac painted by paul helleu on the vaulted ceiling of grand central terminal , in new york city . see new york transit museum and a. w. robins , _ grand central terminal _ , ( new york : stewart , tabori & chang , 2013 ) , p. 90 . i. kant , `` on the first ground of the distinction of regions in space , '' in _ the philosophy of right and left _ , eds . j. van cleve and r. e. frederick , ( dordrecht : kluwer , 1991 [ 1768 ] ) , http://dx.doi.org/10.1007/978-94-011-3736-2_3[pp . 2733 ] . kant s views later evolved into his mature belief famously defended in his _ critique of pure reason _ of 1781 that absolute time and space are not features of the external world , but rather the necessary forms of human intuition . see ref . , pp . xlviii xlix , and j. v. buroker , `` the role of incongruent counterparts in kant s transcendental idealism , '' in _ the philosophy of right and left _ , j. van cleve and r. e. frederick , ( dordrecht : kluwer , 1991 ) , http://dx.doi.org/10.1007/978-94-011-3736-2_22[pp . 315339 ] . a. f. mbius , `` on higher space , '' in _ the philosophy of right and left _ , eds . j. van cleve and r. e. frederick , ( dordrecht : kluwer , 1991 [ 1827 ] ) , http://dx.doi.org/10.1007/978-94-011-3736-2_6[pp . 3941 ] . j. van cleve , `` right , left , and the fourth dimension , '' in _ the philosophy of right and left _ , eds . j. van cleve and r. e. frederick , ( dordrecht : kluwer , 1991 [ 1987 ] ) , http://dx.doi.org/10.1007/978-94-011-3736-2_18[pp . 203234 ] . g. b. kauffman and r. d. myers , `` pasteur s resolution of racemic acid : a sesquicentennial retrospect and a new translation , '' chem . educator http://dx.doi.org/10.1333/s00897980257a[*3*(6 ) , 118 ] ( 1998 ) . j. gal , `` the discovery of biological enantioselectivity : louis pasteur and the fermentation of tartaric acid , 1857a review and analysis 150 yr later , '' chirality http://dx.doi.org/10.1002/chir.20494[*20*(1 ) , 519 ] ( 2008 ) . l. pasteur , `` observations sur les forces dissymtriques , '' in _ uvres de pasteur _ , l. p. vallery - radot , ( paris : masson et cie . , 1922 [ 1874 ] ) , https://archive.org/stream/oeuvresdepasteur01past#page/360/mode/2up/[pp . 360363 ] . particle physicists distinguish between `` helicity , '' which corresponds to the handedness of a particle as seen by a given observer , and `` chirality , '' which is independent of the observer . helicity and chirality are equivalent only for massless particles , which travel at the speed of light . for a simplified discussion , see ref . technically , a relativistic quantum field may be characterized by a pair of positive integers , @xmath3 , corresponding to the respective dimensionalities of two representations of the rotation group . spatial reflection exchanges @xmath4 and @xmath5 . fields with @xmath6 or @xmath7 have definite chirality ( left - handed or right - handed , respectively ) . massive particles with spin correspond to a mixture of chiralities . see t. banks , _ modern quantum field theory _ , ( cambridge : cambridge u. p. , 2008 ) , ch . p. noguez , letter to the abb conti , dated 19 aug . 1726 . quoted in c. albertan and a .- chouillet , `` autographes et documents , '' recherches sur diderot et sur lencyclopdie http://rde.revues.org/417[*34 * , 213230 ] ( 2003 ) . see also ref . , p. 326 . w. stukeley , _ memoirs of sir isaac newton s life _ , ed . a. hastings white , ( london : taylor & francis , 1936 ) . a transcription of the 1752 manuscript is available at http://en.wikisource.org/wiki/memoirs_of_sir_isaac_newton%27s_life ( accessed 28 jul . 2013 ) edmund turnor had a memorial to the newton family installed within the colsterworth church . it shows similar heraldic symbols to those in stukeley s engraving ( fig . [ fig : stukeley ] ) . see `` memorial to the newton family in colsterworth church , '' colsterworth and district parish council , http://parishes.lincolnshire.gov.uk/colsterworthanddistrict/imagedetail.asp?id=71858 ( accessed 28 jul . 2013 ) ; e. turnor , _ collections for the history of the town and soke of grantham _ , ( london : w. miller , 1806 ) , http://books.google.co.cr/books?id=zrw2aqaamaaj&pg=pa154[p . 154 ] . the lancaster herald of arms in ordinary , mr . robert noel , could find no surviving instance of a coat of arms drawn or used by newton personally , and that armigers of the period might not have been careful about details such as which bone was shown surmounting the other . b. juby , personal communication , 10 oct . j. martin , `` turnor , edmund ( _ bap_. 1754 , _ d_. 1829 ) , '' in _ oxford dictionary of national biography _ , ( oxford : oxford u. p. , 2004 ) . online ed . may 2006 , http://www.oxforddnb.com/view/article/27887 ( accessed 28 jul . 2013 ) curiously , ref . mentions the 1798 restoration of woolsthorpe manor and the plaque installed at that time in the room where newton was born ( p. 156 ) , and is otherwise scrupulous in documenting local inscriptions and monuments , yet fails to mention the tablet of fig . [ fig : woolsthorpe ] . did the younger mr . turnor have qualms about the chirality of the coat of arms depicted there ? a bastard child who had been acknowledged as such could adopt the father s arms , but only after adding a suitable difference to avoid infringing on the privileges of the father s legitimate descendants . various different marks could be used for that purpose , but from about the 15th to the 18th centuries the bend sinister was the most common in english heraldry and it increasingly became associated with bastardy in the public mind . see c. given - wilson and a. curteis , _ the royal bastards of medieval england _ , ( london : routledge & kegan paul , 1984 ) , pp . 5153 . in the gregorian calendar , not adopted in britain until 1752 , newton s birthday was 4 january . newton worked on an alternative proposal for reforming the civil and ecclesiastical calendars , never carried out . see a. belenkiy and e. vila echage , `` history of one defeat : reform of the julian calendar as envisaged by isaac newton , '' notes rec . r. soc . http://dx.doi.org/10.1098/rsnr.2005.0096[*59*(3 ) , 223254 ] ( 2005 ) . that newton worked before many of the conventions of modern science ( regular publication , professionalized peer review , error analysis , etc . ) had been established , coupled to his portrayal by de morgan , manuel , westfall , and other biographers who lent more credence to newton s enemies than to his friends , have caused some leading scientists to deal hypercritically with newton in recent years . see ref . and s. l. glashow , `` the errors and animadversions of honest isaac newton , '' contr . http://dx.doi.org/10.2436/20.7010.01.42[*4*(1 ) , 105110 ] ( 2008 ) . m. nauenberg , `` newton s perturbation methods for the three - body problem and their application to lunar motion , '' in _ isaac newton s natural philosophy _ , eds . j. z. buchwald and i. b. cohen , ( cambridge , ma : mit press , 2001 ) , pp . 189224 ; review of n. kollerstrom s _ newton s forgotten lunar theory _ , j. hist . http://articles.adsabs.harvard.edu/full/2001jha....32..162n[*32*(2 ) , 162168 ] ( 2001 ) . this was j. e. e. bessler , _ alias _ orffyreus , whose work is surveyed in a. jenkins , `` the mechanical career of councillor orffyreus , confidence man , '' am . j. phys . http://dx.doi.org/10.1119/1.4798617[*81*(6 ) , 421427 ] ( 2013 ) [ arxiv:1301.3097 [ physics.hist-ph ] ] .

there are basically two approaches to modeling the edge and scrape - off layer ( sol ) regions in magnetically confined plasmas : mean field transport models and turbulence models . in the first approach@xcite , mean field fluid equations are solved in a realistic geometry typically including the divertor and x - point regions . these models have very detailed descriptions of collisions , neutral particles , impurities , and material surfaces , whereas turbulent transport is not calculated self - consistently but modeled by effective turbulent diffusion terms . turbulence models@xcite take a different approach . here , a detailed description of the turbulence is the main goal , whereas the level of detail of collisions , neutral particles , impurities and material surfaces , if included in the models at all , is crude compared with the highly detailed descriptions in mean field transport models . furthermore , in order to reduce the computational requirements the magnetic field geometry is often simplified . both approaches have provided useful results in their respective regimes of validity . nevertheless , several open questions in contemporary fusion plasma research require that these complementary approaches are merged . examples are : the formation of a elevated density shoulder@xcite as the particle density approaches the greenwald density limit and the associated changes of filament properties@xcite , the high particle density front@xcite on the high field side of single null diverted plasmas , and eventually the possible influence of turbulent transport on the transition to the detached@xcite divertor regime . here , we present a step towards more detailed edge and sol turbulence models by showing how classical collisional transport across the magnetic field is included in drift - fluid models@xcite based on the braginskii fluid closure@xcite . compared with neoclassical transport , classical collisional transport across the magnetic field is small in most regions of a magnetically confined fusion plasma . in turbulence dominated systems collisional transport is important because of its diffusive nature which inevitably determines the smallest possible scale length of the turbulence below which the turbulent energy is dissipated . it is well - known that turbulence behaves remarkably different when collisional dissipation is completely neglected compared with the situation where a very small but finite dissipation is present . the results of neoclassical transport theory are concerned with the steady - state and does not take turbulence into account nor resolve the much faster turbulent time - scale . in the pfirsch - schl " uter regime , neoclassical transport is a result of collisional diffusive transport along the magnetic field in combination with guiding - center drifts due to magnetic field inhomogeneity . on the turbulent time - scale , it is therefore not evident that neoclassical effects through parallel collisional transport is the dominant regulator of turbulence in particular when the spatial scale of the turbulence is small as in e.g. , drift wave turbulence . a consistent model for perpendicular collisional transport may therefore be of importance for the interplay between collisions and turbulence and certainly provide a valuable validation tool for models which do not take classical perpendicular collisional transport into account . moreover , classical perpendicular transport is believed to play a role for the perpendicular convection of filaments in the sol region . a filament at the outboard midplane typically has a size which is on the order of @xmath0 ion gyroradii , but when then filament approaches the x - point region the perpendicular scale length can easily be smaller than the ion gyroradius due the strong magnetic shear near the x - point@xcite . in that case the perpendicular conductivity and viscosity are important mechanisms for closing the filament current loop@xcite and thereby modify the filament convection and possibly decouple perturbations above and below the x - point . finally , classical perpendicular transport are important in other fully ionized plasmas without neoclassical transport , e.g. cylindrical plasmas . in drift - fluid models , the components of the momentum density equations perpendicular to the magnetic field are solved iteratively essentially under the assumption that the dynamics is evolving much slower than the ion gyration time scale . in this limit , every force density , including resistivity and collisional viscosity , gives rise to a perpendicular drift@xcite . most of the collisional drifts are included in transport models@xcite but not in a way which provides energy conservation . in turbulence models , on the other hand , the collisional drifts are usually neglected or do only include selected collisional effects@xcite . in a model including ion temperature dynamics , we show how these collisional terms in combination with heat fluxes , heat exchange terms , and viscous heating terms in the electron and ion pressure equations are included in drift - fluid turbulence models in a consistent way . consistency here mainly refers to energy conservation . in this paper we show that collisional transport across the magnetic field can be included in drift - fluid models without altering the energy theorem@xcite . for example , we show that the collisional viscous damping of turbulence and mean flows gives rise to a conservative energy transfer between the kinetic energy and the ion thermal energy . in other words , collisions can be added to drift - fluid models without introducing energy sinks or sources . the inclusion of perpendicular collisional transport in the drift - fluid equation gives rise to diffusion of particle density , momentum , and ion and electron pressure , and hence potentially renders the common use of artificial diffusion terms in turbulence models redundant . the collisonal terms are complex functions of the fluid and the electromagnetic field variables . thus when included in turbulence models , they place significant demands on computing powers . therefore , we also present a partly linearized model for the collisional transport across the magnetic field which allows significantly faster computing times . the model is embedded in the hot edge - sol electrostatic ( hesel ) model@xcite , the successor of the esel model@xcite . hesel is a two - dimensional ( 2d ) model describing interchange turbulence in a plane perpendicular to the magnetic field at the outboard midplane of a tokamak . it evolves the particle density , the vorticity , and the electron and ion pressures . we show that our partly linearised collision terms , including inter species energy exchange , do not alter the energy theorem , and we show by means of 1d numerical simulations without turbulence that the simplified and full models give very similar results . the article is organized as follows . in section [ sec : perpcolltrans ] we describe how collisional effects influence the dynamics perpendicular to the magnetic field in a low - frequency drift - fluid turbulence model . we derive the global energy theorem and discuss energy exchange mechanisms due to collisions . in section [ sec : heselmodel ] a simplified model for collisional effects is derived which is included in the hesel model also presented here . numerical simulations of the full and simplified models for the perpendicular collisional effects are presented , and finally in sec . [ sec : conclusions ] we summarise our findings and draw conclusions . in this section we show how collisional transport across the magnetic field is consistently incorporated into drift - fluid models . our derivation is based on the braginskii@xcite fluid equations describing the time evolution of particle density @xmath1 , velocity @xmath2 and temperature @xmath3 for a collisional , quasi - neutral @xmath4 , electrostatic , simple electron - ion plasma . the index @xmath5 is a species label , which refers to ions and electrons , respectively . we start our derivation from the braginskii momentum equation @xmath6\bm{u}_a = -\nabla p_a - \nabla \cdot \bm{\pi}_a + q_an_a(\bm{e } + \bm{u}_a\times \bm{b } ) + \bm{r}_a . \label{eq : momentum}\end{aligned}\ ] ] here , @xmath7 denotes mass , @xmath8 is charge , @xmath9 is the scalar pressure , @xmath10 is the stress tensor , and @xmath11 denotes the resistive force . the resistivity in the electron momentum equation consists of a frictional force @xmath12 \label{eq : ru}\end{aligned}\ ] ] and a thermal force @xmath13 where the electron - ion collision frequency is defined as @xmath14 the magnitude of the electron gyrofrequency is @xmath15 , and @xmath16 denotes the coulomb logarithm . we also introduced a unit vector parallel to the magnetic field @xmath17 and the magnetic field aligned component of the gradient operator @xmath18 . if @xmath19 is an arbitrary vector , then we introduced the notation @xmath20 for the projection onto the magnetic field unit vector @xmath21 , and @xmath22 for the perpendicular part of the vector . momentum conservation implies @xmath23 . the origin of the thermal force @xmath24 is the velocity dependence of the particle collision frequency ( @xmath25 ) . the stress tensor @xmath10 consists of three parts : i ) a part @xmath26 describing viscosity along the magnetic field due to like - particle collisons , ii ) a gyro - frequency dependent part @xmath27 describing collisional momentum transport across the magnetic field due to like particle collisions , and iii ) a gyro - viscous part @xmath28 . here we consider the @xmath29 and @xmath30 parts of the ion stress tensor . the electron stress tensor is neglected because it is smaller than the ion contributions by the electron - ion mass ratio . in the ion stress tensor we assume that the perpendicular flow is incompressible and assume a constant magnetic field . the pure perpendicular ion viscous tensor@xcite written in a local coordinate system @xmath31 , where @xmath32 is aligned with @xmath21 , then reads : @xmath33 [ eq : visc_perp_coll ] here the ion viscosity coefficient is given as @xmath34 and the ion - ion collision frequency is @xmath35 the parallel - perpendicular parts , e.g. , @xmath36 , do not contribute to the perpendicular dynamics to lowest order and are therefore not treated here since we do not consider the dynamics parallel to the magnetic field . in the parallel direction the parallel - perpendicular part of the stress tensor give rise to perpendicular viscosity . in the drift ordering @xcite the fundamental assumptions are that the characteristic frequencies are much smaller than the ion gyrofrequency @xmath37 and that perpendicular gradient length scales @xmath38 are larger than the ion gyro - radius @xmath39 we further assume @xmath40 , here @xmath41 denotes a characteristic parallel gradient length scale . this ordering permits iterative solutions of the momentum equations for the perpendicular fluid drifts . the first order drifts are given as @xmath42 which are the @xmath43 and diamagnetic drifts , respectively . the second order drifts are @xmath44 in all second order drift we take @xmath45 . the first term @xmath46 is the polarization drift . due to the mass dependence only the ion polarization drift is retained . the second term @xmath47 is the drift@xcite associated with resistivity , which is identical for electrons and ions due to momentum conservation . the resistive drift consists of a friction force drift and a thermal force drift , respectively : @xmath48 here @xmath49 denotes the total scalar pressure . note that only the diamagnetic current appears in the friction force drift @xmath50 because the electron and ion @xmath43-drift contributions cancel . the diamagnetic drift is the fluid representation of the particle grad - b and curvature drifts , and the magnetization current associated with the gyrating charged particles@xcite . therefore , the diamagnetic drift does not represent transport of guiding - centers unless the magnetic field is inhomogeneous . in that sense the friction force drift @xmath50 is a fluid representation of friction between electrons and ions in opposite directed larmor orbits . the direction of thermal force drift @xmath51 is parallel to the electron temperature gradient . as already mentioned , this drift arises because fast particles experience less collisions than slow particles , which result in unbalanced fluxes , and hence implies up - gradient transport . the last first order drift is the viscous drift @xmath52 . as earlier mentioned this term is mass dependent and hence only the ion drift is retained . in the local coordinate system @xmath31 the purely perpendicular parts can be written as @xmath53 being a second order drift only the first order drifts @xmath54 , see eq . ( [ eq : vperp0 ] ) , are retained in the stress tensor @xmath55 . with the algebraic expressions for the perpendicular drifts at hand , we can write down the resulting drift - fluid equations . we will omit the magnetic field aligned parts of the momentum equations since we are here mainly concerned with perpendicular dynamics . by inserting the perpendicular drifts given in equations ( [ eq : vperp0 ] ) and ( [ eq : vperp1 ] ) into the braginskii@xcite particle density and pressure equations , we get : @xmath56\big ) + p_e \nabla \cdot [ \bm{u}_e + \bm{u}_{de } + { \bm{\hat{b}}}u_{\|e}+\bm{u}_r ] \notag \\ + \nabla \cdot \bm{q}_{e } + \nabla_{\perp } \cdot \bm{q}_{e}^ * = -q_{\delta } + \bm{r } \cdot ( \bm{u}_i-\bm{u}_e ) \label{eq : df_pe}\\ \frac{3}{2}{\frac{\partial}{\partialt}}p_i + \frac{3}{2}\nabla \cdot \big(p_i[\bm{u}_e + \bm{u}_{di } + { \bm{\hat{b}}}u_{\|i } + \bm{u}_{pi } + \bm{u}_{\pi i } + \bm{u}_r]\big)\notag \\ + p_i \nabla \cdot [ \bm{u}_e + \bm{u}_{di } + { \bm{\hat{b}}}u_{\|i } + \bm{u}_{pi } + \bm{u}_{\pi i } + \bm{u}_r ] + \nabla \cdot \bm{q } _ { i } + \nabla_{\perp } \cdot \bm{q}_{\perp i}^ * + \bm{\pi}_i^{\perp } : \nabla \bm{u}_{i\perp,1 } = q_{\delta}. \label{eq : df_pi}\end{aligned}\ ] ] equation ( [ eq : df_w ] ) is the vorticity equation which results from subtracting the ion and electron particle density equations assuming quasi - neutrality . in all equations higher order terms are retained even though some of these terms are formally small . the motivation for not neglecting these higher order terms is that they , e.g. , guarantee energy conservation , provide diffusion or are responsible for energy exchange between different plasma species . drift - fluid equations have been derived by several authors@xcite but without self - consistently including perpendicular collisional transport . in the remainder of this section we describe and discuss the collisional terms entering the drift - fluid moment equations . thorough descriptions and discussions of classical transport outside the drift - fluid context is found in e.g. refs . . we start with the electron particle density equation ( [ eq : df_n ] ) where the divergence of the resistive flux is the only collisional term . using equations ( [ eq : uru ] ) and ( [ eq : urt ] ) , the divergence of the resistive flux can be written as @xmath57 \bigg ) , \label{eq : resflux}\end{aligned}\ ] ] demonstrating the origin of the resistive drifts , namely a perpendicular random - walk process with step length @xmath58 and frequency @xmath59 . we see that the perpendicular friction force @xmath60 gives rise to particle density diffusion with a diffusion coefficient @xmath61 in drift - fluid models , similar to the explicit appearance of spatial diffusion in low - frequency ordered collision operators@xcite . the thermal force drift @xmath51 is partly canceled by the electron temperature gradient dependent part of the friction force drift @xmath50 , leaving only the last term on the right hand side . in order to highlight the @xmath62 and @xmath63 dependency of the electron - ion collision frequency @xmath64 we define @xmath65 where @xmath66 and @xmath67 denote a constant reference particle density and a constant reference electron temperature , respectively . equation ( [ eq : resflux ] ) then reads : @xmath68\bigg ) , \label{eq : resfluxii}\end{aligned}\ ] ] showing that the particle density diffusion coefficient has a @xmath69 dependence . next , consider the electron pressure equation ( [ eq : df_pe ] ) . here , collisional effects enter through classical heat fluxes , heat transfer terms , and resistive drift terms . the collisional electron heat flux@xcite consists of two parts : @xmath70 where the heat conductivities@xcite are @xmath71 and where only the first order perpendicular drifts given in eq.([eq : vperp0 ] ) were used in @xmath72 . the origin of the heat flux @xmath72 is the same as the thermal force @xmath73 given in eq . ( [ eq : rt ] ) , which are both associated with the velocity dependence of the electron - ion collision frequency . note the exact onsager@xcite symmetric coefficients in @xmath73 and @xmath72 . it is convenient to express the perpendicular parts of the electron heat flux in terms of the perpendicular resistive drifts given in equations ( [ eq : uru ] ) and ( [ eq : urt ] ) @xmath74 when written in this form it is evident that the divergence of the frictional electron pressure flux cancels the perpendicular part of @xmath72 : @xmath75 on the right hand side of the electron pressure equation ( [ eq : df_pe ] ) , the @xmath76 term transfers energy between the electron and ion fluid kinetic energies an the electron thermal energy . in the drift - fluid electron pressure equation the energy transfer term can be combined with the remaining perpendicular heat flux terms and terms which depend on the resistive drift @xmath77 the two first terms on the right hand side , @xmath78 \bigg ) , \label{eq : pe_diffii } \end{aligned}\ ] ] among others , yield diffusion of electron pressure and in general relax gradients , and hence the up - gradient flux contained in @xmath79 is canceled by perpendicular heat conduction in @xmath80 . the last term in eq . ( [ eq : pe_diff ] ) transfers energy between the electron and ion thermal energy due to resistivity . hence , no explicit energy transfer channel between kinetic and thermal energy due to resistivity remains in the drift - fluid equations . energy exchange between comoving electron and ion fluids is described by the heat exchange term @xmath81 entering the electron and ion pressure equations ( [ eq : df_pe ] ) and ( [ eq : df_pi ] ) , respectively . the equilibration occurs on the slower @xmath82 collision time - scale because the energy transfer in each scattering event is proportional to the mass ratio of the scattering particles @xmath83 . we elaborate further on inter - species energy exchange in section [ sec : energyconservation ] . we now turn to the ion pressure equation ( [ eq : df_pi ] ) where the heat conduction is given as @xmath84 and the thermal conductivities@xcite are @xmath85 the ion heat flux is solely driven by ion temperature gradients and is independent of ion - electron collisions . the ratio between the perpendicular electron and ion heat conductivities is of the order @xmath86 for @xmath87 , whereas @xmath88 . perpendicular collisional heat transport is therefore dominated by the ions , and parallel heat transport is electron dominated . the terms involving the resistive drift in the ion pressure equation can be written as @xmath89\bigg ) - \bm{u}_r \cdot { \nabla_{\perp}}p_i , \label{eq : pi_diff}\end{aligned}\ ] ] which shows that electron - ion collisions through the resistive drift give rise to ion pressure diffusion , but also that the thermal force drift @xmath51 potentially drives the ion pressure up the electron temperature gradient . note that the corresponding effect in the electron pressure equation was canceled by the electron heat conduction . the diffusion coefficient is proportional to the electron diffusion coefficient times the ratio of the ion to the electron temperature . the ratio of the resistive diffusion and the perpendicular ion heat conduction can be estimated as @xmath90 where @xmath91 and @xmath92 are characteristic electron and ion temperature gradient length scales . perpendicular heat conduction therefore dominates resistive diffusion except for plasmas where the ion temperature is significantly higher than the electron temperature or where the electron temperature has much steeper gradients than the ion temperature . in both pressure equations the diamagnetic heat flux is defined as @xmath93 lastly , we consider viscous effects in the vorticity and ion pressure equations ( [ eq : df_w ] ) and ( [ eq : df_pi ] ) , respectively . in both equations viscous effects enter through the viscous drift @xmath94 defined in eq . ( [ eq : vperp1 ] ) , but the viscous tensor only explicitly appears in the ion pressure equation . notice that only the divergence of the viscous particle density flux remains in the vorticity equation . the resistive electron and ion fluxes cancel as a consequence of momentum conservation . the divergence of the perpendicular viscous flux results in diffusion of the magnetic field aligned @xmath43 and ion diamagnetic vorticities due to ion - ion collisions , as we will show in section [ sec : heselmodel ] . in the ion pressure equation the corresponding ion pressure flux gives rise to hyperviscosity `` @xmath95 '' of the ion pressure . however , compared with the perpendicular ion heat conduction the viscous ion pressure flux term is in most cases small @xmath96 the viscous drift and the viscous tensor provide an energy transfer channel between ion thermal energy and kinetic energy through the @xmath97 and the @xmath98 terms . the underlying energy transfer mechanism is the randomization of ordered perpendicular fluid motion due to ion - ion collision . as we will show in the next section , both terms are important for the conservation of energy . the gyro - viscous part vanishes@xcite exactly @xmath99 , when the gyro - viscous tensor @xmath100 is evaluated with the first order drift velocities @xmath101 . in this section explicit expressions for the polarization and gyro - viscous drifts are given . these drifts enter the _ gyro - viscous _ cancellation which eliminates the advection of vorticity by the diamagnetic drift in the polarization equation ( [ eq : df_w ] ) , but also the advection of ion parallel momentum and ion heat fluxes by the diamagnetic drift in their corresponding moment equations@xcite . these equations are significantly altered by the gyro - viscous tensor . the gyro - viscous cancellation is not complete in the sense that numerous small correction terms remains in the vorticity equation , in the parallel momentum equation , and in the heat flux equation . the actual derivation is cumbersome and the level of complication rises with the level of detail included in the polarization drift terms , e.g. , the inclusion of the polarization heat flux and anisotropic pressure@xcite . here , we aim at formulating a workable model for the use in numerical turbulence simulations . therefore , we do not include the polarization heat flux and we do not account for anisotropic pressures . furthermore , we bring the magnetic unit vector under the material derivative in the polarization drift @xmath102 and neglect the corresponding correction terms in the gyro - viscous drift . this approximation yields simpler expressions compared to the full expressions in e.g. refs . . as already discussed , we leave out parallel momentum for the purpose of exposition . after carrying out the gyro - viscous cancellation@xcite , the combined divergences of the ion polarization and gyro - viscous fluxes are : @xmath103 where we introduced the short - hand notation @xmath104 and where the material derivative is defined as @xmath105 . the advecting velocities are @xmath106 here the drift @xmath107 represents the remainder of the gyro - viscous cancellation . @xmath108 is derived from the scalar function @xmath109 given in eq . 37 in ref . the scalar function @xmath109 enters the momentum equation as a correction to the scalar pressure and hence only contributes when the magnetic field is inhomogeneous . the contributions originating from @xmath109 are therefore small . here , we choose to keep only the terms in @xmath109 necessary for energy conservation . furthermore , terms in @xmath109 originating from the purely perpendicular heat flux are neglected because energy conservation requires that these contributions are retained together with the polarization heat flux which is not kept here . as a rule of thumb@xcite all ion drifts retained in the ion continuity equation , except the ion diamagnetic drift , must be kept in the advection part of the polarization drift in order to conserve energy . unfortunately , this leads to a recursive definition of the polarization drift since the polarization drift includes a polarization drift advection term . if exact energy conservation is a crucial demand , this feature makes the model unsuitable for numerical calculations , but the model is convenient when discussing energy conservation in drift - fluid models in general . in this section we present the energy theorem for the drift - fluid model eqs . ( [ eq : df_n])-([eq : df_pi ] ) . without taking collisional effects into account it has previously been shown , e.g. , refs . , that drift - fluid models conserve energy . here , the drift - fluid energy theorem is presented including the perpendicular collisional transport terms and the collisional energy exchange terms described in section [ sec : moments ] . the global energy is obtained by multiplying the vorticity equation ( [ eq : df_w ] ) by @xmath110 and integrate over all space neglecting surface terms @xmath111\notag \\ = \int dv \ , & nm_i \frac{{\nabla_{\perp}}\phi}{b } \cdot d_t^\ # { \nabla_{\perp}}\phi^ * + p_i { \nabla \times \frac{{\bm{\hat{b}}}}{q_ib } } \cdot \nabla { \nabla_{\perp}}\phi^ * \cdot \frac{{\nabla_{\perp}}\phi}{b } -\bm{u}_e \cdot \nabla p + ( \nabla \cdot \bm{\pi}^{\perp}_i ) \cdot \bm{u}_e . \label{eq : energy_kin}\end{aligned}\ ] ] terms involving parallel dynamics are everywhere left out . detailed descriptions of energy conservation and energy transfer channels in the parallel direction in the absence of perpendicular collisional effects is thoroughly described in refs . and . next , the ion and electron pressure equations are integrated over space using eqs . ( [ eq : electron_heatflux_cancel ] ) , ( [ eq : pe_diff ] ) , and ( [ eq : pi_diff ] ) , again neglecting surface terms @xmath112 where @xmath113 and @xmath108 were inserted into the @xmath114 term , resulting in the second and third terms on the right hand side of eq . ( [ eq : energy_pi ] ) , respectively . the energy theorem @xmath115 is obtained by adding eqs . ( [ eq : energy_kin])-([eq : energy_pi ] ) and by bringing the ion particle density under the time derivative in terms involving the ion polarization drift using the particle density equation ( [ eq : df_n ] ) and the vorticity equation ( [ eq : df_w ] ) . as expected , the energy theorem is not altered by the inclusion of perpendicular collisional effects . in other words , the collisional effects do not give rise to sinks no sources in the energy theorem , but solely provide energy transfer channels . in that respect the energy theorem is a sanity check of the model and can be used as a verification tool in numerical simulations . the perpendicular kinetic energy @xmath116 is a peculiar quantity . the @xmath43-drift @xmath117 part is easy to interpret , because the fluid moments in eqs . ( [ eq : df_n])-([eq : df_pi ] ) are all advected by the @xmath43-drift , and hence this part describes the kinetic energy associated with the @xmath43-drift . on the contrary , the terms in the kinetic energy @xmath116 involving the diamagnetic drift are more difficult to interpret because the diamagnetic drift to lowest order does not advect fluid elements . the diamagnetic part of the kinetic energy is more appropriately described as a finite larmor radius ( flr ) correction to the ion pressure and to the ion @xmath43-kinetic energy@xcite . to emphasize this property of the kinetic energy , it is denoted the _ modified kinetic energy_. to elucidate how energy is transferred between the constituents of the energy theorem eq . ( [ eq : cons_energy ] ) it is instructive to consider the time evolution of each term separately @xmath118 terms which enter with opposite signs represent conservative energy transfer channels . the interchange transfer terms , first terms on the right hand sides of eqs . ( [ eq : ddt_ekin])-([eq : ddt_epe ] ) , transfer energy between ion and electron thermal energies and the modified kinetic energy . the compression of the polarization drift , second term on the right hand sides of eqs . ( [ eq : ddt_ekin ] ) and ( [ eq : ddt_epi ] ) , provides an energy transfer between ion thermal energy and the modified kinetic energy through the @xmath119 terms . this energy transfer channel may play an important role in generating and sustaining @xmath43 mean flows@xcite . the heat exchange term @xmath120 , entering eqs . ( [ eq : ddt_epe ] ) and ( [ eq : ddt_epi ] ) with opposite signs , describes energy exchange between electrons and ions due to elastic electron - ion collisions . resistivity provides an additional energy transfer channel through the resistive drift transfer terms @xmath121 entering eqs . ( [ eq : ddt_epe ] ) and ( [ eq : ddt_epi ] ) with opposite signs . the parallel resistivity , on the contrary , gives rise to an energy exchange between the ion and electron parallel kinetic energies @xmath122 and the electron thermal energy . here , resistivity opposes the motion of the charge carrying particles along the magnetic field . in the perpendicular direction , resistivity opposes the diamagnetic current , which to lowest order describes particle gyration around magnetic field lines . since the kinetic energy of the gyrating motion is included in the thermal energies , perpendicular resistivity in the drift - fluid model accordingly give rise to a conservative energy between the ion and electron thermal energies . note that a finite resistive energy transfer requires a finite ion pressure gradient . when @xmath123 , ions are homogeneously distributed and therefore the probability of an electron colliding with a gyrating ion is the same in all directions . only when @xmath124 , electrons do on average experience an average effect of the gyrating ions . the direction of the energy transfer can be deduced by writing out the transfer term in the electron pressure equation explicitly @xmath125 \cdot { \nabla_{\perp}}p_i , \label{eq : r_e_trans}\end{aligned}\ ] ] where the thermal force and the electron temperature gradient dependent part of the friction force were combined . the second and third terms on the right hand side heat the electrons whenever @xmath126 . the heating of electrons due to friction is known as joule heating . however , it can be shown that the second and third terms are also capable of cooling the electrons . necessary conditions are @xmath127 and that the ion temperature gradient length scale exceeds the particle density gradient length scale . the electron temperature gradient dependent term transfers thermal energy from the electron to the ions when @xmath128 , an effect known as the ettingshausen effect@xcite . it is important to keep in mind that both the heat exchange term @xmath120 and the resistive transfer terms occur on the slow time - scale @xmath129 . comparing the magnitudes of these energy exchange terms @xmath130 shows that unless the temperatures are very different , the resistive energy transfer is smaller than the heat exchange when considering profiles , but also that the resistive can be important when pressure gradients are steep , e.g. , in the edge region and in general when turbulent transport is strong . viscosity on the perpendicular flows due to ion - ion collisions gives rise to the energy transfer term @xmath131 between the modified kinetic energy and the ion thermal energy . we should expect that the energy transfer goes from kinetic energy to thermal energy because collisions increase entropy . if the perpendicular electric field vanishes the energy transfer term vanishes accordingly . it can be shown that the @xmath132 dependent part of @xmath133 entering the transfer term gives rise to an unidirectional energy transfer from modified kinetic energy to the ion thermal energy . the direction of the energy transfer by the remaining part , which depends on the ion diamagnetic drift , is not unidirectional . however , recall that the modified kinetic energy does not solely describe kinetic energy . therefore , in order to investigate the direction of the energy transfer one must instead consider the energy transfer from the ( flr ) corrected ion @xmath43 kinetic energy . disentangling the true kinetic energy from the modified kinetic energy is beyond the scope of this work . turbulent transport in the sol region is intermittent and is carried by coherent structures@xcite . these coherent structures of elevated pressure , often referred to as filaments or blobs , are elongated along the magnetic field . they are mainly born in the vicinity of the last closed flux surface ( lcfs ) at the outboard midplane due to ballooning@xcite . once these plasma outbreaks are expelled into the the open field line region , they expand@xcite into the rare sol plasma in the parallel direction with velocities comparable to the ion sound speed @xmath134 , while traversing the sol region radially at velocities reaching a fraction of @xmath135 . typically , the blobs imply fluctuation levels which exceed unity . in this section we derive a reduced model for investigations of blob formation and transport across the sol region of magnetically confined toroidal plasmas . the model is named hesel@xcite and is the successor of the esel model@xcite . like the esel model , the hesel model considers the dynamics in a plane perpendicular to the magnetic field located at the outboard midplane . perpendicular collisional transport is described by partly linearized expressions derived from the more general results presented in the preceding section . parallel transport in the sol region is represented by damping terms . parallel drift - wave dynamics and magnetic perturbations are ignored . turbulent fluctuations are hence driven by the interchange instability due to an inhomogenious toroidal magnetic field . these reductions facilitate long but also computationally inexpensive computer simulations which can provide sufficient data for statistical analysis and comparison with experiment . compared to the esel model the hesel model has the following enhancements : 1 ) a dynamical description of ion pressure dynamics , 2 ) the inertial response to changes in the ion pressure is included , 3 ) sheath dissipation , and 4 ) an improved description of collisional processes , including inter - species energy exchange , based on the results presented in the previous section . the model domain is a plane perpendicular to the magnetic field at the outboard midplane . the magnetic field geometry is approximated by local slab coordinates @xmath31 . @xmath136 and @xmath137 coordinates denote the radial and the poloidal azimuthal positions in the domain and @xmath138 denotes the position along the magnetic field . the toroidal magnetic field magnitude is approximated as @xmath139 , where @xmath140 denotes the inverse aspect ratio , and @xmath141 and @xmath142 are the minor and major radii , respectively . the hesel model consists of two regions joined at the lcfs ( @xmath143 ) modeling the edge and sol regions . besides radial boundary conditions , these regions are only set apart by damping terms due to losses along open field lines which are only active in the sol region . in this section we present and discuss the simplifications of the drift - fluid model equations ( [ eq : df_n])-([eq : df_pi ] ) that lead to the hesel model . first we turn our attention to the vorticity equation ( [ eq : df_w ] ) . here we invoke an approximation commonly known as the thin - layer approximation . this approximation shares many features with the boussinesq approximation often used in the description of incompressible , buoyancy - driven flows in neutral fluid dynamics , but is fundamentally different since the mechanism driving the dynamics in plasma drift - fluid equations is finite compressibility of the plasma fluid drifts . this rather crude approximation neglects particle density variations in the polarization flux entering the vorticity equation ( [ eq : df_w ] ) , and hence assumes a constant inertia of all fluid parcels irrespective of the local particle density . the motivation for this approximation , which in particular is debatable when applied to edge plasma models , is the desire for computational efficiency and that energy conservation requires polarization drift advection in the polarization drift itself as discused in section [ sec : pola_gyr_canc ] . this recursive definition is very inconvenient in numerical computations . we note that gyrofluid models@xcite do not suffer from these difficulties . specifically , we approximate the ion polarization flux defined in eq . ( [ eq : gv_cancel ] ) as @xmath144)= -\nabla \cdot \bigg(\frac{n}{\omega_i } d_t^ \ # { \nabla_{\perp}}\phi^ * \bigg ) + \nabla \cdot ( n\bm{u}_{\chi } ) \simeq -\frac{n_0}{\omega_{i0 } } \nabla \cdot ( d_t^0 { \nabla_{\perp}}\phi^*\big ) , \label{eq : hesel_tl}\end{aligned}\ ] ] where we have introduced @xmath145 and where subscripts `` @xmath146 '' on quantities on the right hand side refer to constant characteristic reference values . the material derivative @xmath147 only includes the @xmath43-drift at constant magnetic field @xmath148 . dictated by energy conservation , all other advection terms in the original polarization flux are neglected . energy conservation requires that the divergence of the polarization drift in the ion pressure equation ( [ eq : df_pi ] ) is approximated in the same way @xmath149 in the final equations the divergence of the polarization drift does not enter the ion pressure equation , because the vorticity equation is substituted into the ion pressure equation . the divergence of the polarization flux @xmath150)$ ] is formally small and does not affect energy conservation . it is therefore neglected . we start by considering the particle density equation ( [ eq : df_n ] ) where the only perpendicular collisional term is the divergence of the resistive flux written out in eq . ( [ eq : resfluxii ] ) . as previously noted , the divergence of the resistive flux gives rise to particle density diffusion with a diffusion coefficient @xmath151 . naturally , the particle density and electron temperature fields in the edge and sol regions of a tokamak plasma will not be completely correlated . however , we expect that the profiles of both fields share features such as being flat in the sol region and having steep gradients near the separatrix . furthermore , measurements in the sol region at the outboard midplane show that @xmath62 and @xmath63 are approximately proportional in blobs@xcite in which case @xmath152 . therefore , we choose to fix the diffusion coefficient in eq . ( [ eq : resfluxii ] ) evaluating it using characteristic constant reference values @xmath66 and @xmath67 for the particle density and the electron temperature , respectively . recall that the thermal force was partly canceled leaving only the last term in eq . ( [ eq : resfluxii ] ) . this residual thermal force is expected to be partly canceled by the ion temperature gradient part whenever the ion and electron temperatures are correlated . therefore , we neglect the temperature gradient terms in the resistive flux eq . ( [ eq : resfluxii ] ) . the divergence of the perpendicular resistive flux in the particle density equation ( [ eq : df_n ] ) then reads @xmath153 where @xmath154 is the constant diffusion coefficient calculated using @xmath66 and @xmath67 , and @xmath155 is the reference ion to electron temperature ratio . usage of the same approximate resistive drift @xmath156 in both the electron and ion pressure equations ( [ eq : df_pe])-([eq : df_pi ] ) avoids undesired temperature dynamics such as temperature anti - diffusion . next we consider the electron pressure equation ( [ eq : df_pe ] ) . as was shown in eq . ( [ eq : pe_diff ] ) , the resistive fluxes and the heat fluxes partially cancel . apart from the energy exchange terms , the remaining terms were given in eq . ( [ eq : pe_diffii ] ) . these terms are approximated as @xmath157 \bigg ) \notag \\ & \simeq \nabla \cdot ( p_e \bm{u}_{r0 } ) - \nabla \cdot \bigg(d_e \frac{11}{12 } n{\nabla_{\perp}}t_e \bigg ] \bigg ) . \end{aligned}\ ] ] the resistive energy transfer terms entering the electron and ion pressure equations are approximated as @xmath158 in the ion pressure equation ( [ eq : df_pi ] ) the collisional transport has a different character . here , the perpendicular ion heat conduction dominate the divergence of the resistive flux . however , the resitive flux is retained in order to handle situations with flat ion temperature gradients properly , see eq . ( [ eq : ion_cond_vs_resi ] ) . as in the electron pressure equation the divergence of the ion pressure resistive flux is approximated as @xmath159 the perpendicular ion heat conduction eq . ( [ eq : ion_heatflux ] ) @xmath160 is approximated by taking the ion diffusion coefficient @xmath161 to be constant . in the electron and ion pressure equations ( [ eq : df_pe])-([eq : df_pi ] ) the heat exchange term defined in eq . ( [ eq : energy_equilibration ] ) is evaluated with a constant electron - ion collision frequency @xmath162 finally , we consider approximations of terms related to the perpendicular collisional ion viscosity tensor @xmath29 , given in equation ( [ eq : visc_perp_coll ] ) , which enters the vorticity and ion pressure equations . in the vorticity equation ( [ eq : df_w ] ) , @xmath29 gives rise to diffusion of vorticity via the divergence of the flux associated with the viscous drift @xmath163 . recall that in order to reduce the computational costs , the thin - layer approximation to the polarization flux was invoked as described in section [ sec : thinlayer ] . the thin - layer approximation implied that the first order perpendicular ion drifts entering the polarization drift itself were approximated as @xmath164 accordingly , only @xmath165 enters @xmath29 . expressed in local coordinates @xmath31 we get @xmath166 furthermore , the thin - layer approximation takes the particle density in the ion polarization flux to be constant . making the same approximation to the viscous drift flux , we get @xmath167 which demonstrates that the perpendicular viscous drift gives rise to diffusion of the modified vorticity @xmath168 . vorticity diffuses due to ion - ion collisions which randomize the ions and hence transfer energy from the modified kinetic energy , see eq . ( [ eq : ddt_ekin ] ) , to the ion thermal energy through the terms @xmath169 and @xmath170 entering the ion pressure equation ( [ eq : df_pi ] ) . energy conservation requires that the perpendicular drifts entering @xmath29 and @xmath163 are approximated according to equation ( [ eq : uperpzero ] ) : @xmath171,\\ p_i\nabla \cdot \bm{u}_{\pi^{\perp } } \simeq p_i\nabla \cdot \bm{u}^0_{\pi^{\perp}_i } & = p_i\frac{3}{10}\frac{d_i}{\omega_{i0}}\nabla_{\perp}^2 \nabla_{\perp}^2 \phi^*. \label{eq : visc_flux_comp_pi}\end{aligned}\ ] ] the @xmath172 term in the ion pressure equation is formally small , see eq . ( [ eq : viscpiflux_to_qi ] ) . this term does not change the energy theorem and is therefore neglected . in a toroidal plasma perpendicular collisional diffusion is enhanced by neoclassical transport . specifically , in a collisional plasma , @xmath173 , the transport along the magnetic field is diffusive ; here @xmath174 denotes the mean free path and @xmath41 is the parallel length scale . despite being interrupted by collisions , guiding - centers drifts , such as the grad - b drift , give rise to transport across magnetic flux surfaces . in a tokamak , the direction , inwards or outwards , depends on whether particles are above or below the outboard midplane . when the diffusive transport along the magnetic field and the perpendicular guiding center drifts are superimposed they give rise to an enhanced pfirsch - schlter@xcite perpendicular diffusion @xmath175 , where @xmath176 is the safety factor . neoclassical transport theory assumes that the plasma evolves on a time - scale much slower than the typical time - scale of turbulent transport . therefore , neoclassical estimates of perpendicular transport do not immediately apply to intermittent transport in the sol region . furthermore , existing literature is only concerned with neoclassical transport on closed field lines . here , we are interested in local transport on closed _ and _ open field - lines . however , it is evident that the physical mechanism driving pfirsch - schlter transport is also present on open field lines and that it will influence profiles as well as transients , but since a rigorous derivation of these effects is beyond the scope of this work , we resort to the existing estimates keeping in mind that these will only hold approximately . specifically , diffusion coefficients are multiplied by a common factor @xmath177 which then include variations of the inverse aspect ratio @xmath178 at the outboard midplane . since the safety factor becomes infinite at the lcfs and is not defined on open field lines we estimate the effect of a finite magnetic field line pitch by the safety factor at @xmath179 of the poloidal magnetic flux @xmath180 . a more detailed discussion is given in fundamenskii et al ref . . it is important to note that since the perpendicular diffusive particle flux is enhanced by neoclassical effects , so is the resistive energy transfer , second last term in eq . ( [ eq : ddt_epe ] ) , between the electron and ion thermal energies@xcite . the hesel model is 2d and therefore parallel losses on open field lines are parametrized . the parametrization is in line with the parametrization in esel model given in ref . , but it is augmented by parametrization of parallel ion pressure dynamics and the addition of a damping term representing sheath currents at material surfaces such as divertors and limiters . as already mentioned , the primary aim of the hesel model is to describe interchange driven turbulence transport at the outboard midplane . as a consequence of ballooning the turbulent plasma source mainly resides on the high field side at the outboard midplane . for this reason , we assume that all parametrized parallel terms in the hesel model act as sinks . blobs expand@xcite in the parallel direction with a velocity comparable to the ion sound speed @xmath181 . the expansion depletes the particle density in the model - domain located at the outboard midplane . under the assumption that both ends of the blob expand with the same velocity , we get the characteristic parallel particle density damping rate @xmath182 where @xmath183 denotes the mach number and @xmath184 is the parallel blob size . we assume that blobs are predominantly born on the high field side in a region centred around the outboard midplane with a @xmath185 poloidal extend@xcite , and hence approximate the filament blob size as @xmath186 in the sol region vorticity losses are due to two mechanisms : first , parallel advection of vorticity enters through the divergence of the ion polarization drift particle density flux @xmath187 . since vorticity is mainly transported and generated by blobs in the sol region , we assume that the parallel gradient length scales of vorticity and particle density are approximately the same . in that case , the damping rate parametrising parallel advection of vorticity is then taken to be identical to the particle density damping rate : @xmath188 secondly , the parallel current in the vorticity equation provides an additional path for current density generated by the diamagnetic current in blobs . the role played by the parallel current is roughly determined by the parallel conductivity , the magnetic topology particularly in the vicinity of the x - point , the sheath in front of material surfaces e.g. the divertor in most tokamaks , and the ratio of the characteristic time @xmath189 it takes a blob to move a distance in the radial direction comparable to its own size to the characteristic time @xmath190 of the parallel communication of current perturbations between the outboard midplane and the sheath . here , we consider blobs in the inertial regime where radial blob convection is barely inhibited by parallel currents . this regime is typically relevant for plasmas with a high particle density , a long connection length @xmath191 between the outboard midplane and the divertor , or a low conductivity e.g. , due to high neutral gas densities in the sol region@xcite . nevertheless , we assume that the divertor sheath influences the background profile of the electrostatic potential . an average in the parallel direction@xcite of the parallel current leaves us with the value of the current at the sheath entrance given by the bohm criterion@xcite . for this reason we only apply the sheath damping to the averaged fields @xmath192 , \label{eq : sheathdamping}\end{aligned}\ ] ] where @xmath193 is the bohm potential and the azimuthal average is defined as @xmath194 here , @xmath195 is the domain size in the @xmath137-direction and @xmath196 is an arbitrary function . it is straightforward to use other closures of the parallel current e.g. , the sheath - connected regime@xcite can be investigated by applying the sheath damping not only to the averaged fields but to the fluctuating parts as well . we now turn to the electron pressure equation ( [ eq : df_pe ] ) . here , the parallel heat fluxes given in eqs . ( [ eq : electron_heatflux_qu])-([eq : electron_heatflux_qt ] ) and the parallel advection term are parametrized . quasi - neutrality only allows small deviations between the ion and electron parallel velocities . consequently , the thermal heat flux @xmath197 given in eq . ( [ eq : electron_heatflux_qu ] ) is therefore small in comparison with the divergence of the parallel pressure flux and is therefore neglected in the model . the spitzer - hrm@xcite parallel electron heat conduction @xmath198 given eq . ( [ eq : electron_heatflux_qt ] ) , on the other hand , is not negligible . it has a strong electron temperature dependence @xmath199 , which attenuates parallel electron temperature gradients . keep in mind that the braginskii closure is only valid in the collisional regime . in particular @xmath198 is not well behaved as we approach the `` collisionless '' regime . this is readily seen when writing the heat conductivity in terms of the mean free path @xmath200 @xmath201 as a rule of thumb the braginskii heat conduction is therefore only valid as long as the electron collisionality is sufficiently high ( @xmath202 ) . at lower collisionalities the spitzer - hrm heat conduction exceeds the free - streaming heat flux@xcite . a remedy for this shortcoming is to introduce _ heat - flux - limiters _ , which in an ad - hoc manner limit the parallel heat flux to a fraction of the free - streaming value . here we restrict ourselves to collisional sol conditions where the spitzer - hrm heat conductivity is valid . conduction is then parametrized as @xmath203 where we introduced the spitzer - hrm damping rate @xmath204 in the parametrization of the electron heat conduction we take the connection length @xmath205 as the parallel gradient scale length and not the parallel filament size @xmath184 to avoid an unphysical damping of the electron temperature background . parallel advection of electron and ion pressure are approximated as a one - dimensional adiabatic expansion ( @xmath206 ) @xmath207 in the ion pressure equation parallel heat conduction is neglected . for the parameters used here ( ion collisionality @xmath208 and @xmath209 , see fig.14 in ref . ) , parallel ion heat transport is dominated by parallel advection . we are now ready to write down the full set of equations which constitute the hesel model . in order to highlight characteristic quantities and for convenience , the model equations are gyro - bohm normalized @xmath210 where @xmath211 is the characteristic ion gyro frequency , @xmath212 is the cold - ion hybrid thermal gyro - radius , @xmath66 , @xmath67 , and @xmath213 are characteristic reference values for the particle density and the temperatures . the full system of equations is : @xmath214 where the advective derivatives are defined as @xmath215 @xmath43-advection is written in terms of the anti - symmetric bracket @xmath216 the curvature operator is defined as @xmath217 and the normalized modified potential reads @xmath218 in the ion pressure equation , the divergences of the polarization and viscous drifts were eliminated by substitution of the vorticity equation . diamagnetic advection in the electron and ion pressure equations is no longer present as a consequence of the diamagnetic cancellation @xmath219 all terms related to diffusion and parallel damping are grouped on the right hand sides of eqs . ( [ eq : n_norm])-([eq : ionpressure_thinlayer ] ) and are given as @xmath220\\ \lambda_{p_i } = & \frac{5}{2}d_e ( 1+\tau ) \nabla \cdot ( t_i{\nabla_{\perp}}n ) -d_e ( 1+\tau){\nabla_{\perp}}\ln n \cdot { \nabla_{\perp}}p_i + 2d_i\nabla \cdot ( n{\nabla_{\perp}}t_i ) \notag \\ & \frac{3}{10}d_i \big [ ( \partial^2_{xx}\phi^ * - \partial^2_{yy}\phi^*)^2 + 4 ( \partial^2_{xy } \phi^*)^2 \big ] + \frac{3 m_e}{m_i}\nu_{ei0}(p_e- p_i ) + \sigma(x)\bigg ( p_i \mathcal{s } - \frac{9}{2}\frac{p_i}{\tau_n}\bigg ) , \end{aligned}\ ] ] where @xmath155 and @xmath221\end{aligned}\ ] ] is a `` smooth step function '' which defines the transition from closed to open field line regions in the model . the normalized diffusion coefficients are @xmath222 and the damping rates due to parallel advection and parallel spitzer - h " arm heat conduction are @xmath223 where @xmath224 is the mach number . the sheath damping term entering the vorticity and ion pressure equations is given as @xmath225.\end{aligned}\ ] ] the thin - layer approximation alters the conserved energy compared to conserved energy of the full drift - fluid model described in sec . [ sec : energyconservation ] . the derivation paths are very similar . first , the pressure equations ( [ eq : electronpressure ] ) and ( [ eq : ionpressure_thinlayer ] ) are integrated neglecting surface terms . next , the vorticity equation ( [ eq : w_norm ] ) is multiplied by `` @xmath226 '' ( [ eq : w_norm ] ) and integrated , again neglecting surface terms . adding the results we obtain the energy theorem @xmath227 . = \int d\bm{x}\ , s_{\|}.\end{aligned}\ ] ] the energy sinks due to parallel losses are contained in @xmath228 as discussed in sec . [ sec : energyconservation ] the electron and ion energy equilibration terms @xmath120 , the frictional heat exchange terms @xmath121 , and the viscous terms exactly cancel and therefore do not enter the energy theorem . just as in the energy theorem for the full model given in eq . ( [ eq : cons_energy ] ) , collisions do not give rise to energy sinks and sources but merely give rise to conservative energy transfer terms in the model . a thorough discussion of energy transfer mechanisms and the impact on mean flows can be found in refs . . in order to test the simplified model for perpendicular collisional tranport applied in the hesel model , we have solved a subset of the governing equations ( [ eq : n_norm])-([eq : ionpressure_thinlayer ] ) numerically using the bout++ framework@xcite . specifically , we assume that : 1 ) the system is homogeneous in the y - direction , 2 ) only terms describing collisional transport are retained , and 3 ) parallel sinks are not active i.e. @xmath229 . in this particular limit , the hesel equations reduce to the 1d coupled set of non - linear equations @xmath230 in order to validate the simplified model , we have solved the full collisional model in the same limit . the full model is described in details in section [ sec : moments ] . in the absence of turbulence the model equations reduce to : @xmath231 \bigg ) + \bm{u}_r \cdot { \nabla_{\perp}}p_i&= -q_{\delta } , \\ \frac{3}{2}{\frac{\partial}{\partialt}}p_i + \nabla \cdot \bm{q}_{i\perp } + \frac{5}{2}\nabla \cdot ( p_i \bm{u}_r ) - \bm{u}_r \cdot { \nabla_{\perp}}p_i & = q_{\delta } \end{aligned } \right\ } \quad \text{full model ( fm ) } \label{eq : fm}\end{aligned}\ ] ] the resistive drift @xmath47 is given in eqs . ( [ eq : uru])-([eq : urt ] ) , the heat exchange term @xmath120 is given in eq . ( [ eq : energy_equilibration ] ) , and the perpendicular ion heat flux @xmath232 is given in eq . ( [ eq : ion_heatflux ] ) . we stress that no terms are linearised and that collision frequencies are evaluated as dynamical functions of temperatures and particle density . for comparison we have also solved simple diffusion equations with constant diffusion coefficients @xmath233 and @xmath234 as in the hesel model @xmath235 the fields are initialized as @xmath236 and the constant reference values used in the hesel and simple diffusion model equations were @xmath237 where @xmath238 denotes the proton mass . in figure [ fig : diff_solutions ] solving the full non - linear model ( fm ) eqs . ( [ eq : fm ] ) ( green ) , the hesel model given in eqs . ( [ eq : hesel ] ) ( blue ) , and a simple diffusion model given in eqs . ( [ eq : sd ] ) ( purple ) . solid black lines shows the initial condition @xmath239 given in eq . ( [ eq : initialcondition ] ) . only one third of the simulation domain is shown.,title="fig:",scaledwidth=100.0% ] [ fig : n ] solving the full non - linear model ( fm ) eqs . ( [ eq : fm ] ) ( green ) , the hesel model given in eqs . ( [ eq : hesel ] ) ( blue ) , and a simple diffusion model given in eqs . ( [ eq : sd ] ) ( purple ) . solid black lines shows the initial condition @xmath239 given in eq . ( [ eq : initialcondition ] ) . only one third of the simulation domain is shown.,title="fig:",scaledwidth=100.0% ] [ fig : pe ] solving the full non - linear model ( fm ) eqs . ( [ eq : fm ] ) ( green ) , the hesel model given in eqs . ( [ eq : hesel ] ) ( blue ) , and a simple diffusion model given in eqs . ( [ eq : sd ] ) ( purple ) . solid black lines shows the initial condition @xmath239 given in eq . ( [ eq : initialcondition ] ) . only one third of the simulation domain is shown.,title="fig:",scaledwidth=100.0% ] [ fig : te ] solving the full non - linear model ( fm ) eqs . ( [ eq : fm ] ) ( green ) , the hesel model given in eqs . ( [ eq : hesel ] ) ( blue ) , and a simple diffusion model given in eqs . ( [ eq : sd ] ) ( purple ) . solid black lines shows the initial condition @xmath239 given in eq . ( [ eq : initialcondition ] ) . only one third of the simulation domain is shown.,title="fig:",scaledwidth=100.0% ] [ fig : pi ] solving the full non - linear model ( fm ) eqs . ( [ eq : fm ] ) ( green ) , the hesel model given in eqs . ( [ eq : hesel ] ) ( blue ) , and a simple diffusion model given in eqs . ( [ eq : sd ] ) ( purple ) . solid black lines shows the initial condition @xmath239 given in eq . ( [ eq : initialcondition ] ) . only one third of the simulation domain is shown.,title="fig:",scaledwidth=100.0% ] [ fig : ti ] we show solutions of all three set of equations at @xmath240 for @xmath241 . @xmath242 is the time at which the particle density amplitude is half the initial value according to the analytical solution of the 1d diffusion equation . we have investigated various initial conditions but have chosen this specific case where the initial temperatures are not identical in order to demonstrate the energy exchange mechanisms within the model . the size @xmath243 of the simulation domain is @xmath244 . in the simulations the time evolution of the particle density @xmath62 does not vary significantly between the three models . the sd and hesel models give almost identical results whereas the fm model shows a slightly flatter profile . for the electron and ion temperatures and pressures the fm model and the hesel models agree quite well . the sd model profiles , on the other hand , significantly differs from the profiles of the hesel and fm models . the presence of energy exchange terms in the hesel and fm models is evident . here , the electron and ion temperatures are equilibrated whereas the absence of an energy coupling in the sd model is evident , particularly in the ion temperature . in conclusion , the hesel model matches the full model quite well . the simulations also demonstrate that a simple diffusion model should be used with care . the implications of the choice of perpendicular collisional model in turbulence simulations is not investigated here but is left for future work . lastly , in the 1d simulations we have observed that solving the full model takes significantly longer time than solving the hesel model . a comparison between the computational requirements of the full model fm and hesel is likewise left for future work . in this paper we have demonstrated how perpendicular collisional effects can be incorporated in low - frequency drift - fluid turbulence models . based on the electrostatic limit of the braginskii fluid equations@xcite , it was shown how resistivity and viscosity give rise to perpendicular fluid drifts which when inserted into the fluid moment equations provide an energy conserving closed model governing the time - evolution of the particle density , vorticity , and electron and ion pressure . we demonstrated that resistivity gives rise to energy exchange between the electron and ion thermal energies and that viscous effects due to ion - ion collisions transfer energy between the modified kinetic and the ion thermal energies . it was also shown that collisions do not imply energy sinks and sources in the global energy theorem , but merely provide conservative energy transfer terms . in conclusion , it was shown that the inclusion of perpendicular collisional transport in drift - fluid models gives rise to perpendicular diffusion in all moment equations , which obviates the need for explicit artificial diffusion . based on the general results , we also derived a simplified 2d model aiming at describing interchange driven turbulent transport in the vicinity of the outboard midplane . the model is named hesel and , as the predecessor esel , it is aimed at being computational efficient allowing the model equations to be time integrated for multiple turbulence de - correlation times . the validity of the simplified collisional dynamics in hesel was investigated by means of numerical simulations without turbulence . the simulations showed good agreement between hesel and the full model . this work has been carried out within the framework of the eurofusion consortium and has received funding from the euratom research and training programme 2014 - 2018 under grant agreement no . the views and opinions expressed herein do not necessarily reflect those of the european commission . gunn , c. boucher , m. dionne , i. uran , v. fuchs , t. loarer , i. nanobashvili , r. pnek , j .- y . pascal , f. saint - laurent , j. stckel , t. van rompuy , r. zagrski , j. admek , j. bucalossi , r. dejarnac , p. devynck , p. hertout , m. hron , g. lebrun , p. moreau , f. rimini , a. sarkissian , and g. van oost , , * 363 * , 484 ( 2007 ) .

from the early studies of brownian motion , more than a hundred years ago @xcite , the use of stochastic differential equations to model a wide variety of dynamical systems has grown dramatically . applications can be found along a wide area of scientific research , from physics and chemistry @xcite , through biology and ecology @xcite , to economy and social sciences @xcite . in the original formulation of brownian motion , the influence of the medium on a diffusive particle is modeled by splitting its effect in two parts : a deterministic one , given by an homogeneous viscous force , and a stochastic part , given by a random force with zero expectation value . in this way , fluctuations exhibit as an _ additive _ noise and , consequently , the considered model represents an _ additive stochastic process _ described by a langevin equation . however , the viscous force could have non - homogeneous contributions , for instance , in the presence of boundary conditions , such as a diffusion of a brownian particle near a wall @xcite . if the diffusion is not homogeneous , fluctuations could depend on the state of the system and they can be regarded as the product of a random force and a function of the state variable . in that situation , _ multiplicative _ noise is defined and the model is known as a _ multiplicative stochastic process_. another interesting example of multiplicative noise is the stochastic landau - lifshitz - gilbert equation @xcite , used to describe dynamics of classical magnetic moments of individual magnetic nano - particles . in this case , the noisy fluctuations of the magnetic field couple the magnetic moment in a multiplicative way . the classification of additive or multiplicative noise should be considered in the more general context of external and internal noise @xcite . in the former case , dissipation and noise can be considered as two effects with different microscopic origin . in other words , in the absence of noise , the classical deterministic system is perfectly well defined . however , in the latter case , dissipation and fluctuation have the same intrinsic origin and it is not possible to `` turn off '' one of these effects . the theory of stochastic evolution provides a beautiful connection between dynamics and statistical physics . in general , einstein relation or , more generally , fluctuation - dissipation relations associate dynamical properties of the system with thermodynamical equilibrium . however , stochastic dynamics not necessarily model physical systems , where the long time evolution should conduce to thermodynamical equilibrium . in fact , it is possible to have more general stationary state distributions that represent an equilibrium state in the stochastic dynamical sense , not related to thermodynamics . moreover , stochastic dynamics provide an interesting approach to out - of - equilibrium statistical mechanics @xcite . interestingly , it is possible to attribute thermodynamical concepts like heat , entropy or free energy to each trajectory of an stochastic evolution , given rise to the research field usually called stochastic thermodynamics @xcite . in this work , we would like to present a study of equilibrium properties of markovian multiplicative white - noise processes . a prototype of these processes is represented by the langevin equation ( [ eq.langevin ] ) . it is well known that , due to the gaussian white - noise distribution , the continuum limit of the discretized time evolution of multiplicative processes is not unique . in fact , there are different prescriptions to perform this limit . perhaps , the most popular prescriptions are the it @xcite and the stratonovich @xcite ones , each one producing a different stochastic evolution and forcing different rules of calculus . we will discuss in detail equilibrium properties in the more general prescription called generalized stratonovich convention @xcite ( also called `` @xmath0-convention '' @xcite in the field theory literature ) . in this prescription , a continuum parameter @xmath1 is defined in such a way that each value of the parameter corresponds to a different discretization rule of the stochastic differential equation . the value @xmath2 corresponds with the it prescription while @xmath3 is the stratonovich one . the concept of equilibrium is tightly related with the concept of time reversal symmetry . for this reason , it is essential to carefully define the time reversal transformation of the stochastic process . in multiplicative white - noise processes , the forward and backward stochastic trajectories evolve with different prescriptions ( except when considering stratonovich convention ) . for instance , if it convention ( @xmath2 ) is defined for the forward evolution , the backward trajectory evolves with the so called hnggi - klimontovich @xcite prescription ( @xmath4 ) . in general , we will show that the prescription @xmath0 is the time reversal conjugate of @xmath5 . in addition , since the asymptotic stationary probability distribution also depends on the chosen prescription , the correct definition of time reversal transformation , compatible with a unique equilibrium distribution , is quite involved . it is very useful to use , instead of the langevin or fokker - planck approach , a path integral formalism @xcite , in which the central object of the theory is a set of stochastic trajectories . using this formalism , we will analyze equilibrium properties , such as detailed balance relations , microscopic reversibility and entropy production . we will show that , due to a careful definition of equilibrium distribution and taken into account the appropriate time reversal transformation , usual equilibrium properties are satisfied for _ any value of @xmath0_. even thought the path integral representation of the stochastic process is suitable for the computation of correlation functions and responses , specific calculations are very cumbersome , since each value of @xmath0 defines different differentiation and integration rules . in particular , the `` chain rule '' to compute total derivatives of a stochastic process depends on @xmath0 . we make a generalization of the it formula for any value of the discretization parameter . in this work , we explicitly show the importance of this `` generalized chain rule '' , in the path integral formalism . the stochastic process can also be represented as a path integral in an extended functional space , by introducing auxiliary commutative as well as anti - commutative grassman variables . the main advantage of this extension for multiplicative noisy systems resides in the prescription independent character of the formulation . while in the original path integral formulation the prescription appears as a continuum limit ambiguity , in the functional grassman formalism , the ambiguity appears in the definition of equal - time grassman green functions @xcite . in this way , provided we do not integrate the grassman variables , we can perform any calculation without specifying a particular prescription . in some sense , all the difficulty introduced by the generalized it calculus is being codified in a simpler grassman algebra . as a by - product , it is possible to study hidden symmetries of the stochastic process . indeed , the class of systems we have studied in this paper is invariant under linear transformations in the extended functional space . this hidden symmetry , called supersymmetry ( susy ) , was recognized in other stochastic processes several years ago @xcite . susy properties have been extensively studied for additive stochastic processes @xcite and for non - markovian multiplicative processes @xcite . in another hand , for markovian multiplicative white - noise systems , important progresses have recently been reported @xcite . the main difficulty in the susy formulation of a markov multiplicative process is related with the great variety of prescriptions available to define the wiener integral , which produces several stochastic evolutions with different final steady states . moreover , time reversal transformations mix different prescriptions . in this paper , we deduce a covariant supersymmetric formulation of a multiplicative markov process showing in detail the importance of the prescription dependent equilibrium distribution and the correct definition of time reversal transformation . physically , susy encodes equilibrium properties of the system . some of the constraints it imposes on correlation functions ( ward - takahashi identities ) are related to fluctuation - dissipation theorems @xcite . this property has acquired a refreshing interest due to the growing importance of stochastic out - of - equilibrium systems @xcite . in this sense , it is possible to understand out - of - equilibrium dynamics as a symmetry breaking mechanism . we analyze in detail ward - takahashi identities specially related with fluctuation - dissipation theorems . we show that this theorem is fulfilled independently of the chosen prescription to define the process . however , the linear response , fluctuations , as well as the equilibrium distribution , do depend on the specific prescription . the paper is organized as follows : in sect . [ model ] we define our model by means of a langevin equation with multiplicative noise . in sect . [ timereversal ] we carefully study the problem of time reversal symmetry and the equilibrium probability distribution . with these results , we analyze properties such as detailed balance and entropy production in sect . [ detailedbalance ] . in sect . [ susy ] , we present a covariant supersymmetric formulation of the stochastic process and we deduce the fluctuation - dissipation theorem for arbitrary prescriptions . finally , we discuss our conclusions in sect . [ discussion ] , leading for [ appendix ] some details of the calculations . the main purpose of this section is to define the model and to establish the concepts and the notation we use in the rest of the paper . for simplicity , we consider a single random variable @xmath6 satisfying a first order differential equation given by @xmath7 where @xmath8 is a gaussian white noise , @xmath9 the drift term @xmath10 and the square root of the diffusion function @xmath11 are , in principle , arbitrary smooth functions of @xmath6 . the only restrictive condition is that @xmath11 should be `` invertible '' , _ i. e. _ , @xmath12 . as it is very well known , to completely define equation ( [ eq.langevin ] ) , it is necessary to give sense to the ill - defined product @xmath13 , since @xmath8 is delta correlated . the problem can be easily understood looking at the integral @xmath14 where we have defined the wiener process @xmath15 as @xmath16 . by definition , the riemann - stieltjes integral is @xmath17 where @xmath18 is taken in the interval @xmath19 $ ] and the limit is taken in the sense of _ mean - square limit _ @xcite . for a smooth measure @xmath15 , the limit converges to a unique value , regardless the value of @xmath18 . however , @xmath15 is not smooth , in fact , it is nowhere integrable . in any interval , white noise fluctuates an infinite number of times with infinite variance . therefore , the value of the integral depends on the prescription for the choice of @xmath18 . there are several prescriptions to define this integral that can be summarized in the so called `` generalized stratonovich prescription '' @xcite or `` @xmath0-prescription '' @xcite , for which we choose @xmath20 in this way , @xmath2 corresponds with the pre - point it interpretation and @xmath3 coincides with the ( mid - point ) stratonovich one . moreover , the post - point prescription , @xmath4 , is also known as the kinetic or the hnggi - klimontovich interpretation @xcite . in principle , each particular choice of @xmath0 fixes a different stochastic evolution . in many physical applications , a weakly colored gaussian - markov noise with a finite variance @xcite is considered . in this case , there is no problem with the interpretation of equation ( [ eq.langevin ] ) and we can take the limit of infinite variance at the end of the calculations . this regularization procedure is equivalent to the stratonovich interpretation , @xmath3 @xcite . however , in other applications , like chemical langevin equations @xcite or econometric problems @xcite , the noise can be considered principally white , since it could be a reduction of jump - like or poisson like processes . in such cases , the it interpretation ( @xmath2 ) should be more suitable . hence , the interpretation of equation ( [ eq.langevin ] ) depends on the physics behind a particular application . once the interpretation is fixed , the stochastic dynamics is unambiguously defined . summarizing , in order to completely define the stochastic process described by equation ( [ eq.langevin ] ) we need to fix a couple of functions @xmath21 and the parameter @xmath1 . the @xmath0-prescription introduces some difficulties in the proper definition of time reversal evolution that we describe in this section . using equation ( [ eq.langevin ] ) and the @xmath0-prescription , it is immediate to obtain @xcite a fokker - planck equation for the probability distribution @xmath22 : @xmath23 where @xmath24 . this equation can be cast into a continuity equation , @xmath25 whith the probability current given by @xmath26 p(x , t)- \frac{1}{2}g^2(x ) \frac{\partial p(x , t)}{\partial x}. \label{eq.j(x , t)}\ ] ] then , the probability distribution @xmath22 is given by the solution of the fokker - planck equation ( [ eq.fp ] ) supplemented with the initial condition @xmath27 and two boundary conditions . we can fix , for instance , the values of @xmath28 at two boundary points @xmath29 , where eventually they can be taken as @xmath30 . however , due to the fact that @xmath22 is normalized to one , for any value of @xmath31 , then @xmath32 . we suppose that , at long times , the probability rapidly converges to a steady state @xmath33 , given by @xmath34 with the normalization constant @xmath35 . in this state , the stationary current gets the form @xmath36 and the stationary fokker - planck equation acquires the simpler form @xmath37 or , simply , @xmath38 . thus , for a given value of @xmath39 , we could have a stationary state characterized by the function @xmath40 . the stationary probability density @xmath41 does not represent , in general , an equilibrium state . an equilibrium state is characterized by a net zero stationary current ( @xmath42 ) . when @xmath43 , although the current is already conserved , there is a stationary probability flux , characterizing an out - of - equilibrium regime . in most physical applications , the diffusion function @xmath11 is a polynomially growing function for large @xmath44 . in these cases ( and for a single variable ) , the only possible solution is the equilibrium one @xmath45 . however , for exponentially growing functions , out - of - equilibrium steady state solutions are possible . let us focus on the equilibrium state , defined as the solution of the stationary fokker - planck equation with zero current probability . thus , for @xmath42 , there is an obvious solution of equations ( [ eq.js ] ) and ( [ eq.stationaryfp ] ) , given by @xmath46 therefore , given the functions @xmath10 and @xmath11 , supplemented by a value of @xmath47 , the probability density tends at long times to the equilibrium distribution @xmath48 where @xmath49 is given by equation ( [ eq.equilibriumpotential ] ) , provided @xmath50 is integrable to define the normalization constant @xmath51 . in the case that equation ( [ eq.langevin ] ) models a conservative physical system , _ i. e. _ , if the otherwise deterministic system is characterized by an energy potential @xmath52 , the drift function @xmath10 does not depend explicitly on time and is given by @xmath53 this equation relates fluctuations ( given by @xmath11 ) with dissipation , therefore it can be considered as a local generalization of einstein relation for brownian motion . using this expression , the equilibrium `` potential '' has the form : @xmath54 as expected , the equilibrium distribution depends , not only on the given functions @xmath55 or , alternatively , on @xmath56 , but also on the value of the @xmath0-prescription which defines the wiener integral . if equation ( [ eq.langevin ] ) is used to model a physical system , which is expected to asymptotically converge to thermodynamic equilibrium ( @xmath57 ) , the only possible choice is the hnggi - klimontovich interpretation @xmath4 ; for this reason , the election of @xmath4 is sometimes called `` thermal prescription '' . recently , an experimental evaluation of @xmath0 was reported @xcite in the study of a falling colloidal particle near a wall . in this work , the value of @xmath4 was recognized as the proper one to make the correctly identification of the microscopic forces acting on the colloidal particle . with any other choice , the equilibrium potential @xmath58 , including the usual it ( @xmath2 ) and stratonovich ( @xmath3 ) prescriptions . except for the stratonovich prescription @xmath3 , each value of @xmath0 is associated with different calculus rules . for this reason , in most physical literature , stratonovich prescription is preferred , because it enables `` normal '' rules of derivation and integration . despite the hnggi - klimontovich convention being the only one which guarantees the convergence to the termodynamic equilibrium , calculations can be particularly cumbersome in this interpretation . however , if we prefer to work in any other different fixed prescription , and we insist in modeling a physical system which converges to boltzmann s equilibrium , we need to modify equation ( [ eq.langevin ] ) . it is done by adding a `` spurious drift force '' @xcite which cancels the last term of equation ( [ eq.u ] ) . that is , if we define a new @xmath0-dependent `` potential '' @xmath59 , then , the equilibrium distribution @xmath60 , for any value of @xmath0 . in this work , we will not adopt this point of view . we will consider a stochastic process completely defined by fixing @xmath61 in eq . ( [ eq.langevin ] ) . thus , the probability distribution @xmath22 flows , at long times , to an @xmath0-dependent equilibrium distribution given by equation ( [ eq.equilibriumpotential ] ) or alternatively equation ( [ eq.u ] ) . in this way , we are able to study , in a unique formalism , model systems with general equilibrium distributions that go from boltzmann thermal equilibrium to power - law distributions . a simple example of the latter case is to consider a pure noisy system with @xmath62 . in that case , the equilibrium distribution is @xmath63 . we will show that this set of distributions , characterized by asymptotically zero current probability , is consistent with all the usual properties of equilibrium evolution , like , for instance , detailed balance and entropy production . consider , for instance , the langevin equation ( [ eq.langevin ] ) , in the simplest case of @xmath64 . naively , we could compute the backward evolution of the stochastic variable @xmath6 by just changing the sign on the velocity @xmath65 , in such a way that , for a given noise configuration , the particle turns back on its own feet . however , the situation is not so simple . let us look more closely . consider a time interval @xmath66 . the forward evolution is obtained by integrating equation ( [ eq.langevin ] ) between the initial and final times @xmath31 and @xmath67 respectively , @xmath68 while the backward evolution is simply obtained by changing the initial and final integration limits , @xmath69 where we use the notation @xmath70 , just to differentiate backward from forward evolution . if the integrals were `` normal '' integrals , we could use the trivial property @xmath71 . in that case , there would be no need to differentiate backward and forward variables , since equations ( [ eq.fwevolution ] ) and ( [ eq.bwevolution ] ) would be the same equation and @xmath72 . however , the integrals are wiener integrals and need to be carefully defined . as we saw in the last section , @xmath73 with @xmath74 the time reversal integral is obtained by changing @xmath75 in equation ( [ eq.wienerfw ] ) . the important point is that @xmath76 also depends on @xmath77 . then , @xmath78 with @xmath79 where @xmath80 was obtained from equation ( [ eq.prescriptionfw ] ) , by replacing @xmath75 or , equivalently , @xmath81 . therefore , the time reversed stochastic evolution is characterized by the transformations @xmath82 in this sense , we say that the prescription @xmath5 is the time reversal conjugate ( trc ) of @xmath0 . then , the post - point hnggi - klimontovich interpretation is the trc of the pre - point it one , and vice versa . the only _ time reversal invariant prescription _ is the stratonovich one , @xmath3 . this means that , except for the stratonovich case , the backward and forward stochastic paths do not have the same end points . this is illustrated in figure ( [ timereversal ] ) , where we compute a `` time - loop '' evolution . consider we want to compute the evolution of the system starting at @xmath6 , going forward a time interval @xmath83 and , then , turning back in time the same interval @xmath84 . the forward path in figure ( [ timereversal ] ) can be computed from equation ( [ eq.fwevolution ] ) considering @xmath6 as an initial condition , @xmath85 then , the backward path is computed from equation ( [ eq.bwevolution ] ) considering @xmath86 as an initial condition @xmath87 replacing equation ( [ eq.path1 ] ) into ( [ eq.path2 ] ) , @xmath88 and using equations ( [ eq.wienerfw ] ) and ( [ eq.wienerbw ] ) we find @xmath89\left(w(t_{j+1})-w(t_j)\right ) . \label{eq.loop}\ ] ] we see that , in general , @xmath90 since the backward and forward evolutions develop with different dual prescriptions . in the stratonovich case , @xmath3 , @xmath91 and @xmath92 . ) . the forward evolution @xmath93 is performed with a prescription @xmath0 , while the backward trajectory @xmath94 evolves with the time reversal conjugate prescription @xmath5 . except for the stratonovich convention @xmath3 , @xmath95.,height=245 ] this fact is reflected in the @xmath0-dependence of the equilibrium stationary state @xmath96 ( equation ( [ eq.equilibriumpotential ] ) or equation ( [ eq.u ] ) ) . while the stationary state in the forward evolution is @xmath97 , the stationary state reached by the backward one is given by @xmath98 , and ( except for the stratonovich prescription ) @xmath99 . thus , this is not the backward evolution we are interested in . the key point is to determine the backward stochastic process which , at long times , converges to the same equilibrium distribution as the forward one . to obtain it we write the time reversed fokker - planck equation as @xmath100 where we made the substitutions @xmath101 , @xmath81 and @xmath102 in equation ( [ eq.fp ] ) . in equation ( [ eq.fpbw ] ) , @xmath103 represents the backward probability distribution of the event @xmath104 . now , we need to specify the drift @xmath105 which produces a backward evolution that asymptotically converges to the equilibrium distribution @xmath106 , with @xmath96 given by equation ( [ eq.u ] ) . setting @xmath107 and @xmath108 , we solve equation ( [ eq.fpbw ] ) for @xmath109 , obtaining @xmath110 therefore , the time reversal transformation @xmath111 which makes physical sense , meaning that it produces a backward evolution which converges to a unique equilibrium distribution , is given by @xmath112 notice that , since @xmath113 is odd under the transformation @xmath114 , the time reversal operator defined in this way satisfies @xmath115 , as it should be . as we have previously stressed , in the stratonovich prescription @xmath116 and @xmath117 , so , the time reversal operator @xmath111 simply corresponds to change @xmath101 . in any other prescription the definition of @xmath111 is more involved , given by equation ( [ eq.t ] ) . although we have defined the time reversal transformation to match the correct equilibrium distribution in the long time limit @xmath118 , this transformation also gives the correct answer in the stationary out - of - equilibrium case @xmath43 . in this case , the stationary forward and backward probabilities are different , and they are related by @xmath119 . that is , @xmath120 as can be easily seen by applying the transformation ( [ eq.t ] ) to the fokker - planck equation ( [ eq.continuity ] ) and ( [ eq.j(x , t ) ] ) . in this section we check that the concept of @xmath0-dependent equilibrium and the time reversal transformation introduced in the previous sections are consistent with the usual concepts of detailed balance , microscopic reversibility and entropy production . to do this , we use the path integral formalism , which turns out to be very convenient for handling stochastic trajectories . the transition probability @xmath121 represents the conditional probability of finding the system in the state @xmath122 at time @xmath123 provided the system was in the state @xmath124 at time @xmath125 . in the path integral approach , it can be written as @xcite @xmath126 } \ ; , \label{eq.pfw}\ ] ] with the boundary conditions @xmath127 and @xmath128 and the `` action '' @xmath129 $ ] given by @xmath130 ^ 2+\alpha f ' \right\rbrace . \label{eq.sfw}\ ] ] the probability for the time reversal process @xmath131 is obtained by applying the time reversal transformation @xmath111 of equation ( [ eq.t ] ) to the forward probability , equations ( [ eq.pfw ] ) and ( [ eq.sfw ] ) , @xmath132 } \ ; , \label{eq.pbw}\ ] ] with the boundary conditions @xmath133 and @xmath134 and the time reversed `` action '' @xmath135={\cal t } s[x]$ ] . in order to compute @xmath136 , it is interesting to write @xmath129 $ ] in another form to make explicit its properties under the time reversal transformation . expanding the integrand of equation ( [ eq.sfw ] ) and considering a conservative system @xmath137 ( equation ( [ eq.conservative ] ) ) we find @xmath138}{dx } \ ; \frac{dx}{dt } \right\rbrace . \label{eq.sfwv}\end{aligned}\ ] ] the last term can be transformed in a total time derivative . however , for arbitrary values of @xmath0 , we need to apply the proper chain rule , since usual calculations rules are valid only for the stratonovich convention @xmath3 . in the general case , for an arbitrary differentiable function of a random variable @xmath139 , the chain rule reads @xmath140 clearly , for @xmath3 , equation ( [ eq.chainrule ] ) is the usual chain rule . for @xmath2 , this formula is known as the _ it formula_. for the hnggi - klimontovich prescription , @xmath4 , the latter differentiation rule differs from the it formula just in the sign of the last term . this is because both prescriptions are time reversal conjugate . in fact , the second derivative term is odd under @xmath81 , as it should be , since all terms in ( [ eq.chainrule ] ) should be odd under time reversal . applying equation ( [ eq.chainrule ] ) to the last term of equation ( [ eq.sfwv ] ) , we find @xmath141 where @xmath142 the most important property which results from equations ( [ eq.sst ] ) and ( [ eq.stilde ] ) is that @xmath143 is _ time reversal invariant _ , @xmath144 . therefore , by using equation ( [ eq.sst ] ) , the time reversed action @xmath136 can be easily determined : @xmath145 thus , for an arbitrary @xmath0 , the variation of the action @xmath129 $ ] under a time reversal transformation is just a total derivative term . using equations ( [ eq.sst ] ) and ( [ eq.sbw ] ) , it is immediate to show that @xmath146 with @xmath49 given by equation ( [ eq.u ] ) . one of the consequences of the time reversal operator ( [ eq.t ] ) is that the conditional probabilities @xmath147 . however , using equations ( [ eq.pfw ] ) , ( [ eq.pbw ] ) , ( [ eq.sst ] ) and ( [ eq.sbw ] ) , it is simple to show that @xmath148_{t_i}^{t_f}},\ ] ] satisfying the detailed balance relation @xmath149 with @xmath150 given by equations ( [ eq.peq ] ) and ( [ eq.u ] ) . so , we have shown that forward and backward conditional probabilities , with the definition of equilibrium state given by the effective potential of equation ( [ eq.u ] ) , satisfy the detailed balance relation , equation ( [ eq.detailedbalance ] ) , for arbitrary values of @xmath1 . in the special case of @xmath3 , @xmath151 , and equation ( [ eq.detailedbalance ] ) is the usual detailed balance relation . it is not difficult to analyze the previous results from the perspective of stochastic thermodynamics @xcite . the detailed balance relation ( [ eq.detailedbalance ] ) represents a connection between a transition probability and its time reversed one . moreover , the transition probability was obtained by integrating over all possible stochastic trajectories . however , there is also a relation between the probabilities associated with each individual stochastic trajectory , that is usually called microscopic reversibility @xcite . to show up this relation explicitly , we can assign to each stochastic trajectory @xmath6 , with end points @xmath152 and @xmath153 , a weight @xmath154 } \ ; , \label{eq.pxtfw}\ ] ] the heat dissipated into the environment @xmath155 , and thus , the increase of entropy in the medium @xmath156 associated with one specific trajectory , is given by @xcite @xmath157=\ln\left[\frac{{\cal p}(x(t)|x_it_i , x_ft_f)}{\hat { \cal p}(x(t)|x_it_f , x_ft_i)}\right ] . \label{eq.q}\ ] ] hence , for each stochastic trajectory beginning in a state with initial distribution @xmath158 and ending with a different distribution @xmath159 , the total entropy production is @xmath160 or , using equation ( [ eq.sts ] ) , @xmath161 thus , in the absence of an explicit time dependent driving force , the stochastic entropy is a state function which depends only on the initial and final states . moreover , if we prepare the initial state in equilibrium @xmath162 , we immediately conclude that @xmath163 for each stochastic trajectory and for arbitrary value of @xmath0 . the relevance of the potential @xmath164 in the entropy production was recently pointed out in @xcite for the stratonovich prescription . here , we generalize this concept to the general case of the @xmath0-prescription where the time reversed process is not trivial . it is interesting to note that a similar expression to ( [ eq.deltas ] ) , however quite cumbersome , could be obtained for a stationary out - of - equilibrium potential . the complication in this case is that the stationary potential depends on @xmath39 , and the action does not transform as easily as in equation ( [ eq.sts ] ) under time reversal , since @xmath165 . in order to push forward our analysis of equilibrium properties it is convenient to have a formalism to compute correlations and response functions . however , the action of equation ( [ eq.sst ] ) and the difficulties introduced by the generalize it calculus ( eq . ( [ eq.chainrule ] ) ) make the functional formalism in @xmath6 quite cumbersome . for this reason it is convenient to introduce auxiliary stochastic variables ( a function @xmath166 and two grassman functions @xmath167 ) extending in this way the functional space of stochastic functions . the immediate consequence is a formalism that does not depend on the explicit value of @xmath0 , simplifying its mathematical properties . as a by - product , hidden symmetries of the system can be put in evidence . these symmetries impose non - perturbative constraints on correlations functions and are related to equilibrium properties of the system like , for instance , the fluctuation - dissipation theorem . we are interested in the calculation of correlations of functions of the stochastic variable at equal or different times . for instance , we could want to compute @xmath168 , where @xmath169 is an arbitrary function of @xmath44 , or two point functions @xmath170 . in the path integral approach these mean values can be written as @xmath171 } \ ; , \label{eq.a}\ ] ] where @xmath129 $ ] is given by equation ( [ eq.sfw ] ) and @xmath172 . this equation takes a simpler form if we consider that the system is prepared in equilibrium at @xmath173 . in this case , @xmath174 . assuming an ergodic evolution , we consider that @xmath175 for an arbitrary value of @xmath176 . therefore , using equation ( [ eq.a ] ) and equation ( [ eq.ergodic ] ) we can write the mean value as @xmath177 } \ ; , \ ] ] where we have extended the time integration from @xmath178 to @xmath179 and , now , there are no more constraints in the functional integration . it is important to recall that this simpler expression is only possible because we sample the initial condition @xmath180 with the equilibrium distribution @xmath181 . provided this is satisfied , we can compute any mean value or correlation function from the generating functional @xmath182+\int_{-\infty}^{\infty}dt ' j(t ' ) x(t ' ) } \ ; , \label{eq.zs}\ ] ] where @xmath183 is a source with compact domain , that is , it adiabatically goes to zero away from an interval @xmath184 where we will compute the correlation functions . in this context , any total derivative term in @xmath129 $ ] does not contribute to the dynamics of any observable . a total derivative term only contributes to a constant prefactor of a correlation function . in this way , a system described by equation ( [ eq.zs ] ) is automatically invariant under time reversal since @xmath185 and @xmath186 differ just in a total time derivative term . on the other hand , if we impose the constraint @xmath187 , the action is truly time reversal invariant @xmath129=\hat s[x]=\tilde s$ ] . again , it is important to observe that this happens because we have prepared the system at equilibrium at some initial time @xmath125 and we are computing correlation functions at later times @xmath188 . for any other initial distribution we need to go back to equation ( [ eq.a ] ) to compute mean values . equation ( [ eq.zs ] ) can be used to study general correlation and response functions of a white - noise multiplicative stochastic process whose dynamics is driven by equation ( [ eq.langevin ] ) . the @xmath0-dependence of the final equilibrium distribution , the structure of @xmath143 and the fact that for general @xmath0 the usual rules of calculus do not apply , make this method cumbersome . nevertheless , we have shown @xcite that , introducing an auxiliary `` bosonic '' variable @xmath166 , and a couple of conjugate grassman variables @xmath189 , we can avoid this inconvenience by alternatively using the generating functional formalism in this extended space . the generating functional can be cast in the following form , @xmath190 = \int { \cal d}x { \cal d}\varphi { \cal d}\xi { \cal d}\bar{\xi } \ e^{-s[x,\varphi,\xi,\bar\xi ] + \int dt j(t)x(t ) } \ , \label{eq.funcionalgenerador}\ ] ] where the `` action '' @xmath185 is given by @xmath191\right\rbrace . \label{eq.sgrassman}\end{aligned}\ ] ] this action in the extended functional space @xmath192 is completely equivalent to the usual formalism presented in [ pathintegral ] . in fact , functional integrating over grassman variables and over the variable @xmath193 we get the generating functional ( [ eq.zs ] ) , with the action given by equation ( [ eq.sfw ] ) . the advantage of this extended formulation is that it does not depend explicitly of the @xmath0-parameter and then , the calculus rules are the usual ones for any @xmath0 . hence , all the complication of the stochastic calculus associated with the definition of the wiener integral is now codified in the anti - commuting grassman algebra , implying , for instance , that @xmath194 . of course , the parameter @xmath0 reappears when we need to properly define grassman retarded green s functions . in fact , equal - time correlations functions are ill - defined , forcing us to set @xmath195 . one of the advantages of this procedure is that , while the grassman variables are not integrated , we can work out any calculation without explicitly indicate an specific convention . this is particularly useful to study markovian white - noise processes , where the equilibrium distribution depends on @xmath0 . this situation is very different from non - markov processes @xcite , in which @xmath196 . the time reversal transformation , equation ( [ eq.t ] ) , has a simpler form in terms of grassman variables , @xmath197 where @xmath198 are the time reversed grassman variables . there is a subtlety in these transformations which involves the extremal values of the grassman variables . while @xmath199 , the time reversed variables satisfy @xmath200 . this fact has no importance at all when computing correlation functions . however , it is relevant to analyze entropy production and other relations that depend on total derivatives . the real advantage of the grassman representation is that almost any important symmetry becomes a linear transformation . it is simple to check that , in this representation , @xmath201 and @xmath202 then , the entropy production @xmath203 , associated with each trajectory , is given by @xmath204 we see that , in the extended space of trajectories @xmath205 , the entropy production associated with each trajectory is not zero . this is due to the apparently extra degrees of freedom introduced by the grassman variables . on the other hand , upon functional integration we have that @xmath206 due to fermionic number conservation and the total entropy production is zero , as it should be in an equilibrium evolution . this opens the interesting possibility of interpreting an entropy production as a spontaneous breaking of fermionic phase ( gauge ) invariance like , for instance , in superconducting phase transition . this effect can not happen in our example of one stochastic variable , but in multiple variables systems @xcite , there are quartic interactions between grassman variables opening this attractive possibility . it is interesting to compute the linear response of the system to an external perturbation . to see this , we slightly perturb the system out of equilibrium @xmath207 and compute the dynamic susceptibility @xmath208 introducing ( [ eq.perturb ] ) into ( [ eq.sgrassman ] ) and computing equation ( [ eq.chi ] ) we find @xmath209 in the case of additive noise , where @xmath210 , the response function has the simpler form , @xmath211 . for this reason , the auxiliary function @xmath166 is usually called the response variable . however , for multiplicative noise , the response is more complex since involves , not only correlations functions of @xmath11 , but also contributions from the grassman sector of the model . this is a direct consequence of multiplicative white - noise since a variation of the `` external '' potential is modifying the fluctuations properties of the system . this means that the susceptibility explicitly depends on @xmath0 . from equation ( [ eq.response ] ) , it is immediate to recognize a _ `` natural response variable '' _ @xmath212 in such a way that , @xmath213 we will show that @xmath214 has a key role in the supersymmetric formulation of the model . the path integral formalism is useful to make evident symmetries of the stochastic process . for instance , the action given by eq.([eq.sgrassman ] ) is invariant under the transformation , @xmath215 where @xmath216 is an anticommuting parameter . this nilpotent transformation ( @xmath217 ) is the famous brs @xcite symmetry , discovered in the context of quantization of gauge theories . in the present context , it simply enforces probability conservation , @xmath218 . there is another set of important symmetries related with equilibrium properties which is called supersymmetry . to display it explicitly , it is convenient to work with the natural response variable introduced in equation ( [ eq.naturalresponsevariable ] ) . therefore , we make in equations ( [ eq.funcionalgenerador ] ) and ( [ eq.sgrassman ] ) the funcional change of variables @xmath219 , defined by @xmath220 the last two transformations make the jacobian trivial , @xmath221 , since @xmath222 and the fermionic measure reads @xmath223 . the generating functional is then given by @xmath190 = \int { \cal d}x { \cal d}\tilde\varphi { \cal d}\eta { \cal d}\bar{\eta } \ e^{-s[x,\tilde\varphi,\eta,\bar\eta ] + \int dt j(t)x(t ) } \ , \label{eq.zeta}\ ] ] where the action @xmath224 $ ] reads @xmath225 + \frac{g'}{g^3}\left(\dot x - f\right)\bar\eta\eta\right\rbrace , \label{eq.seta}\end{aligned}\ ] ] where we have drop - off function s arguments just to simplify notation . at first sight , this action written in the space @xmath226 seems to be more complex than the preceding one . however , these new physical variables can be rearranged in order to explicitly display susy . first , it is convenient to write @xmath10 and @xmath11 in terms of two `` potentials '' , @xmath52 and @xmath227 , defined as follows @xmath228 then , we collect the variables @xmath229 , in the definition of the superfield @xmath230 where we have introduced two `` temporal '' grassman variables @xmath231 and @xmath232 . the generating functional , equation ( [ eq.seta ] ) , can be re - written in terms of @xmath233 as @xmath190 = \int { \cal d}\phi \ e^{-\tilde s[\phi ] + \int dt d\theta d\bar \theta\ ; j(t,\theta,\bar\theta)\phi(t,\theta,\bar\theta ) } \ , \label{eq.gfphi}\ ] ] where @xmath234 $ ] depends on the potentials @xmath235 and @xmath236 by @xmath237 = \int dtd\theta d\bar\theta\ ; l(\phi)=\int dtd\theta d\bar\theta\ ; \left\{\bar d\gamma[\phi ] d\gamma[\phi]+\frac{1}{2 } v[\phi]\right\}. \label{eq.ssusy}\ ] ] we have defined the covariant derivatives @xmath238 which satisfy @xmath239 and @xmath240 . the action ( [ eq.ssusy ] ) , when written in components form , is completely equivalent to equation ( [ eq.sgrassman ] ) or , after integration of the grassman variables , to equation ( [ eq.stilde ] ) , as it is shown in the appendix . it is now immediate to verify that ( [ eq.ssusy ] ) is invariant under transformations of the supersymmetry group @xcite , whose generators are @xmath241 the symplicity of the superfield representation resides in the fact that each susy transformation is a translation of the temporal variables @xmath242 . then , the invariance of the last term of equation ( [ eq.ssusy ] ) is trivial , since @xmath235 does not depend explicitly on @xmath242 . to demonstrate the invariance of the kinetic part of the action , it is useful to have in mind the graded algebra , @xmath243 thus , @xmath244 d\gamma[\phi]\right\}=\bar d\delta\gamma[\phi ] d\gamma[\phi]+\bar d\gamma[\phi ] d\delta\gamma[\phi].\ ] ] on the other hand , @xmath245=\epsilon q_\beta \gamma[\phi]$ ] , where @xmath246 is an arbitrary grassman parameter and @xmath247 may represent @xmath155 or @xmath248 defined in equation ( [ eq.generators ] ) . in this way , @xmath244 d\gamma[\phi]\right\}=\epsilon\left\{-\bar d\left ( q_\beta\gamma[\phi]\right ) d\gamma[\phi]+\bar d\gamma[\phi ] d\left(q_\beta\gamma[\phi]\right ) \right\}\ ] ] finally , using the algebra ( [ eq.algebra ] ) , we find , @xmath244 d\gamma[\phi]\right\}=\epsilon q_\beta\left\{\bar d\gamma[\phi ] d\gamma[\phi]\right\},\ ] ] which is a total derivative term , demonstrating in this way , the invariance of @xmath143 . for additive processes , the diffusive potential is linear , @xmath249 , and eq.([eq.ssusy ] ) reduces to the usual action defined with a single superpotential @xmath235 . in this case , the tadpole theorem @xcite guarantees that the stochastic evolution does not depend on @xmath0 . however , multiplicative processes ( non - linear @xmath250 ) induce derivative couplings in the superfield . these couplings are responsible for the @xmath0-dependent evolution which leads to the equilibrium distribution of equation ( [ eq.u ] ) . the study of the symmetries of the action enables us to deduce properties of equilibrium dynamics of the system . each of the three generators of susy , @xmath251 , imposes several non - perturbative constraints on correlation functions . to obtain these constraints , we will use the ward - takahashi identities @xmath252 where @xmath246 is an arbitrary parameter and @xmath253 is one of the generators of susy . note that , if @xmath253 is @xmath155 or @xmath248 , then @xmath246 is a grassman parameter , while , if @xmath254 , @xmath246 is a commutative parameter . the ward - takahashi identity related with the generator @xmath255 reads , @xmath256 implying time - translation invariance as @xmath257 then , any two - point correlation function depends on time differences @xmath258 . this is a necessary condition for an equilibrium evolution . on the other hand , the generator of @xmath232 translations , @xmath155 , induces the identity @xmath259 that , when written in components , imposes the relations @xmath260 all components , involving only one of the grassman variables , @xmath261 and @xmath262 , vanish because of fermionic number conservation . the first equation relates the physical linear response of the system with the grassman two - point correlation function . since the physical response is causal , _ i. e. _ , @xmath263 for @xmath264 , ( see equation ( [ eq.naturalresponse ] ) ) , we are forced to choose the retarded prescription in computing green s functions of grassman variables . on the other hand , the equal - time green function is not well defined , forcing us to adopt a prescription @xmath265 , where @xmath266 is an arbitrary initial value . this presciption , when written in the original variables , is @xmath195 , allowing us to identify @xmath0 with the needed prescription to define the wiener integral . notice that , in the case of non - markovian processes , @xmath196 , without any ambiguity . finally , the invariance of the action generated by @xmath248 , which arise from the identity @xmath267 results in the non - perturbative constraints , @xmath268 this constraints imply another important equilibrium property , the fluctuation - dissipation theorem , which relates the spontaneous fluctuations of the system with its response to an external perturbation . to explicitly see this , we substitute relation ( [ eq.linearresponse_eta ] ) into ( [ eq.wt-fdt ] ) , obtaining @xmath269 using the equilibrium time translations invariance of the correlation function @xmath270 , this identity becomes @xmath271 which , bearing causality in mind , can be rewritten as @xmath272 where @xmath273 is the heaviside step function . equation ( [ fdt ] ) is a traditional form of the fluctuation - dissipation theorem . a more commonly used form of equation ( [ fdt ] ) is given in terms of the fourier transform of the response function , @xmath274 then , the fluctuation - disipation theorem can be understood as a direct consequence of supersymmetry . we have shown that , even thought the correlation functions and the responses depend on @xmath0 in a multiplicative process , the fluctuation - dissipation theorem is satisfied for any value of @xmath0 , provided we correctly identify the equilibrium distribution , equation ( [ eq.u ] ) and the time reversal transformation , equation ( [ eq.t ] ) . these physical concepts are mathematically codified in the invariance under supersymmetry transformations , that we have explicitly shown in this section . it is very interesting to note that extensions of the fluctuation - dissipation theorem , for out - of - equilibrium steady states , can be formulated @xcite . a natural arising question is whether a related susy exists for these out - of - equilibrium states . we have presented a detailed study of equilibrium properties of a _ markovian multiplicative white - noise stochastic process_. the stochastic process was modeled in its simplest version of a single stochastic variable , by means of the langevin equation ( [ eq.langevin ] ) , defined by a drift term @xmath10 and a diffusion function @xmath11 . we have completed the definition of the stochastic differential equation by using the _ generalized stratonovich convention _ , characterized by a parameter @xmath1 . the asymptotic equilibrium distribution was found by solving the stationary fokker - planck equation , imposing an asymptotic zero current probability . this naturally conduce to an equilibrium potential @xmath49 ( eq . ( [ eq.u ] ) ) which explicitly depends on @xmath0 . for @xmath4 ( hnggi - klimontovich interpretation ) , the potential coincides with the classical deterministic one , @xmath52 , in agreement with the boltzmann distribution for thermodynamical equilibrium . however , for any other convention , the equilibrium potential is more general , including , for instance , power - law distributions . thus , the definition of equilibrium in the stochastic dynamical sense not necessarily coincides with the thermodynamical equilibrium concept . we have shown that , even for this generalized definition of equilibrium , the system satisfies the usual equilibrium properties such as detailed balance , microscopic reversibility and the fluctuation - dissipation theorem . time reversal properties of the stochastic process are quite interesting . we have shown that , if the forward stochastic trajectory evolves with a definite value of @xmath0 , the time reversed trajectory evolves with the conjugated prescription @xmath5 . therefore , in order to have a unique equilibrium distribution , the definition of time reversed stochastic process is given by the transformation ( [ eq.t ] ) , where we need to change not only the velocity sign , but also the prescription and the drift force . we have shown that , using the equilibrium potential @xmath164 and the time reversal transformation of equation ( [ eq.t ] ) , the stochastic process satisfies detailed balance relations for any value of the parameter @xmath0 . it is convenient to note that , in equation ( [ eq.detailedbalance ] ) , the forward transition probability is not equal to the backward one , except for the stratonovich convention @xmath3 , where forward and backward trajectories coincide . we have also shown that , if the initial and final states of a finite time evolution are prepared in equilibrium , the entropy production of each stochastic trajectory vanishes for arbitrary prescriptions , verifying the related principle of microscopic reversibility . in order to show these properties , we have used the path integral representation of the stochastic process , where the central object is the hole stochastic trajectory . also , we were forced to intensively use an @xmath0-generalized it calculus , summarized in the @xmath0-dependent `` chain rule '' of equation ( [ eq.chainrule ] ) . related with equilibrium properties , there are hidden symmetries in the stochastic process that can be better analyzed in the path integral approach , formulated in an extended functional space composed of commutative as well as anti - commutative grassman variables . the advantage of this formulation is that it is prescription independent . that is , ambiguities in the definition of the multiplicative stochastic processes can be understood , not as a continuum limit ambiguity , but as a necessity to define equal time grassman green s functions . then , provided the grassman variables are not integrated out , it is not necessary to fix any prescription . in this way , we can use ordinary calculus rules , but instead we need to work with anti - commuting variables . in some sense , all the complexity of the it calculus is codified in the grassman algebra in this formulation . in this context , we have computed the entropy for each stochastic trajectory ( in the extended space ) and we have shown that it is not zero , but proportional to a time derivative of the grassman variables density @xmath275 ( equation ( [ eq.entropy ] ) ) . this is due to the extra degrees of freedom we have added to extend the functional space . however , integrating over the grassman variables , we immediately recognize that the entropy production is zero , as it should be for an equilibrium evolution . interestingly , this cancellation is due to `` _ fermionic number conservation _ '' , in other words , due to the invariance under phase transformations . this allows us to speculate about the very attractive possibility of interpreting an out - of - equilibrium evolution , _ i. e. _ , a non - zero entropy production , as a consequence of a gauge symmetry breaking , very similar with a superconducting transition . of course , this can not happen in our present example , since we only have one stochastic variable and , consequently , the grassman variables are not - interacting . nevertheless , in multiple variable systems , quartic terms in the grassman variables are certainly possible . another interesting symmetry which appears in this formulation is supersymmetry ( susy ) , expressed as a linear transformation , mixing commuting as well as anti - commuting variables . we have presented a supersymmetric covariant representation of the stochastic process . while this representation was known for additive processes and multiplicative non - markov processes , the present case of _ markovian multiplicative white noise _ was quite difficult to treat due to the different prescriptions available to define it , connected by a time reversal transformation . in this representation , the non - trivial time reversal transformation is represented by a simple linear transformation . the difficulty of the time conjugated prescriptions are confined to boundary conditions , irrelevant in the computation of correlation functions . using this susy covariant formalism we have analyzed two - point ward - takahashi identities . one of them relates linear responses and fluctuations . we have shown that , while the linear response function and fluctuations are @xmath0-dependent , they satisfy the fluctuation - dissipation theorem for arbitrary values of the prescription @xmath0 . therefore , similar to the additive case , supersymmetry is a consequence of equilibrium evolution , even thought we have different equilibrium evolutions for different prescriptions . summarizing , we have presented a compact path integral formalism to deal with _ markovian multiplicative white - noise _ systems , independently of the prescription used to define the wiener integral . in its supersymmetric covariant form , it automatically encodes equilibrium properties . therefore , it is very appropriate to organize perturbative calculations , since supersymmetry imposes non - perturbative constraints that should be verified at any order of perturbation theory . on the other hand , keeping in checked equilibrium properties , we can safely go forward in the study of out - of - equilibrium fluctuation relations described by this type of stochastic processes . the brazilian agencies , _ fundao de amparo pesquisa do rio de janeiro _ , faperj and _ conselho nacional de desenvolvimento cientfico e tecnolgico _ , cnpq are acknowledged for partial financial support . was partially supported by the latin american center of physics , claf , under the collaborative doctoral program ictp - claf . [ [ appendix ] ] in this appendix we explicitly show that the action , written in the super - field formalism , equation ( [ eq.ssusy ] ) , is in fact equivalent to the time reversal invariant action @xmath143 , given by equation ( [ eq.stilde ] ) . let us define the scalar super - field @xmath230 and the super - potentials @xmath235 and @xmath236 , that we expand in powers of the temporal grassman variables @xmath231 and @xmath232 as , @xmath276 and @xmath277 the covariant derivatives ( [ eq.cov_derivs ] ) acting on the super - potential @xmath236 take the form @xmath278 and @xmath279 the system lagrangian is given by the @xmath280 components in the form @xmath281 then , using equations ( [ eq.vphi ] ) , ( [ eq.dgammabar ] ) and ( [ eq.dgamma ] ) , @xmath282 and @xmath283 replacing these expressions into equation ( [ eq.lagrangian ] ) we find @xmath284 we make the following change of variables in the functional integral : first , we shift the variable @xmath285 , @xmath286 whose jacobian is one , since it is just a translation . then , we rescale @xmath287 in the following way , @xmath288 the jacobian of the first two transformations is @xmath289 and the jacobian of the last transformation is @xmath290 , then the two jacobians cancel each other and the lagrangian in the new variables reads , @xmath291 considering that , by definition , @xmath292 and , thus , @xmath293 , we obtain @xmath294 the last term is a total derivative . this term appears making a trivial transformation like @xmath295 in the @xmath235 term . this is because a supersymetric lagrangian is invariant up to a total derivative under susy transformations . this is a general feature of susy . a transformed supersymmetric action is invariat up to boundary terms . rewriting this last expression in terms of the original functions @xmath296 and @xmath297 we finally get @xmath298 upon integration and realizing that the last term is a total derivative one , we find @xmath299 where @xmath185 is the action of equation ( [ eq.sgrassman ] ) and , then , @xmath143 is equal to equation ( [ eq.stilde ] ) , as can be seen comparing ( [ eq.lfinal ] ) with ( [ eq.sst ] ) , as we want to demonstrate . to compute correlation functions the range of integration is infinite and @xmath300 . however , for finite time we see that the supersymmetric action is equivalent to the time reversal invariant action @xmath143 . 10 url # 1#1urlprefix[2][]#2 hnggi p and marchesoni f 2005 _ chaos : an interdisciplinary journal of nonlinear science _ * 15 * 026101

the statistical character of our sky s population of clusters of galaxies , viewed from radio to gamma - ray wavelengths , is sensitive to models of cosmology , astrophysics , and large - scale gravity . galaxy clusters are cosmographic buoys that signal locations of peaks in the large - scale matter density . the population is shallow and finite . surveys in the coming decades will definitively map our universe s terrain as defined by the highest @xmath0 peaks . current maps have advanced to the stage where abell 2163 , a cluster at redshift @xmath1 with a plasma virial temperature @xmath2 @xcite and galaxy velocity dispersion @xmath3 @xcite , has been nominated a candidate for _ the _ most massive cluster in the universe @xcite , the cosmic equivalent of mount everest . physically , galaxy clusters are manifested in the most massive of the bound structures termed _ halos _ ( or _ haloes _ ) that emerge in the cosmic web of large - scale structure ( lss ) . the lss web is a gravitationally amplified descendant of a weak noise field seeded by quantum fluctuations during an early , inflationary epoch @xcite . its evolutionary dynamics have been well studied into the non - linear regime by n - body simulations @xcite . locally bound regions ( the halos ) emerge , initially via coherent infall within a narrow mass range and , subsequently , via a combination of infall and hierarchical merging that widen the dynamic range , pushing to increasingly larger halo masses . the merging process is of considerable interest for cluster studies , driving astrophysical signatures that can test physical models from the nature of dark matter @xcite to the magnetohydrodynamics of hot , dilute plasmas ( e.g. * ? ? ? merging also potentially confuses cosmological studies , by creating close halo pairs that may appear as one cluster in projection and by introducing variance into observable signals . halos are multi - component systems consisting of dark matter and baryons in several phases : black holes ; stars ; cold , molecular gas ; warm / hot gas ; and non - thermal plasma . after decades of study via n - body and hydrodynamic simulation and related methods ( see recent review by * ? ? ? * ) , models for the detailed evolution of the baryons in clusters are growing in capability to describe an increasingly large and rich volume of observations . what is clear empirically is that the galaxy formation process is globally inefficient : a recent study by @xcite finds that stellar mass accounts for only @xmath4 percent of the total baryon budget in the most massive halos . radiative cooling of gas is overcome by feedback from various sources , including mechanical and radiative input from supernova winds and black hole jets , thermal conduction and other plasma processes , and ablation and harassment during gravitational encounters . while the hierarchical nature of structure formation implies that galaxy and cluster formation are deeply intertwined and , therefore , that detailed understanding of cluster structure and evolution _ requires _ that we understand galaxy formation , the scales separating the most massive clusters from the largest galaxies roughly a factor of 100 in length and 1000 in mass allow progress to be made by approximate physical treatments . the dark matter kinematic structure , including remnant , fine - scale _ sub - halos _ @xcite , as well as the morphology and scaling behaviors of the hot , intracluster medium ( icm ) that dominates the baryonic component @xcite , are examples of areas where direct simulations made good , early progress . a key aspect of their multi - component nature is the fact that clusters offer multiple , observable signals across the electromagnetic spectrum ( e.g. * ? ? ? * ) . at x - ray wavelengths , the hot icm emits thermal bremsstrahlung and line emission from ionized metals injected into the plasma by stripping and feedback processes . stellar emission from galaxies and intracluster light dominates the optical and near - infrared . at millimeter wavelengths , inverse compton scattering within clusters distorts the spectrum of the cosmic microwave background ( cmb ) . gravitational lensing offers a unique probe into the total matter distributions in clusters . synchrotron emission from relativistic electrons is visible at radio frequencies . these and other signatures discussed below provide physically coupled , and often observationally independent , lines of evidence with which to test astrophysical models of cluster evolution . a challenge to cluster cosmology is the construction of accurate statistical models that address survey observables explicitly while incorporating intrinsic property covariance . the use of clusters to study cosmology has a history dating to zwicky s discovery of dark matter in the coma cluster @xcite . brightest cluster galaxies were later employed as standard candles to study the local expansion history of the universe ; @xcite actually derived ( with low significance ) a negative deceleration parameter using this approach , implying accelerated expansion consistent with present findings . in the 1980 s , measurement of the enhanced spatial clustering of clusters relative to galaxies supported the model of gaussian random initial conditions expected from inflation @xcite . in the early 1990 s , an apparent discrepancy between local baryon fraction measurements of clusters @xcite with primordial nucleosynthesis expectations helped rule out a model with critical matter density @xcite . the revelation of hot clusters at high redshift later that decade @xcite presaged the ultimate discovery of dark energy from type ia supernova ( snia ) surveys . the turn of the millennium witnessed a flurry of activity aimed at measuring the amplitude of the matter power spectrum from cluster counts . x - ray studies in particular showed that the amplitude was lower than had been accepted previously ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , a result later confirmed by cmb and cosmic shear measurements . these studies also exposed the importance of understanding systematic effects associated with the use of directly observable quantities as proxies for mass @xcite . recent studies have used cluster counts or the icm mass fraction in very massive systems ( both methods described in more detail below ) to constrain cosmological parameters . these studies are consistent with other observations that find a universe dominated by dark energy ( @xmath5 ) , with sub - dominant dark matter ( @xmath6 ) , and a small minority of baryonic material ( @xmath7 ) @xcite . a detailed pedagogical treatment of how cluster studies helped establish this reference cosmology is given in the review of @xcite . @xcite review x - ray studies of clusters from the rosat satellite era . explaining the nature of the dark energy and dark matter are core problems of physics . the consensus ` concordance ' cosmological model , @xmath8cdm , postulates that dark energy ( de ) is associated with a small , non - zero vacuum energy , equivalent to a cosmological constant term in einstein s equations . another possibility is that de arises from a light scalar field ( or fields ) that evolves over cosmic time . a third option is that de is essentially an apparition , not a source term of einstein s general relativistic equations but a reflection of their breakdown at length and time scales of cosmic dimensions ( e.g. * ? ? ? sky surveys of cosmic systems , from supernovae to galaxies to clusters of galaxies , provide the means to discriminate among these alternatives . forthcoming cluster surveys at mm , optical / near - infrared , and x - ray wavelengths , discussed in section [ sec : new_surveys ] , have the potential to find hundreds of thousands of groups and clusters . figure [ fig : surveyhistory ] puts these efforts into historical perspective , by plotting size against year of publication for cluster samples that generated cosmological constraints discussed in this review . symbol size is proportional to median sample redshift , and symbol types encode the selection method . the stars at far right show theoretical estimates of the all - sky number and median redshift of halos with masses above @xmath9 and @xmath10 . the former mass limit roughly marks the transition from galaxy groups to galaxy clusters , while the latter marks the deepest potential wells with icm temperatures @xmath11 . current surveys have made good progress , but the full population of clusters remains largely undiscovered . optical and x - ray surveys have the longest histories , but these traditional methods are being complemented by new approaches . space - based surveys in the near - infrared extend optical methods to @xmath12 @xcite , and the first few clusters identified by their gravitational lensing signature have been published @xcite . ongoing mm surveys have released the first sets of clusters discovered through the sunyaev - zeldovich ( sz ) effect @xcite , with the promise of much more to come . panoramic , multi - wavelength surveys of common sky areas offer profound improvements to our understanding of clusters as astrophysical systems , which in turn further empowers their use for cosmological studies . and while considerable challenges to interpretation and modeling of survey data certainly exist , a halo model framework , discussed in section [ sec : theory ] , is rising to meet this task . a feature common to many techniques that study de is the nature of the input data , which consists of catalogs of properties , @xmath13 , of discrete objects that lie along our past light - cone . upcoming wide - field surveys will generate @xmath13-catalogs of large dimension that will be distilled to constrain perhaps tens of cosmological and astrophysical parameters . such catalogs may contain internal support through the use of complementary methods : besides galaxy clusters ( cl ) , the same data set can be analyzed for baryon acoustic oscillations ( bao ) and , for optical surveys , weak lensing ( wl ) . ( in the case of repeat observations , optical surveys can also be analyzed for type ia supernovae and gravitational time delay signatures . ) science processing leads to a compressed set of statistical signals , @xmath14 , where @xmath15 indicates an aforementioned method . for large cluster surveys , @xmath16 might consist of counts of clusters binned by sky area , redshift and detected signal . extracting accurate constraints on a set of cosmological model parameters , @xmath17 , from these surveys requires sophisticated likelihood analyses . the critical ingredient is @xmath18 , the underlying likelihood that the cl ( and other ) statistics of the observed sky would be realized within a particular universe . key capabilities that enable such likelihood analysis are : 1 . to predict statistical expectations , @xmath18 , for _ many _ universes , @xmath17 ; + 2 . to extract _ unbiased _ statistical signals from the raw catalog , @xmath19 ; + 3 . to understand the expected signal covariance , cov(@xmath20 genuine understanding of cosmological models from observed cluster data is dependent on the degree to which theory and simulation can provide robust predictions for the observed signals . while numerical simulations of lss can predict catalog - level yields for a given cosmology ( e.g. * ? ? * ) , such predictions necessarily entail additional astrophysical assumptions , meaning @xmath18 is actually @xmath21 , where @xmath22 represents degrees of freedom introduced by an assumed astrophysical model . recovering cosmological information from survey data therefore necessitates marginalization over a reasonable range of astrophysical assumptions . on the other hand , as cosmological constraints from all methods improve , the cluster community can potentially invert the problem , recovering constraints on astrophysical models after marginalizing over cosmology . we begin this review by describing the theoretical basis for cluster cosmology ( section [ sec : theory ] ) , and include there an opening discussion of important sources of systematic error . key observational windows are described in section [ sec : observations ] , and recent cosmological constraints are reviewed in section [ sec : cosmo_constraints ] . in section [ sec : physics ] , cluster contributions to particle physics and gravity are examined . in section [ sec : opportunities ] , we highlight opportunities for important , near - term progress . in closing , we emphasize some essential considerations in survey modeling and analysis ( section [ sec : modeling ] ) before presenting our conclusions ( section [ sec : conclusions ] ) . this section sketches a theoretical description of the halo framework that supports cluster cosmology . there is considerable richness to the galaxy formation problem that we omit here ; the recent review of @xcite provides substantial detail . simulation studies of halo evolution into the strongly non - linear regime are becoming increasingly powerful , but finite resolution and uncertainty in astrophysical treatments limit predictive power . while not yet robust enough to offer sharp prior characterization of the astrophysics required for cosmological studies , simulations offer key insights into the structure and physics sensitivity of the functions that relate observable signals to halo mass and epoch . ample evidence now supports the picture that lss formed via gravitational amplification of initially small density fluctuations , @xmath23 . cosmic microwave background anisotropy measurements are consistent with expectations from a large class of basic inflationary models ( e.g. * ? ? ? such models are characterized by an instantaneous primordial power spectrum , @xmath24 , with spectral index , @xmath25 , expected to be close to unity . here , @xmath26 is the fourier transform of the density fluctuations , @xmath27 . after inflation ceases , fluctuations in the coupled photon - baryon - dark matter fluid evolve in ways that are now well understood from linearized boltzmann treatments @xcite . for the standard case of adiabatic fluctuations , and on scales above the baryon jeans mass , the post - recombination matter ( dark matter and baryons ) power spectrum exhibits a growing mode that scales with the cosmic expansion parameter , @xmath28 , as @xmath29 here , @xmath30 is a transfer function that encapsulates evolution before recombination at @xmath31 , @xmath32 is the density perturbation growth factor from linear theory , and @xmath33 is the controlling parameter set of the background cosmological model . dark energy models that involve modifications to general relativity may introduce @xmath34-dependence into the growth function above . the power spectrum in the reference @xmath8cdm model is set by present - epoch energy densities , @xmath35 , where @xmath36 is the dimensionless hubble constant and @xmath37 is the density of component @xmath38 relative to the critical density , @xmath39 . the curvature density , @xmath40 , is zero to within @xmath41 @xcite , consistent with a flat spatial metric on cosmic scales . our notation uses ` b ' for baryons , ` c ' for cold , dark matter ( cdm ) , and ` m ' for all matter : @xmath42 . in the minimal model , the dark energy is a vacuum energy with equation of state , @xmath43 . we use @xmath44 for this case and employ @xmath45 when referring to models wherein the dark energy equation of state , @xmath46 , differs from @xmath47 . [ tab : wmapcosm ] cc + parameter & value + + @xmath44 & @xmath48 + @xmath49 & @xmath50 + @xmath51 & @xmath52 + @xmath53 & @xmath54 + @xmath25 & @xmath55 + @xmath56 & @xmath57 + @xmath58 & @xmath59 + @xmath60 from @xcite . current constraints for a flat @xmath8cdm model from cmb measurements , combined with angular clustering of red galaxies and local measurements of @xmath61 , are shown in table [ tab : wmapcosm ] @xcite . the parameter @xmath62 is the variance in density fluctuations evaluated at horizon crossing , which is independent of @xmath34 for @xmath63 , and the wavenumber @xmath64 corresponds to a large comoving length scale , @xmath65 . in the minimal model , the matter density fluctuations filtered within a sphere of comoving radius @xmath66 are gaussian distributed with zero mean . the comoving radius defines a mass , @xmath67 , of matter within that radius in the young universe , when @xmath68 . early observations that the variance in galaxy counts is near unity on a scale of @xmath69 led to this as a conventional choice of scale at which to quote the fluctuation amplitude ( see table [ tab : wmapcosm ] ) . the corresponding mass , @xmath70 , is characteristic of rich clusters of galaxies . the variance of linearly - evolved , cdm fluctuations , filtered on mass scale @xmath71 , has the form @xmath72 where the filter function is @xmath73 $ ] for the typical case of sharp ( or top - hat ) spatial filtering within radius @xmath66 . evaluating equation [ eq : sigmasq ] at @xmath74 and @xmath75 produces the oft - quoted matter power spectrum normalization parameter , @xmath58 . we will see below that @xmath76 serves as a similarity variable for expressing model - independent forms of the halo space density and clustering . the evolution of the fluctuation spectrum , equation [ eq : powspect ] , is valid at early times or at scales sufficiently large so that @xmath77 at all times . on small scales , where cdm power spectra are generically maximum , fluctuation growth produces @xmath78 , and linear theory breaks down . mode - mode coupling terms become important to the dynamics , and solutions in fourier space become difficult . while higher - order perturbation theory solutions can extend analytic evolution to later times than linear theory ( e.g. * ? ? ? * ; * ? ? ? * ) , the full problem is typically treated using n - body simulations , discussed below . a recent analytical advance considers lss as an effective fluid . @xcite show that integrating out small - scale , non - linear structures renormalizes the cosmological background and introduces dissipative terms , of order @xmath79 , into the dynamics of large - scale modes , with @xmath80 the typical velocity dispersion of collapsed halos . since even the most massive halos have @xmath81 , the magnitude of these effects is very small . furthermore , @xcite show that virialized halos decouple completely from large - scale dynamics , at all orders in the post - newtonian expansion . astrophysical structures , from the first stars at high redshift to galaxy clusters at low redshift , tend to emerge from local maxima of the filtered density field . while density peaks are generally non - spherical @xcite , a first - order description considers them spherical and isolated from their surroundings . birkhoff s theorem then implies that the expansion histories of radial mass shells within a peak follow trajectories perturbed from the overall background , with sufficiently dense shells expanding to a maximum size and then contracting . the traditional _ ansatz _ assumes collapse by a radial factor of two @xcite , after which a quasi - virialized and quasi - hydrostatic structure a perfectly spherical _ halo _ is born . the collapse criterion is that the _ linearly - evolved _ perturbation amplitude reach a critical value , @xmath82 , with @xmath83 the conventional choice . applying this idea to the cdm spectrum , equation [ eq : sigmasq ] , leads to a characteristic mass scale , @xmath84 , defined by @xmath85 = { { \ensuremath{\delta_{\mathrm{c}}}}}$ ] . at a given epoch , a spectrum of halo masses exist , with masses above ( below ) @xmath84 forming from perturbations with amplitudes above ( below ) the _ rms _ level of the filtered gaussian spectrum . considerable literature ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * and many others ) has established this picture as the _ halo model _ of large - scale structure . we review here only aspects relevant for cluster cosmology ; a more thorough review can be found in @xcite . the basic element of the halo model is the population mean space density , @xmath86 , in units of number per unit comoving volume , commonly referred to as the _ mass function_. expressed as a differential function of mass , it takes the form @xmath87 where @xmath88 is the comoving mean matter density and @xmath89 is a model - dependent function of the filtered perturbation spectrum , equation [ eq : sigmasq ] . analytic forms for @xmath89 capture much , but not all , of the behavior seen in n - body simulations , as discussed below . the spatial clustering of halos is described by a modified version of the matter power spectrum . on large spatial scales , or low wavenumbers , the halo autocorrelation power spectrum is modified , @xmath90 where @xmath91 , the halo _ bias function _ , is independent of @xmath34 , for the case of gaussian fluctuations , but dependent on mass and epoch . while this expression applies to the spatial autocorrelation of systems with fixed mass @xmath71 , it generalizes to the cross - correlation between sets of halos at different masses , @xmath92 . the theory of peaks in gaussian random fields expresses the bias as a function of the normalized peak height , @xmath93 @xcite . below , we show that n - body simulations support the forms of equations ( [ eq : massftnbasic ] ) and ( [ eq : corrftnbasic ] ) , but a precise fit to the mass function requires that @xmath89 be adjusted to include explicit redshift dependence , @xmath94 . there are subtleties to the definition of mass in simulations that must also be taken into account . various astrophysical processes play out within the photon - baryon components of the evolving cosmic web , including hydrodynamic , magnetohydrodynamic and radiative transfer effects ; star and black hole formation with associated feedback of momentum , energy and entropy ; and so on . except for the immediate vicinity of black holes , these processes involve classical physics that is largely known . but the fully three - dimensional and non - linear nature of the problem , the wide dynamic range in length and time scales , and the non - trivial couplings among the constituent physical processes introduce tremendous complexity into baryon evolution . galaxy formation is truly a grand challenge computational problem . we touch on select issues relevant to the observable features of galaxy clusters . _ shocks and turbulent mhd heating . _ during halo formation , gravitational potential energy in the baryonic component is thermalized via shocks . the highest mach numbers , of tens or more , should occur in the accretion shocks at the edges of clusters ( e.g. * ? ? ? * ) . while these strong shocks are expected to be efficient particle accelerators , recent observations place tight limits on the volume - averaged pressure contributions from relativistic particles @xcite shocks with mach numbers of a few are also associated with major mergers : a spectacular example is the narrow radio relic in the cluster ciza j2242.8 + 5301 , for which @xcite use multi - frequency radio and polarization observations to infer a mach number @xmath95 in a shock located @xmath96 from the cluster center . most of the energy thermalized during cluster formation , however , is dissipated in weak shocks that are persistently driven by dissipating sub - structures and ongoing minor mergers . shocks are also driven by jets from agns and , at earlier times , by winds from star forming galaxies . details of the nano - parsec scale physics that drive thermalization remain under active study , especially the roles of magnetic fields , turbulence and plasma instabilities ( e.g. * ? ? ? observations and simulations discussed below indicate that thermalization is efficient ; thermal pressure supplies the bulk of support against gravity within the halo potential except during brief periods near periapsis of major mergers . _ radiative cooling . _ since the intracluster plasma ( and , to a lesser extent , its interstellar counterpart ) is optically thin at most wavelengths , radiation loss is the primary cooling mechanism for the baryonic component of halos . indeed , the classic criterion for setting an upper bound on galaxy size comes from balancing the gas cooling time against the halo dynamical time @xcite . the first generation of stars form at @xmath97 , aided by molecular hydrogen line emission , within halos of mass @xmath98 @xcite . by @xmath99 , atomic line cooling in halos with virial temperatures above @xmath100 k produces the first generation of galaxies , which grow hierarchically for a time determined by the large - scale environment . proto - cluster regions have more efficient cooling at high redshift than do proto - voids , but the feedback from vigorous , early production of compact sources helps to quench star formation before a large fraction of baryons are converted to stars . the cooling timescale of the gas in massive halos is typically longer than a hubble time , except for a subset of systems that exhibit cool cores . the central @xmath101 region of such systems tends to be x - ray bright and typically contains a dominant elliptical galaxy . we discuss aspects of cool core phenomenology in section [ sec : future : coreevol ] . _ star and black hole formation . _ cold , molecular gas fuels star formation . the star formation rate can roughly be considered as proportional to the local rate of gas cooling below @xmath100 k , but there are other considerations . different venues for star formation exist , ranging from quiescent disks to the bulges of tidally - triggered starburst galaxies , and it is not yet clear whether a single model based on local gas conditions captures the full range of observed behavior . supermassive black hole ( smbh ) growth occurs through mergers and accretion in galactic cores , and these central engines drive quasar and radio jet activity ( e.g. * ? ? ? sloan digital sky survey ( sdss ) quasar studies indicate that smbhs of mass @xmath102 exist at @xmath103 @xcite , and processes for forming such large black holes in the first few hundred million years of the universe have been proposed @xcite . in gaussian random fields , small - scale peaks are more abundant when embedded within large - scale peaks , so the largest galaxies and quasars at high redshift represent the progenitors of massive galaxies observed in low redshift clusters . _ feedback from compact sources . _ feedback of mass , momentum , and entropy from stellar / smbh sources is important at all stages of the lss hierarchy . photoionization and supernova - driven winds serve to limit cooling and star formation in low - mass halos @xcite . jets driven by accretion onto the central smbh appear to be required to limit the maximum size of galaxies ( e.g. * ? ? ? * ; * ? ? ? formulations for this feedback typically tie the energy input to the mass accretion rate which , in turn , is governed by the local rate of cooling and/or cold accretion . the end result of this competition between cooling and heating is that heating largely wins . the overall efficiency of star formation is small , and peaks in halos of roughly galactic scale ( e.g. * ? ? ? figure [ fig : dai2010fstarvcirc ] shows a recent compilation of stellar mass fraction ( @xmath104 ) measurements as a function of halo circular velocity , @xmath105 , with @xmath71 the total halo mass and @xmath106 its radius @xcite . the horizontal lines show the cosmic baryon fraction , @xmath107 , derived from wilkinson microwave anisotropy probe ( _ wmap _ ) data analysis @xcite . the stellar mass fraction is maximized at a few tens of percent of the cosmic mean in halos with @xmath108 , equivalent to a mass of @xmath109 at @xmath110 . in cluster - sized halos , the stellar fraction declines with mass , taking on values @xmath111 of the global baryon fraction at the highest masses . yet , the largest galaxies are found in the cores of massive clusters , and their very old stellar populations produce a characteristically narrow ` red sequence ' in a color - magnitude diagram of cluster members . _ dynamical and thermodynamical equilibrium . _ in the context of the evolving cosmic web , the processes above must contend with conditions imposed by halo merging . at any given time , major mergers , such as those involving progenitor pair mass ratios larger than @xmath112 , occur in @xmath111 of the population , concentrated toward the highest masses . these rare events can drive the mass contents of a halo considerably out of equilibrium . minor mergers , while much more frequent , are also less damaging . current simulations and observations indicate that the dynamical and thermodynamical response of halos is quite fast . hydrostatic and virial equilibrium assumptions are typically valid to within roughly ten percent for the majority of the cluster population ( e.g. * ? ? ? * ; * ? ? ? all this astrophysical evolution offers a treasure trove of observational possibilities . uniquely in massive clusters , all of the matter is readily observed , allowing a complete census to be taken . stars make up 13% @xcite ; @xmath113 resides in the hot , diffuse , intergalactic gas @xcite ; and the rest is in the form of non - baryonic cdm ( section [ sec : physics_dm ] ) . as tracers of massive halos , galaxy clusters provide a number of signatures that are sensitive to the underlying cosmology . we review here the principles underlying key methods . a typical set of cosmological parameters for such studies might consist of the primordial spectrum amplitude and slope , the present - epoch densities of the three energy components dominant at late times , the dimensionless hubble constant , and the de equation of state parameters , @xmath114 where the last two parameters define a linearly - evolving de equation of state , @xmath115 this particular set is meant to be illustrative . there is considerable variation in the literature , and many works restrict analysis to a flat cosmology , which removes one degree of freedom from the above through the condition @xmath116 . the yield of upcoming cluster surveys will be sufficiently large to enable disaggregation by angular position , redshift , and the observed signal , @xmath117 . ( note the latter is also referred to in the literature as the _ mass proxy _ , or sometimes the _ observable mass _ , @xmath118 . ) complications associated with the signal mass likelihood and with redshift estimation are discussed below . as a starting point , consider a perfect tracer of mass , @xmath119 , with error - free redshifts , @xmath120 . within a given survey , the expected number of halos , @xmath121 , in a cell described by mass bin @xmath28 and redshift bin @xmath15 with solid angle @xmath122 is @xmath123 cosmology enters this expression through the mass function and the volume element , @xmath124 . the counts in each large spatial bin will deviate from the mean by an excess number , @xmath125 , determined by the local large - scale density field , @xmath27 . following @xcite , the spatial covariance of the counts is @xmath126 where @xmath127 describes the spatial correlation between mass - redshift bins , @xmath128 here , @xmath129 is the window function for cell @xmath15 ( that , when present , can include the effects of redshift estimate uncertainties ) and @xmath130 is a geometric term that depends on the comoving separation , @xmath131 , between cells @xmath15 and @xmath132 . when cells @xmath15 and @xmath132 sample different redshifts , an accurate approximation uses their geometric mean to evaluate @xmath133 @xcite . combining the spatial clustering with a diagonal shot noise term forms the full covariance for a survey sample . derivatives of the mean counts and covariance with respect to model parameters form the fisher information matrix used in survey forecasts . expressions for the fisher matrix can be found in @xcite . equations ( [ eq : mbar ] ) through ( [ eq : xipixel ] ) serve as the foundation of likelihood analysis of large cluster surveys . to be useful in practice , these expressions must undergo a number of modifications , including : transformation from mass to the signal used for cluster detection , @xmath134 ; inclusion of counting errors arising from incompleteness ( missed sources ) and impurities ( false sources ) ; and inclusion of photometric uncertainties , @xmath135 . we discuss these issues in section [ sec : theorytopractice ] and summarize current results in section [ sec : constraints : growth ] . the mass fraction of hot gas , @xmath136 , measured within a characteristic radius of a halo at redshift @xmath137 can be written as @xmath138 where @xmath139 accounts for star formation and other baryon effects within that radius . at large radii in the most massive halos , where the hot icm dominates the baryon budget and the impacts of feedback processes are modest , baryon losses are small and @xmath140 is a reasonable expectation . motivated by the growing body of measurements of @xmath136 from the _ rosat _ x - ray satellite , @xcite and @xcite recognized that a mismatch in the dependence on metric distance , @xmath141 , between gas mass ( @xmath142 ) and total mass ( @xmath143 ) measured from x - ray observations implied that gas fraction measurements in massive clusters could be exploited as a distance estimator , with @xmath144 . like type ia supernovae , massive clusters serve as standard calibration sources that test the expansion history of the universe . key benefits , relative to survey counts , are the ability to perform this test with a relatively small number of clusters and the relative insensitivity to cluster selection . we summarize results from this exercise in section [ sec : constraints : fgas ] . in a similar vein , @xcite noted that x - ray and sz measurements could be combined to determine distances to clusters . the cmb spectral shift is governed by the compton y - parameter , a measure of the electron pressure along the line of sight , @xmath145 . given an observed sz signal , @xmath146 , and a predicted signal based on x - ray measurements of the icm density and temperature , @xmath147 , the angular diameter distance scales as @xmath148 the cosmological constraint originates from the distance dependence of the x - ray measurements , @xmath149 , and the requirement that @xmath150 . accurate sz and x - ray flux and temperature calibration are particularly important to this method , referred to below as xsz . the thermal and kinetic sz signals from clusters ( section [ sec : obs_sz ] ) cause distortions in the cmb at small angular scales ( @xmath151 ) . if the distortion pattern from a single halo of mass @xmath71 at redshift @xmath137 is described by an angular fourier transform , @xmath152 , then the full halo population will generate a fluctuation spectrum @xcite @xmath153 adding halo spatial correlations gives a small correction to this estimate @xcite . this approach to testing cosmology is limited by degeneracy with astrophysical assumptions , as the interplay between @xmath152 and @xmath154 makes clear . measurements of the cosmic peculiar velocity field contain additional cosmological information ( e.g. @xcite and references therein ) . the kinetic sz effect ( section [ sec : obs_sz ] ) in principle offers a way to measure the peculiar velocities of galaxy clusters . although some initial results based on such measurements have been reported ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , the technique has not yet reached the maturity of those discussed above and is not discussed further in this review . n - body simulations of a single , collisionless dark matter fluid offer the means to investigate non - linear evolution of lss under an implicit ` light - traces - mass ' assumption . the technology supporting such simulations has advanced to the state where @xmath155 is available ( but not yet realized ) on peta - scale computational platforms @xcite . employing larger-@xmath156 simply to model bigger volumes is a natural mode of growth , since parallelization is relatively simple ( large - volume domain decomposition minimizes the particle transfer among computational nodes ) , the number of timesteps is independent of @xmath156 , and the light - traces - mass assumption is easier to justify under modest mass and force resolution . large - volume simulations produce generous halo population realizations with which to calibrate the mass function and clustering of halos , and current state - of - the - art studies employ ensembles of @xmath157-particle simulations . coupled n - body and gas dynamic simulation methods enable multi - fluid studies that break free of the light - traces - mass assumption . indeed , the first application of this class of codes tested the possible separation of baryons and neutrinos within clusters formed in a universe dominated by massive neutrinos @xcite . the field has advanced considerably since then , and we refer the reader to @xcite for a recent review . we discuss primarily dark matter simulations here , with some relevant multi - fluid simulations results presented in the next section . through the mass function , halo mass provides the critical measure that connects observables to the underlying cosmology . but halos are complex , dynamic structures that confound attempts at a unique definition of mass . in the model of spherical collapse applied to initial density peaks , the halo edge and interior mass are readily defined by the outermost caustic in dark matter or by the location of the shock in cold baryonic accretion @xcite . in both cases , this radius marks an abrupt transition in the mean radial velocity , separating a nearly hydrostatic interior from an infall - dominated exterior . halos forming in 3-d simulations deviate from this ideal case in important ways , some of which can be described by higher - order analytic approaches to peak evolution @xcite . the collapse process is more ellipsoidal than spherical , and merging competes with smooth accretion as the dominant mode of halo growth ( e.g. , * ? ? ? defining centers and boundaries in this complex environment has become a matter of convention . two common algorithmic conventions have emerged : ( i ) percolation , also known as friends - of - friends ( fof ) , and ( ii ) spherical overdensity ( so ) . fof first links all pairs of particles within a given distance , @xmath158 , then merges them into groups based on a shared link condition ( ` a friend of a friend is a friend ' ) . the so approach first filters the particle field to identify peaks , then grows spheres around peaks with sizes determined by an interior density threshold , @xmath159 . the threshold density @xmath160 is typically chosen to be either the background matter density , @xmath161 , or the critical density , @xmath162 . unless otherwise specified , we adopt the latter convention in this article . = 4.75 cm = 4.75 cm = 4.75 cm several studies discuss the relative merits of these approaches and argue values for the parameters @xmath158 and @xmath163 ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? figure [ fig : lukic09 ] provides a visualization of three simulated halos spanning a range of dynamical and morphological behaviors . in each panel , white particles are members of the fof halo with @xmath164 while green are so members using @xmath165 against @xmath166 . these are typical of parameter values used in the literature . the left panel shows a relatively isolated system where the two methods give fairly consistent results . the other panels show two discrepant cases ; in the middle is a highly - structured , active merger while , at right , percolation across a filamentary bridge links two similarly sized systems that are just beginning to merge . the discrepant cases do not dominate in number , but neither are they uncommon . for cosmological studies , what is important is to establish an accurate accounting process to enumerate observable halo features . roughly speaking , observers viewing the systems in figure [ fig : lukic09 ] would be likely to identify one dominant cluster in the left and middle panels , and two in the right . an fof accounting system would need to admit a non - unitary condition ( one halo maps to two clusters ) when converting mass to observable signals . in contrast , so masses map to integrated aperture observations more directly . for this reason , so masses see more frequent use for survey data analysis . the original multiplicity function paper of @xcite used the clustering of particles in n - body experiments with @xmath167 to support their analytic form for @xmath89 in equation [ eq : massftnbasic ] . later , @xcite used @xmath168 simulations to set free parameters of their @xmath89 model derived using an ellipsoidal , rather than spherical , collapse approximation . using a suite of simulations of open and flat cosmologies with @xmath169 ranging from @xmath112 to 1 , @xcite found a unique , three - parameter form for @xmath89 that produced a mass function accurate to @xmath170 across the suite of models . a recent study by @xcite employs 22 large ( @xmath171 ) simulations produced with three independent n - body codes to calibrate a functional form motivated by @xcite , @xmath172 \ , e^{-c/\sigma^2 } . \label{eq : massftntinker}\ ] ] this study was the first to open the density threshold degree of freedom ; their fitting parameters are published as functions of @xmath163 ( against @xmath173 ) for @xmath174 $ ] . with the high statistical power of their simulation ensemble , @xcite achieve a fit with @xmath175 statistical precision in halo number at @xmath176 for a @xmath8cdm cosmology . maintaining this precision for redshifts @xmath177 requires the introduction of mild redshift dependence into the fit parameters , @xmath178 , @xmath179 and @xmath180 . the theoretically expected halo counts above masses @xmath181 and @xmath10 in the reference @xmath8cdm cosmology , shown in figure [ fig : surveyhistory ] , are based on the tinker form for threshold @xmath165 against the mean mass density ( see fitting formulae in * ? ? ? on the other hand , the bias function measured in the same simulation ensemble shows no need for such redshift - dependent corrections . framed in terms of the normalized linear perturbation amplitude , @xmath182 , @xcite find a robust fit of the form @xmath183 with a single set of parameters @xmath184 that are written only as functions of @xmath163 . for the case @xmath185 ( i.e. , @xmath186 peaks ) , the value of the bias is large , @xmath187 , for @xmath165 . the cluster power spectrum , equation [ eq : corrftnbasic ] , can be enhanced by factors of several tens over the mass power spectrum . the very massive end of the fof mass function was recently revised by @xcite using @xmath188-particle simulations in @xmath8cdm cubic volumes up to @xmath189 in scale . above @xmath190 , their fit lies up to @xmath191 above prior calibrations @xcite . gravitational relaxation drives the phase - space structure of halos to a common structure that applies from small galactic satellites to the most massive galaxy clusters . the form of the radial density , @xmath192 is known as the navarro - frenk - white ( nfw ) profile @xcite . here , @xmath193 is the scale radius , @xmath194 is the concentration parameter ( with @xmath195 ) and @xmath196}$ ] . simulations show that concentration and mass are weakly correlated . in the mass range of galaxies to clusters , @xmath197 , with @xmath198 at @xmath110 and @xmath199 at @xmath200 ( e.g. , * ? ? ? that study finds that a fixed concentration , @xmath201 , applies in the mean to high mass halos , independent of redshift . tracking the mass accretion histories of halos in simulations , @xcite find a common functional form , and show that the formation epoch correlates strongly with concentration . the concentration mass relation can be understood as a result of adiabatic contraction of differently - shaped peaks in the linear density field @xcite . cluster cosmology originates from phenomena observed on the sky , in the 2 + 1 space of angular coordinates and redshift . the observables employed for a likelihood analysis must be predicted under a set of combined cosmological and astrophysical parameters , @xmath202 . for constraints based on cluster counts , the mass function , @xmath86 , written in terms of spherical overdensity or percolation measures from simulations needs to be translated into a signal function , @xmath203 , for one or more signals , @xmath204 . we use the terms _ signal _ and _ observable _ interchangeably , and generically they refer to bulk measures at mm ( sz decrement @xmath205 ) , optical ( richness , @xmath206 , or velocity dispersion , @xmath207 ) , or x - ray ( luminosity , @xmath208 ; temperature , @xmath209 ; gas mass , @xmath210 ; and/or gas thermal energy , @xmath211 ) wavelengths ( see section [ sec : observations ] ) . an ideal experiment would measure all of these observables within apertures optimally matched to the underlying halo sizes , @xmath212 . this ideal is often frustrated by signal - to - noise constraints and confused by projection effects and foreground / background contamination . scaling relations for cluster signals , based on assumptions of virial equilibrium and self - similar internal structure , were first published by @xcite . in this model , halos at fixed mass and redshift are identical , and scalings with mass and redshift follow calculated power - law behaviors . observations generally support power - law behavior , but not always with the self - similar slope ( section [ sec : constraints : scaling ] ) . we describe here a non - self - similar model that incorporates arbitrary mass scaling and allows for variations at fixed mass and redshift . for compactness of notation , let @xmath213 , for each of the @xmath156 observables , @xmath204 , and let @xmath214 . the power - law assumption transforms to log - linear scaling @xmath215 where the average is over a very large cosmic volume . the elements of @xmath216 are the slopes of the individual mass - observable relations , and the intercepts @xmath217 reflect the evolution at fixed mass . at a fixed epoch , we can always choose units such that @xmath218 ( as we do below ) . for cosmological studies , a measure of merit is the equivalent mass scatter in each signal , @xmath219 . various processes , including different formation histories and the stochastic nature of mergers , generate deviations from the mean . taking these as gaussian in the log leads to a form for the _ conditional signal likelihood _ , @xmath220^\dagger \mathbf{\psi}^{-1 } \left [ { \mathbf{s}}-{\mathbf{\bar{s}}}(\mu , z)\right ] \right\ } . \label{eq : pofs}\ ] ] the elements of the covariance matrix , @xmath221 , could have mass or redshift dependence , but a first - order approach considers them as constants . when the mass variance of signals is small , @xmath222 , then the above expressions can be convolved with a locally power - law approximation to the mass function , @xmath223 , to obtain the local _ signal space density _ function , @xmath224 , \label{eq : nofs}\ ] ] where @xmath225 is the variance about the mean log - mass selected by the set of signals @xmath226 , @xmath227 the first term above is the mean mass for the case of a flat mass function , @xmath228 . the second term , represents the ( eddington ) mass bias induced by asymmetry in the mass function convolution . upscattering of low - mass systems dominates when @xmath229 , and the high - mass end of the @xmath8cdm mass function is steep , @xmath230 @xcite . these equations make explicit the degeneracy between cosmology ( e.g. @xmath231 and @xmath28 ) and astrophysics ( e.g. @xmath216 and @xmath232 ) inherent in cluster counts . they provide the means to compute biases , relative to a mass complete sample , associated with signal - limited cluster samples ( discussed further in section [ sec : theory : selection ] ) . figure [ fig : stanek10 ] provides support for this model from millennium gas simulation analysis @xcite . the left panel shows deviations about the mean behavior of four intrinsic ( 3-dimensional ) properties measured within @xmath233 for @xmath234 halos with mass @xmath235 at @xmath110 . the lower diagonal and red histograms show results from a cooling and preheating ( ph ) treatment of the baryons , where the entropy is instantaneously raised to @xmath236 at @xmath237 . only a small fraction of baryons cool into stars in this model @xcite . the upper diagonal and blue histograms are from a gravity - only ( go ) treatment , where the gas is heated only by shocks and does not cool . the internal properties generally have modest variance , and pairs tend to be positively correlated with typical correlation coefficient @xmath238 . halos identified by a pair of properties will have mass variance @xmath239 shown by the areas of the off - diagonal circles in the right panel of figure [ fig : stanek10 ] . individual properties lie along the diagonal . the intrinsic gas thermal energy , @xmath205 , selects mass with @xmath240 dispersion , the best individual measure for both physics cases . this level is also seen in the simulations of @xcite which include cooling , star formation and feedback . pairs of intrinsic measurements always improve mass selection , and the strong correlation between @xmath241 and @xmath205 combines with the large mass variance of @xmath241 to achieve mass selection with @xmath242 scatter in the ph model . applying bayes theorem to this model allows one to write the likelihood of mass and an observable , @xmath243 , for a sample selected on observable @xmath244 . when the two signals are correlated , one can show that the scaling with mass of the non - selection signal will be @xmath245 , \label{eq : s2bargs1}\ ] ] which is biased relative to the naive expectation of @xmath246 . the intrinsic correlation between signals at fixed mass is relatively challenging to constrain from current data , but first measurements have been made for samples selected using optical @xcite and x - ray @xcite observations . clusters on the sky relate to halos through selection on one or more observables . matching cluster detections ( which originally reside in a 2 + 1 space of angular position and signal - to - noise ) to halos can sometimes be complex ; two halos along nearly the same line of sight may be blended into a single cluster , or a single halo may be fragmented into more than one cluster . the frequency of these occurrences is typically not large , @xmath247 , but the exact values are sensitive to a number of factors , particularly detection method and mass , and so are best modeled via direct sky realizations ( e.g. * ? ? ? * ) . the selection observable can be distorted from its intrinsic value ( equation [ eq : pofs ] ) by triaxiality , by additional sources along the line - of - sight , by mis - centering and/or mis - estimation of the radial scale , and by other effects . telescope / instrument calibration and data processing methods also contribute to the error budget . for upcoming studies using cluster counts , photometric redshift errors have an important , but not dominant , effect ( section [ sec : new_surveys ] ) . testing cosmology with halo counts and clustering requires that the theoretical mass function be transformed , via the scaling relations and a model of the selection process , into a prediction for the distribution of clusters in the space of survey observables ( e.g. redshift and x - ray flux ) . the scaling relation parameters set the space density portion of the survey yield ( equation [ eq : nofs ] ) in terms of the ( cosmologically dependent ) local amplitude , @xmath248 , and logarithmic slope , @xmath249 , of the mass function . sample selection must be well understood to avoid perturbing @xmath248 and @xmath249 from their true values , biasing cosmological results . fortunately , such effects can be mitigated by survey self - calibration @xcite or by calibration using follow - up observations , as discussed below . the task of empirically constraining the scaling relations is complicated by the fact that the clusters targeted for follow - up observations are themselves subject to selection effects related to their original discovery . in an x - ray flux - limited sample , for example , higher x - ray luminosity at a given mass leads to a larger probability of detection ( commonly known as malmquist bias ) . the effects of selection bias must therefore be accounted for in the calibration of scaling relations , much as in the cosmological analysis ( e.g. @xcite ) . figure [ fig : constraints : selection ] illustrates the influence of selection on observed scaling relation data in a cartoon case . the full population ( black crosses and green points ) obeys a scaling law ( red line ) with non - trivial intrinsic scatter . in the simple case where detection requires a particular threshold luminosity ( the dashed , blue line ) , it can be seen that , even if every detected cluster is followed up to obtain precise measurements of the mass and luminosity , the resulting data set will be a biased representation of the full population . while complete at the highest masses , the sample is increasingly incomplete at low masses , with the low - luminosity systems absent . a closely related consideration is the effect of the underlying mass function on the observed scaling relation data . the distribution of the relation s independent variable(s ) ( in this case cluster masses ) within the full population generically influences constraints on scaling laws ( e.g. @xcite ) . neglecting to account for this influence corresponds to the assumption of uniformly distributed independent variables ; often this approximation is sufficient , but the exponentially steep slope of the cluster mass function suggests that we should take the issue seriously in the context of cluster cosmology @xcite . figure [ fig : constraints : selection ] illustrates how the steepness of the mass function influences the fraction of the observed data which are strongly biased relative to the underlying scaling relation . given the need to solve for both the slope and scatter of the scaling relation , accounting for the disparity in the number of high - mass and low - mass systems is critical . we note that simply conditioning the sampling distribution on cluster detection , as some authors have done , is not sufficient to rigorously recover all the scaling information . note that this effect has a floor set by non - zero intrinsic scatter in the scaling relations , but the effect can in principle be enhanced by measurement error . however , measurement errors in current x - ray and optical cluster surveys are typically smaller than the intrinsic dispersion , even at the survey limit . thus , re - measurement of the survey observables through deeper , follow - up observations ( e.g. to improve the signal - to - noise of x - ray or sz flux ) does not circumvent the issue of selection bias in the scaling relation analysis . while selection bias clearly influences scaling relations involving the selection observable , it also influences relations of other signals with which the selection observable has non - zero intrinsic correlation ( equation [ eq : s2bargs1 ] ) . this is illustrated in figure [ fig : constraints : selection2 ] , for a signal which is correlated with the selection observable with coefficient 0.1 ( left panel ) and 0.9 ( right panel ) . the red line shows the true scaling law and the points shown correspond to the detected clusters from figure [ fig : constraints : selection ] ( right panel ) . with relatively mild intrinsic correlation , as has been found for temperatures and soft x - ray flux detection @xcite , the distribution of data points closely follows the underlying relation ; for more extreme values of the correlation coefficient , as might be expected , e.g. , between temperature and sz signal , deviations due to selection bias become evident . note that the severity of the effect also depends on the covariance of the signals rather than only on the correlation coefficient ( i.e. the size of the marginal scatter in each signal is also important ) . cluster samples are often characterized in terms of _ completeness _ and _ purity _ @xcite . completeness is used in many ways , but its simplest form for cluster cosmology refers to the fraction of halos above mass @xmath71 at redshift @xmath137 that are identified in a survey with some observable limit , @xmath250 . completeness of unity is achievable at high masses when the survey limit , @xmath250 , lies sufficiently far in the signal likelihood s negative tail . impurity is a measure of false positive sources in the sample . fewer conventions for its definition exist in the literature . generically , one can write the observed counts above some signal limit @xmath117 as a sum , @xmath251 , where the first term represents genuine cluster systems manifestations of a single massive halo along the line of sight and the second expresses detections of other origin . zero impurity , @xmath252 , is a desired goal . telescopes aimed at a distant halo necessarily collect photons that originate elsewhere along the multi - gigaparsec sightline than within the target system . due to their softer angular profiles , sz , lensing and optical cluster signals can be blended more readily than x - ray . chance orientations of two or more halos within local supercluster regions create an asymmetric tail to high signal values . considered in terms of mass selection , the effect produces a tail to low masses in the distribution of halo mass selected at a given signal ( e.g. * ? ? ? * ) . since the matter components of halos are generally ellipsoidal rather than spherical , orientation variations also produce scatter in signals observed in halos of fixed mass . signals are generally maximized when viewed along the long axis and minimized along the short axis . orientation can affect cluster selection , with prolate systems oriented along the line - of - sight being preferentially included . since its collisional nature drives the x - ray emitting gas toward equipotential surfaces , it tends to be rounder than the dark matter and so less susceptible to orientation bias . as discussed in section [ sec : observations ] , the density squared dependence of the x - ray emissivity means that x - ray selection is less prone to projected confusion . optical richness measurements roughly trace mass density and are therefore more easily confused by projection and orientation effects . sz measurements are intermediate , since the sz effect depends on electron pressure , the product of density and temperature . it is important to keep in mind that theory offers many potential deviations from the reference @xmath8cdm cosmology sketched above . key model assumptions that the dark matter is a weakly interacting massive particle , that inflation produced a gaussian spectrum of initial density fluctuations with a power - law initial spectrum , that small - amplitude metric perturbations are well described by newtonian , weak field expansions in general relativity , and so on need to be rigorously tested . in section [ sec : physics ] , we discuss ways in which clusters can be used to test a number of proposed modifications to the reference model . in this section we review briefly the physics underlying multiwavelength observations of galaxy clusters . we summarize efforts to construct cluster catalogs , with an emphasis on surveys that have led to cosmological constraints . we discuss techniques used to measure the masses of clusters , and observable proxies that correlate tightly with mass . most of the baryons in the universe are in diffuse gas . typically , this gas is very difficult to observe . within galaxy clusters , however , gravity squeezes the gas , heating it to virial temperatures of @xmath253@xmath254 , which causes it to shine brightly in x - rays . galaxy clusters therefore ` light up ' at x - ray wavelengths as luminous , continuous , spatially - extended sources ( figure [ fig : multiwavelength ] ) . = 4.5 cm = 4.5 cm = 4.5 cm the primary x - ray emission mechanisms from the diffuse icm are collisional : free - free emission ( bremsstrahlung ) ; free - bound ( recombination ) emission ; and bound - bound emission ( mostly line radiation ) . the emissivities of these processes are proportional to the square of the electron density , which ranges from @xmath255 in the centers of bright ` cool core ' clusters to @xmath256 in cluster outskirts . at these low densities , the x - ray emitting plasma is optically thin and in the coronal limit , which makes modeling straightforward . for survey observations , the primary x - ray observables are flux , spectral hardness and spatial extent . using deeper , follow - up observations of individual clusters , modern x - ray satellites allow the spatially - resolved spectra of clusters to be determined precisely , permitting measurements of the density , temperature and metallicity profiles of the icm , and a host of derived thermodynamic quantities . for reviews of the principles underlying x - ray observations of clusters see , e.g. , @xcite and @xcite . the optical and near - ir emission from galaxy clusters is predominantly starlight . the galaxy populations of clusters are dominated by ellipticals and lenticulars ( i.e. early - type galaxies ) . this is particularly true in the central regions , where the largest and most luminous galaxies are found ( figure [ fig : multiwavelength ] ) . the old and relatively homogeneous nature of their stellar populations leads to the majority of the galaxies in clusters occupying relatively tight loci in color - magnitude diagrams ( e.g. * ? ? ? this characteristic has proved important to modern cluster finding algorithms . for optical surveys of clusters , the main observables are the richness ( i.e. the number of galaxies within the detection aperture ) , luminosity and color . for follow - up observations of individual clusters , aimed in particular at measuring their masses , the primary observables are the galaxy number density , luminosity , and velocity dispersion profiles . typical velocity dispersions for large clusters are of order 1000@xmath257 . for reviews of optical studies of galaxy clusters including discussions of the development of the field , see @xcite and @xcite . as cmb photons pass through a galaxy cluster they have a non - negligible chance to inverse compton scatter off the hot icm electrons . this scattering boosts the photon energy and gives rise to a small but significant frequency - dependent shift in the cmb spectrum observed through the cluster known as the thermal sunyaev - zeldovich ( hereafter sz or tsz ) effect @xcite . the magnitude of the effect is proportional to the line of sight integral of the product of the gas density and temperature . the kinetic sz ( ksz ) effect is an additional , smaller distortion of the cmb spectrum due to the peculiar motion of a cluster with respect to the hubble flow ( i.e. the cmb rest frame ) . the magnitude of the ksz effect is proportional to the peculiar velocity . for a review see @xcite . according to general relativity , the gravity associated with a mass concentration will bend light rays passing near to it in a phenomenon known as gravitational lensing . this can both magnify and distort the images of background galaxies . with modern data , gravitational lensing can be detected clearly in the statistical appearance of background galaxies observed through clusters ( weak lensing ) , and in the field ( often termed cosmic shear ) . occasionally , lensing can also lead to strong distortions and multiple images of individual sources ( strong lensing ) . for a galaxy cluster and background galaxies of known redshifts , the measured gravitational shear can be used to infer the cluster mass . for a recent review of gravitational lensing , see @xcite . a well - designed cluster survey should meet requirements in terms of angular scale , flux sensitivity and redshift coverage . the survey should be as complete ( i.e. not have missed clusters that it should have detected ) and pure ( i.e. not have detected spurious clusters ) as possible and the selection function describing the completeness and purity as a function of signal , position and redshift should be known precisely . the survey observables should correlate as tightly as possible with mass . in tension with these requirements , surveys must also be constructed within the context of limited resources . x - ray observations currently offer the most mature and powerful technique for constructing cluster catalogs . the primary advantages of x - ray surveys are their exquisite purity and completeness , and the tight correlations between x - ray observables and mass . galaxy clusters are simple to identify at x - ray wavelengths , being the only x - ray luminous , continuous , spatially extended , extragalactic x - ray sources . clusters have typical soft x - ray band luminosities of @xmath258 erg / s or more , and spatial extents of several arcmin or larger , even at high redshifts . given modest angular resolution , e.g. @xmath259 arcmin ( full width half maximum , fwhm ; easily achievable ) and tens of detected counts , the x - ray emission from galaxy clusters can be detected against a background populated otherwise only sparsely with point - like active galactic nuclei . the first x - ray cluster catalogs constructed for cosmological work ( ( * ? ? ? * often called the brightest 50 or b50 catalog ) ; @xcite ) were based on the ariel v and heao-1 all - sky surveys , and pointed observations made with the _ einstein observatory _ and _ exosat _ ( see also @xcite ) . these catalogs provided early evidence for evolution in the x - ray luminosity function of clusters @xcite and were used subsequently in a series of pioneering cosmological works ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? these catalogs were eventually superseded by surveys carried out with the _ rosat _ satellite . this mission , launched in june 1990 , had two main parts : the _ rosat _ all - sky survey ( rass ; * ? ? ? * ) , spanning the first 6 months ; and pointed observations , which took place over the next 8 years . the main instrument aboard _ rosat _ , the position sensitive proportional counter , had a modest point spread function ( psf ; @xmath260 arcmin fwhm in survey mode ) , but low background and a wide field of view ( @xmath261 degree diameter ) . the main cluster catalogs constructed from the rass and used in cosmological studies include the rosat brightest cluster sample ( bcs ; * ? ? ? * ) , which covered the northern hemisphere at high galactic latitudes and low redshifts ( @xmath262 ) to a flux limit of @xmath263 @xmath264(0.12.4 kev ) ; the rosat - eso flux - limited x - ray galaxy cluster survey ( reflex ; * ? ? ? * ) , which covered the southern sky at low redshifts to a flux limit of @xmath265 @xmath264 in the same band ; the hiflugcs sample @xcite of the x - ray brightest clusters at high galactic latitudes , with @xmath266 @xmath264(0.12.4kev ) ; and the massive cluster survey ( macs ; * ? ? ? * ) , which extended this work to higher redshifts @xmath267 and slightly fainter fluxes @xmath268 @xmath264 . other cluster surveys have been constructed from the rass , or are in the process of being constructed , but have not yet been used to derive rigorous cosmological constraints . a number of x - ray cluster catalogs have also been constructed based on serendipitous discoveries in the pointed phase of the rosat mission . notable among these are the rosat deep cluster survey and the 400 square degree rosat pspc galaxy cluster survey ( 400d ; * ? ? ? * ) , which have been used to derive cosmological constraints . these catalogs cover much smaller areas than the rass , but reach an order of magnitude or more fainter in flux ( figure [ fig : xray_surveys ] ) . a second major advantage of x - ray surveys is the observed strong correlation between x - ray luminosity and mass across the entire flux and redshift range of interest . these quantities follow a simple power law relation ( section [ sec : constraints : scaling ] ) , with a dispersion in luminosity at a given mass of @xmath269 and no significant outliers @xcite . the density - squared dependence also makes the x - ray survey signal from clusters relatively insensitive to projection effects . thus , an x - ray survey of sufficient depth can be translated straightforwardly into statistical knowledge of the distribution of massive halos . in principle , x - ray surveys could be constructed using even lower - scatter mass proxies ( section [ sec : obs_masses_proxies ] ) such as temperature or center - excised luminosity as the survey observable . however , given the ease and depth to which total x - ray luminosity can be measured , these lower - scatter mass proxies are typically used as auxiliary data ( e.g. * ? ? ? * ; * ? ? ? the primary disadvantage of x - ray cluster surveys is that they can only be carried out from space , which makes their construction relatively expensive . the first extensive cluster catalog was constructed at optical wavelengths by george abell @xcite based on visual inspection of photographic plates from the palomar observatory sky survey . abell identified clusters as concentrations of 50 or more galaxies in a magnitude range m3 to m3 + 2 ( where m3 is the magnitude of the third brightest cluster member ) and radius @xmath270 mpc ( with distance estimated based on the magnitude of the tenth brightest galaxy ) . clusters were further characterized into richness and distance classes . abell s catalog was updated and extended to the southern sky by @xcite ( hereafter aco ) . the final aco sample has more than 4000 clusters . an additional , early optical cluster catalog extending to poorer systems was compiled by zwicky and collaborators ( see e.g. * ? ? ? * ) , although the search criteria were less strict than abell s . @xcite applied a percolation algorithm to an early cfa redshift catalog to identify a set of 92 nearby groups and clusters . using 4-m class telescopes and a mix of photographic plate and ccd observations , @xcite opened high redshift cluster studies by identifying 418 systems over @xmath271 deg@xmath272 extending out to @xmath273 . spatial and photometric matched filter methods ( e.g. * ? ? ? * ) as well as the introduction of n - body simulations to calibrate projection effects ( e.g. * ? ? ? * ) marked the beginning of the modern era of optical cluster cosmology . because the cores of galaxy clusters are dominated by red , early - type galaxies , an effective way to reduce the impact of projection effects is to use color information to select for overdensities of red galaxies ( e.g. * ? ? ? * and references therein ) . the red - sequence cluster survey ( rcs ) , a sample of 956 clusters identified with a single ( @xmath274 ) color , provided the first modern cosmological constraints using optical selection @xcite . to cover a broad range of redshifts , multi - color photometry is needed to track the intrinsic 4000 angstrom break feature of old stellar populations as it reddens . the five - band photometry of the sloan digital sky survey ( sdss ) enabled such selection . the maxbcg catalog @xcite of 13,823 clusters with optical richness @xmath275 was produced using @xmath276 colors and spans the redshift range @xmath277 . cosmological constraints from this sample @xcite are discussed below . recently , larger sdss clusters samples have become available , identified using photo - z clustering @xcite , a gaussian mixture modeling extension of the maxbcg method @xcite , and an adaptive matched filtering approach @xcite . these catalogs contain between 40000 and 69000 clusters spanning @xmath278 , and cover roughly 8000 deg@xmath272 of sky . a primary challenge to cosmological analysis using such catalogs is the definition of robust mass proxies that possess minimal and well - understood scatter across the full mass and redshift ranges of interest . projection of filamentary structures and small groups along the line of sight has a greater impact on optical cluster catalogs than x - ray , and these effects introduce a degree of skewness into the mass - observable relations @xcite . uncertainty in modeling this and other selection effects currently limits the constraining power offered by the large sample sizes of optical cluster catalogs . the first large catalogs of galaxy clusters selected from observations of the sz effect are currently under construction , using measurements made with the south pole telescope ( spt ; * ? ? ? * ; * ? ? ? * ) , the atacama cosmology telescope ( act ; * ? ? ? * ; * ? ? ? * ) and the _ planck _ satellite @xcite . the primary advantage of sz surveys is that , in contrast to x - ray and optical measurements , the sz signal of a cluster does not undergo surface brightness dimming . sz surveys are therefore well - suited , in principle , to searches for massive clusters at high redshifts . the surveys mentioned above are each expected to produce catalogs of hundreds of massive systems at intermediate - to - high redshifts . challenges for these projects include determining the optimal observables ( i.e. the best mass proxies ) to measure from the survey data in the current , low signal - to - noise ratio regime ; calibrating the mass scaling for these observables ; and understanding in detail the impact of contamination by radio and infrared sources @xcite . projection effects are also expected to be more significant for sz surveys than for x - rays @xcite . accurate measurements of cluster masses provide a cornerstone of cosmological work . x - ray mass measurements are based on the assumption of hydrostatic equilibrium ( hse ) in the icm . for a spherically symmetric system in hse , the measured gas density and temperature profiles can be related to the total mass ( e.g. * ? ? ? * ) , @xmath279,\ ] ] where @xmath280 is the mass within radius @xmath106 , @xmath281 is the icm temperature , @xmath282 is the gas particle density , @xmath283 is newton s constant , @xmath34 is the boltzmann constant , and @xmath284 is the mean molecular weight . note that the mass within radius @xmath106 depends more strongly on the temperature than the density at that radius . hydrostatic equilibrium requires that the gravitational potential remain stationary on a sound crossing time ; that all motions in the gas be subsonic ; and that forces other than gas pressure and gravity are unimportant . the hydrostatic method can therefore not be applied robustly to systems undergoing major merger events , nor to regions of otherwise relaxed clusters where these assumptions break down , e.g. in their central regions where strong agn feedback effects are commonly observed @xcite . out to intermediate radii , measurements of the gas temperature and density profiles with _ chandra _ or _ xmm - newton _ are straightforward . at large radii ( @xmath285 ) , however , where the x - ray emission is faint , such measurements become challenging . recent advances in this regard have been made with the _ suzaku _ satellite , and opportunities for additional progress remain ( section [ sec : future_outskirts ] ) . potentially increased levels of non - thermal pressure support ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and gas clumping @xcite can also complicate measurements at large radii . a number of approaches have been used in implementing the hydrostatic method . the most common , which employs relatively strong priors , uses parameterized fits to the observed , projected surface brightness and temperature profiles ; these are then used to calculate the appropriate partial derivatives at each radius to determine the mass profile ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? a second , arguably preferable , approach employs a non - parametric deprojection of the brightness and temperature data , but assumes that the mass distribution follows a well - motivated parameterized form ( e.g. equation [ eq : nfw ] ; @xcite , @xcite ) ; this approach simultaneously provides a framework for testing the validity of various mass models . in the case of very high quality x - ray data , a fully non - parametric deprojection of the surface brightness and temperature data can be employed , without additional , regularizing assumptions ( e.g. * ? ? ? x - ray mass measurements are relatively insensitive to triaxiality @xcite . for dynamically relaxed clusters , and for measurements out to intermediate radii , simulations indicate that hydrostatic x - ray masses should exhibit modest scatter ( @xmath286% ) and be biased low by @xmath287% ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , due primarily to kinetic pressure arising from residual gas motions . like the x - ray method , optical - dynamical mass measurements are based on the assumption of dynamical equilibrium , with the galaxies used as test particles in the cluster . the mass enclosed within radius @xmath106 is given by the jeans equation ( e.g. * ? ? ? * ; * ? ? ? * ) @xmath288 , \label{eq : jeans}\ ] ] where @xmath289 is the galaxy number density , @xmath290 the 3-dimensional velocity dispersion , and @xmath291 the velocity anisotropy parameter . these quantities can be determined under model assumptions from the projected galaxy number density and velocity dispersion profiles . an advantage of the optical dynamical method over the x - ray method is that it is insensitive to several forms of non - thermal pressure support that affect x - ray mass measurements ( e.g. magnetic fields , turbulence , and cosmic ray pressure ) . the galaxy population can also be observed at high contrast out to large radii . however , where the x - ray gas is a collisional fluid that returns rapidly to equilibrium following a disruption , the galaxies are collisionless and relax on a longer timescale @xcite . whereas x - ray mass measurements are relatively insensitive to triaxiality , the galaxy velocity anisotropy must be accounted for . the precision of optical dynamical measurements is also limited by the finite number of galaxies . while identifying the center of a cluster is straightforward at x - ray wavelengths , at optical wavelengths this can be a source of significant uncertainty . the infall regions of clusters form a characteristic trumpet - shaped pattern in radius - redshift phase - space diagrams , the edges of which are termed caustics . the identification of these caustics enables mass measurements out to larger radii ( up to @xmath292 mpc ) , to an accuracy of @xmath293% ( e.g. * ? ? ? in contrast to x - ray and optical dynamical methods , gravitational lensing offers a way to measure the masses of clusters that is free of assumptions regarding the dynamical state of the gravitating matter @xcite . weak lensing methods have an important role in cosmological work : while triaxiality is expected to introduce scatter in individual ( deprojected ) mass measurements at the level of tens of per cent @xcite , for statistical samples of clusters and using suitable mass estimators , working over optimized radial ranges and with good knowledge of the redshift distribution of the background population , weak lensing measurements are expected to provide almost unbiased results on the mean mass @xcite . the most common technique employed in weak lensing mass measurements is fitting the observed , azimuthally - averaged gravitational shear profile with a simple parameterized mass model ( e.g. * ? ? ? stacking analyses of clusters detected in survey fields have also proved successful in calibrating the mean mass observable scaling relations down to relatively low masses ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? strong lensing enables precise measurements of the projected masses through regions enclosed by gravitational arcs . in combination with weak lensing , strong lensing constraints can improve significantly the absolute calibration of projected mass maps ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? . deprojected strong lensing measurements are particularly sensitive to triaxiality ( e.g. * ? ? ? * ; * ? ? ? nonetheless , detailed strong lensing studies of small samples of highly relaxed clusters have reported mass measurements in good agreement with x - ray results ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? some recent works have used clusters with multiple systems of strongly - lensed arcs to constrain the geometry of the universe ( * ? ? ? * and references therein ) . this approach is related to the shear ratio test also discussed for weak lensing measurements with clusters @xcite . a good mass proxy should be straightforward to measure and correlate tightly with mass , exhibiting minimal dispersion across the mass and redshift range of interest . robust , low - scatter mass proxy information for just a small fraction ( appropriately selected ) of the clusters in a survey can boost significantly its constraining power with respect to self - calibration alone ( e.g. * ? ? ? * ; * ? ? ? the total x - ray luminosity has an intrinsic dispersion at fixed mass of @xmath269 @xcite . the scatter in optical richness at fixed mass for the maxbcg catalog is also @xmath294 , with a modest inferred non - gaussian tail toward low masses @xcite . the scatter in the projected , integrated sz flux at fixed mass is expected to be somewhat smaller ( @xmath295% ; @xcite ) , although this is yet to be measured robustly from data . ( note that the scatter in the observed , projected sz flux is significantly larger than for the predicted , intrinsic signal ; figure [ fig : stanek10 ] . ) since total x - ray luminosity , optical richness and integrated sz flux are all survey observables , their scatter versus mass tends to play a significant role in tests based on cluster counts even when more precise mass measurements for a fraction of the clusters are available from follow - up data . although not the basis for cluster surveys , other observables that correlate tightly with mass can provide powerful mass proxies . for the most massive clusters , the x - ray emitting gas mass is strongly correlated with total mass , with an observed scatter @xmath296 at fixed mass @xcite . the x - ray temperature , and the product of gas mass and temperature ( @xmath297 ) , have observed scatters @xmath298 ( e.g. @xcite ) ; it is thought that the tightness of the @xmath299@xmath71 relation may extend to lower masses than for either the @xmath209@xmath71 or @xmath210@xmath71 relations @xcite . center - excised x - ray luminosity also traces mass extremely well , with an observed scatter @xmath296 ( @xcite ; see section [ sec : future : coreevol ] ) . weak lensing mass measurements exhibit larger scatter ( tens of per cent ) . however , the minimal predicted bias in the mean mass for statistical samples of lensing measurements ( section [ sec : obs_masses_lensing ] ) makes the combination with x - ray measurements particularly promising . while the relation between the three - dimensional dark matter velocity dispersion and mass is predicted to be tight @xcite , velocity anisotropy and projection effects cause the one - dimensional velocity dispersion to exhibit larger scatter ( @xmath300% , implying a scatter in mass at fixed velocity dispersion of @xmath269 ; @xcite ) . the relationship between galaxy and dark matter velocity dispersion , the velocity bias , also remains uncertain . the past decade has seen marked improvements in the data and analysis techniques employed in cluster cosmology . _ chandra _ observations of relaxed clusters provided important measurements of the distance scale , confirming the recent acceleration of the cosmic expansion . improvements in mass measurements led to a convergence in estimates of @xmath58 from cluster counts , and x - ray surveys extending to @xmath301 provided the first constraints on dark energy from the growth of structure . we begin this section with a review of the latest results from measurements of the number and growth of clusters , as well as the closely related problem of constraining scaling relations . we then review the state of other cluster - based probes of cosmology . we conclude by summarizing the latest constraints on dark energy from the combination of galaxy cluster data with other , independent probes . the application of cluster measurements to other areas of fundamental physics are discussed in section [ sec : physics ] . because the cosmological model influences predictions for scaling relation observations , and vice versa ( section [ sec : theory : selection ] ) , the two must be constrained simultaneously . the statistical framework needed for this approach , in which a subset of detected clusters are targeted for more detailed observations including mass estimation , is described by @xcite . without reproducing the details here , we enumerate the components of the statistical model : 1 . the mass function and expansion history , which together predict the number of clusters as a function of mass and redshift , @xmath302 . 2 . stochastic scaling relations which describe the ( multivariate ) distribution of observables as a function of mass and redshift . measurements for each cluster and associated sampling distributions , describing the probability of obtaining particular measured values given true cluster properties ( i.e. a model for the measurement errors ) , accounting for any covariances in measured quantities . we note that all quantities entering the scaling relations need not be measured for every cluster , apart from ( necessarily ) those determining cluster selection ( e.g. , mass follow - up need not be complete ) . 4 . the selection function , quantifying the probability of a cluster being detected and included in the sample . using the likelihood function corresponding to this formalism ( see @xcite ) , constraints on cosmological and scaling relation parameters are obtained by marginalizing over the true , unknowable properties of each cluster , constrained by the measured values . straightforward adaptations of this approach , for example incorporating a cluster bias model ( section [ sec : theory : countsclustering ] ) or stacked gravitational lensing data for poor clusters and groups , are possible . we concentrate our review on three recent , independent results based on measurements of the local abundance and evolution of clusters from x - ray and optical surveys , as well as preliminary results from new sz surveys . @xcite studied the bcs @xcite , reflex @xcite and bright macs @xcite cluster samples compiled from the rass . this survey strategy , covering a large fraction of the sky to relatively shallow depth , is optimized to the task of finding the largest clusters at the expense of depth in redshift . the mantz et al . data set consists of 238 clusters with masses @xmath303 distributed over the redshift range @xmath304 . pointed _ chandra _ or _ rosat _ follow - up observations were used to measure @xmath210 for 94 clusters , while the remaining 144 had measurements only of the cluster redshift and survey flux . the @xmath305@xmath306 relation was marginalized over using hydrostatic mass estimates for relaxed clusters from @xcite . the data set was used to simultaneously constrain cosmological parameters and the @xmath208@xmath71 and @xmath209@xmath71 relations using the method described above and detailed in @xcite . @xcite also used x - ray selected clusters , but pursued a different survey strategy . their data set consists of disjoint low - redshift and high - redshift samples , with 49 clusters ( originally culled from several rass samples ) at redshifts @xmath307 ( nearly all @xmath308 ) and an additional 36 clusters serendipitously detected in the 400 square degree survey @xcite at @xmath309 . this survey strategy , covering a smaller area of the sky to greater depth , naturally finds fewer of the most massive clusters , but extends to lower masses ( @xmath310 ) and higher redshifts . all 85 clusters in the full data set were followed up with _ chandra _ , and their masses were estimated using either @xmath210 or @xmath299 as a proxy ; the proxy mass relations were calibrated using hydrostatic mass estimates for a sample of well observed , low - redshift clusters . the analysis includes empirical constraints on the scaling relations and corrections for selection effects , though not using exactly the approach described in section [ sec : constraints : procedure ] . the top panels of figure [ fig : constraints : mfcn_fgas ] provide a simplified visualization of how constraints on dark energy arise from these data , comparing the observed mass function in two redshift ranges to model predictions for different cosmological models . in particular , an open universe with no dark energy clearly under - predicts the evolution of the mass function over the redshift range of the data . = 5.63 cm = 5.6 cm the optically selected maxbcg sample @xcite employed by @xcite probes a different part of the cluster population ; it is restricted to lower redshifts than the x - ray samples described above ( @xmath277 ) , but extends to lower masses ( @xmath311 ) . this lower effective mass limit , which changes less strongly with redshift compared to x - ray surveys , makes the maxbcg sample significantly larger than the others , with @xmath312 clusters divided into 9 bins based on optical richness . mean masses for 5 richness ranges were estimated through a weak gravitational lensing analysis of stacked clusters , providing information from which to constrain the richness mass relation . the cosmological analysis accounts for the covariance between cluster counts in each richness bin and the mean lensing mass estimates . the results obtained by these three groups on flat @xmath8cdm and constant @xmath313 models are summarized in table [ tab : constraints : basic ] . note that , for the two works which fit @xmath313 models , the results on @xmath169 and @xmath58 are dominated by the low - redshift data and so are not degraded noticeably by the introduction of @xmath313 as a free parameter ; thus all three sets of constraints are directly comparable . the agreement between the different works , as well as others listed in table [ tab : constraints : basic ] , is encouraging ; in particular , the close agreement in the constraints on @xmath58 reflects the relatively recent convergence in cluster mass estimates using different techniques , and our improved understanding of the relevant systematics ( section [ sec : obs_masses ] ; see also , e.g. , @xcite ) . importantly , the concordance @xmath8cdm model provides an acceptable fit to the data in each case . [ tab : constraints : basic ] ccccccc + reference@xmath314 & & @xmath58 & @xmath315 & @xmath316 & @xmath313 & @xmath53 + + + m10 & x - ray & @xmath317 & @xmath318 & @xmath319 & @xmath320 & + v09 & x - ray & @xmath321 & @xmath322 & @xmath319 & @xmath323 & + + r10 & optical & @xmath324 & @xmath325 & @xmath319 & @xmath47 & + h09 & x - ray & @xmath326 & @xmath112 & @xmath319 & @xmath47 & + + s03 & x - ray & @xmath327 & @xmath328 & @xmath319 & @xmath47 & + + a08 & x - ray & & @xmath329 & @xmath330 & @xmath47 & + a08 & x - ray & & @xmath331 & @xmath319 & @xmath332 & + e09 & x - ray & & @xmath333 & @xmath319 & @xmath334 & + l06 & x - ray+sz & & @xmath335 & @xmath319 & @xmath47 & + + b06 & x - ray+sz & & @xmath112 & @xmath319 & @xmath47 & @xmath336 + s04 & x - ray+sz & & @xmath112 & @xmath319 & @xmath47 & @xmath337 + + @xmath60 entries @xmath338 indicate the assumption of global spatial flatness ( or use of _ wmap _ priors requiring flatness at the few percent level ) ; other entries without error bars indicate parameters that were fixed in the corresponding analysis . @xmath339 error bars are marginalized ( single - parameter ) 68.3% confidence intervals , and include each author s estimate of the systematic uncertainties ( with the exception of s04 ) . @xmath314 a08 = @xcite ; b06 = @xcite ; e09 = @xcite ; h09 = @xcite ; l06 = @xcite ; m10 = @xcite ; r10 = @xcite ; s03 = @xcite ; s04 = @xcite ; v09 = @xcite . @xmath340 cluster surveys extending to redshifts @xmath341 are required to constrain @xmath313 from the evolution of the mass function . note that the @xmath315 and @xmath58 constraints from these works are essentially unchanged , whether or not @xmath313 is allowed to differ from @xmath47 . figure [ fig : constraints : rozo_wm_s8 ] ( left panel ) shows the joint constraints on @xmath169 and @xmath58 obtained by ( * ? ? ? * solid lines ) , which display the typical degeneracy between those parameters from cluster survey data . ( the degeneracy can be broken , for example , by including cluster @xmath342 data ; see section [ sec : constraints : fgas ] . ) also shown are results from 5 years of _ wmap _ data ( * ? ? ? * dashed lines ) , which are tight for the assumed flat @xmath8cdm model . nevertheless , it is evident that the combination of the two types of data ( shaded regions ) is significantly improved over either one individually . = 5.5 cm the constraints on constant @xmath313 models from @xcite and @xcite are respectively shown in figures [ fig : constraints : rozo_wm_s8 ] ( right panel ) and [ fig : constraints : mantz_w ] , along with results from various other cosmological data sets . the cluster results are in good agreement with one another , as well as with the other , independent data . the @xmath343 precision constraint on @xmath313 from cluster growth alone ( including systematic uncertainties ) is clearly competitive , and further constraining power may be available from theoretical advances ( see section [ sec : physics : lcdm ] ) . the strong degeneracy in cmb constraints evident in figure [ fig : constraints : mantz_w ] also arises in many other models that are more complex than flat @xmath8cdm ; the ability of cluster data to break this degeneracy by providing precise and independent constraints on @xmath58 ( right panel of figure [ fig : constraints : mantz_w ] ) makes the combination particularly powerful ( e.g. section [ sec : physics : nu ] ) . although their respective surveys are not yet complete , preliminary results have been reported from sz - detected clusters from the south pole telescope ( spt ; * ? ? ? * ) and atacama cosmology telescope ( act ; * ? ? ? as the number of cluster detections used in these works is small , respectively 21 and 9 at redshifts @xmath344 , the data are not capable of producing interesting constraints on their own , but both groups demonstrate consistency with results from the _ wmap _ satellite @xcite . neither group has yet obtained simultaneous constraints on cosmology and the relevant sz scaling relation , making their final error budgets strongly dependent on the priors chosen to constrain the scaling relation . to obtain scaling relations appropriate for constraining cosmology , the analysis of the two must be simultaneous and must account properly for the influence of the survey selection function ( e.g. @xcite and @xcite ) and the cluster mass function , as in ( * ? ? ? * see also sections [ sec : constraints : procedure ] and section [ sec : theory : selection ] ) . some authors have included corrections for the expected malmquist bias given a flux limit or selection function @xcite , but most analyses of scaling relations in the literature employs least - squares regression without detailed consideration of sample selection . since various cluster samples , with different selection functions , have been used , it is not surprising that results on the slopes , scatters and evolution of the scaling relations have varied widely compared with the formal uncertainties . another statistical issue is that the covariance that comes about when the scaling quantities are measured from the same observations , for example temperature and hydrostatic masses from x - ray data , is commonly ignored ( see section [ sec : modeling ] ) . the method used to estimate masses has also varied . most authors have used the assumption of hydrostatic equilibrium applied to x - ray data , regardless of the dynamical state of the cluster , which must introduce spurious scatter due to departures from equilibrium and non - thermal support ( e.g. @xcite ) . more recently , mass proxies such as gas mass ( e.g. @xcite ) and x - ray thermal energy ( @xmath299 ; e.g. @xcite ) , or gravitational lensing signal ( e.g. @xcite ) have been employed . these variations in mass estimation and analysis methods , in addition to changes to instrument calibration over the years , makes a comprehensive and fair census of scaling relation results problematic . here we focus on the cosmological importance of scaling relation measurements , citing examples from recent work where the issues mentioned above are at least partially mitigated . for x - ray and sz observables , under the assumption of strict self - similarity ( no additional heating or cooling ) , @xcite derived specific slopes and redshift dependences for the power - law form of equation [ eq : sbarmu ] : @xmath345^{4/3 } , \nonumber \\ { \ensuremath{kt_{\mathrm{mw } } } } & \propto & \left [ e(z)m \right]^{2/3 } , \nonumber \\ e(z)\,y & \propto & \left [ e(z)m \right]^{5/3},\end{aligned}\ ] ] where the factors of @xmath346 are appropriate for measurements made at a fixed critical - overdensity radius . the subscripts ` bol ' and ` mw ' reflect the fact that these predictions apply to the bolometric luminosity and mass - weighted temperature . optical richness is more complex to predict , but empirical studies that map galaxies to sub - halos in simulations support a power law richness mass relation for groups and clusters @xcite . figures [ fig : constraints : scaling ] and [ fig : constraints : other_scaling ] show a few examples of recent scaling relation measurements . @xcite present a @xmath71@xmath208 relation for x - ray selected clusters in the cosmos field by measuring stacked weak lensing masses ( left panel of figure [ fig : constraints : scaling ] ) . under the common assumption of symmetric scatter in the log , stacking on @xmath208 allows the mean @xmath347 relation to be recovered , at the cost of losing information about the intrinsic scatter . as the authors note , transforming these results to an @xmath348 relation introduces a dependence on the mass function . stacked lensing was also used to determine the mass richness relation of optically selected clusters in sdss by @xcite . their results are shown in the right panel of figure [ fig : constraints : scaling ] ; red points in the figure show that compatible masses were derived from galaxy velocity dispersion measurements . = 5.8 cm = 5.8 cm for more massive systems , lensing can provide mass estimates on a cluster - by - cluster basis . the left panel of figure [ fig : constraints : other_scaling ] shows such a mass temperature relation from @xcite for x - ray selected clusters . since temperature only weakly influences x - ray detectability , selection bias should be relatively unimportant here , although intrinsic correlation with luminosity can still produces subtle biases relative to a mass - limited sample ( section [ sec : theory : selection ] ) . = 5.9 cm an sz scaling relation from @xcite , using clusters detected by spt , appears in the right panel of figure [ fig : constraints : other_scaling ] ; in this case , masses are estimated from measurements of the x - ray thermal energy , @xmath299 , and an assumed @xmath299@xmath71 relation . this case is instructive in that the plotted @xmath205 values are derived from the survey data , with suggestions of malmquist bias in the flattening observed at the lowest masses ( compare with figure [ fig : constraints : selection ] ; this potential bias is accounted for in the fit in figure [ fig : constraints : other_scaling ] ) . recently , the _ planck _ collaboration released an early sz - selected cluster catalog together with an analysis of sz scalings for existing x - ray and optical cluster samples ( * ? ? ? * and references therein ) . measurements of the scaled sz signal from _ planck _ as a function of x - ray luminosity for @xmath349 x - ray selected galaxy clusters are shown in the left panel of figure [ fig : constraints : planck_scaling ] , along with the stacked , mean sz signal in luminosity bins and model expectations derived from the observed x - ray properties @xcite . there is good agreement between the model expectations and the _ planck _ observations , validating the simple model description of the hot icm over the mass and redshift range probed . = 7.5 cm = 7.5 cm the right panel of figure [ fig : constraints : planck_scaling ] shows the result of a similar exercise , stacking the _ planck _ sz signal as a function of richness for the optically selected maxbcg catalog . also shown is the expected signal , based on the observed mean mass richness relation for the maxbcg sample @xcite and a standard x - ray @xmath205@xmath71 scaling relation @xcite . in this case , the observed _ planck _ sz signal lies well below model predictions . for an x - ray detected sub - sample of the maxbcg catalog , however , relatively good agreement between the _ planck _ sz measurements and model predictions is observed @xcite . an important difference between the @xmath208 and richness comparisons is that the centers and radii used to extract the clusters sz signals from the ( relatively low resolution ) _ planck _ data were respectively based on x - ray and optical estimates . besides optical center and scale mis - estimates , other potential sources for the discrepancy in figure [ fig : constraints : planck_scaling]b include lower than expected purity in the optical sample ( with lower mass halos being enhanced in richness by projection ) ; unmodeled biases in the scaling relations used ( e.g. the impact of non - thermal pressure support on the x - ray relations ) ; and the presence of an intrinsic anti - correlation between the galactic and hot gas mass fractions in the clusters . these and other possibilities are under investigation . in general , most current scaling results are consistent with power - law relations over a wide range in mass ( e.g. @xcite ) . however , the exponents often differ from the values in equation [ eq : constraints : virial_scaling ] , due at least in part to the impact of astrophysical processes , mainly star / smbh formation and associated feedback , discussed in section [ sec : theory_astrophysics ] . for example , the @xmath208@xmath71 slope is commonly found to exceed @xmath350 , with the value depending on the mass range explored . there is less consensus on the @xmath209@xmath71 slope , with most estimates consistent with self - similarity ( e.g. @xcite ; @xmath351 ) , but some finding a shallower slope ( e.g. @xcite ; @xmath352 ) ; systematics related to correlated observables , mass estimation ( see section [ sec : modeling ] ) , instrument calibration , selection , and differences in the mass range studied may play a role . there are fewer empirical estimates of the @xmath205@xmath71 or @xmath299@xmath71 slopes , but most are broadly compatible with the predicted value . we note that departure from self - similar scaling with mass does not necessarily imply departure from self - similar evolution with redshift ; for example , the high - mass halos in the preheated simulations of @xcite have x - ray luminosities that evolve within @xmath353 of the self - similar expectation at @xmath354 , despite having a slope with mass that is @xmath355 steeper than self - similar . the marginal intrinsic scatter in each relation determines the degree of selection bias in surveys and each observable s usefulness as a mass proxy ; current results are reviewed in section [ sec : obs_masses_proxies ] . the evolution of the scaling relations is of great importance , since it is potentially degenerate with the cosmological signal of cluster growth . because the impact of survey limits can vary strongly with redshift for x - ray surveys , it is particularly crucial to account for them in this context . both @xcite and @xcite investigate evolution in the @xmath208@xmath71 relation , finding only marginal @xmath356 departures from the self - similar redshift dependence in equation [ eq : constraints : virial_scaling ] . @xcite additionally found no evidence for departures from self - similar evolution in the @xmath209@xmath71 or center - excised @xmath208@xmath71 relations ( see section [ sec : future : coreevol ] ) . as current and upcoming surveys expand the reach of cluster samples to @xmath357 , obtaining precise constraints on scaling relation evolution will be imperative ( see also section [ sec : modeling ] ) . the effects of survey bias on scaling relations can be _ partially _ mitigated by selecting clusters based on an observable other than the observable of immediate interest ( e.g. @xcite ) . however , as sections [ sec : theory : mvscaling ] and [ sec : theory : selection ] make clear , the amount of residual bias depends on the intrinsic correlation between the two observables at fixed mass . to date , the only empirical estimates of such correlation are for mass and x - ray luminosity at fixed optical richness , @xmath358 @xcite ; and for soft - band x - ray luminosity and temperature at fixed mass , @xmath359 @xcite . large , overlapping surveys , and/or surveys coupled with multi - wavelength follow - up campaigns , are required to better constrain the property covariance of clusters ; ultimately , such constraints will provide a new level of robustness to cosmological work . in principle , the confirmed existence of even a single galaxy cluster of implausibly high mass would challenge the standard @xmath8cdm model with gaussian initial conditions . recently , the discovery of high redshift , massive systems such as xmmuj2235.3 - 2557 at @xmath360 @xcite and spt - clj0546 - 5345 at @xmath361 @xcite have led to reports of possible tension with gaussian @xmath8cdm @xcite . such a test is in some sense an attractively simple alternative to the more involved work discussed above . however , despite the fact that the test involves only one or a few very massive clusters detected in a survey rather than the complete sample , a robust assessment of the likelihood still requires a detailed understanding of the selection function and survey biases , as well as a full accounting for the effects of scatter in the mass observable scaling relations ( sections [ sec : theory : selection ] and [ sec : constraints : procedure ] ) . the accuracy and precision of the cluster mass measurements are also critical , due to the steepness of the high mass tail of the cluster mass function . errors in mass measurements at the tens of per cent level , for example ( as might be expected for weak lensing measurements of an individual cluster ) , can modify the likelihood of such a cluster being observed by up to an order of magnitude . @xcite estimate confidence limits for the exclusion of the gaussian @xmath8cdm model based on the properties of the most massive galaxy cluster , or @xmath156 most massive galaxy clusters , detected in a given survey , employing constraints on the expansion history from current data . these authors conclude that none of the presently known high mass , high redshift clusters are in significant tension with the standard gaussian @xmath8cdm paradigm . the framework described in section [ sec : constraints : procedure ] for simultaneously constraining cosmology and scaling relations has yet to be applied to the spatial clustering of clusters . however , @xcite have obtained cosmological constraints using a method based on the karhunen - love eigenvectors of 428 clusters from the x - ray selected reflex sample above a luminosity of @xmath362 ( see @xcite for the theory underlying this method of estimating the power spectrum ) . the constraints are relatively weak compared with more recent results from other methods : @xmath363 and @xmath364 at 95.4% confidence , fixing @xmath365 and without including the systematic uncertainty due to cosmic variance . in @xcite , the analysis was extended to include cluster abundance using an empirical x - ray luminosity mass scaling relation from @xcite , demonstrating that the combination of the two methods breaks the degeneracy between @xmath169 and @xmath58 , in particular reducing the uncertainty on @xmath58 to the point where systematics related to mass estimation dominate ( table [ tab : constraints : basic ] ) . in principle , the spatial distribution of clusters can be employed in a simpler way , as has been done with individual galaxies , by using the baryon acoustic oscillation signature in the power spectrum as a probe of cosmic distance . @xcite and @xcite respectively analyzed the correlation function and power spectrum of clusters in the optically selected maxbcg catalog , finding a weak ( @xmath366 ) detection of the bao peak . more recently , @xcite found no significant evidence for a bao feature in the smaller , x - ray selected reflex ii catalog . cluster gas mass fractions can be measured robustly using x - ray , or the combination of x - ray and sz , data for dynamically relaxed clusters ( section [ sec : observations ] ) . when measured from x - ray data , @xmath342 values within a given angular aperture depend on cosmology as @xmath367 , while the predicted @xmath342 for a given cosmology is given by equation [ eq : constraints : fgas_model ] . ( a more detailed expression for the comparison of measured and predicted @xmath342 values is given by @xcite , who include terms accounting for instrument calibration , non - thermal pressure , and the relationship between the characteristic radius of the model , e.g. @xmath368 , and the aperture of the measurement . ) the apparent redshift dependence of @xmath342 measurements on the cosmological background is illustrated in the bottom panels of figure [ fig : constraints : mfcn_fgas ] . in principle , @xmath342 measurements can be made at radii corresponding to any overdensity . in practice , this overdensity should be low enough ( i.e. the radius large enough ) that non - gravitational feedback effects do not introduce prohibitive scatter ; but not so small ( i.e. the radius not so large ) that the measurements become dominated by the systematic limitations of the instruments . a variety of simulations indicate that radii @xmath369 are sufficiently large to benefit from low @xmath342 scatter ( e.g. @xcite ) , while at radii @xmath370 uncertainties in the x - ray background and the impact of gas clumping can become a concern ( @xcite ; see also section [ sec : future_outskirts ] ) . using this method , ( * ? ? ? * see also @xcite ) obtained cosmological constraints using @xmath342 measurements at @xmath368 from _ chandra _ observations of 42 hot ( @xmath371 ) , relaxed clusters covering the redshift range @xmath372 . their analysis incorporated weak priors on @xmath53 and @xmath373 from hubble key project @xcite and big bang nucleosynthesis data , priors on the baryonic depletion and stellar mass fraction of clusters and their evolution ( the term @xmath139 in equation [ eq : constraints : fgas_model ] ) , and marginalized over systematic allowances accounting for instrument calibration and non - thermal pressure . from equation [ eq : constraints : fgas_model ] , we see that the normalization of the @xmath374 curve , combined with the priors on @xmath53 and @xmath373 , provides a constraint on @xmath169 ; while the shape of @xmath374 allows dark energy parameters to be constrained via the apparent dependence of @xmath342 on distance . the results for spatially flat , constant @xmath313 models and non - flat @xmath8cdm models are respectively shown in figures [ fig : constraints : mantz_w ] and [ fig : constraints : allen_wm_wl ] . for the non - flat models , the presence of dark energy is detected at high ( @xmath375 ) confidence , comparable to current snia results ; the constraints , including systematic uncertainties , are @xmath376 and @xmath377 , with the flat @xmath8cdm model yielding an acceptable goodness of fit . for constant @xmath313 models , the @xmath342 data are again competitive with other cosmological results , obtaining @xmath378 . the systematic , cluster - to - cluster scatter in @xmath342 is small , @xmath379 , corresponding to only 5% in distance ; this high precision results from the restriction to hot , dynamically relaxed systems for which total masses can be accurately estimated ( section [ sec : obs_masses_xray ] ) . interestingly , for clusters with @xmath380 , the measured @xmath342 values show no dependence on temperature ( right panel of figure [ fig : constraints : allen_wm_wl ] ) , indicating that , for the most massive clusters , the self - similar expectation of constant @xmath342 with mass is realized . at lower temperature and mass ( extending to the group scale ) , a trend of increasing @xmath342 with temperature and mass is observed ( e.g. @xcite ) . = 5.5 cm = 5.5 cm @xcite have also applied x - ray data to the @xmath342 test , using _ measurements at @xmath381 for 52 clusters in the redshift range @xmath382 , adopting similar priors on @xmath53 and @xmath373 and marginalizing over a prior on @xmath383 ( assumed constant with redshift ) . however , their data set was not restricted to dynamically relaxed systems , resulting in significantly weaker constraints ( table [ tab : constraints : basic ] ) . @xcite employed _ chandra _ x - ray and ovro / bima sz observations of 38 clusters in the redshift range @xmath384 ( with no restriction dynamically relaxed systems ) , finding @xmath342 values consistent with previous x - ray work . ( their analysis did not take advantage of the relative normalization of the x - ray and sz signals to simultaneously provide a second distance constraint ; see section [ sec : constraints : xsz ] . ) they adopted the simpler approach of assuming constant @xmath342 and marginalizing over its value , incorporating a _ wmap _ prior on the total density , @xmath385 . although this explicitly ignores the information available from the normalization of @xmath374 , their results clearly disfavor a dark matter dominated universe , preferring a low - density universe with dark energy ( table [ tab : constraints : basic ] ) . the different dependence on distance of the gas density inferred from x - ray and sz observations of clusters can be exploited in a conceptually similar way to @xmath342 data ( section [ sec : theory : xsz ] ) . the most recent contribution is that of @xcite , who measured distances to 38 clusters at redshifts @xmath384 . this cosmological test is intrinsically less sensitive to distance than the @xmath342 test , with the signal proportional only to @xmath386 ( equation [ eq : daxraysz ] ) , and currently can constrain only one free parameter . assuming spatial flatness and fixing @xmath387 , @xcite obtained a constraint on the hubble parameter , @xmath388 , consistent with results from other data such as the hubble key project @xcite or the combination of @xmath342 and cmb data @xcite . we note that other works using the same method have typically found somewhat lower best fitting values ( @xmath3890.7 ; e.g. @xcite ) , but these discrepancies are not significant given the systematic uncertainties . the cmb temperature power spectrum at multipoles @xmath390 encodes the thermal sz signature of unresolved clusters at all masses and redshifts ( section [ sec : theory : tsz ] ) . although the primary cmb power decreases rapidly at these scales , extracting this cosmological information from the tsz spectrum has proved challenging due to uncertainties in , e.g. , the relevant observable mass scaling relation at low masses and high redshifts ; the population of infrared and radio point sources ; the magnitude of the integrated kinetic sz effect ; and the form of the electron pressure profile at large cluster radii , where it is poorly constrained by current x - ray data ( @xcite ; section [ sec : future_outskirts ] ) . the _ wmap _ , spt and act collaborations have all detected excess power at large multipoles , @xmath391 . subject to the systematic uncertainties mentioned , their results are broadly in agreement , and are consistent with estimates of @xmath58 obtained from studies of resolved clusters and the primary cmb ( @xcite ) . as discussed above , current cluster growth and @xmath342 data can constrain spatially flat models with constant @xmath313 , finding consistency with the cosmological constant model ( @xmath392 ) . constraints on constant @xmath313 models from the combination of these cluster data are shown in the left panel of figure [ fig : physics : wevol ] . to go beyond this simple description of dark energy , it is necessary to include cosmological data from additional sources in the analysis . equation [ eq : wwa ] provides a simple and commonly adopted model of evolving dark energy , in which the equation of state takes the value @xmath393 at @xmath110 and approaches @xmath394 at high redshift . constraints on this model were obtained from cluster growth data by @xcite and @xcite , assuming spatial flatness and in combination with external cmb , snia and bao ( and @xmath342 , in the case of @xcite ) data . in both studies , the results are consistent with the @xmath8cdm model ( @xmath395 and @xmath396 ; right panel of figure [ fig : physics : wevol ] ) . a slightly more general model due to @xcite , @xmath397 also makes a smooth transition from one value at the present day ( @xmath393 ) to another at early times ( @xmath398 ) , but has the advantage that the transition redshift , @xmath399 , can be marginalized over . ( when @xmath400 , equation [ eq : physics : lcdm_wetmodel ] reduces to equation [ eq : wwa ] with @xmath401 . ) @xcite used the data sets above to obtain @xmath402 and @xmath403 , again consistent with @xmath8cdm . absent a concrete physical model for dark energy , phenomenological models of evolving dark energy can take any form ; for example , one possibility that has not yet been investigated using cluster data expands @xmath404 in principal components ( e.g. @xcite ) . although these ad - hoc descriptions of dark energy are perfectly straightforward to apply to measurements of the expansion history , their applicability to an analysis based on the cluster mass function or power spectrum is less clear . generically , descriptions of dark energy as a fluid with @xmath405 should include the effects of spatial variations in dark energy density ( e.g. @xcite ) . however , the results from cluster growth so far have made use of mass functions from @xmath8cdm simulations , accounting for the value of @xmath313 only in the expansion history ( @xcite ) or , at most , including the effect of density variations on the linear matter power spectrum ( @xcite ) . in principle , the cluster mass function must also be adjusted to account for the behavior of fluid dark energy on smaller scales and in higher density environments . encouragingly , preliminary work in this area suggests that the influence of dark energy on the mass function might be readily measurable , resulting in additional constraining power from clusters ( e.g. @xcite ) . such improvements , however , might come at the cost of requiring additional sophistication in the theoretical description of dark energy ( e.g. the dark energy sound speed and viscosity ; @xcite ) . the @xmath8cdm paradigm , while providing an excellent model for the large scale structure of the universe , incorporates little information on the physical nature of dark matter . it assumes only that dark matter is non - baryonic ; that it interacts weakly with baryonic matter and itself ; that it emits and absorbs no detectable electromagnetic radiation ; and that the dark matter particles move at sub - relativistic speeds . as clusters merge under the pull of gravity , their dark matter halos and x - ray emitting gas can become separated temporarily , as the gas experiences ram pressure and is slowed . the observed offsets between the dark matter and x - ray peaks in the bullet cluster ( 1e0657 - 558 ; @xcite ) and macsj0025.4 - 1222 ( @xcite ; figure [ fig : bullet_macs0025 ] ) , both massive merging systems with relatively simple geometries , require conservatively that the scattering depth for the merging dark matter can not be greater than one . using gravitational lensing data to estimate the dark matter column densities through these clusters , @xcite and @xcite use the observed dark matter and x - ray peak separations to derive limits on the velocity - independent dark matter self - interaction cross - section per unit mass of @xmath406@xmath407g@xmath408 and @xmath409@xmath407g@xmath408 for 1e0657 - 558 and macsj0025.4 - 1222 , respectively . @xcite additionally use the non - detection of an offset between the lensing peaks and the galaxy centroids for the bullet cluster to refine this constraint to @xmath410@xmath407g@xmath408 . upcoming surveys ( section [ sec : new_surveys ] ) , should provide hundreds of similar examples , removing the current systematic limitations set by small number statistics . in combination with multiwavelength follow - up observations and improved numerical simulations ( e.g. * ? ? ? * ) , this should allow the properties of dark matter in merging clusters to be studied in a robust , statistical manner . one of the most remarkable predictions of the cdm model is that the density profiles of relaxed dark matter halos , on all resolvable mass scales , can be approximated by a simple , universal profile ( equation [ eq : nfw ] ) with an inner , density slope @xmath411 . in contrast , for dark matter models with significant self - interaction cross sections , halos are expected to exhibit flattened , quasi - isothermal cores @xcite . in the absence of significant rotational support , these cores are also expected to be approximately spherical . using chandra x - ray data for a sample of massive , dynamically relaxed galaxy clusters , @xcite measure a mean central density slope of @xmath412 ( 95 per cent confidence limits ) , in good agreement with @xmath8cdm . detailed strong - plus - weak lensing analyses for a subset of these systems yields consistent constraints @xcite . on smaller scales , using stellar velocity dispersion profiles for the dominant cluster galaxies , @xcite ( see also @xcite ) infer a possible flattening of the central dark matter halos in abell 383 and ms2137.3 - 2353 . however , on these small scales , the impact of baryonic physics becomes important . @xcite use the lack of a dark matter core in x - ray and lensing data for the relaxed cluster ms1358 + 6245 , to place a limit on the velocity independent dark matter particle - scattering cross section @xmath413cm@xmath272g@xmath408 . @xcite use constraints on the ellipticity in the central regions of ms2137.3 - 2353 to infer @xmath415cm@xmath272g@xmath408 . to improve these constraints , combined multiwavelength observations for large samples of relaxed clusters , coupled with improved simulations modeling fully the interactions between dark matter and baryons , are required . certain dark matter candidates , including sterile neutrinos , possess a two - body radiative decay channel that produces a photon with energy @xmath416 , where @xmath417 is the dark matter particle mass ( e.g. * ? ? ? galaxy clusters have been the targets of searches for emission lines associated with such decays . the soft x - ray ( kev ) regime is particularly interesting , marking the lower limit of masses consistent with constraints from large scale structure formation . to date , all searches for monochromatic x - ray emission lines associated with non - baryonic matter in clusters ( as well as other dark matter rich objects ) have proved negative ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? gamma - ray observations of clusters with the fermi gamma - ray space telescope have been used to place interesting upper limits on dark matter annihilation , and the lifetimes of particles for a range of masses and decay final states @xcite . dark energy , though a key component of the standard cosmological model , provides by no means the only possible explanation for cosmic acceleration . various non - standard gravity models can also produce acceleration on cosmological scales ( for reviews , see @xcite ) . these include frameworks that consistently parametrize departures from general relativity ( gr ; @xcite ) ; full , alternative theories such as the dvali - gabadadze - porrati ( dgp ) braneworld gravity @xcite ; @xmath418 modifications of the einstein - hilbert action @xcite ; and modifications of gravity based on the mechanism of ghost condensation @xcite . a critical requirement for any modified gravity model is that it should mimic gr in the relatively small scale , high density regime where gr has been tested precisely . thus , in addition to investigating whether dark energy is well described by a cosmological constant , we are simultaneously interested in asking whether gr provides the correct description of gravity , and indeed whether dark energy is needed at all . to discriminate among these possibilities , and between particular dark energy and modified gravity models , it is important to combine expansion history data with measurements of the growth and scale - dependence of cosmic structure . galaxy clusters provide some of our strongest constraints on structure growth . to utilize these constraints robustly , however , accurate predictions for the halo mass function are required . recently , a few mass functions for specific modified gravity models have been constructed and calibrated using n - body simulations . these include the self - accelerated branch @xcite and normal branch @xcite of dgp gravity , and an @xmath418 model @xcite . constraints on the latter model using the observed local cluster abundance and other data are presented by @xcite . an alternative to evaluating specific gravity theories is to adopt a convenient , parameterized description for the growth of structure . this can then be used to constrain departures from the predictions of @xmath8cdm+gr @xcite . at late - times , the linear growth rate can be simply parametrized as ( e.g. @xcite ) @xmath419 where @xmath420 is the density contrast and @xmath421 the growth index . conveniently , gr predicts a nearly constant and scale - independent value of @xmath422 for models consistent with current expansion data . as in the case of @xmath313 for dark energy models , constraining @xmath421 constitutes a phenomenological approach to studying gravity . @xcite report constraints on departures from gr on cosmic scales using this parameterization with cluster data . their results are simultaneously consistent with gr ( @xmath423 ) and @xmath8cdm ( @xmath392 ) at the 68 per cent confidence level ( left panel of figure [ fig : physics : gamma_s8 ] ) . the mass of neutrinos directly influences the growth of cosmic structure , since any particle with non - zero mass at some point cools from a relativistic state , in which it effectively suppresses structure formation , to a non - relativistic state , in which it actively participates in the growth of structure ( details are reviewed in @xcite ) . in the standard scenerio where the neutrino species have approximately degenerate mass , the species - summed mass , @xmath424 , is sufficient to describe their cosmological effects . although current data lack the precision to directly detect the effect of neutrino mass on the time - dependent growth of clusters , cluster data do play a key role in cosmological constraints on neutrinos when combined with cmb observations . on its own , the cmb can place only a relatively weak upper bound on the mass , @xmath425 at 95% confidence for a @xmath8cdm model ( e.g. @xcite ; works discussed here used 5 years of _ wmap _ data , but their conclusions are not significantly changed by the 7-year update ) . incorporating cosmic distance measurements improves this to @xmath426 , with the results displaying a strong degeneracy between @xmath424 and the value of @xmath58 predicted from the amplitude of the primordial power spectrum , due to the integrated effect of neutrinos on the growth of structure . cluster data at low redshift provide a direct measurement of @xmath58 , breaking this degeneracy ( right panel of figure [ fig : physics : gamma_s8 ] ) , improving the upper limit by a further factor of two , to @xmath427 ( @xcite ) . the degeneracy - breaking power of cluster observations also significantly improves the robustness of neutrino mass limits to the assumed cosmological model ( e.g. marginalizing over global curvature ; @xcite ) . in combination with _ planck _ data , expected near - term improvements in cluster mass measurements from high - quality lensing data could reduce the limit on @xmath428 to the point of distinguishing between the normal and inverted neutrino mass hierarchies . in the near future ( over the next 23 years ) , the completion of the spt , act and _ planck _ sz catalogs will extend our detailed , statistical knowledge of galaxy clusters out to @xmath357 . together , these projects expect to find @xmath429 new clusters , mostly at intermediate - to - high redshifts @xcite . used in combination with existing low - redshift x - ray and optical catalogs , they should provide significant improvements in our knowledge of cluster growth , and corresponding improvements in the constraints on dark energy and gravity models . as discussed in section [ sec : obs_surveys_sz ] , challenges for these surveys will be defining and calibrating the optimal survey mass proxies , and understanding the impact of contamination from associated infrared sources and agn . looking further ahead , the development of experiments with improved spatial resolution , expanded frequency coverage and improved sensitivity , such as the cerro chajnantor atacama telescope ( ccat ) , will be of advantage . at optical and near - infrared wavelengths , a suite of powerful , new ground - based surveys are about to come on - line . these include the panoramic survey telescope and rapid response system ( pan - starrs ) ; the dark energy survey ( des ) ; the kilo - degree survey ( kids ) and complementary vista kilo - degree infrared galaxy survey ( viking ) ; the subaru hyper suprime - cam survey ( hsc ) ; and , later , the large synoptic survey telescope ( lsst ) . these experiments offer significant potential for finding clusters , and will provide critical photometric redshift and lensing data . a primary challenge in constructing optical cluster catalogs will be the definition of robust mass proxies with minimal , well - understood scatter across the mass and redshift ranges of interest . planned , space - based survey missions such as the wide field infrared survey telescope ( wfirst ) and euclid also offer outstanding potential for cluster cosmology , complementing the ground - based surveys in providing lensing masses and photometric redshifts , and extending cluster search volumes out to higher redshifts . in the near term , x - ray cluster samples constructed from serendipitous detections in _ chandra _ and _ xmm - newton _ observations offer the potential for important gains ( e.g. * ? ? ? * ; * ? ? ? the main advances at x - ray wavelengths , however , will be provided by the erosita telescope on the spektrum - roentgen - gamma mission . scheduled for launch in 2012 , erosita will perform a four year all - sky survey that should detect an estimated 50,000100,000 clusters with excellent purity and completeness . as discussed in section [ sec : obs_surveys_xray ] , the unambiguous detection of clusters at x - ray wavelengths requires angular resolution to distinguish point sources from extended cluster emission . at modest redshifts and high fluxes , this should be straightforward for erosita ; at high redshifts ( @xmath357 ) and faint fluxes , however , separating cool - core clusters from agn will be challenging and will require follow - up observations . here , we note more than a dozen known examples of powerful agn surrounded by x - ray bright clusters at intermediate - to - high redshifts ( e.g. @xcite ) . looking further ahead , the development of improved x - ray mirrors with high spatial resolution across a wide field of view , such as those proposed for the wide field x - ray telescope , would be a major advance , rendering trivial the removal of contaminating agn emission and allowing surveys to take full advantage of the center - excised x - ray luminosity as a low - scatter mass proxy ( section [ sec : future : coreevol ] ) . for all of these surveys , accurate calibration will be important ; this is a particular challenge for space - based missions . in comparison to other cosmological probes , the demands on photo-@xmath137 calibration will be relatively modest . this is due both to the pronounced 4000 break in early - type galaxy spectra and the ability to combine measurements for many galaxies per cluster . fisher matrix studies by @xcite suggest targets of @xmath430 for bias error and @xmath431 for error in the scatter in surveys with @xmath0 clusters . for surveys with fewer counts , shot noise dominates , and photo-@xmath137 errors become less important . extensive programs of follow - up observations using high resolution , high throughput telescopes will also be essential . x - ray observatories like _ chandra _ , _ xmm - newton _ , _ suzaku _ and _ astro - h _ are likely to remain cornerstones of this work , providing excellent , low - scatter mass proxy measurements for individual clusters . we note that such information need only be gathered for a fraction of the clusters in a survey to gain a significant boost in constraining power with respect to self - calibration alone ( e.g. * ? ? ? statistical calibration of the mean masses of clusters in flux and redshift intervals from weak lensing measurements will also be critical : in order for the intrinsic power of large surveys not to be impacted severely , calibration of the mean mass at the few per cent level is required @xcite . to achieve this accuracy , detailed redshift information for the lensed , background sources will be needed ; indeed , the use of full redshift probability density functions rather than simple color cuts may be required . detailed simulations will also be needed to probe the systematic limitations of these measurements , and to advise on the best analysis approaches ( section [ sec : future_sims ] ) . inflation predicts a near scale invariant power spectrum and nearly gaussian distribution for the primordial curvature inhomogeneities that seed lss . for slow - roll , single - field inflation , departures from gaussianity are currently unobservable ( by at least four orders of magnitude ; @xcite ) . however , other multi - field and single - field inflation models predict observable non - gaussianity ( ng ) . examples include certain brane models such as the dirac - born - infeld ( dbi ) inflation ; single - field models with non - trivial kinetic terms ; ghost inflation ; models in which density perturbations are generated by another field such as the curvaton ; and models with varying inflation decay rate . ( see @xcite and @xcite for recent reviews . ) any detection of ng would provide critical information about the physical processes taking place during inflation . in particular , a convincing detection of ng of the ` local ' type ( referring to particular configurations in fourier space ) would rule out not only slow - roll but _ all _ classes of single field inflation models @xcite . current measurements of cmb anisotropies and lss are consistent with gaussianity @xcite . in principle , measurements of the clustering and abundance of galaxy clusters can also be used to place powerful , complementary constraints . galaxy clusters trace the rare , high - mass tail of density perturbations in the universe ( section [ sec : theory ] ) and are uniquely sensitive to ng . n - body simulations have been used to study cluster formation under non - gaussian initial conditions ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? analytical mass functions have also been calculated using the press - schechter formalism @xcite or excursion set theory @xcite . to date , cmb and lss studies have only been used to place scale - independent constraints on ng . however , certain models that produce relatively large ng ( see above ) also present strong scale - dependence of the signal , due to , e.g. , a changing ( effective or otherwise ) sound speed @xcite . the addition of measurements at the smaller , galaxy cluster scale should provide important constraints on such models @xcite . several recent works @xcite have investigated quantitatively the potential of future optical and/or x - ray cluster surveys for constraining ng . to date , robust thermodynamic measurements have , in general , only been possible for the inner parts of clusters ( @xmath432 ) , where the x - ray emission is brightest and the sz signal strongest ; a large fraction of their volumes remain practically unexplored . precise , accurate measurements of the density , temperature , pressure and entropy out to the virial radii of clusters provide important insight into the physics of clusters and ( ongoing ) large - scale structure formation , and can improve the precision and robustness of cosmological constraints . recently , the japanese - us _ suzaku _ satellite has opened a new window onto the outskirts of clusters . due to its lower instrumental background than flagship x - ray observatories like _ chandra _ and _ xmm - newton _ , which orbit beyond the earth s protective magnetic fields , _ suzaku _ can study the faint , outer regions of clusters more reliably . the primary challenge with _ suzaku _ is the relatively large point spread function of its mirrors , which limits detailed , spatially - resolved studies to systems at modest redshift ( @xmath433 ) . over the past two years , a series of ground breaking measurements of the outskirts of clusters with _ suzaku _ have been reported @xcite . these have confirmed the presence of smoothly decreasing density and temperature profiles out to large radii , as was qualitatively expected from theoretical models and earlier data ( e.g. * ? ? ? * ; * ? ? ? interestingly , these observations have also suggested a possible flattening of the outer entropy profiles . of particular interest are the advances made with the _ suzaku _ key project study of the perseus cluster ( @xmath434 ) , the nearest , massive galaxy cluster and brightest , extended extragalactic x - ray source . the temperature and metallicity profiles for the northwestern ( nw ) and eastern ( e ) arms of perseus measured with _ suzaku _ are shown in figure [ fig : suzaku_kt ] @xcite . also plotted , for comparison , are the results from earlier , deep _ observations of the cluster core @xcite . the _ suzaku _ and _ chandra _ data show excellent agreement where they meet , and measure the thermodynamic structure of the icm over three decades in radius , out to @xmath233 . models of large scale structure formation show that gas is shock heated as it falls into clusters . entropy is an important tracer of this virialization process . numerical simulations predict that in the absence of non - gravitational processes such as radiative cooling and feedback , the entropy , @xmath435 , should follow a power - law with radius , @xmath436 , with @xmath437@xmath438 ( e.g. * ? ? ? . with the exception of a cold front seen along the eastern arm at @xmath439 , the entropy profile in perseus roughly follows the expected trend out to @xmath440 ( figure [ fig : suzaku_thermo ] ) . beyond this radius , however , both arms flatten away from the predicted power - law shape . the pressure profile shows good agreement between the nw and e arms . within @xmath441 , the _ suzaku _ pressure results show good agreement with the predictions from numerical simulations @xcite , and previous measurements with _ xmm - newton _ @xcite . at larger radii , however , the observed pressure profile is shallower than a simple extrapolation of the _ xmm - newton _ results . = 6.5 cm = 6.5 cm accurate estimates of the gas masses and total masses out to large radii are of particular importance for cosmological studies . the _ suzaku _ observations of the perseus cluster provide the first such measurements for a massive cluster . the best - fit nfw mass model determined from a hydrostatic analysis of the ( relatively relaxed ) nw arm has parameters in good agreement with the predictions from cosmological simulations ( @xmath442 , @xmath443 ) and provides a good description of the data @xcite . of particular interest is the cumulative @xmath342 profile , shown in the right panel of figure [ fig : suzaku_thermo ] . within @xmath368 ( @xmath444 ) , the observed @xmath342 profile is consistent with previous _ chandra _ and sz measurements for other massive , relaxed clusters @xcite . from @xmath445@xmath446 ( i.e. excluding the central cooling core ) the @xmath342 profile is also consistent with the predictions from recent numerical simulations , incorporating a simple model of agn feedback @xcite . at @xmath447 , the enclosed @xmath342 value approximately matches the mean cosmic baryon fraction , as measured from the cmb @xcite . beyond @xmath448 , however , where the entropy also flattens away from the expected power - law behavior , the @xmath342 apparently _ exceeds _ the mean cosmic value . the most plausible explanation for the apparent excess of baryons at large radii is gas clumping : in x - rays , the directly measurable quantity from the intensity of the emission is the average of the square of the electron density , rather than the average electron density itself . if the density is not uniform , i.e. the gas is clumpy , then the average electron density estimated from the x - ray intensity will overestimate the truth ; this will lead to an overestimate of the gas mass and gas mass fraction , and will flatten the apparent entropy and pressure profiles . the amount of gas clumping in perseus , as inferred from comparison of the observed and expected @xmath342 profiles , is shown in the lower panel of the right of figure [ fig : suzaku_thermo ] . using this clumping profile to correct the pressure and entropy measurements gives the solid red curves in the left panel . the clumping - corrected entropy profile shows good agreement with the expected power - law form out to @xmath233 . likewise , the clumping - corrected pressure profile matches the form predicted by simulations @xcite . importantly , the _ suzaku _ results provide no evidence for the puzzling deficit of baryons at @xmath449 inferred from some previous studies of massive clusters using lower quality x - ray data ( at large radii ) and stronger modeling priors ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the perseus data suggest that beyond the innermost core but within @xmath450 , x - ray measurements for massive , relaxed clusters can be used simply and robustly for cosmological work . at larger radii , the effects of gas clumping become increasingly important and must be accounted for . in principle , the combination of x - ray and sz observations can also be used measure gas clumping , offering an important cross - check of the suzaku results . the origin of these density fluctuations is also an important , open question . although numerical simulations predict that the gas in the outskirts of clusters should be clumpy ( e.g. * ? ? ? * ) , the degree of inhomogeneity predicted depends in detail on a range of uncertain physical processes , including cooling , conduction , viscosity , and the impact of magnetic fields . over the next few years , _ suzaku _ and sz studies of other bright , nearby clusters , complemented later by higher spectral resolution x - ray data from _ astro - h _ , can be expected to stimulate significant progress . in particular , @xmath342 measurements out to large radii for a statistical sample of nearby clusters , and measurements of the dispersion in this and other properties from region - to - region and system - to - system , should provide a robust low - redshift anchor for cosmological work and powerful constraints on astrophysical models . cluster cores typically describing the central region of @xmath451100 radius often host a variety of astrophysical processes including efficient radiative cooling , agn outbursts , modest star formation , and sloshing or other bulk motions ( see , e.g. , @xcite for a review ) . in the cores of even the most dynamically relaxed clusters , small - scale departures from hydrostatic equilibrium are often apparent ( @xcite ) . the astrophysics of cluster cores is especially important for cosmological studies at x - ray wavelengths . in particular , clusters for which radiative cooling in the core is efficient form an easily distinguishable sub - population with significantly enhanced central gas density and luminosity , also commonly accompanied by a drop in icm temperature @xcite . the sharp , central density peaks of these ` cool core ' clusters enhance their detectability at x - ray wavelengths . however , without spatial resolution @xmath286 arcsec or additional follow - up data , cool - core clusters at high redshift lying close to the survey flux limit can not be distinguished easily from non - cluster x - ray point sources . ( this selection bias has affected some studies in the literature . ) the prevalence and evolution of cool cores in the cluster population thus plays a significant role in determining the shape and evolution of the scatter in x - ray luminosity at fixed mass . confirmation of this fact can be found in the dramatic reduction in scatter obtained when emission from a central region of radius @xmath452 is excised from luminosity measurements in forming the @xmath208@xmath71 scaling relation , from @xmath453 to @xmath296 ( figure [ fig : corescatter ] ; see also @xcite ) . ( while the excision radius of @xmath452 has become somewhat conventional , a similar reduction in scatter is evident when excising a fixed metric radius of similar size , e.g. 150 . ) @xcite found that roughly half of the intrinsic scatter can be attributed to radii @xmath454 , the typical scale of cool , dense cluster cores , with most of the remainder due to variations in the gas density profile at @xmath455 . the fraction of the total flux coming from @xmath456 can be as large as 50% , and correlates strongly with other observable signatures of cool cores such as central cooling time and cuspiness of the surface brightness profile ( @xcite ; see also @xcite ) . interestingly , the fractions of cool - core systems identified using this criterion are comparable in x - ray flux - selected samples , at least within @xmath457 ( @xmath269 ; see figure [ fig : corescatter ] ; @xcite ) , although these fractions are biased relative to the full population due to selection effects . cool cores are also common in x - ray selected samples at high redshift ( @xcite ; @xmath458 ) . = 6.7 cm although the number of massive clusters found in rass is too small to detect evolution or departures from log - normality in the luminosity mass scatter , future , deeper x - ray surveys such as erosita will provide the larger samples necessary for these investigations . conversely , a better understanding of cluster cores will be required to fully exploit these surveys for cosmology . such studies will need to employ the full statistical apparatus described in section [ sec : constraints : procedure ] to obtain unbiased results . due to the weaker dependence of the sz signal on the central gas density ( section [ sec : obs_observations ] ) , sz surveys are expected to be less sensitive to the presence of cool cores than x - ray surveys . this is consistent with the relatively low fraction of cool core systems observed among the newly discovered clusters in the _ planck _ early release cluster catalog @xcite . we note , however , that current sz surveys are not necessarily immune to biases associated with cool cores ; depending on the selection techniques employed , infrared emission associated with star formation in the central regions of cool core clusters may actually diminish the measured sz signal . unleashing the full statistical power of upcoming surveys will require careful control of theoretical uncertainties . for cluster abundance and evolution tests , uncertainties in scaling relation parameters are currently more important than uncertainties in the halo mass and bias functions @xcite . with aggressive mass calibration efforts to reduce the former , the mass function and bias errors will need to be limited to the percent level in order to avoid significant degradation in cosmological parameter constraints @xcite . simulation campaigns will be needed to address this challenge . over the next decade , increased computing power will enable models with new capabilities ( e.g. multi - scale simulation in place of sub - grid models ) and will vastly expand the size of current simulation suites . by densely sampling a large control space of cosmological and astrophysical parameters , simulation ensembles can support survey analysis via functional interpolation , a method termed emulation @xcite . an initial application of this technique uses 38 @xmath459-particle n - body simulations to predict the non - linear matter power spectrum to @xmath460 accuracy ( for @xmath461 ) in a 5-dimensional space of cosmological parameters @xcite . precise n - body calibrations must be treated with some caution , as the baryons representing @xmath462 of the total mass undergo different small - scale dynamics than cdm . back - reaction effects of cooling and star formation could be important @xcite and should be systematically investigated . as discussed in sections [ sec : physics : lcdm ] and [ sec : physics ] , studies of lss in non - standard cosmologies , including those for which dark energy may cluster or in which weak - field gravity is non - newtonian , should also be pursued . full solution of the galaxy formation problem from first principles remains challenging . direct simulation methods will continue to improve , though not likely to the point where multi - fluid simulations will offer definitive predictions for next - generation survey analysis . semi - analytic methods are increasingly informative ; for example , the frantic early merging that forms the central galaxies in clusters is becoming understood @xcite . but many elements required to predict the spectrophotometric properties of galaxies remain poorly understood , including stellar population synthesis and dust evolution @xcite . empirically tuned statistical approaches , essentially assigning galaxy properties to halos or sub - halos in a manner informed by observed clustering as a function of luminosity and/or color , will continue to provide a valuable complement to physical methods . upcoming surveys will require sophisticated , automated reduction and analysis pipelines for their large data streams , along with quality assurance to validate accuracy . simulations of sky expectations , often referred to as ` mock ' or ` synthetic ' surveys , will be required to provide realistic testbeds in which the answers are known . mock surveys provide key quality assurance support by : ( i ) incorporating line - of - sight projection effects on measured properties ; ( ii ) including distortions and noise from telescope / camera optics and other sources ; and ( iii ) validating image processing and data management pipelines . since the effort involved in producing such data is non - trivial , mechanisms to publish and enable their broader use within the community need to be pursued . with such significant improvements in data quality expected and such profound questions to be addressed , modeling considerations will become increasingly important . the key issues will be modeling and mitigating all important sources of bias and systematic error in the analyses , and using the information efficiently . for tests based on the mass function and clustering of clusters and their associated mass observable scaling relations , the impact of survey biases must be accounted for ( section [ sec : theory : selection ] ) . typically this requires simultaneous modeling of the cluster population and scaling relations in a single likelihood function . such an approach also facilitates understanding the covariance between model parameters , and provides a structure within which to examine the impact of residual systematic uncertainties which often correlate with model parameters . to a large degree , the statistical frameworks required for such analyses have been developed ( section [ sec : constraints : procedure ] ) and their application to existing data are discussed in this review . the application of judicious , blind ( see below ) cuts can improve the balance of statistical versus systematic uncertainties . for example , with the @xmath342 test ( section [ sec : constraints : fgas ] ) , cosmological constraints are best derived from the most massive , dynamically relaxed clusters , which can be identified easily from short , snapshot _ chandra _ or _ xmm - newton _ x - ray observations . these clusters are also ideal targets for the xsz test ( section [ sec : constraints : xsz ] ) . for the mass function and clustering tests , one must determine the optimal mass / flux / redshift range over which to compare the data and models , given the characteristics of the survey and follow - up data and one s understanding of mass observable scaling relations . strong priors should be used with caution and their implications understood . in cluster cosmology , common assumptions include log - normal scatter in the mass observable scaling relations , and negligible evolution of this scatter with redshift . when measuring masses from x - ray or optical dynamical data , the use of parameterized temperature , velocity and gas density models is also common . where such strong assumptions are not necessary , they should be avoided . where priors do become necessary , for example in parameterizing the impacts of known astrophysical effects , the validity of these assumptions should be checked empirically , where possible . the covariance between different measured quantities is an important consideration that is often wrongly neglected . the simplest example of this is simply the statistical correlation of quantities measured from the same observation ( e.g. due to poisson noise ) . a more subtle case follows from the definition of cluster radius for scaling relations in terms of the mean density enclosed : because the measured mass and radius covary , other quantities measured within that radius necessarily covary with mass , even if measured from independent observations . fortunately , within the framework of bayesian analysis , it is straightforward to account for such measurement correlations . for example , given a set of scaling relation parameters to be tested , the data likelihood can be integrated over all possible true values of , e.g. , the mass and temperature , with one of the terms in the integrand being the probability of the true mass and temperature values producing the observed values ( the sampling distribution ; see section [ sec : constraints : procedure ] ) . the form of this probability density can be as general as is required a multidimensional gaussian or log - gaussian in the simplest case and in particular may have non - zero correlation . * see also @xcite ) discusses this general approach in the astrophysical context , and provides useful tools for bayesian linear regression . in other situations , unnecessary covariance can be introduced by the model applied to the data . an example is the historically common practice of fitting parametric temperature and gas density profiles to x - ray data , and inferring overdensity radii and hydrostatic masses from these model profiles . here , the procedure introduces an artificial covariance between mass and temperature ( i.e. a prior on the scaling relation ) which could be avoided simply by modeling the temperature and gravitating mass profiles independently . where binning will result in a loss of relevant information , it should be avoided . for example , the binning of x - ray surface brightness data to determine an integrated x - ray luminosity results in a significant loss of information ; the center - excised luminosity provides a tighter mass proxy ( section [ sec : future : coreevol ] ) . hypothesis testing will remain a critical element of future analyses . wherever a model is fitted to data , one should check that it provides an adequate description ( i.e. that the goodness of fit is acceptable ) ; if it does nt , then the model can be ruled out . this is particularly important in the context of cosmological surveys and scaling relations that are subject to selection bias , since no straightforward , visual check of the goodness of fit is typically possible . where the simplest models fail to describe the data , one must evaluate carefully the additional degrees of freedom needed to alleviate the tension , in particular considering both astrophysical and cosmological possibilities ; simulations will play a critical role in motivating and validating alternatives to the simplest scaling models . a related situation arises in the combination of constraints from independent experiments : before combining , one should ensure that the model in question provides an adequate description of the data sets individually , and that the parameter values are mutually consistent , i.e. that their multi - dimensional confidence contours overlap . here it can be helpful to include the impact of all known systematics in the contours ; where contours do not overlap , the model is incomplete and/or unidentified systematic errors are present . historically , the combination of mutually _ inconsistent _ data sets has sometimes led to unphysically tight formal constraints . training sets can be useful to tune analyses . however , the biases introduced by this training must be understood and accounted for . the inclusion of sufficient redundancy into experiments ( i.e. having more than one independent way to make a measurement ) can also enhance significantly the robustness of conclusions . in cosmological studies , for example , we require more than a single way to measure both the growth and expansion histories , e.g. clusters and lensing , and snia and bao , respectively . a final modeling consideration is the impact of experimenter s bias : the subjective , subconscious bias towards a result expected by an experimenter . this is important where we are seeking to measure a quantity for which prior expectations exist , e.g. @xmath313 or the growth index , @xmath421 . evidence for experimenter s bias can be found readily in the literature , e.g. in historical measurements of the speed of light or the neutron lifetime ( see @xcite for a review ) . the best approach to overcoming experimenter s bias is blind analysis . a range of blind analysis techniques are commonly employed in the particle and nuclear physics communities . these are used to mask the ability of scientists to determine quantities of interest until all methods are finalized , the systematic errors that can be identified have been , and the final measurement is ready to be made . a technique with good applicability to cosmological studies is _ hiding the answer _ , wherein a fixed ( unknown to the experimenter ) offset is added to the parameter(s ) of interest , and only revealed once the final measurements are made . before un - blinding , it is advisable to think through how to proceed afterward , and what additional checks to employ . double - blind analyses , where two independent teams repeat the same process ( often used in biomedical research ) target additional sources of error . such techniques can be used powerfully , where resources allow . our article has summarized the status of cluster cosmology and the methods used to extract cosmological information from galaxy cluster observations . these methods have been applied successfully to samples of tens to thousands of massive clusters . much of this work has been pioneered at x - ray and optical wavelengths , but sz surveys and gravitational lensing studies are also now poised to play central roles . the availability of powerful , new cluster surveys will help to address profound mysteries such as the origin of cosmic acceleration , inflation and the nature of gravity . this work will require a multiwavelength approach , combining the strengths of the available techniques for finding clusters , calibrating their masses and obtaining low - scatter mass proxies . the analysis of these data will require dedicated efforts by large teams of researchers . the demands for follow - up observations will be high , and will require the support of time allocation committees . building on the progress made , we can expect clusters to remain at the forefront of cosmological work through the next decade . by combining cluster measurements with other , powerful probes such as snia , cmb , bao and cosmic shear , we can be optimistic of having the precision , complementarity and redundancy required to allow robust conclusions to be drawn . we are grateful to our colleagues for their many valuable insights . in particular , we thank roger blandford , hans bhringer , marusa brada , joanne cohn , carlos cunha , megan donahue , harald ebeling , andy fabian , pat henry , dragan huterer , andrey kravtsov , dan marrone , tim mckay , glenn morris , daisuke nagai , aaron roodman , eduardo rozo , robert schmidt , tim schrabback , neelima sehgal , aurora simionescu , alexey vikhlinin , mark voit , anja von der linden , risa wechsler and norbert werner . we particularly thank david rapetti for his detailed input to the material on gravity tests and non - gaussianities . swa was supported in part by the u.s . department of energy under contract number de - ac02 - 76sf00515 . aee acknowledges support from nasa astrophysics theory grant nnx10af61 g . abm was supported by an appointment to the nasa postdoctoral program at the goddard space flight center , administered by oak ridge associated universities through a contract with nasa . this research was supported in part by the national science foundation under grant no . nsf phy05 - 51164 .

current observations are more consistent with low - density cold dark matter ( cdm ) dominated cosmogonies . for recent discussions see , e.g. , ratra et al . ( 1999a ) , kravtsov & klypin ( 1999 ) , doroshkevich et al . ( 1999 ) , colley et al . ( 2000 ) , freudling et al . ( 1999 ) , sahni & starobinsky ( 1999 ) , bahcall et al . ( 1999 ) , and donahue & voit ( 1999 ) . the simplest low - density cdm models have either flat spatial hypersurfaces and a constant or time - variable cosmological constant " @xmath3 ( see , e.g. , peebles 1984 ; peebles & ratra 1988 , hereafter pr ; efstathiou , sutherland , & maddox 1990 ; stompor , grski , & banday 1995 ; caldwell , dave , & steinhardt 1998 ) , or open spatial hypersurfaces and no @xmath3 ( see , e.g. , gott 1982 , 1997 ; ratra & peebles 1994 , 1995 ; grski et al . 1998 ) . while these models are consistent with most recent observations , there are notable exceptions . for instance , recent applications of the apparent magnitude versus redshift test based on type ia supernovae favor a non - zero @xmath3 ( see , e.g. , riess et al . 1998 , hereafter r98 ; perlmutter et al . 1999a , hereafter p99 ) , although not with great statistical significance ( drell , loredo , & wasserman 1999 ) . on the other hand , the open model is favored by : * measurements of the hubble parameter @xmath4 (= @xmath5 ) which indicate @xmath6 at 2 @xmath7 ( see , e.g. , suntzeff et al . 1999 ; biggs et al . 1999 ; madore et al . 1999 ) , and measurements of the age of the universe which indicate @xmath8 gyr at 2 @xmath7 ( see , e.g. , reid 1997 ; gratton et al . 1997 ; chaboyer et al . this is because the resulting central @xmath9 value is consistent with a low - density open model with nonrelativistic - matter density parameter @xmath10 and requires a rather large @xmath11 in the flat - constant-@xmath3 case . * analyses of the rate of gravitational lensing of quasars and radio sources by foreground galaxies which require a large @xmath12 at 2 @xmath7 in the flat - constant-@xmath3 model ( see , e.g. , falco , kochanek , & muoz 1998 ) . * analyses of the number of large arcs formed by strong gravitational lensing by clusters ( bartelmann et al . 1998 ; meneghetti et al . 1999 ) . * the need for mild anti - biasing to accommodate the excessive intermediate- and small - scale power predicted in the dmr normalized flat - constant-@xmath3 cdm model with a scale - invariant spectrum ( see , e.g. , cole et al . 1997 ) . while the time - variable @xmath3 model has not yet been studied to the same extent as the open and flat-@xmath3 cases , it is likely that it can be reconciled with most of these observations ( see , e.g. , pr ; ratra & quillen 1992 , hereafter rq ; frieman & waga 1998 ; perlmutter , turner , & white 1999b ; wang et al . 1999 ; efstathiou 1999 ) . we emphasize that most of these observational indications are tentative and certainly not definitive . this is particularly true for constraints derived from a @xmath13 comparison between model predictions and cosmic microwave background ( cmb ) anisotropy measurements ( see , e.g. , ganga , ratra , & sugiyama 1996 ; lineweaver & barbosa 1998 ; baker et al . 1999 ; rocha 1999 ) , see discussions in bond , jaffe , & knox ( 1998 ) and ratra et al . ( 1999b ) . more reliable constraints follow from models - based maximum likelihood analyses of the cmb anisotropy data ( see , e.g. , grski et al . 1995 ; yamamoto & bunn 1996 ; ganga et al . 1998 ; ratra et al . 1999b ; rocha et al . 1999 ) , which make use of all the information in the cmb anisotropy data and are based on fewer approximations . however , this technique has not yet been used to analyze enough data sets to provide robust statistical constraints . kamionkowski & kosowsky ( 1999 ) review what might be expected from the cmb anisotropy in the near future . in this paper we examine constraints on a constant and time - variable @xmath3 that follow from recent type ia supernovae apparent magnitude versus redshift data and recent measurements of @xmath4 and @xmath14 . we focus here on the favored scalar field ( @xmath0 ) model for a time - variable @xmath3 ( pr ; ratra & peebles 1988 , hereafter rp ) , in which the scalar field potential @xmath15 , @xmath2 , at low redshift . this scalar field could have played the role of the inflaton at much higher redshift , with the potential @xmath16 dropping to a non - zero value at the end of inflation and then decaying more slowly with increasing @xmath0 ( pr ; rp ) . see peebles & vilenkin ( 1999 ) , perrotta & baccigalupi ( 1999 ) and giovannini ( 1999 ) for a specific model and observational consequences of this scenario . a potential @xmath17 could arise in a number of high energy particle physics models , see , e.g. , bintruy ( 1999 ) , kim ( 1999 ) , barr ( 1999 ) , choi ( 1999 ) , banks , dine , & nelson ( 1999 ) , brax & martin ( 1999 ) , masiero , pietroni , & rosati ( 1999 ) , and bento & bertolami ( 1999 ) for specific examples . it is conceivable that such a setting might provide an explanation for the needed form of the potential , as well as for the needed very weak coupling of @xmath0 to other fields ( rp ; carroll 1998 ; kolda & lyth 1999 , but see periwal 1999 and garriga , livio , & vilenkin 1999 for other possible explanations for the needed present value of @xmath3 ) . a scalar field is mathematically equivalent to a fluid with a time - dependent speed of sound ( ratra 1991 ) . this equivalence may be used to show that a scalar field with potential @xmath15 , @xmath2 , acts like a fluid with negative pressure and that the @xmath0 energy density behaves like a cosmological constant that decreases with time . this energy density could come to dominate at low redshift and thus help reconcile low dynamical estimates of the mean mass density with a spatially - flat cosmological model . alternate mechanisms that also rely on negative pressure to achieve this result have been discussed ( see , e.g. , zer & taha 1986 ; freese et al . 1987 ; turner & white 1997 ; chiba , sugiyama , & nakamura 1997 ; zer 1999 ; waga & miceli 1999 ; overduin 1999 ; bucher & spergel 1999 ) . however , these mechanisms are either inconsistent or do not share a very appealing feature of the scalar field models . for some of the scalar field potentials mentioned above , the solution of the equations of motion is an attractive time - dependent fixed point ( rp ; pr ; wetterich 1995 ; ferreira & joyce 1998 ; copeland , liddle , & wands 1998 ; zlatev , wang , & steinhardt 1999 ; liddle & scherrer 1999 ; santiago & silbergleit 1998 ; uzan 1999 ) . this means that for a wide range of initial conditions the scalar field @xmath0 evolves in a manner that ensures that the cosmological constant contributes a reasonable energy density at low redshift . of course , this does not resolve the ( quantum - mechanical ) cosmological constant problem . there have been many studies of the constraints placed on a time - variable @xmath3 from type ia supernovae apparent magnitude versus redshift data ( see , e.g. , turner & white 1997 ; frieman & waga 1998 ; garnavich et al . 1998 ; hu et al . 1999 ; cooray 1999 ; waga & miceli 1999 ; p99 ; perlmutter et al . 1999b ; wang et al . 1999 ; efstathiou 1999 ) . however , except for the analysis of frieman & waga ( 1998 ) , who use the earlier perlmutter et al . ( 1997 ) data , these analyses have mostly made use of the time - independent equation of state fluid approximation to the scalar field model for a time - variable @xmath3 ( perlmutter et al . 1999b and efstathiou 1999 do go beyond the time - independent equation of state approximation by also approximating the time dependence of the equation of state , however , they do not analyze the exact scalar field model ) from gravitational lensing statistics ( see , e.g. , bloomfield torres & waga 1996 ; jain et al . 1998 ; cooray 1999 ; waga & miceli 1999 ; wang et al . 1999 ) , with the exception of the analysis of rq . ] . a major purpose of this paper is to use the newer supernovae data of r98 and p99 to derive constraints on a time - variable @xmath3 without making use of the time - independent equation of state fluid approximation to the scalar field model . in the analyses here we use the most recent r98 and p99 data to place constraints on a constant and time - variable @xmath3 . we note , however , that this is a young field and insufficient understanding of a number of astrophysical processes and effects ( the mechanism responsible for the supernova , evolution , environmental effects , intergalactic dust , etc . ) could render this a premature undertaking ( see , e.g. , hflich , wheeler , & thielemann 1998 ; aguirre 1999 ; drell et al . 1999 ; umeda et al . 1999 ; riess et al . 1999 ; wang 1999 ) . furthermore , other cosmological explanations ( time - variable gravitational or fine structure constants " ) could be consistent with the data ( see , e.g. , amendola , corasaniti , & occhionero 1999 ; barrow & magueijo 1999 ; garca - berro et al . 1999 ) . in addition to analyzing just the supernovae data , we also incorporate constraints from recent measurements of @xmath4 and @xmath14 and derive a combined likelihood function which we use to constrain both a constant and a time - variable @xmath3 . we also examine the effect on the model - parameter constraints caused by varying the prior for @xmath18 . we emphasize that the tests examined in this paper are not sensitive to the spectrum of inhomogeneities in the models considered . this spectrum ( possibly generated by quantum mechanical zero - point fluctuations during inflation , see , e.g. , fischler , ratra , & susskind 1985 ) is relevant for some of the other tests mentioned above ( cmb anisotropy , anti - biasing , etc . ) . in @xmath192 we summarize the computations . results are presented and discussed in @xmath193 . conclusions are given in @xmath194 . we examine three sets of supernovae apparent magnitudes . we use the mlcs magnitudes for the r98 data , both including and excluding the unclassified sn 1997ck at @xmath20 ( with 50 and 49 sne ia , respectively ) . the third set are the corrected / effective stretch factor magnitudes for the 54 fit c sne ia of p99 . in all three cases we assume that the measured magnitudes are independent . we also assume that sne ia at high and low @xmath21 are not significantly different ( drell et al . 1999 , riess et al . 1999 , and wang 1999 consider the possibility and consequences of evolution ) , and that intergalactic dust has a negligible effect ( aguirre 1999 considers the effects of dust ) . our analysis of the r98 data is similar to that described in their @xmath194 , with a few minor modifications . for the time - independent @xmath3 model we compute the likelihood function @xmath22 for a range of @xmath18 spanning the interval 0 to 2.5 in steps of 0.1 , for a range of @xmath23 spanning the interval @xmath24 to 3 in steps of 0.1 , and for a range of @xmath4 spanning the interval 50 to 80 km s@xmath25 mpc@xmath25 in steps of 0.5 km s@xmath25 mpc@xmath25 . our analysis of the p99 data is similar to theirs , with some modifications . specifically , we work with their corrected / effective magnitudes and so need only compute a three - dimensional likelihood function @xmath26 . here @xmath27 is their hubble - constant - free " b band absolute magnitude ( see p99 ) related to @xmath4 by @xmath28 ( determined by us from results in p99 ) . for the time - independent @xmath3 model we compute this likelihood function for a range of @xmath18 spanning the interval 0 to 3 in steps of 0.1 , for a range of @xmath23 spanning the interval @xmath29 to 3 in steps of 0.1 , and for a range of @xmath27 spanning the interval @xmath30 to @xmath31 in steps of 0.05 ( which corresponds to about the same interval in @xmath4 used in our analyses of the r98 data ) . note that in our p99 time - independent @xmath3 model plots we do not show the likelihood for the whole @xmath32 region over which we have computed it . the spatially - flat time - variable @xmath3 model we consider is the scalar field model with potential @xmath15 , @xmath2 . when @xmath33 the model tends to the flat - constant-@xmath3 case and when @xmath34 it approaches the einstein - de sitter model . it is discussed in detail in pr , rp , and rq , and we derive the predicted distance moduli for the sne using expressions given in these papers . as shown in these papers , the scalar field behaves like a fluid with a constant ( but different ) equation of state in each epoch of the model . for instance , in the cdm and baryon dominated epoch of the model , it obeys the equation of state @xmath35 ( relating the pressure and energy density of the scalar field ) , where @xmath36 see , e.g. , rq eq . ( 2 ) or see zlatev et al . ( 1999 ) for a more recent derivation . we shall also have need for an average equation of state parameter , used by perlmutter et al . ( 1999b ) and wang et al . ( 1999 ) , @xmath37 where @xmath38 is the scalar field density parameter and @xmath39 is the scale factor ( with @xmath40 being the present value ) . in this model , for the r98 data we compute the likelihood function @xmath41 , and for the p99 data the likelihood function @xmath42 . in both cases we evaluate the likelihood function for a range of @xmath18 spanning the interval 0.05 to 0.95 in steps of 0.025 , for a range of @xmath43 spanning the interval 0 to 8 in steps of 0.5 , and for the same range of @xmath4 or @xmath27 as in the time - independent @xmath3 cases discussed above . we marginalize these three - dimensional likelihood functions by integrating over @xmath4 ( for the r98 data ) or @xmath27 ( for the p99 data ) and derive two - dimensional likelihood functions , @xmath44 for the constant @xmath3 model and @xmath45 for the spatially - flat time - variable @xmath3 case . these two - dimensional likelihood functions are used to derive highest posterior density limits ( see ganga et al . 1997 and references therein ) in the @xmath46 or @xmath47 planes . in what follows we consider 1 , 2 , and 3 @xmath7 confidence limits which include 68.27 , 95.45 , and 99.73% of the area under the likelihood function . when marginalizing over a parameter or deriving a limit from the likelihood functions , we consider a number of different priors . we first consider a uniform prior in the parameter integrated over , set to zero outside the range considered for the parameter . we also incorporate constraints from measurements which indicate @xmath48 at 1 @xmath7 ( see , e.g. , biggs et al . 1999 ; madore et al . 1999 ) , by using the prior @xmath49 , \ ] ] and from measurements which indicate @xmath50 gyr at 1 @xmath7 ( i.e. , 0.5 gyr added to the globular cluster age estimate of chaboyer et al . 1998 ) , by using the prior @xmath51 , \ ] ] with @xmath43 replacing @xmath23 in this expression in the spatially - flat time - variable @xmath3 case . note that figure 2 of chaboyer et al . ( 1998 ) may be used to establish that a gaussian prior of the form of eq . ( 4 ) is a good approximation to the shifted monte carlo globular cluster age distribution . see ganga et al . ( 1997 ) for a discussion of gaussian priors in a related context . this method of incorporating constraints from measurements of @xmath4 and @xmath14 is not identical to the classical @xmath9 cosmological test ( see , e.g. , @xmath1913 of peebles 1993 ) . finally , since @xmath18 is a positive quantity , we also consider the non - informative prior ( berger 1985 ) here ( see @xmath19b.2 of drell et al . 1999 ; @xmath19vii of jaynes 1968 gives a more general discussion ) . it is however still of interest to determine how the conclusions depend on the choice of prior . ] , @xmath52 since @xmath18 is bounded from below by the observed lower limit on the baryon density parameter ( @xmath53 , depending on data used , see , e.g. , olive , steigman , & walker 1999 ) , this prior does not result in an infinity . figure 1 shows the posterior probability density distribution function ( pdf ) confidence contours for the time - variable @xmath3 scalar field model , derived using the three data sets discussed above . figure 1@xmath54 shows that constraints from the three different data sets are quite consistent . at 2 @xmath7 a large region of the parameter space of these spatially - flat models ( with a constant or time - variable @xmath3 ) is consistent with the sne ia data . this data favors a smaller @xmath18 as @xmath43 is increased from zero . a fluid with a time - independent equation of state @xmath55 , @xmath56 , has often been used to approximate the scalar field with potential @xmath15 in the time - variable @xmath3 model . the solid lines in figure 2@xmath39 show the confidence contours for such a fluid model , derived using the p99 fit c data . these confidence contours are consistent with those shown in figure 1 of perlmutter et al . ( 1999b ) ( but see discussion below ) , figure 10 of wang et al . ( 1999 ) , and figure 4 of efstathiou ( 1999 ) . the dashed lines in figure 2@xmath39 show the exact scalar field model contours of figure 1@xmath57 , transformed using the relation between @xmath58 and @xmath43 in the cdm and baryon dominated epoch ( eq . [ 1 ] ) , which comes to an end just before the present . the two sets of contours agree near @xmath59 , as they must since at @xmath60 this is the flat-@xmath3 model and @xmath3 does behave exactly like a fluid with a time - independent equation of state . however , the two sets of confidence contours differ significantly at larger @xmath61 . since the scalar field @xmath58 eventually switches over to @xmath24 ( in the scalar field dominated epoch , rp ) , it is unclear what significance should be ascribed to this difference . we stress , however , that since the time - independent equation of state fluid model is an approximation to the time - variable @xmath3 scalar field model , constraints on model - parameter values that are based on the fluid model approximation are only approximate ( and possibly indicative ) . in passing , we note that perlmutter et al . ( 1999b ) use @xmath62 ( eq . [ 2 ] ) and not @xmath61 to parametrize the fluid model constraints . figure 2@xmath63 shows contours of constant @xmath62 as a function of @xmath43 and @xmath18 in the time - variable @xmath3 scalar field model . in a large part of model parameter space @xmath62 is a sensitive function of both @xmath43 and @xmath18 and hence is not the best parameter to use to describe the time - variable @xmath3 scalar field model . figure 3 shows the effects of incorporating constraints based on @xmath4 and @xmath14 measurements ( eqs . [ 3 , 4 ] ) . panel @xmath64 shows that adding the @xmath4 constraint does not significantly alter the contours derived from the sne ia data alone . this is expected since the value of @xmath4 used here is very close to the value that is indicated by the sne ia data ( see r98 ) . however , incorporating the @xmath14 constraint , @xmath50 gyr at 1 @xmath7 , does significantly shift the contours , panel @xmath65 . this is because the sne ia data alone favor a higher @xmath14 , @xmath66 gyr ( r98 ) , or @xmath67 gyr ( p99 ) . figure 4 shows constraints on the time - variable @xmath3 model , from the three different sne ia data sets used in conjunction with the @xmath4 and @xmath14 measurements ( eqs . [ 3 , 4 ] ) . the main effect of incorporating the @xmath4 and @xmath14 constraints is to increase the favored values of @xmath18 ; a weaker effect is the disfavoring of larger values of @xmath43 . even this extended set of data does not tightly constrain model - parameter values . figure 5 shows the corresponding constraints on the constant @xmath3 model ( from the sne ia , @xmath4 , and @xmath14 measurements ) . again , the major effect of including the @xmath4 and @xmath14 data is to increase the favored values of @xmath18 . a weaker effect is that it reduces the odds against reasonable open models ( grski et al . 1998 ) , but not by a large factor . figures 6 and 7 show the effects of using the non - informative prior , @xmath68 , of eq . the major effect is a decrease in the favored values of @xmath18 . in the time - variable @xmath3 case there is also a slight increase in the favored values of @xmath43 ( see figure 6 ) while in the constant @xmath3 case there is a mild reduction in the odds against reasonable open models ( see figure 7 ) . if the pdf was narrower ( i.e. , if the error bars on the data were smaller ) , changing from the flat to the non - informative prior would not result in as large a change in the confidence contours . recent sne ia data do favor models with a constant or time - variable @xmath3 over an open model without a @xmath3 . however , this is not at a very high level of statistical significance . also , the incomplete understanding of a number of astrophysical effects and processes ( evolution , intergalactic dust , etc . ) means that these results are preliminary and not yet definitive . the constraints on the time - variable @xmath3 model derived here are based on the exact scalar field model equations of motion , not on the widely used time - independent equation of state fluid approximation equations of motion . we acknowledge the advice and assistance of r. gott , b. kirshner , l. krauss , v. periwal , s. perlmutter , a. riess , e. sidky , m. vogeley , and i. wasserman , and are specially indebted to j. peebles and t. souradeep . we acknowledge helpful discussions with i. waga who has also computed the pdf confidence contours for the time - variable @xmath3 scalar field model , with results consistent with those found in this paper . we thank the referee , i. wasserman , for a prompt and detailed report which helped us improve the manuscript . we acknowledge support from nsf career grant ast-9875031 . aguirre , a.n . 1999 , , 512 , l19 amendola , l. , corasaniti , s. , & occhionero , f. 1999 , astro - ph/9907222 bahcall , n.a . , ostriker , j.p . , perlmutter , s. , & steinhardt , p.j . 1999 , science , 284 , 1481 baker , j.c . , et al . 1999 , mnras , submitted banks , t. , dine , m. , & nelson , a.e . 1999 , j. high ener . phys . , 9906 , 014 barr , s.m . 1999 , phys . b , 454 , 92 barrow , j.d . , & magueijo , j. 1999 , astro - ph/9907354 bartelmann , m. , huss , a. , colberg , j.m . , jenkins , a. , & pearce , f.r . 1998 , a&a , 330 , 1 bento , m.c . , & bertolami , o. 1999 , gr - qc/9905075 berger , j.o . 1985 , statistical decision theory and bayesian analysis ( new york : springer - verlag ) , 82 biggs , a.d . , browne , i.w.a . , helbig , p. , koopmans , l.v.e . , wilkinson , p.n . , & perley , r.a . 1999 , , 304 , 349 bintruy , p. 1999 , , 60 , 063502 bloomfield torres , l.f . , & waga , i. 1996 , mnras , 279 , 712 bond , j.r . , jaffe , a.h . , & knox , l. 1998 , phys . d , 57 , 2117 brax , p. , & martin , j. 1999 , phys . b , in press bucher , m. , & spergel , d. 1999 , , 60 , 043505 caldwell , r.r . , dave , r. , & steinhardt , p.j . 1998 , phys . , 80 , 1582 carroll , s.m . 1998 , phys . lett . , 81 , 3067 chaboyer , b. , demarque , p. , kernan , p.j . , & krauss , l.m . 1998 , apj , 494 , 96 chiba , t. , sugiyama , n. , & nakamura , t. 1997 , , 289 , l5 choi , k. 1999 , hep - ph/9902292 cole , s. , weinberg , d.h . , frenk , c.s . , & ratra , b. 1997 , , 289 , 37 colley , w.n . , gott , j.r . , weinberg , d.h . , park , c. , & berlind , a.a . 2000 , , 529 , in press copeland , e.j . , liddle , a.r . , & wands , d. 1998 , , 57 , 4686 cooray , a.r . 1999 , , 341 , 653 donahue , m. & voit , g.m . 1999 , , 523 , l137 doroshkevich , a.g . , mller , v. , retzlaff , j. , & turchaninov , v. 1999 , , 306 , 575 drell , p.s . , loredo , t.j . , & wasserman , i. 1999 , , in press efstathiou , g. 1999 , , submitted efstathiou , g. , sutherland , w.j . , & maddox , s.j . 1990 , nature , 348 , 705 falco , e.e . , kochanek , c.s . , & muoz , j.a . 1998 , apj , 494 , 47 ferreira , p.g . , & joyce , m. 1998 , phys . d , 58 , 023503 fischler , w. , ratra , b. , & susskind , l. 1985 , nucl . b , 259 , 730 freese , k. , adams , f.c . , frieman , j.a . , & mottola , e. 1987 , nucl . b , 287 , 797 freudling , w. , et al . 1999 , , 523 , 1 frieman , j.a . , hill , c.t . , stebbins , a. , & waga , i. 1995 , , 75 , 2077 frieman , j.a . , & waga , i. 1998 , phys . d , 57 , 4642 ganga , k. , ratra , b. , gundersen , j.o . , & sugiyama , n. 1997 , , 484 , 7 ganga , k. , ratra , b. , lim , m.a . , sugiyama , n. , & tanaka , s.t . 1998 , apjs , 114 , 165 ganga , k. , ratra , b. , & sugiyama , n. 1996 , , 461 , l61 garca - berro , e. , gaztaaga , e. , isern , j. , benvenuto , o. , & althaus , l. 1999 , astro - ph/9907440 garnavich , p.m. , et al . 1998 , , 509 , 74 garriga , j. , livio , m. , & vilenkin , a. 1999 , astro - ph/9906210 giovannini , m. 1999 , astro - ph/9903004 grski , k.m . , ratra , b. , stompor , r. , sugiyama , n. , & banday , a.j . 1998 , apjs , 114 , 1 grski , k.m . , ratra , b. , sugiyama , n. , & banday , a.j . 1995 , , 444 , l65 gott , j.r . 1982 , nature , 295 , 304 gott , j.r . 1997 , in critical dialogues in cosmology , ed . n. turok ( singapore : world scientific ) , 519 gratton , r.g . , fusi pecci , f. , carretta , e. , clementini , g. , corsi , c.e . , & lattanzi , m. 1997 , apj , 491 , 749 hu , w. , eisenstein , d.j . , tegmark , m. , & white , m. 1999 , phys . d , 59 , 023512 hflich , p. , wheeler , j.c . , & thielemann , f.k . 1998 , , 495 , 617 jain , d. , panchapakesan , n. , mahajan , s. , & bhatia , v.b . 1998 , apj , submitted jaynes , e.t . 1968 , ieee trans . ssc-4 , 227 kamionkowski , m. & kosowsky , a. 1999 , ann . nucl . part . sci , in press kim , j.e . 1999 , j. high . phys , 9905 , 022 kolda , c. & lyth , d.h . 1999 , b , 458 , 197 kravtsov , a.v . , & klypin , a.a . 1999 , , 520 , 437 liddle , a.r . , & scherrer , r.j . 1999 , phys . d , 59 , 023509 lineweaver , c.h . , & barbosa , d. 1998 , , 496 , 624 lucchin , f. , & matarrese , s. 1985 , , 32 , 1316 madore , b.f . , et al . 1999 , , 515 , 29 masiero , a. , pietroni , m. , & rosati f. 1999 , , submitted meneghetti , m. , bolzonella , m. , bartelmann , m. , moscardini , l. , & tormen , g. 1999 , , submitted olive , k.a . , steigman , g. , & walker , t.p . 1999 , , in press overduin , j.m . 1999 , , 517 , l1 zer , m. 1999 , , 520 , 45 zer , m. , & taha , m.o . 1986 , phys . b , 171 , 363 peebles , p.j.e . 1984 , , 284 , 439 peebles , p.j.e . 1993 , principles of physical cosmology ( princeton : princeton university press ) peebles , p.j.e . , & ratra , b. 1988 , apj , 325 , l17 ( pr ) peebles , p.j.e . , & vilenkin , a. 1999 , , 59 , 063505 periwal , v. 1999 , astro - ph/9906253 perlmutter , s. , et al . 1997 , , 483 , 565 perlmutter , s. , et al . 1999a , , 517 , 565 ( p99 ) perlmutter , s. , turner , m.s . , & white , m. 1999b , , 83 , 670 perrotta , f. , & baccigalupi , c. 1999 , , 59 , 123508 ratra , b. 1991 , , 43 , 3802 ratra , b. 1992 , , 45 , 1913 ratra , b. , ganga , k. , stompor , r. , sugiyama , n. , de bernardis , p. , & grski , k.m . 1999a , apj , 510 , 11 ratra , b. , & peebles , p.j.e . 1988 , , 37 , 3406 ( rp ) ratra , b. , & peebles , p.j.e . 1994 , , 432 , l5 ratra , b. , & peebles , p.j.e . 1995 , phys . d , 52 , 1837 ratra , b. , & quillen , a. 1992 , mnras , 259 , 738 ( rq ) ratra , b. , stompor , r. , ganga , k. , rocha , g. , sugiyama , n. , & grski , k.m . 1999b , apj , 517 , 549 reid , i.n . 1997 , aj , 114 , 161 riess , a.g . , et al . 1998 , aj , 116 , 1009 ( r98 ) riess , a.g . , filippenko , a.v . , li , w. , & schmidt , b.p . 1999 , aj , in press rocha , g. 1999 , in dark matter in astrophysics and particle physics 1998 , ed . h.v . klapdor - kleingrothaus & l. baudis ( bristol : institute of physics publishing ) , 238 rocha , g. , stompor , r. , ganga , k. , ratra , b. , platt , s.r . , sugiyama , n. , & grski , k.m . 1999 , apj , 525 , in press , astro - ph/9905127 sahni , v. , & starobinsky , a. 1999 , int d , in press santiago , d.i . , & silbergleit , a.s . 1998 , phys . a , submitted stompor , r. , grski , k.m . , & banday , a.j . 1995 , , 277 , 1225 suntzeff , n.b . , et al . 1999 , , 117 , 1175 turner , m.s . , & white , m. 1997 , phys . d , 56 , 4439 umeda , h. , nomoto , k. , kobayashi , c. , hachisu , i. , & kato , m. 1999 , , 522 , l43 uzan , j .- 1999 , , 59 , 123510 waga , i. , & miceli , a.p.m.r . 1999 , , 59 , 103507 wang , l. , caldwell , r.r . , ostriker , j.p . , & steinhardt , p.j . 1999 , apj , submitted wang , y. 1999 , , submitted wetterich , c. 1995 , , 301 , 321 yamamoto , k. , & bunn , e.f . 1996 , apj , 464 , 8 zlatev , i. , wang , l. , & steinhardt , p.j . 1999 , , 82 , 896 fig . 1. pdf confidence contours for the spatially - flat time - variable @xmath3 scalar field model , with potential @xmath15 , derived using the three sne ia data sets . the @xmath69 axis corresponds to the spatially - flat time - independent @xmath3 case . confidence contours in panels @xmath70 run from @xmath71 to @xmath72 @xmath7 starting from the lower left hand corner of each panel . panel @xmath64 shows those derived from all the r98 sne , while @xmath65 is for r98 sne excluding the @xmath20 one , and @xmath73 is for the p99 fit c data set . panel @xmath74 compares the @xmath75 @xmath7 limits from the three data sets : all r98 sne ( long - dashed lines ) ; r98 sne excluding the @xmath20 one ( short - dashed lines ) ; and the p99 fit c sne ( dotted lines ) . fig . 2. @xmath64 pdf confidence contours derived from the p99 fit c sne , for a spatially - flat fluid model with equation of state @xmath55 ( solid lines show contours from @xmath76 to @xmath72 @xmath7 starting from the lower left hand corner ) and for the spatially - flat time - variable @xmath3 scalar field model with potential @xmath15 which behaves like a fluid model with equation of state @xmath35 in the cdm and baryon dominated epoch ( dotted lines show contours from @xmath71 to @xmath72 @xmath7 starting from the lower left hand corner ) . the scalar field model contours were derived by using eq . ( 1 ) to transform the vertical axis of figure 1@xmath57 . the fluid model likelihood function was computed for the full range of @xmath61 and @xmath18 shown . the dot - dashed lines bound the @xmath77 region that corresponds to the @xmath78 region of figure 1@xmath57 over which the scalar field model likelihood was computed . @xmath65 contours of constant @xmath62 ( eq . [ 2 ] ) in the @xmath78 plane for the time - variable @xmath3 scalar field model with potential @xmath15 . starting at the lower left hand corner , the contours correspond to @xmath79 and @xmath80 . 3. pdf confidence contours derived from the p99 fit c sne , for the spatially - flat time - variable @xmath3 scalar field model with potential @xmath15 . the dotted lines in panels @xmath70 show the @xmath71 to @xmath72 @xmath7 contours of figure 1@xmath57 . the solid lines in these panels show the corresponding contours derived from the sne data in conjunction with : @xmath64 @xmath4 measurements ( using eq . [ 3 ] ) ; @xmath65 @xmath14 measurements ( using eq . [ 4 ] ) ; and @xmath73 @xmath4 and @xmath14 measurements ( using eqs . [ 3 , 4 ] ) . panel @xmath74 compares the @xmath75 @xmath7 confidence contours from the sne data in conjunction with : the @xmath4 constraint ( long - dashed lines ) ; the @xmath14 constraint ( short - dashed lines ) ; and both constraints ( dotted lines ) . 4. pdf confidence contours for the spatially - flat time - variable @xmath3 scalar field model , with potential @xmath15 , derived using the three sne ia data sets . panel @xmath64 shows those derived from all the r98 sne , while @xmath65 is from the r98 sne excluding the @xmath20 one , and @xmath73 is for the p99 fit c data . solid ( dotted ) lines in @xmath70 are the 1 , 2 , and 3 @xmath7 contours from the sne data with ( without ) the @xmath4 and @xmath14 constraints ( eqs . [ 3 , 4 ] ) ; the dotted lines here are the solid lines in figure 1@xmath81 . panel @xmath74 compares the @xmath75 @xmath7 limits from the @xmath4 and @xmath14 constraints used in conjunction with : all r98 sne ( long - dashed lines ) ; r98 sne excluding the @xmath20 one ( short - dashed lines ) ; and the p99 fit c sne ( dotted lines ) . 5. pdf confidence contours for the time - independent @xmath3 model , derived using the three sne ia data sets . panel @xmath64 shows those derived from all the r98 sne , while @xmath65 is from the r98 sne excluding the @xmath20 one , and @xmath73 is for the p99 fit c sne . solid ( dotted ) lines in @xmath70 are the 1 , 2 , and 3 @xmath7 contours from the sne data with ( without ) the @xmath4 and @xmath14 constraints ( eqs . [ 3 , 4 ] ) . panel @xmath74 compares the @xmath75 @xmath7 limits from the @xmath4 and @xmath14 constraints used in conjunction with : all r98 sne ( long - dashed lines ) ; r98 sne excluding the @xmath20 one ( short - dashed lines ) ; and the p99 fit c sne ( dotted lines ) . in all panels , models with parameter values in the upper left hand corner region bounded by the diagonal dot - dashed curve do not have a big bang . the horizontal dot - dashed line demarcates models with a zero @xmath3 and the diagonal dot - dashed line running from @xmath82 to @xmath83 corresponds to spatially - flat models . 6. pdf confidence contours ( 1 , 2 , and 3 @xmath7 ) derived from the p99 fit c sne , for the spatially - flat time - variable @xmath3 scalar field model . panel @xmath64(@xmath65 ) ignores ( accounts for ) the @xmath4 and @xmath14 constraints ( eqs . [ 3 , 4 ] ) . the solid ( dotted ) lines use the non - informative @xmath68 ( flat ) prior . the dotted lines in @xmath64(@xmath65 ) are the same as the solid lines in figure 1@xmath57 ( 4@xmath57 ) . 7. pdf confidence contours ( 1 , 2 , and 3 @xmath7 ) derived from the r98 sne excluding the @xmath20 one , for the time - independent @xmath3 model . panel @xmath64(@xmath65 ) ignores ( accounts for ) the @xmath4 and @xmath14 constraints ( eqs . [ 3 , 4 ] ) . the solid ( dotted ) lines use the non - informative @xmath68 ( flat ) prior . the dotted lines in @xmath64(@xmath65 ) are the same as the dotted ( solid ) lines in figure 5@xmath63 . the dot - dashed lines are described in the caption of figure 5 . in the non - informative prior cases the likelihood function is computed down to @xmath84 .

when evaluating justification for new facilities , one may question whether and by how much the new technique is superior to already existing ones . we describe a general approach which leads to a quantitative answer , expressed as a relative figure of merit of two observational techniques . the approach is based on the comparison of the corresponding errors in the model parameters determination . the derived figure of merit can also be used in the closely related problem of optimization of complex observations , thus contributing in maximizing their scientific output . we illustrate the developed method by comparing the technique of optical long baseline interferometry to the classical spectroscopy , both used to find the parameters of the gaseous envelope around the p cyg star . for this first trial application , we use simulated data , obtained by solving the radiative transfer problem for grids of physical models of the envelope and then adding the photon noise . the chosen example of the interferometry vs spectroscopy comparison appears to be timely . indeed , in the recent years , interferometry has been undergoing a steady progress witnessed by publication of fringe visibilities for circumstellar envelopes of various types sometimes combined with a good spectral resolution ( mourard et al . @xcite , vakili et al . @xcite , @xcite ; quirrenbach et al . @xcite , @xcite , @xcite ; harmanec et al . @xcite ; stee et al . @xcite ; see also expected performances of interferometric arrays under construction , vlti , von der lhe et al . @xcite , petrov et al . @xcite , and chara , mcalister et al . @xcite ) . while this effort is fully justified , for the knowledge of the flux distribution on small angular scales can unveil important new features of astrophysical objects , it is also justified to wonder whether this effort is well payed for all classes of objects and for all kinds of physical problems . as a matter of fact , interferometric observations as compared to those at a single telescope are more complex , more expensive , more time consuming , and , when possible to compare , have a larger accumulated measurement error due to their complexity . furthermore , in the foreseeable future the instruments of this type will remain orders of magnitude less numerous and by all given reasons much less available for the community than single telescopes . thus , it makes sense to evaluate in quantitative terms what are the targets for which the interferometric instrumentation has the highest potential in terms of astrophysical information as compared to single telescopes . also , the optimization of interferometric observations may be a crucial aspect . indeed , the existing and forthcoming long baseline interferometers comprise only few individual apertures , covering a small set of spatial frequencies during one observing cycle . on the other hand , the information provided by observations can depend critically on spatial frequencies . certainly , in such a situation the scientific yield of the interferometric array could be greatly improved if thorough model calculations can indicate the configurations expected to be the most useful for determination of aimed physical parameters of the object . not surprisingly , the relative figure of merit derived in the present article depends not only on the experimental techniques , but also on the chosen set of model parameters , reflecting the fact that the physical model makes a necessary part of the merit problem . however , in modern astronomy , designing and building instrumentation on the one hand , and modeling objects on the other hand , were pushed to such a level of sophistication that they are as a rule two distinct activities . the optimum use of interferometry needs a close collaboration of two communities . after these preliminary remarks , let us specify that in the present first trial application , our discussion will be limited to the comparison of observations with a two - aperture optical long baseline interferometer ( hereafter olbi ) and the classical single telescope spectroscopy ( hereafter shortly spectroscopy ) , both applied to study a circumstellar gaseous envelope formed by the wind of a massive star . the olbi observations are assumed to have the same spectral resolution as spectroscopy . the discussion will be further limited to spectral profiles and visibilities in the h@xmath1 line of the p cyg envelope , allowing to use published high quality observations both in olbi and spectroscopic , and thus to provide a clear numerical illustration . in our earlier attempts to solve the olbi vs spectroscopy merits problem ( burgin & chalabaev @xcite ; bourguine & chalabaev @xcite ) , we used qualitative comparison of observables , visibility and spectral profile , computed for a small number of envelope models . the results were ambiguous . analising them , we arrived to the firm conviction that meaningful conclusions can be obtained only if ( 1 ) intercomparison of different types of observations is performed using a clearly defined figure of merit , indicating how much information on the studied object is provided by various techniques , and ( 2 ) a sufficiently large ranges in the space of envelope model parameters are analyzed . in the case of p cyg , the required number of computed models ( see sect . [ results ] ) turned out to be @xmath2 . an efficient simplified envelope model code , computing the emergent spectral profile and visibilities for the h@xmath1 line , was developed and is described in sect . [ model ] . we considered only spherically symmetric outflow models , the choice which is justified in sect . [ model ] . we consider astrophysical observations and their interpretation in the framework of the theory of parameter estimation ( e.g. kendall & stuart @xcite ) . that is , the observed object is assumed to be exactly described by a fixed multiparametric physical model , the errors being solely those of the experimental measurements ( see however sect . [ systerrs ] ) . any observable quantity is then a function of the vector of @xmath3 model parameters @xmath4 . an observation consists in finding the values of @xmath5 observables @xmath6 , where each observable @xmath7 is either measured directly or can be calculated in a model - independent way as a known function of directly measurable quantities . obviously , observations with instruments of different types provide observables of different number and nature . the @xmath8-th observable can be represented as @xmath9 , where @xmath10 is the `` theoretical '' value of corresponding observable in the absence of observational errors and @xmath11 is the random vector of observational errors , which is assumed here to be distributed normally . we consider the case when @xmath12 and the parameters @xmath13 could have been uniquely determined if the vector @xmath14 were known . we shall call the parameters to be determined the `` target parameters '' and denote a set of @xmath15 target parameters , or target set , as @xmath16 . it is possible that @xmath17 and the set @xmath16 is a proper subset of the set of all parameters of the model . the situation of this kind may arise for two reasons . first , a model parameter may be set to some a priori fixed value because , say , it has been measured earlier by entirely different methods . this reduces the dimension of the problem and computational difficulties , the general method remaining the same . let @xmath18 denote the set of @xmath19 fixed parameters , and @xmath20 be the value a priori assigned to a parameter @xmath21 . secondly , some of the unknown physical parameters may eventually be considered `` not interesting '' if their values are not relevant to the astrophysical problem under study . following the theory of parameter estimation , we will call them the `` nuisance '' parameters and denote the set of nuisance parameters , or `` nuisance set '' , as @xmath22 , their number being @xmath23 . as a rule , their influence can be separated in the error analysis only at the final stage of computations , for their values are to be calculated along with the values of target parameters . interpretation of an observation consists in calculating the vector @xmath24 that provides the best fit to the observed values . here @xmath25 is the index of the component of vector @xmath26 corresponding to the same model parameter as the @xmath8-th target parameter ( @xmath27 ) . the precision of the values @xmath28 resulting from interpretation of the observations is characterized by @xmath29 and @xmath30 , where for any random vector @xmath31 expressions @xmath32 and @xmath33 denote its mean and its covariance matrix respectively , and @xmath34 in what follows , we assume that @xmath35 is an unbiased estimate of @xmath14 , i.e. @xmath36 , and that the measurement errors are small enough , so that the error analysis can be performed using the linearized version of the least squares method . we additionally assume that all a priori fixed parameters are set to their true values ( see however [ systerrs ] ) , that is @xmath37 for any @xmath21 , then @xmath38 and statistical properties of errors in parameter determination are completely characterized by the covariance matrix of errors c = d. according to kendall & stuart ( ( * ? ? ? * chap.19 ) ) , the covariance matrix of errors c is related to @xmath14 and @xmath39 by the following expression : @xmath40 where @xmath41 is the @xmath42 matrix with elements @xmath43 for @xmath44 and @xmath45 is the vector of experimental errors defined in sect . [ general - theory ] . the natural scalar characteristics of the precision of parameter determination for a given set @xmath16 of target parameters is the following principal subdeterminant of the covariance matrix : @xmath46 geometrically , this subdeterminant is proportional to the hypervolume of the scattering ellipsoid in the space of target parameters . it depends not only on the physical model and errors of measurements , but also on the analytical form used for description of the model : two physically equivalent but mathematically different ( e.g. interrelated by a reversible substitution of variables ) analytical representations could yield entirely different values of @xmath47 . however , a pair of observational methods can well be compared if one makes use of the ratio @xmath48 where @xmath49 and @xmath50 are the subdeterminants calculated for the observation of the same object with instruments i and ii . this ratio depends only on the physical model used , and the obtained precision on the model parameters . we shall call the quantity @xmath51 , introduced by the eq . ( [ rdef ] ) , the `` random error ratio '' , or the `` relative figure of merit '' . if the instruments are of the same kind , and differ from each other only in precision , the value @xmath51 is merely the ratio of observational errors . if the instruments are of different kinds , providing observables of different nature and number , and the target set consists of only one model parameter , the value @xmath51 is merely the ratio of resulting random errors in the parameter determination . however , in the general case of instruments of arbitrary kinds and multiparametric models , no simple ratio of errors exists , and the evaluation of relative merits can be done only using the quantity @xmath52 defined by eq . ( [ rdef ] ) . it allows to compare various observational techniques applied to objects described by various multiparametric models : equality @xmath53 means that the instrument i is better suited for determination of parameters from the target set than the instrument ii . the fact that the description of an object by a multiparametric physical model is only an approximation to the reality implies that we have to consider the robustness of the method , that is the stability of the results it yields with respect to deviations of the real situation from the model . the present framework offers a way to obtain certain quantitative characteristics of the robustness . indeed , let us consider a multiparametric model with @xmath54 . if @xmath55 for some @xmath21 then , in general , @xmath56 for @xmath57 that is , in addition to random errors of observational origin , the result is biased by systematic errors due to inaccurate interpretation . the value of that bias characterizes the robustness of the method with respect to deviations of @xmath20 from its true value . in the linear approximation , @xmath58 , where @xmath59 is the @xmath60 matrix with elements @xmath61 where @xmath62 . a comprehensive study of systematic errors requires a joint analysis of individual elements of the matrix @xmath59 and uncertainties in model parameters from @xmath63 . in the present paper , we will develop a simplified approach providing semiquantitative indications concerning the relative robustness of different observational techniques . let us first define for each observational method under consideration the value @xmath64 when the estimates for target parameters are uncorrelated with estimates for parameters from @xmath63 , equality @xmath65 takes place : the method is robust ( with respect to deviations of specified form from the model ) . when such a correlation exists , @xmath66 exceeds unity ; the closer the correlation , the larger its value , eventually implying poor robustness of the method . further , to compare the robustness of methods i and ii with respect to inaccuracies in parameters from @xmath63 , we will define the `` robustness ratio @xmath67 '' as follows : @xmath68 where function @xmath69 is defined in eq . ( [ rdef ] ) . note that once the values of @xmath70 , which are necessary for analysis of random errors , are found , comparison of robustness is straightforward . in this section we obtain the expressions for the elements of the covariance matrix @xmath71 in the form that can be directly used to compare the olbi and spectroscopy . first , the measurement errors have to be specified . for this first trial application , the only considered source of random errors is the photon noise . other sources of error , in particular those in olbi arising from atmospheric seeing and calibrations of the modulation transfer function ( see roddier & lna @xcite ; mourard et al . @xcite ) are important and has to be incorporated in a future work . also , the spectral coverage and the spectral resolution have to be specified . we assume that they are the same for both kinds of observations ; the covered spectral region is @xmath72 $ ] , the spectral resolution is sufficiently high , i.e. the monochromatic intensity received from each point of the observed object remains nearly constant across each spectral channel . in other words , we assume that @xmath73 , the width of the spectral channels , in terms of velocity is lesser than the thermal velocity of the emitting atoms . for the p cyg type envelopes this corresponds to @xmath74 , which is easily satisfied with modern spectrographs . then one can formally consider the convenient limit case @xmath75 . in this case , the resulting covariance matrices tend to a limit which is independent of the spectral resolution ( cf . ( [ cspec ] ) and ( [ colbi ] ) ) . the physical quantity provided by spectroscopy is the flux density as a function of wavelength . in terms of observables , it is given by the vector @xmath76 with the components @xmath77 where @xmath78 is the number of photons recorded in the _ i_-th spectral channel during the exposure , and @xmath79 is the number of channels . the random values @xmath78 are related to the physical parameters of the object by @xmath80 where @xmath81 @xmath82 is the monochromatic flux from the object at wavelength @xmath83 , @xmath84 is the width of @xmath8-th spectral channel , and @xmath85 is the measurement error due to the photon noise , with the variance given by @xmath86 finally the factor @xmath87 is given by @xmath88 where @xmath89 is the instrument effective aperture and @xmath90 is the exposure time . hereafter we suppose that @xmath91 , where @xmath92 and @xmath93 pertain correspondingly to the spectrometer and interferometer . since the values @xmath92 and @xmath93 enter the final results only through the ratio @xmath94 , we can for the sake of simplicity set @xmath95 . thus , @xmath96 and the elements of the covariance matrix of measurement errors are given by @xmath97 substituting eqs . ( [ ys])([nncovar ] ) into eq . ( [ covmateval ] ) and proceeding to the limit @xmath98 we obtain : @xmath99 for spherically symmetric objects considered in the present paper , @xmath100 where @xmath101 is the monochromatic intensity at the angular distance @xmath102 from the center of the object . we consider the case of a two - aperture olbi with an adjustable baseline . the physical quantity eventually provided by olbi measurements is the fringe visibility v as a function of the baseline vector . the visibility v is equal to the real part of the complex degree of coherence , and therefore is the fourier transform of the brightness distribution in the focal plane by virtue of the van cittert - zernicke theorem ( e.g. mariotti @xcite , textbooks of perina @xcite , and goodman @xcite ) . however , the value estimated in practice is @xmath103 . we refer the reader for details to the thorough discussion by mourard et al . ( @xcite ) . in the present analysis we consider the estimates @xmath104 of the quantity @xmath105 where @xmath8 refers as previously to the @xmath8-th spectral channel , and @xmath106 is the projected baseline length of the interferometer . in the case of spherically symmetric objects the intensity distribution is circularly symmetric and @xmath107 is given by the normalized hankel transform as follows ( bracewell @xcite ) : @xmath108 where @xmath109 is the bessel function , and @xmath110 . statistical errors @xmath111 obey the equations @xmath112 and @xmath113 the validity of eqs . ( [ wwcovar ] ) and ( [ nwcovar ] ) for @xmath114 is evident . the case @xmath115 is treated in [ olbidiag ] . as a rule , the spectrum of the object , i.e. the values @xmath78 , is also recorded . consequently , the vector of observables for the olbi and the elements of its covariance matrix of errors are given by @xmath116 and @xmath117 substituting this expressions into eq . ( [ covmateval ] ) , we obtain in the limit @xmath98 that @xmath118 where @xmath119 is defined in eq . ( [ cspec ] ) . as discussed in sect . [ randerr ] , the relative figure of merit of olbi as compared to spectroscopy is the random error ratio : @xmath120 from eqs . ( [ colbi ] ) and ( [ risdef ] ) , it follows that @xmath121 , which merely reflects the fact that the observables of spectroscopy constitute a subset of the olbi observables . consequently , in the adopted comparison , the statistical errors in parameter determination by olbi are always lesser than those obtained when only spectroscopic data are used , if however the spectroscopic data are of equal precision . as explained in sect . [ spectroscopy ] , the eq . ( [ risdef ] ) is obtained in the assumption that @xmath91 . to compare a pair of instruments for which that equality does not hold , the rhs of eq . ( [ risdef ] ) should be multiplied by @xmath122 . the star p cyg ( b1 iape ) is characterized by a high mass - loss rate and exhibits bright optical emission lines , formed in the dense and nearly fully ionized stellar wind . it belongs to the class of luminous blue variables , a short - living transition phase in the evolution of a massive star at the end of the hydrogen burning ( see e.g. humphreys & davidson @xcite , langer et al . @xcite , maeder @xcite ) . there were at least two strong reasons to choose p cyg as the astrophysical object of the present study . first , this bright object , @xmath123 , is one of the best studied emission line stars . although in this first trial we use simulated data , it is important to note that for p cyg there exists a rich literature providing not only high quality spectroscopic data ( e.g. scuderi et al . @xcite ) , but also interferometric data ( vakili et al . @xcite ) , as well as thorough theoretical analysis ( drew @xcite ; pauldrach & puls @xcite ) . this insures that the present work can be followed up by a practical application . secondly , it happens that in the growing list of outflows known to be non - spherical ( e.g. wolf , stahl , fullerton @xcite ) , that of p cyg is an exception exhibiting spherical symmetry to a good degree of accuracy ( nota @xcite ) , becoming clumpy only on short time and small flux scales ( taylor et al . @xcite ; vakili et al . @xcite ; nota @xcite ) . this implies that the assumption of spherical symmetry of the envelope , used in the present work , is realistic . let us remind that it allows to compute a model in a reasonable amount of time , and thus to explore a large domain of parametric space , which is a requirement for the evaluation results to be meaningful . for the distance to the star @xmath124 , its radius @xmath125 and the effective temperature @xmath126 we adopted the following values : @xmath127 , @xmath128 , @xmath129 ( lamers et al . @xcite ; pauldrach & puls @xcite ) . the spectrum of the star was assumed to be blackbody . our study is limited to the hydrogen h@xmath1 line , which is the most prominent and the best studied feature in the spectrum of the star . the mass distribution within a spherically symmetric and stationary outflow is described by its mass loss rate @xmath130 and the outflow velocity @xmath131 , which we assume to depend on radial distance @xmath132 in the following way : @xmath133 where @xmath134 is the terminal velocity of the wind , @xmath135 is velocity of the wind at the base of the photosphere , and @xmath1 is a dimensionless parameter that characterizes the rate at which the velocity approachs its asymptotic value at large @xmath132 . the radial distance @xmath132 is measured in units of stellar radius @xmath125 , while @xmath130 , @xmath135 , @xmath134 , and @xmath1 are parameters of a model . the temperature in the envelope is often assumed to be constant . however , as it was shown by drew ( @xcite ) , across the region of the h@xmath1 line formation @xmath136 the temperature in the envelope can vary by as much as 6000k . therefore , along with isothermal models , we also computed the emergent emission for non - isothermal models . to keep a finite number of scalar parameters , the dependence of envelope temperature @xmath137 on radial distance is approximated in the following manner : it is assumed that @xmath138 for @xmath139 , that @xmath137 is a linear function of @xmath140 at each interval @xmath141 $ ] for @xmath142 , and that @xmath143 for @xmath144 . the values @xmath145 were fixed for each model of the family . we consider only non - increasing temperature laws @xmath137 , that is @xmath146 for @xmath147 , where @xmath148 . since the computations are organized in such a way that the model parameters vary independently on each other , it is the values @xmath149 that are used along with @xmath150 as the parameters defining the envelope temperature , and the vector of model parameters is given by @xmath151 the `` supersonic '' sobolev approximation is a well known efficient method of calculating line spectral profiles ( e.g. castor @xcite ) . however , it would have a low accuracy if applied in the p cyg case , for a noticeable fraction of h@xmath1 flux is emitted in the inner part of envelope , where the outflow velocity is comparable to that of sound . we adopted a mixed method computing the source function in the sobolev approximation , and then finding the emergent intensities by exact numerical solution of the transfer equation . as it was shown by hamann ( @xcite ) , the errors in line profiles computed in the original sobolev approximation come mainly from calculations of the emergent intensities , whereas the source function is accurate for a wide range of physical conditions . the choice of the model of the hydrogen atom was based on the fact that the regions of the envelope emitting the major fraction of the h@xmath1 flux are nearly completely ionized and opaque to lyman continuum and l@xmath1 , so that direct recombinations to and photoionizations from the ground level cancel out , and l@xmath1 is saturated ( drew @xcite ) . the populations of levels @xmath152 and @xmath153 are mainly defined by collisional and radiative transitions between these two levels , radiative ionizations due to stellar radiation , and radiative recombinations ( including indirect ) to the level @xmath153 . we therefore adopted the three - level + continuum model of hydrogen atom . the balance equations for @xmath154 and @xmath155 , the number densities of hydrogen atoms respectively at levels 2 and 3 , take then the following form : @xmath156 where @xmath157 is the spontaneous emission coefficient , @xmath158 and @xmath159 are collisional excitation and de - excitation coefficients for the transition , @xmath160 and @xmath161 are respectively ionizations coefficients and recombination rates for the levels involved , and @xmath162 is the escape probability . since the hydrogen is nearly completely ionized , the values @xmath158 , @xmath159 , @xmath163 and @xmath164 can be considered independent on @xmath154 and @xmath155 , the only source of non - linearity in eqs . ( [ balance ] ) being the terms containing @xmath165 . the escape probability @xmath162 is a rather complex function of the population of the lower level of the transition , so that commonly the eqs . ( [ balance ] ) are solved by iterations , and this and , in particular , the recalculation of the escape probability , is by far the most time consuming operation . to speed up computations , we developed a new approximate method for solving the equations of statistical equilibrium . it is based on the fact that , as shown in [ epapprox ] , the function @xmath166 can be approximated with sufficiently good accuracy by a simple analytic expression as follows : @xmath167 where @xmath168 @xmath169 and @xmath170 are statistical weights of the levels , and @xmath83 is the line wavelength . introducing the dimensionless variables @xmath171 where @xmath172 and @xmath173 is the energy of transition , and substituting the approximation ( [ betapp ] ) into eqs . ( [ balance ] ) we obtain after some algebra that @xmath174 is the positive solution of the quadric equation @xmath175 where @xmath176 here @xmath177 and @xmath178 when @xmath174 is calculated , the value of @xmath179 can be found using the relation @xmath180 the source function is then easily computed from level populations , which can be obtained using eq . ( [ pdef ] ) . integration of the transfer equation was performed using the code developed by bertout(@xcite ) , who kindly provided it to the authors . in computing the line profile , the code assumes the envelope to be isothermal , so that its use for a non - isothermal case requires some comments . the envelope temperature enters the calculations at two points : ( 1 ) in calculating the source function , through coefficients @xmath158 and @xmath159 of eq . ( [ balance ] ) . since the bertout code is applicable to arbitrary source function , variations in temperature does not cause any difficulties here . ( 2 ) in integrating the transfer equation , @xmath137 enters the result through the local doppler line width . since we consider only the envelopes with relatively low temperature contrast @xmath181 , and dependence of the doppler width on temperature is rather week , the error induced by this isothermal code structure is still negligible . once the intensity @xmath101 is computed , the observable quantities , i.e. the emergent flux @xmath182 and the visibility squared @xmath183 are obtained using eqs . ( [ fdef])([vaxisym ] ) . the relative figure of merit @xmath184,@xmath16,@xmath185,@xmath186 and the robustness ratio @xmath187,@xmath188,@xmath63,@xmath186 are calculated in two stages , implemented as separate programs . first , for the given grids of model parameters @xmath189 and the projected baseline lengths @xmath190 , we calculate and store the matrices @xmath191 and @xmath192 , defined in eqs . ( [ cspec ] ) and ( [ colbi ] ) . the defining parameters of the grids are entered as input data , the grid of models being constructed as the direct product of uniform grids for individual parameters . the derivatives entering definitions of the matrices are approximated by finite differences . all the model parameters that are not constant on the grid are treated as adjustable , that is either target or nuisance . the distinction between those two types of parameters is irrelevant at this stage . at the second stage , for given partitions of set @xmath193 of all adjustable parameters on the subsets @xmath16 and @xmath22 , we compute the values of @xmath184,@xmath16,@xmath22,@xmath186 and @xmath187,@xmath16,@xmath63,@xmath186 ( see eqs . ( [ rdef ] ) and ( [ sdef ] ) ) on the grid . in this way , the dependence of the results on @xmath16 and @xmath22 can be studied without repeating the time consuming physical modeling of the envelope . in total , the values @xmath194 and @xmath195 were calculated as a function of projected baseline length for various sets of target and nuisance parameters at more than @xmath0 points of the parametric space . table [ grids ] presents the principal characteristics of the model grids . the ranges of variations of model parameters on the grids were chosen so as to include the values of parameters from earlier works on p cyg cited in sect . [ stpars ] . lrllllllrrrrrrrr grid & @xmath196&@xmath197&@xmath1 & @xmath130&@xmath134&@xmath135 & @xmath150&@xmath163&@xmath198&@xmath164&@xmath199&@xmath200&@xmath201&@xmath202 & @xmath203 + no . & & & & @xmath204 & & & k & & k & & k & & k & & k + 1 & 10000 & 3&3.5 & 1.010@xmath205 & 200 & 15.0 & 11000 + & & & 4.5 & 2.010@xmath205 & & 55.0 & 15000 + 2 & 10000&3 & 3.5 & 1.010@xmath205 & 200 & 15.0 & 11000 & 2.0 & 1300 & 3.4 & 2100 & 5.0 & 200 & 10.0 & 3000 + & & & 4.5 & 2.010@xmath205 & & 55.0 & 15000 + 3 & 32805&4&3.5 & 1.010@xmath205 & 150 & 15.0 & 11000 + & & & 4.5 & 2.010@xmath205 & 250 & 55.0 & 15000 + 4 & 15625 & 6&3.5 & 1.010@xmath205 & 200 & 15.0&11000 & 2.0 & 0 & 5.0 & 0 + & & & 4.5 & 2.010@xmath205 & & 55.0 & 15000 & & 1300 & & 2300 + 5 & 46656 & 6&3.25 & 1.010@xmath205 & 200 & 15.0&11000 & 2.0 & 0 & 3.4 & 0 + & & & 4.25 & 2.010@xmath205 & & 55.0 & 15000 & & 1300 & & 2100 + 6 & 6561 & 8&3.5 & 1.010@xmath205 & 200 & 15.0 & 11000 & 2.0 & 0 & 3.4 & 0 & 5.0 & 0 & 10.0 & 0 + & & & 4.5 & 2.010@xmath205 & & 55.0 & 15000 & & 1300 & & 2100 & & 200 & & 3000 + 7 & 65536 & 8&3.5 & 1.010@xmath205 & 200 & 15.0 & 11000 & 2.0 & 0 & 3.0 & 0 & 4.0 & 0 & 5.0 & 0 + & & & 4.5 & 2.010@xmath205 & & 55.0 & 15000 & & 1300 & & 2100 & & 200 & & 3000 + 8 & 6561 & 8&4.0 & 1.510@xmath205 & 180 & 20.0 & 14000 & 2.0 & 0 & 3.0 & 0 & 4.0 & 0 & 5.0 & 0 + & & & 4.5 & 2.310@xmath205 & 220 & & 16000 & & 1000 & & 1000 & & 1000 & & 1000 + 9 & 19683 & 9&3.5 & 1.510@xmath205 & 220 & 15.0 & 16000 & 2.0 & 0 & 3.0 & 0 & 4.0 & 0 & 5.0 & 0 + & & & 4.5 & 2.310@xmath205 & 300 & 25.0 & 18000 & & 1300 & & 2100 & & 200 & & 3000 + the main result of the present study is that the increase in accuracy provided by olbi strongly depends on the number of free parameters in the model : it varies from almost negligible for models of low dimension to very significant for models of high dimension . since in our case the boundary between `` low '' and `` high '' is nearly coincident with division onto models with fixed and adjustable thermal structure , we discuss the results pertaining to those two classes of models separately . among our grids , the grid 1 most closely corresponds to the models often used in interpreting h@xmath1 observations of p cyg stars , where @xmath134 is excluded from the set of adjustable parameters of the model ( see e.g. scuderi et al . @xcite ) and its value is taken from the analysis of absorption lines of metals in the ultraviolet ( casatella et al . @xcite ; lamers et al @xcite ) . the typical results for an individual point in the parametric space are illustrated by fig . [ isothtpc ] , where the random error ratio @xmath206 is plotted as a function of the normalized baseline length @xmath207 for various @xmath16 and @xmath22 . here @xmath106 is the projected baseline length , @xmath208 , and @xmath209 is the angular radius of the central star . for the adopted values of @xmath210 , @xmath127 , and @xmath211 one gets @xmath212 . the principal features of dependence of @xmath213 on @xmath106 are independent on details of physical model used and can be easily explained qualitatively . as shown in sect . [ relfigmer ] , inequality @xmath121 holds for all @xmath214 . when the projected baseline length is very small , the interferometry does not provide any additional information as compared with the spectroscopy , for the object gets unresolved . consequently , @xmath215 for @xmath216 . in the opposite case of large @xmath106 the fringe contrast is close to 0 for all physically realistic intensity distributions and its dependence on model parameters is impossible to measure because of observational errors . that is in that case the interferometry again provides no additional information and @xmath215 for @xmath217 . it follows from the above stated general properties of @xmath213 that for any @xmath16 and @xmath22 it has a ( possibly non - unique ) minimum at a certain finite value of @xmath106 . the location of the minimum indicates the projected baseline length at which interferometric observations are most informative , and the corresponding value of @xmath213 characterize to what degree the interferometry can reduce the errors in model parameters determination as compared to spectroscopy . let us note that it would be insecure to compare the merits of the methods using values of @xmath218 obtained for an isolated point or even for few arbitrarily chosen points of parametric space . since the model parameters are not known a priori , it is necessary to use integral characteristics describing the behavior of @xmath213 on the whole grid . we will use the following three values calculated as a function of @xmath219 for various @xmath16 and @xmath22 : @xmath220 @xmath221 where @xmath222 is the number of points on the grid for which function @xmath223 reaches its minimum at @xmath224 . the function defined in eq . ( [ poptdef ] ) is immediately related to the problem of optimal choice of the baseline length : for given @xmath16 and @xmath22 , the higher the value @xmath225 , the higher the probability that the interferometric observations at the baseline length @xmath106 would yield most accurate model parameters . the dependence of @xmath226 and @xmath227 on the normalized baseline length and set of target parameters for grid 1 is shown in fig . [ isothrdt ] . for this grid , the value of @xmath228 is close to unity for all @xmath16 and @xmath22 and is not shown in the plot . it can be easily seen that for this grid of models olbi provides only relatively small reduction of the random error in parameter determination as compared with spectroscopy . this conclusion remains also valid if we add the terminal velocity @xmath134 to the set of adjustable parameters ( grid 3 ) . as @xmath229 for all @xmath16 and @xmath22 , the same approximate equality is evidently valid for the robustness ratio @xmath230 , which indicate ( see sect . [ systerrs ] ) that , for models with a priori fixed thermal structure , interferometric data will neither reduce significantly the level of systematic errors . the number of possible target sets is an exponentially increasing function of the dimension @xmath197 of the model parameter space , so that when the values @xmath149 are added to the set of unknowns , it becomes impossible to make an exhaustive presentation of even the integral results , i.e. the functions @xmath231 , + @xmath232 , and @xmath225 . in fig.[3lmerr ] , we show the dependence of @xmath226 , @xmath228 , and @xmath227 on the number of nuisance parameters for one of our simplest grids of models with adjustable thermal structure , the target set consisting of one parameter , the mass loss rate @xmath130 . from many points of view , it is the most important parameter of the stellar wind , characterizing both its effect on the evolution of the star and the influence of the outflowing matter on the surrounding interstellar medium . as it can be seen , the lower limit of the random error ratio @xmath226 rapidly decreases as the number of adjustable temperature parameters increases , so that when the thermal structure of the envelope becomes a nuisance parameter , interferometry can reduce the random error in @xmath130 by as much as an order of magnitude . note , that the decrease in @xmath226 is caused by the fact that parameters @xmath149 are adjustable , rather then by deviation from isothermicity . this can be seen from the upper two plots in fig . [ 3lmerr ] , where @xmath233 for a grid containing high proportion of models with noticeable temperature gradient , and is also supported by the results obtained for grid 2 , which differs from grid 1 only in that its thermal structure approximates that of drew s model b , which corresponds to @xmath234 , @xmath235 , @xmath236 , and @xmath237 in our notations , instead of being isothermal . for other single - parameter target sets the dependencies of @xmath226 and @xmath228 on @xmath22 are qualitatively the same . [ 3l1 + 5 ] displays the results for the most important case when all the parameters of the model except for @xmath134 are adjustable . as one can see , again , in favorable circumstances interferometry can increase accuracy by an order of magnitude . in contrast to @xmath226 , the value @xmath228 is close to unity for all projected baseline lengths and combinations of sets @xmath238 and @xmath22 that we have studied . this signifies that for any projected baselength there exists a combination of model parameters for which interferometry practically do not provide reduction in error . it is interesting to notice that for target sets containing several parameters ( see fig . [ 3lmultyt ] ) , the influence of individual parameters on @xmath226 and @xmath228 to some extent averages out : @xmath226 is systematically higher , and @xmath228 is systematically lower than the corresponding values for individual parameters . for this reason , when more than one model parameter is to be determined , interferometry at a single projected baselength is unlikely to yield overall gain in accuracy in excess of factor of two even at optimal baselines . we studied further the effect of increasing the number of sublayers @xmath239 , which permits to investigate the influence of fine details of the thermal structure of the envelope . this appears to be justified for drew ( @xcite ) showed that in certain cases this structure is quite complex , and gradient of @xmath240 varies strongly with radial distance . consequently , a model aiming to closely approximate the realistic envelope should be parameterized using a large @xmath239 ( see sect . [ stpars ] ) and hence requires computations on grids of high dimension @xmath197 . since the number of mesh points of the grid depends on @xmath197 exponentially , for computations based on such complex models ( grids 59 from table [ grids ] ) the limited computer resources dictate sparser mesh points along each axis of the parameter space and a tuning of parameters controlling the computational process in such a way as to increase the speed of computation at the expense of precision . although numerically less precise than computations for simpler models , our results for more realistic grids indicate that the conclusions derived for low @xmath239 can be safely extrapolated for higher @xmath239 . this point is illustrated in fig . [ 5l1 + 7 ] showing the results for single - parameter target sets for grid 7 . as it can be seen in figs . [ 3lmerr][5l1 + 7 ] and is confirmed by our other results not presented here , the distribution of optimal projected baseline lengths in the majority of cases has a maximum in the range 0.40.8 in dimensionless units . in some cases this maximum shifts to higher baseline lengths , but this occurs only for a very special choice of @xmath16 that correspond to measurements of wind temperature gradient at the distances of several stellar radii from the star with all other model parameters considered nuisance ( see e.g. bottom plots in figs . [ 3l1 + 5 ] and [ 5l1 + 7 ] ) . thus , the aforementioned range , corresponding to 4590 m in linear scale , can be recommended as the optimal choice for interferometric observations aimed at determination of global characteristics of the p cyg wind . the typical dependence of the robustness ratio + @xmath241 on the baseline length and the set of fixed parameters for a model with adjustable thermal structure is shown in fig . [ rbst ] . at nearly all points of parametric space @xmath242 , and @xmath243 is generally lower when the set of fixed parameters @xmath63 includes the values @xmath149 . for a small fraction of models and for short projected baselines where interferometry can not significantly reduce random error , the value @xmath244 by a negligible amount . thus , our results indicate that if the baseline length and the model used for interpretation are chosen in such a way as to reduce random errors , olbi appears also able to reduce the systematic errors . using the theory of model parameter determination , we developed a general method which permits a quantitative comparison of observational techniques and optimization of complex observations through the relative figure of merit , defined as a generalization of the ratio of random errors of model parameters . the method was applied to compare the outcome of olbi and classical spectroscopic observations , both used to determine the parameters of the outflow of the p cyg star . the observable quantities were computed using an efficient radiative transfer code and realistic grids of the envelope model . the olbi and spectroscopic measurements errors were simulated assuming that the only source of errors is the counting statistics , i.e. we considered a nearly - ideal interferometer . other main assumptions and simplifications of the present work were as follows : ( 1 ) the p cyg wind is stationary and spherically symmetric ; ( 2 ) the only olbi observable considered here is the visibility modulus ; ( 3 ) only the hydrogen h@xmath1 line is considered . the main conclusions of the present work are as follows : 1 . a meaningful evaluation of the olbi vs spectroscopy relative figure of merit requires exploration of large domain of the model parametric space ( about @xmath0 points computed in the present work ) . 2 . if the p cyg outflow can be accurately described by a simple model of a stationary spherically symmetric isothermal envelope , then the use of interferometry does not substantially reduce the random error in the model parameter determination . 3 . if the p cyg outflow requires more complex models , e.g. with a priori unknown dependence of the envelope temperature on radial distance , use of a nearly - ideal interferometer can reduce errors in model parameters by an order of magnitude as compared to the spectrometer of the same collecting power and noise level . the study of the relative robustness , i.e. the stability of the obtained model parameter determination with respect to systematic biases in the model , indicates that when the olbi and the physical model reduce random errors , they also tend to reduce the systematic error . the optimal projected baseline lengths for observation of p cyg with a nearly ideal interferometer lie in the range 4590 m. 6 . if several model parameters are to be determined simultaneously , the olbi observations at a single projected baseline can not provide a substantial error reduction . in this section we discuss to what extent our results are relevant for real observations , and how the methods developed here could be generalized to more complex cases . although general formalism presented in sect . [ statth ] remains valid for any source of errors provided that the linearized least - square method is applicable , eqs . ( [ cspec ] ) and ( [ colbi ] ) need to be modified if the accuracy of measurements is not shot noise limited . in the simplest case of uncorrelated errors in spectral channels , this modification reduces to multiplication of the integrands in eqs . ( [ cspec ] ) and ( [ colbi ] ) by corresponding ratios ( shot noise error)/(total error ) . if this ratios are nearly constant across the observed spectral range , then after appropriate scaling our results can be directly applied to real observations . the assumption of uncorrelated errors may be invalid for interferometric observations , the observables @xmath104 being obtained as a result of much more complicated processing of raw observational data ( cf . mourard et al @xcite ) then that considered in [ olbidiag ] . this point must be further analysed , and in future works , it may be necessary to use directly eq . ( [ covmateval ] ) instead of eq . ( [ colbi ] ) . in any case , a generalization on arbitrary measurement errors could be easily incorporated into existing code with no considerable increase in the required computer resources . the theory developed in sect . [ statth ] can be applied to the case of multiple observed lines without any modifications . formally , it reduces to changing integration domains in eqs . ( [ cspec ] ) and ( [ colbi ] ) . rapid evaluation of source function , on the contrary , would require a detailed study and development of efficient methods specific to each spectral line involved , because different lines can be emitted in different regions of the envelope and originate from different physical processes . the method used here can be generalized to visibility measurements done simultaneously at more than one baseline : the only modification required is the replacement of rhses of eq . ( [ colbi ] ) by the sum of corresponding expressions over given set of projected baseline lengths , the directions of the baselines being unimportant because of circular symmetry of intensity distribution . this can be done by modifying only the code performing the second stage of calculations ( see sect . [ compfinres ] ) and using the data prepared for the single - baseline case during the most time consuming first stage of computation . although time - averaged structure of p cyg wind can be considered spherically symmetric , polarimetric ( taylor et al . @xcite ) and interferometric ( vakili et al . @xcite ) observations revealed existence of time - dependent `` clumpy '' structures in the envelope , giving rise to deviation from spherical symmetry . in this case , measurements of the fringe phase can yield important information . this contrasts with the spherically symmetric case , where the fringe phase is irrelevant to physical modeling of the object . on the other hand , a unique determination of model parameters using only spatially unresolved spectroscopy appears impossible for the complex physical models resulting in two - dimensional brightness distributions , so that even the formulation of the comparison problem for spectroscopy and interferometry can be hardly done . the choice of the optimal set of baseline lengths for interferometry still remains the important problem , and the method developed here can be used for such an optimization after certain modifications . first , the fringe phases ( or observable linear combinations thereof ) should be included into the set of observables defined in eq . ( [ yolbi ] ) , and eq . ( [ colbi ] ) should be modified accordingly . this would require comprehensive error analysis depending on particular method used for extracting phase information from measurements . second , errors in parameter determination for observations performed with interferometer configurations @xmath245 and @xmath246 ( here @xmath245 and @xmath246 denote corresponding sets of simultaneously used projected baseline lengths ) can be compared using the ratio + @xmath247 instead of @xmath248 . the authors are grateful to a. labeyrie , d. mourard and f.vakili for encouraging discussion at the beginning of this project . we thank c.bertout who provided the code for computation of the emergent intensities . msb is grateful to the staff of observatoire de grenoble for the hospitality during his stay in grenoble . this research made use of simbad astronomical database , created and maintained by the cds , strasbourg , and nasa s astrophysics data system abstract service . part of this work was supported by cnrs through the grants pics @xmath249 194 france russie , astronomie haute rsolution angulaire " , gdr milieux circumstellaires " , and gdr and pn haute rsolution angulaire en astronomie " . in this appendix , we compute the non - trivial elements of the covariance matrix of errors used in sect . [ interferometry ] . since we consider here a single spectral channel , the index @xmath8 will be omitted everywhere . thus , the values @xmath252 , @xmath253 , @xmath254 , @xmath255 , @xmath256 , and @xmath257 in this appendix are identical with the values denoted as @xmath258 , @xmath259 , @xmath85 , @xmath260 , @xmath261 , and @xmath262 respectively in sect . [ statth ] . in practice , the error in @xmath263 depends on too many details of experimental techniques , atmospheric conditions and data processing methods to be analyzed in general form ( mourard et al @xcite ) . the present analysis is restricted to a highly idealized situation . namely , as well as in sect . [ spectroscopy ] , the only source of measurement errors taken into account is the photon shot noise . we assume that the static fringes are detected by counting the photons in nonintersecting channels that cover in total the interval of length @xmath264 of optical path difference ( opd ) . if z is the opd , which is measured here in units of @xmath265 , and @xmath266 is the photon counting rate per unit @xmath267 , then @xmath268 where @xmath269 is unknown fringe phase shift and @xmath252 is proportional to the total received flux . if we define @xmath270 and @xmath271 then assuming that @xmath272 and neglecting the terms that decrease as @xmath273 for @xmath274 we obtain that @xmath275 and @xmath276 an estimate for @xmath277 can be constructed by replacing the values in the rhs of eq . ( [ wvalue ] ) by their estimates that can be easily deduced from eqs . ( [ fcos])([nolbi ] ) . if the interval of opds covered by measurements is divided in @xmath278 subintervals of length @xmath279 and @xmath280 for @xmath281 , then estimate @xmath263 of @xmath255 based upon the number of photons received in each of these subintervals is given by @xmath282 where @xmath283 , @xmath284 is the number of photons received in the @xmath102-th subinterval , @xmath285 , and @xmath286 . if we assume further that @xmath263 as a function of @xmath287 can be linearized in the vicinity of zero and take into account that @xmath288 and @xmath289 , then after some algebra we obtain for the statistical characteristics of @xmath263 @xmath290 which proves the validity of eqs . ( [ wwcovar ] ) and ( [ nwcovar ] ) for @xmath115 . in the sobolev approximation , the escape probability for a line photon in the spherically symmetric accelerating outflow is given by ( see e.g. mihalas @xcite , sect . 14.2 ) @xmath291 where @xmath292 @xmath293 @xmath294 , @xmath131 and @xmath295 are the outflow and scattering atoms thermal velocities respectively , @xmath296 , @xmath297 is the line frequency , @xmath298 is the oscillator strength , and @xmath254 is the number density of atoms on the lower level of the transition . here and in what follows we consider the fixed point of the envelope at the distance @xmath132 from the center and do not show explicitly the dependence of @xmath299 , @xmath107 , @xmath254 , and other physical parameters of the envelope on @xmath132 . also , in expression for @xmath300 we neglected the stimulated emission , which do not introduce a noticeable error when the formation of h@xmath1 in the p cyg envelope is considered . our purpose is to find an approximation @xmath301 to @xmath299 that permits fast evaluation . in order that in solving the equations of statistical equilibrium the approximation could be used instead of exact definition , the former should retain principal mathematical properties of the latter reflecting underlying physics . namely , we require that @xmath302 and @xmath303 should monotonically decrease to zero as @xmath304 . the simplest function that obey the above stated requirements is @xmath305 where @xmath306 is a parameter . to choose the optimal value of @xmath306 , we consider the behavior of @xmath307 for large @xmath254 . when @xmath304 , the numerator of the integrand in eq . ( [ epdef ] ) can be replaced by 1 and asymptotic form of @xmath299 can be easily computed . comparison of that asymptotic form with eq . ( [ epappdef ] ) shows that if we set @xmath308 then @xmath309 up to the terms of order @xmath310 when @xmath304 . here @xmath311 and @xmath312 are statistical weights of the lower and upper levels of the transition respectively , @xmath83 is the line wavelength , and @xmath313 is the spontaneous emission coefficient . to study the precision of approximation ( [ epappdef ] ) , we performed numerical computations for a wide range of values @xmath314 and @xmath315 ( it can be easily shown that the error is uniquely determined by the values of these two parameters ) . our results show that the relative error reaches its maximum of 0.23 for @xmath316 , that is when @xmath317 , and @xmath318 . the approximation found here is applicable if the escape probability is given by eq . ( [ epdef ] ) , that is if the frequency change in scattering is described by complete frequency redistribution . of course , it is useful only if the usual conditions of applicability of the sobolev approximation are satisfied . the expressions ( [ epappdef ] ) and ( [ nas ] ) can be generalized in two ways . first , the error can be reduced if rational approximations of higher order are used instead of eq . ( [ epappdef ] ) . second , a similar approximation can be obtained for a general three - dimensional flow as it will be showed in a future work . p. , morand f. , bonneau d. , jiang y. , yang s. , guinan e.f . , hall d.s . , mourard d. , hadrava p. , bozic h. , sterken c. , tallon - bosc i. , walker g.a.h . , mccook g.p . , vakili f. , stee p. , le contel j.m . , 1996 , a&a 312 , 879

in recent years the sensitivity of the _ spitzer space telescope _ has made it possible to expand our knowledge of dusty disk emission dramatically , enabling statistically significant studies of disk emission as a function of mass and age over a much wider wavelength range than previously possible ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? in addition to confirming the rapid decrease of optically - thick disk emission previously inferred from near - infrared observations @xcite , _ spitzer _ observations have provided strong evidence that disk dissipation timescales depend significantly on host star mass @xcite , with a timescale for inner disk dissipation of 5 - 7 myr @xcite for low mass stars ( @xmath21m@xmath3 ) , and @xmath43 myr @xcite for intermediate mass stars ( 2 - 10@xmath5 ) . _ spitzer _ observations have also provided new clues to disk structure through spectral energy distribution ( sed ) analysis , from slight indications of faster dust settling in inner disks @xcite to strong evacuation of the inner disks in some stars that retain optically thick outer disks @xcite . spitzer data have also been used to show that debris disks around intermediate mass stars are more frequent and have larger 24 @xmath6 m excesses around 10 - 13 myr , when models predict that dust is produced by collisional cascades due to the formation of large solid bodies ( @xmath01000 km ) in the disk @xcite . the possible effects of environment on disk frequency and structure are less clear . in clusters with high mass stars , disk photoevaporation due to the strong ultraviolet radiation field can be important , and is directly observed in the orion nebula cluster @xcite . @xcite found a reduction in disk frequency of about a factor of two in the relatively young cluster ngc 2244 , but only in the innermost regions ( d@xmath20.5 pc ) . ngc 2244 is very rich and contains seven o type stars , so that the ultraviolet radiation field should be very strong . the discovery of a young cluster of low - mass stars around the @xmath1 vel system @xcite , with the central binary consisting of an o7.5 star and a wolf rayet star @xcite , provides an opportunity to study the possible effects of central star ultraviolet radiation fields on disk evolution . the wolf rayet star is classified as wc8 ( carbon - rich ) , implying a progenitor mass of @xmath740@xmath5 @xcite . in this paper we report _ spitzer _ observations of infrared emission in this cluster . in [ sec : obs ] we present the observational data used in this contribution , and discuss membership in [ s : mem ] . we analyze the infrared data [ sec : disks ] , and compare with disk frequencies in other young stellar populations in [ sec : comp ] , with our main results summarized in [ sec : conc ] . we have obtained near - infrared ( nir ) and mid - infrared photometry of the @xmath1 velorum cluster using the four channels ( 3.6 , 4.5 , 5.8 & 8.0 ) of the infrared array camera ( irac , * ? ? ? * ) and the 24 band of the multiband imaging spectrometer for spitzer ( mips ; * ? ? ? * ) on board the spitzer space telescope . these data were collected during 2005 february 23 ( irac ) and april 10 ( mips ) as part of a gto program conducted by the irac and mips instrument teams . the irac observations were done using a standard raster map with 280 `` offsets , to provide maximum areal coverage while still allowing @xmath020 overlap between frames in order to facilitate accurate mosaicking of the data . the mapped region was covered by a 9x10 position mosaic , with three dither exposures at each position . images were obtained in high dynamic range ( hdr ) mode , in which a short integration ( 1 s ) is immediately followed by a long integration ( 26.8 s ) . standard basic calibrated data ( bcd ) products from version s14.0.0 of the _ spitzer _ science center s irac pipeline were used to make the final mosaics . post - bcd data treatment was performed using custom idl software @xcite that includes modules for detection and correction of bright source artifacts , detection and removal of cosmic ray hits , construction of the long and short exposure hdr mosaics , and the merger of those mosaics to yield the final science images . the final mosaics have a scale of 1''.22 pixel@xmath8 . point source detections were carried out individually on each irac band using photvis ( version 1.09 ) , an idl gui - based photometry visualization tool developed by r. gutermuth using the daophot modules ported to idl as part of the idl astronomy user library @xcite . more than 40,000 sources were detected in at least one irac band . we extracted the photometry of these objects using the _ apphot _ package in iraf , with an aperture radius of 3.7 and a background annulus from 3.7 to 8.6 . we adopted zero - point magnitudes for the standard aperture radius ( 12 ) and background annulus ( 12 - 22.4 ) of 19.665 , 18.928 , 16.847 and 17.391 in the [ 3.6 ] , [ 4.5 ] , [ 5.8 ] and [ 8.0 ] bands , respectively . aperture corrections were made using the values described in irac data handbook @xcite . final photometric errors include the uncertainties in the zero - point magnitudes ( @xmath00.03 mag ) . mips observations were made using medium scan mode with full - array cross - scan overlap , resulting in a total effective exposure time per pointing of 40 seconds . the images were processed using the mips instrument team data analysis tool ( dat ) , which calibrates the data and applies a distortion correction to each individual exposure before combining it into a final mosaic @xcite . a second flat correction was also done to each image using a median of all the images in order to correct for dark latents and scattered light background gradients . the second flat correction is determined by creating a median image of all the data in a given scan leg , and then dividing each individual image in that scan leg by that median . bright sources and extended regions are masked out of the data before creation of the median , so the nebulosity seen in the map has a negligible effect on any noise that might be added by the second flat . we obtained point source photometry at 24 with iraf/_daophot _ point spread function fitting , using an aperture size of about 5.7 and an aperture correction factor of 1.73 derived from the stinytim psf model . the aperture size corresponds to the location of the first airy dark ring , and was chosen to minimize effects of source crowding and background inhomogeneities . the sources were identified using an idl photometry routine , based on the iraf/_daophot _ package , with a 10@xmath9 cutoff . any obvious point sources missed by the find routine were added in by hand . the absolute flux calibration uncertainty is @xmath04% @xcite . the photometric uncertainties are dominated by the background / photon noise . our final flux measurements are complete down to about 0.8mjy . ccd photometry was obtained with the 0.9 m telescope at the cerro tololo interamerican observatory ( ctio ) using a tek 2048x2048 ccd ( 13.5x13.5 arcmin@xmath10 ) . a region of 0.93x0.95 deg@xmath10 centered on the @xmath1 velorum system was surveyed in bvi@xmath11 with short ( 20 , 10 , 6s ) and long ( 300 , 120 , 60 s ) exposure times ( more details in * ? ? ? * r. d. jeffries et al , in preparation ) . photometry was performed on the reduced data using the optimal extraction algorithm detailed in @xcite and @xcite . stars closer than 3 to the central object have poor photometry because of the brightness of the @xmath1 velorum system . the optical catalog includes 24,730 sources in the irac field ranging from v=10 to v=21 and with photometric errors less than 0.1 magnitudes . we augmented the v , b optical data set toward the brightest objects using the @xcite catalog . a preliminary list of 15,712 objects ( hereafter sample 1 ) was created by selecting the optical sources that have photometric measurements in all irac bands within 3 of matching radius . we have cross - correlated sample 1 with stars in the 2mass point source catalog . about 30% of these sources do not have 2mass counterparts within a 3 matching radius ; however , many of these objects without a 2mass counterpart are likely to be background sources ( see [ s : mem ] ) . in general , optical , 2mass , and irac positions agree within 1 . sample 1 was cross - correlated with the second xmm - newton serendipitous source catalog ( 2xmm ) using a correlation radius of 3 , which is the maximum astrometric error expected for most of the sources in 2xmm . in an area of @xmath00.25 square degrees around the @xmath1 velorum system about 174 objects in sample 1 have x ray counterparts . finally , 32 stars were confirmed as members of the @xmath1 velorum cluster using the presence of the 6707 line in absorption , the equivalent width of the h@xmath12 line and/or radial velocity in a narrow range around 17 km / s ( * ? ? ? * ; jeffries r.d . et al in preparation ) . the spectroscopic data were obtained at ctio using the 4-m blanco telescope in conjunction with the hydra fiber spectrograph ( resolving power @xmath0 25000 ) . figure [ fig : map ] shows the field covered by the optical survey of the @xmath1 velorum cluster . solid lines indicate the region studied in this paper ( irac field ) , which is the area covered by the four irac channels and where sample 1 is located . the mips observation ( dashed lines ) covers most ( @xmath090% ) of this area . the 2xmm catalog covers about 30% of the irac field ; x ray sources are indicated on the image ( crosses ) . most of the members confirmed spectroscopically ( open squares ) are x ray sources . only three spectroscopic members do not have x - ray counterparts , and they are all located outside the 2xmm field of view . the ring - like diffuse emission structure around the @xmath1 velorum cluster confirms that some of the gas from which the cluster formed is still present . our membership selection is based on using the sequence defined by the spectroscopic members ( * ? ? ? * ; jeffries r.d . et al , in preparation ) and the x - ray sources from 2xmm . as most of the stars in sample 1 do not have spectroscopic data , we use photometric criteria to make a selection of potential members based on their colors and magnitudes . figure [ fig : cmd1 ] shows color magnitude diagrams ( cmds ) , v versus v - j ( left panels ) and v versus v - i ( right panels ) , illustrating the procedure to select photometric candidates of the @xmath1 velorum cluster . we assume a distance of 350 pc ( see [ sec : comp ] ) to plot the zero age main sequence ( solid line ) and the 5 myr isochrone ( dashed line ) from @xcite . since , it is well known that theoretical , non - birthline isochrones do not accurately match empirical isochrones for intermediate mass stars @xcite and that the opacity tables at low temperatures are incomplete @xcite , these theoretical isochrones are plotted as reference only and are not used for the selection of the photometric candidates ( however , evidence in favor of this age is detailed in [ sec : comp ] ) . open squares and crosses represent the spectroscopic members and the x ray sources , respectively . bright stars with optical photometry from @xcite are plotted as open triangles . the upper left panel shows that x - ray sources are highly concentrated along the nominal empirical isochrone , with a scattering of stars below this region in the cmd probably representing background low - mass stars or other non - members . all the spectroscopic members within the 2xmm field are x ray sources ( see figure [ fig : map ] ) . essentially none of the objects above the nominal isochrone are x - ray sources , and thus are mostly likely heavily - reddened giants . in addition , a number of the bright objects from @xcite are also far above the nominal isochrone so we reject them as members as well . as shown in the upper right panel of figure [ fig : cmd1 ] , stars without 2mass counterparts ( open diamonds ) are located well below the nominal isochrone and even below the zams ( at 350 pc ) , demonstrating that they are background objects . in the lower panels we plotted the regions of probable members ( large error bars ) . these regions were calculated using the median colors ( v - i and v - j ) of the member sample ( spectroscopic members plus x ray sources above the zams ) for 1.0 mag bins in the visual band . we used the differences between the observed colors and the expected colors ( the median ; represented by thick solid lines ) to calculate the standard deviation ( @xmath9 ) of the member sample . the regions of probable members are defined using 2.5@xmath9 limit from the median colors ( see * ? ? ? * ; * ? ? ? the lower left panel shows the stars above the zams on the v versus v - j diagram ( sample 2 ) , roughly defined using a straight line ; while the lower right panel shows the v versus v - i diagram for the stars selected using the region of probable members in the v versus v - j diagram ( sample 3 ) . our final list of photometric candidates includes 579 stars located within the probable member region of the cmds . using the 5 myr isochrone @xcite at 350 pc and a reddening of = 0.2 , we estimate that we can detect excess at [ 5.8 ] and [ 8.0 ] with signal to noise larger than 5 for a star with spectral type m5 or earlier ( at [ 3.5 ] and [ 4.5 ] the corresponding signal to noise is larger than 40 ) . for mips observations , the completeness limit roughly corresponds to spectral type early k ( see [ sec2:ir ] ) . table [ tab : members ] shows the optical , irac and mips photometry of the photometric candidates of the @xmath1 velorum cluster . column ( 1 ) shows the internal running identification number in our sample ; column ( 2 ) provides the 2mass object name ; columns ( 3 ) , ( 4 ) , and ( 5 ) give the optical magnitudes in the bands b , v and ic , respectively ; column ( 6 ) indicates the reference for the optical data ; columns ( 7 ) , ( 8) , ( 9 ) and ( 10 ) give the irac magnitudes in the bands [ 3.6 ] , [ 4.5 ] , [ 5.8 ] and [ 8.0 ] , respectively ; column ( 11 ) gives the flux at the 24 mips band ; column ( 12 ) indicates whether the star is a 2xmm source ; column ( 13 ) shows the disk classification based on the irac and mips analysis ( see [ sec : disks ] ) the giant branch of background stars clearly contaminates one portion of the cluster member region defined in figure [ fig : cmd1 ] . we plotted in figure [ fig : cmd2 ] the v versus v - j diagram for the photometric members illustrating the contamination level expected in several color ranges . since the reddening of the @xmath1 velorum cluster is quite low ( @xmath40.2 ; * ? ? ? * ; jeffries in preparation ) , we can estimate spectral types for the sources using their v - j color and the standard colors @xcite indicated in the figure . the upper panel shows the v - j distributions for the photometric candidates ( open histogram ) and for the spectroscopic or x - ray selected member sample ( solid histogram ) . within the 2xmm field , more than @xmath080% of the photometric members with v - j@xmath73.5 are x ray sources ; most stars without x - ray counterparts are very faint ( v@xmath719.5 ) or located near the 2xmm field border . this suggests that most photometric candidates with v - j@xmath73.5 are real members of the @xmath1 velorum cluster . normalizing the member sample histogram to the photometric candidates histogram in the color range v - j@xmath73.5 , we estimate the number of members expected as function of v - j color ( dotted histogram ) . the deviation between the expected member distribution and the photometric candidates distribution indicates the level of contamination for a given range of color ( or spectral types ) . the number of photometric candidates increases between v - j@xmath01.5 and v - j@xmath03.5 , where the giant branch crosses the member sequence . in this range of color , the contamination level by non - members of the @xmath1 velorum cluster is quite high ( @xmath1368% ) . since the number of members of stellar groups normally decreases toward higher stellar masses , the relative contamination at v - j@xmath41.5 can still be significant . figure [ fig : ccd ] shows the 2mass color color diagram , j - h versus h - k , for the photometric members of the @xmath1 velorum cluster . since the locus of giants separates from main sequence late type stars @xcite , we can identify background m type giants in this diagram . we have selected 27 stars in the locus of giants with j - h@xmath140.8 , these stars are labeled in table [ tab : members ] as m type giant candidates . taking into account this m giant sample , the contamination level where the giants branch crosses the @xmath1 velorum sequence in figure [ fig : cmd2 ] is still high ( @xmath1366% ) . figure [ fig : disk ] shows three diagrams used to identify and characterize stars bearing disks in the @xmath1 velorum cluster . the top panel shows the sed slope , determined from the [ 3.6]@xmath15[8.0 ] color ( sed slope=@xmath16/dlog[\lambda]$ ] ) , versus the [ 8.0 ] magnitude . dotted lines represent the 3 @xmath9 boundaries ( photospheric level ) , which is calculated using the photometric errors propagated from the [ 3.6]@xmath15[8.0 ] color @xcite . stars with excess emission at 8 can be identified in this diagram above the photospheric limit . in general , disk bearing stars in taurus @xcite are located above the dotted line , with an irac sed slope @xmath7 -1.8 ( see * ? ? ? * ; * ? ? ? * ) ; this limit is used to identify objects with optically thick primordial disks ( class ii ) . half of the stars with irac excesses are located below the class ii region suggesting that the disks around these stars have evolved significantly , and are in an intermediate phase between stars with optically thick disks and debris disk stars . we call these evolved disks . the bottom left panel shows the v - j versus k@xmath15[24 ] color - color diagram , in which we identify stars with 24 infrared emission above the photospheric level ( e.g. ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the k-[24 ] color distribution of stars with k-[24]@xmath41.0 describes a gaussian centered at k-[24 ] = 0.05 with @xmath9=0.14 . the 3 @xmath9 boundaries ( dotted lines ) represent the photospheric region at 24 ; stars with 24 excesses are located red - ward from this region . except for the three faintest stars with excesses at 8 ( inverse triangles in the upper panel ) , all stars with irac excess have excess at 24 . our sample of disk - bearing candidates includes 14 stars with excess at 8 and 24 , 3 stars with excess at 8 and with no mips detection , and 12 stars with excess at 24 and photospheric color in the irac bands . using the standard v - j colors in figure [ fig : cmd2 ] ( or in the bottom left panel ) , we can estimate the spectral types of stars with infrared excesses . we use three bins : early type stars ( v - j@xmath170.54 ; f0 or earlier ) ; solar type stars ( 0.54@xmath4v - j@xmath42.0 ; from early - f to early - k ) ; and late - type stars ( v - j@xmath72.0 ; k5 or later ) . the bottom right panel of figure [ fig : disk ] displays the distribution of the disk - bearing stars in a sed slope diagram , generated using the k-[5.8 ] and [ 8.0]-[24 ] colors . the [ 8.0]-[24 ] sed slope indicates whether the star has a flat or a rising sed ( sed slope @xmath14 0 ) at wavelengths greater than 8.0 , and hence whether it should be classified as a flat - spectrum or a class i source . we can also compare the properties of disks in the @xmath1 velorum cluster to those disks with strong evidence of dust evolution . primordial disks with a high degree of inner disk clearing have been called transitional disks ; observationally they are characterized by relatively small excess below 5.8 and a large excess at longer wavelengths implying a a sed slope [ 8.0]-[24 ] @xmath7 0 @xcite . we show in figure 4 the location of four well known transitional disks in taurus , chameleon , and the tw hya association : tw hyd ( 4 ) , gm aur ( 5 ) , coku tau ( 6 ) and cs cha ( 7 ) @xcite . in addition , we show in figure 4 the location of the newly discovered class of pre - transitional disks , in which gaps in primordial disks rather than holes have been identified @xcite . these are lkca 15 ( 1 ) , ux tau ( 2 ) , and hk tau ( 3 ) @xcite . in contrast to the disk population in the @xmath9 orionis cluster , the orion ob1b association , and the 25 orionis aggregate ( with ages of @xmath0 3 , 5 and 8 myr , respectively * ? ? ? * ; * ? ? ? * ) , there are no transitional disk candidate detected in the @xmath1 velorum cluster . however , there is a pre - transitional disk candidate ( star 52 ; labeled as ptd in table [ tab : members ] ) with infrared excesses similar to the pre - transitional disk systems ux tau and lk ca 15 @xcite . the late - type stars exhibit less infrared excess than the disk population in taurus , suggesting that the disks in the @xmath1 velorum cluster have evolved substantially . except for one solar type star ( see [ s : gtype ] ) , which exhibits irac and mips excesses , the early - type and solar - type stars show excesses only at 24 . the relatively small 24 excess ( k-[24]@xmath43 ) of these objects are consistent with their being debris disks , in which second generation dust is produced by collisions between planetesimals ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? . however , theoretical models of collisional cascade at 30 - 150 au @xcite do not predict enough excess emission to explain the observed excesses at 5 myr . a small dust contribution from collisional cascade in the inner disk ( @xmath430 au ) or remaining primordial dust in the disk could explain these differences ( see * ? ? ? * ) . since , the timescale for primordial disk dissipation decreases as the stellar mass increases ( e.g. * ? ? ? * ; * ? ? ? * ) , it is more likely to observe remaining primordial dust in the disk around solar - type stars than in early - type stars . visual inspection of the irac and mips images reveals that stars with infrared excesses are not contaminated by artifacts , nebulosity or extragalactic sources . in general , the stars with infrared excesses are isolated sources . only stars 115 and 359 have near faint sources ( within the background photometric annulus ) that could affect the photometry at [ 3.6 ] and [ 4.5 ] ; however , these objects look like isolated objects at longer wavelengths , where infrared excesses are more obvious . the star 8797 ( with elongated psf ) can be a close binary not resolved by irac . for this source , irac and mips photometry includes contributions from both components . we detected six early type stars with excess at 24 . figure [ fig : sed1 ] shows the seds of these objects . except for the star 285 ( @xmath1 vel d , * ? ? ? * ) , a binary star near the @xmath1 velorum system , all stars have spectral types from @xcite . @xcite classified @xmath1 vel d as a metallic - lined binary star with spectral type ranging from a2 to a8 ; we assumed a spectral type of a5 in figure [ fig : sed1 ] . color excesses ( e@xmath18 ) were obtained by interpolating spectral types in the table of standard v - j color given by @xcite . seds were corrected for reddening ( ) using e@xmath18 and the extinction relation from @xcite , with the value of total - to - selective extinction for normal interstellar reddening ( @xmath19 ) . the small values of obtained ( average=0.2@xmath200.1 ) support the low reddening reported for the @xmath1 velorum cluster ( * ? ? ? * ; * ? ? ? * jeffries r.d . et al in preparation ) . we use the color k-[24 ] to determine the excess ratio at 24 ( e.g. , * ? ? ? * ) , @xmath21 - 0.05)/2.5}$ ] , where 0.05 is the mean value of k-[24 ] for the photometric candidates with mips detections . star 105 exhibits a very small excess at 24 . as it is also an emission - line star @xcite and its 2mass colors are similar than those observed in classical be stars @xcite , this excess is likely to be free - free emission in a hot envelope - that is , this is probably the excess of a classical be star . the remaining stars with 24 excesses are likely to be debris disk systems with @xmath22 similar to those observed in debris disks candidates of the @xmath9 orionis cluster and the orion ob1b subassociation @xcite . most of stars with v - j@xmath170.54 ( f0 or earlier ) have spectral types from @xcite or @xcite . except for the stars 277 and 572 , the published spectral type and the photometric spectral type calculated from the v - j color ( see figure [ fig : cmd2 ] ) agrees within 5 spectral subclasses . however , additional studies ( e.g radial velocity ) are necessary to confirm membership of our early type sample . assuming that stars 277 and 572 are not members of the @xmath1 velorum cluster , we find 8 early type candidates without infrared excess ( diskless ) , 5 debris disk candidates and one classical be star . since some diskless could not be members of the cluster , the debris disk frequency estimated in this work ( 5 debris disks / 14 early type stars ; @xmath036@xmath2016% ) is a lower limit . @xcite report that mechanisms of gas dispersal , like photoevaporation , can disrupts planet formation in two important ways : ( 1 ) if the gas is dispersed before the disk dust has grown to centimeter size , the formation of planetesimals will be curtailed as the dust is removed with the gas ; ( 2 ) if the gas is dispersed before large rocky planets are formed , the formation of gas giant planets like jupiter will be suppressed . the debris disk frequency ( @xmath1336% ) and the infrared emission from the debris disks in the @xmath1 velorum cluster are comparable to the debris disk population of stellar groups with similar ages . for instance , using the same method to detect stars with excess at 24 , @xcite reported debris disk frequency of @xmath038%@xmath203% and @xmath046%@xmath204% for the orion ob1b subassociation ( @xmath05 myr ) and for the 25 orionis stellar aggregate ( @xmath08 myr ) , respectively . since the presence of debris disks implies the formation of km size objects @xcite , proposal ( 1 ) appears unlikely . we detected six solar - type stars with excesses at 24 . figure [ fig : sed2 ] shows the seds for this subsample . each panel shows the corresponding photospheric fluxes @xcite . spectral types were estimated by interpolating dereddened v - j colors using the standard table given by @xcite . we assumed a reddening value of = 0.15 ( e@xmath23=0.06 ; * ? ? ? the values of @xmath22 in this subsample are slightly smaller than the debris disks around fgk stars observed in the 15 myr - old scorpius - centaurus - lupus sub - association @xcite and comparable to the excesses of debris disks around fgk stars in the 30 myr old cluster ngc 2547 @xcite . assuming the same excess detection threshold of e@xmath24=1.41 ( k-[24]=0.42 ) , we have calculated a median e@xmath24 of 2.1@xmath25 , 2.8@xmath26 and 2.0@xmath27 for f g type debris disk candidates in the @xmath1 velorum cluster , scorpius - centaurus - lupus sub - association and the ngc 2547 cluster , respectively . the errors represent the upper and lower quartiles of each sample . if interpreted as an evolutionary sequence , these disk emissions would be consistent with a growth period of dust production and a following decrease as dust particles are removed , as predicted by theory @xcite . similar trends have been seen in a type stars @xcite . however , given the small number of stars , the statistical significance of this result is low . only star 316 shows an excess at shorter wavelengths ; it represents the earliest star exhibiting excess in the irac bands . the excesses around this star can be explained if the disk is in an intermediate phase evolving from a primordial to a debris disk ( evolved disk ) . another possibility is that this star has a very massive debris disk produced recently by a catastrophic collision between two planetary - size bodies @xcite . figure [ fig : sed3 ] shows the seds of late type stars sorted by brightness in the v band . we display the median sed of optically thick disk stars in taurus @xcite and the photospheric fluxes for a star with spectral type m2 @xcite . the diversity of disks in the @xmath1 velorum cluster is apparent . we distinguish between class ii systems ( cii , with sed slopes k-[5.8]@xmath71.8 ) , and evolved disk ( ev ) systems exhibiting smaller irac / mips excesses than class ii systems ( k-[5.8]@xmath41.8 ) . two stars have photometric colors between k5 and m0 ; one of them ( star 52 ) shows infrared excesses similar to pre - transitional disks objects @xcite . the other star ( 106 ) exhibits larger infrared excesses than the optically thick median in taurus . the next two objects ( stars 559 and 78 ) , with photometric spectral type @xmath0 m3 , exhibit modest infrared emission , which could be explained by decreasing the height of the irradiation surface in the disks due to a higher degree of settling producing flatter disk structures @xcite . the remaining 13 stars with infrared excesses have photometric spectral types m4 or later ( see figure [ fig : cmd2 ] ) suggesting , as in other young stellar population @xcite that primordial disk frequencies depend on the mass of the central object . using the number of members expected in our photometric sample ( dashed histogram in figure [ fig : cmd2 ] ) , we calculated disk fractions of 4.3@xmath203.0% for the color range 2@xmath4v - j@xmath43 ( @xmath0k5-m1 ) , 3.5@xmath202.4% for the color range 3@xmath4v - j@xmath44 ( @xmath0m1-m3.5 ) and 7.1@xmath202.0% for stars in with v - j@xmath74 ( later than m3.5 ) . stars 100 , 154 , 155 are too faint to be detected by mips . two of them show excesses in all irac bands , making identification of these stars ( 100 and 155 ) as having dust disks fairly certain . however , star 154 shows excess only at 8 , which could be produced by either disk emission or pah contamination . figure [ fig : space ] shows the spatial distribution of the stars with infrared excesses in our sample . different spectral type ranges are represented with different symbols . as reference , we identify x - ray sources and photometric candidates with v - j@xmath73.5 ( little contamination by non - members is expected in this color range ) . the projected distances from the @xmath1 velorum system are displayed in intervals of 0.5 pc . one debris disk is located inside 0.25 pc , while the only optically - thick disks are in systems located at projected distances of @xmath130.5 pc . since , about 10% of the stars bearing disks are expected to have transitional disks ( * ? ? ? * ; muzerolle j. , in preparation ) , we can expect one or two transitional disks in the @xmath1 velorum cluster . however , we did not detect any transitional disk objects in the cluster . we can speculate that external mechanisms of disk dispersal ( like photodissociation and/or photoionization by strong stellar winds and/or strong uv radiation ) , affecting mainly the disk surface and the outer region of the disk , combined with the fast inner clearing characterizing transitional disks , decrease the probability of observing a disk in this phase . given the small number of stars bearing disks , the statistical significance of this speculation is very low . to compare the disk frequencies in @xmath1 vel with other regions , we need to estimate the cluster age , which depends to some extent upon the assumed distance to the cluster . @xcite confirmed that this group is a stellar cluster belonging to the vela ob2 association , analogous to the @xmath9 orionis cluster in the orion ob1b subassociation @xcite . the primary star in the @xmath1 vel system is the only wr star with a distance from hipparcos ( 258@xmath28pc ; * ? ? ? * ) ; but this distance is in marked contrast to distances estimated by other methods ( 350 - 450 pc ) and to the hipparcos distance of the vela ob2 association ( 410@xmath2012 pc ; * ? ? ? * ) . using interferometric observations to estimate the geometrical parameters of the orbit of the @xmath1 vel system , @xcite and @xcite give distances of 368@xmath29 and 336@xmath30 , respectively . these values are in agreement with the distance obtained by ( * ? ? ? * pc ) with the new reduction of the hipparcos data that includes modeling of the satellite dynamics and eliminates the data correlation in the original catalog caused by attitude - modeling errors @xcite . the age estimated by ( 3.5 myr ; * ? ? ? * ) for the o star is in agreement with the age estimated by ( @xmath25 myr ; * ? ? ? * ) for the cluster and the age estimated by ( 3.6 myr ; * ? ? ? * ) for the @xmath1 vel system . we assume the corrected hipparcos distance of 350 pc , which is in agreement within 1.5@xmath9 from the distance calculated by @xcite and @xcite and a reference age of 5 myr . to support the assumed age , we plot in figure [ fig : cmd3 ] the median absolute magnitudes versus the median v - j color for members of several different stellar groups . regardless of the uncertainties introduced by theoretical evolutionary tracks , figure [ fig : cmd3 ] shows that the @xmath1 velorum cluster is in a similar evolutionary stage as the other stellar groups with ages normally quoted as @xmath05 myr : the @xmath31 orionis cluster @xcite , the cluster ngc 2362 @xcite , and the orion ob1b subassociation @xcite . we estimate an error of 1.5 myr comparing the standard deviation in the color v - j for the member sample ( see [ s : mem ] ) to the standard deviation obtained using the theoretical isochrones from @xcite with a reference age of 5 myr . figure [ fig : fracd ] shows the disk frequencies of late type stars ( stars k middle or later ) with near - infrared disk emission in different stellar groups , as a function of age @xcite . using the number of members with v - j@xmath72.0 ( @xmath0k5 or later ) expected in our photometric sample ( dashed histogram in figure [ fig : cmd2 ] ) and the disk detected in [ s : late ] , we calculated a primordial disk frequency of 6@xmath202% for the @xmath1 velorum cluster . we include recent spitzer results for the young stellar clusters ngc 1333 @xcite , ngc 2068/71@xcite , taurus @xcite , ngc 7129 @xcite , chameleons @xcite , tr 37 and ngc 7160 @xcite , ic 348 @xcite , ngc 2244 @xcite , ngc 2264 @xcite , @xmath9 ori @xcite , ngc 2362 @xcite , @xmath31 ori @xcite , upper scorpius @xcite , and orion ob1b and 25 ori @xcite . the disk frequencies decrease toward older ages with a time scale for primordial disk dissipation of @xmath05 myr . it is apparent that the disk frequency found in the @xmath1 velorum cluster is lower than those found in young stellar populations with similar ages , and comparable to the disk frequency in older stellar groups . this could indicate that the low disk presence observed in the @xmath1 velorum cluster is abnormal for its evolutionary stage , and environmental effects , such as strong stellar winds and/or strong radiation fields from the @xmath1 velorum system , could provide the physical mechanism for the low disk frequency . moreover , @xmath075% of the disk bearing stars in the @xmath1 velorum cluster show irac sed slopes smaller than the median values of other 5 myr old stellar groups plotted in figure [ fig : cmd3 ] suggesting that the disks of the @xmath1 velorum cluster have a higher degree of dust settling . in particular , the median irac sed slopes for the disk population of ngc2362 , the @xmath31 orionis cluster , the ob1b subassociation and the @xmath1 velorum cluster are -1.72 , -1.60 , -1.70 and -1.82 , respectively ( with a typical error of 0.06 ) . studying the young ( 2 - 3 myr ) open cluster ngc2244 , @xcite showed that high - mass stars ( o type stars ) can affect the primordial disks of lower mass members only if they are within @xmath00.5 pc of the high - mass star . we find similar results for the @xmath1 velorum cluster . using the photometric members with v - j@xmath73.5 ( @xmath0m2 or later ) , we find a disk frequency of 4@xmath203% at projected distance of 0.25 - 1.0 pc ; inside 0.25 pc the central objects of the cluster contaminate the optical photometry used to select photometric candidates ( [ s : mem ] ) . the closest primordial disk is located at projected distance of @xmath00.5 pc . at larger projected distance ( @xmath71.0 pc ) , the disk frequency is larger ( 8@xmath202% ) . this suggests that the relative fast dispersion of disks in the @xmath1 velorum cluster is produced by the strong radiation fields and strong stellar winds from the central objects . this result must still be considered tentative given the small number of stars with disks , which results in large errors in disk frequencies . alternatively , figure [ fig : fracd ] indicates that the disk frequency drops rapidly at @xmath05 myr and the relative low frequency of disks observed in the @xmath1 velorum cluster could be explained if the cluster is slightly older than 5 myr . additional studies of photometric members presented in table [ tab : members ] are necessary to explain the disk population found in @xmath1 velorum cluster . we have used the irac and mips instruments on board the spitzer space telescope to conduct a study of disks around the @xmath1 velorum cluster . since the central object is a binary system consisting of the closest known wolf rayet star and a high mass o star , a strong uv radiation field and stellar winds are present in the cluster . using optical photometry of x ray sources ( 2xmm ) and members confirmed by spectroscopy ( jeffries r.d . et al , in preparation ) , 579 photometric candidates were selected as possible members of the cluster . the level of contamination by non - members depends on the v - j color range , showing the highest level of contamination ( @xmath068% ) at v - j = 1.5 - 3.5 , where the field giant branch crosses the young stellar population . combining optical , 2mass and spitzer data we have detected infrared excess in 29 stars . one of the infrared excess stars is a be star . we report 5 debris disks around a type stars , 5 debris disks around solar type stars ( spectral type range f - early - k ) and one solar type stars with infrared excess produced by a very massive debris disks or by a primordial disk with a high degree of dust settling . seventeen disk bearing low mass stars ( k5 or later ) were found in the cluster with a range of disk properties . we classified these objects in three classes using the infrared excess at 5.8 and the sed slope [ 8.0]-[24 ] : 9 class ii stars , 7 evolved disks star and one pretransitional candidate . we found that 76% of the stars bearing primordial disks have color v - j@xmath74 ( @xmath0later than m3.5 ) , indicating a mass dependent timescale for disk dissipation in the @xmath1 velorum cluster , similar to results in other young star populations @xcite . we found that the disk frequency observed in the @xmath1 velorum cluster is lower than in other young stellar groups with similar ages ( @xmath05 myr ) , and comparable to values found in older populations ( @xmath010myr ) . empirical isochrones support the age of @xmath05 myr for the overall population of the @xmath1 velorum cluster . the relative low frequency observed in the cluster is consistent with : ( 1 ) rapid disk dissipation due to the influence of the strong radiation fields and winds from the wr and o star cluster members or ( 2 ) the age of the cluster is slightly older and the low frequency reflects the fast dropping in disk frequency observed at @xmath05 myr . more than 75% of the primordial disk stars have inner disk emission lower than the median observed in the disk population of other 5 myr clusters , indicating that the disks of the @xmath1 velorum cluster have a larger degree of dust settling . we find no primordial disks closer than @xmath00.75 pc from the high mass central binary . at projected distance @xmath71 pc , the disk fraction seems to be constant around 5 - 7% . this result is similar to that found in the ngc 2244 cluster @xcite in which primordial disks are affected only in the immediate vicinity of the o stars ( @xmath40.5pc ) . our results reinforce an emerging trend in studies of disk frequency , specifically that optically thick disk dissipation is rapid for @xmath32% of systems , lasting less than 5 myr , while @xmath33% of disks can last for 8 myr or more . we thank to aletta tibbetts r.d . for helping us in the initial steps of the irac reduction data . an anonymous referee provided many insightful comments . this publication makes use of data products from two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center / california institute of technology . this work is based on observations made with the _ spitzer space telescope _ ( go-1 0037 ) , which is operated by the jet propulsion laboratory , california institute of technology under a contract with nasa . support for this work was provided by university of michigan and nasa grants nag5 - 13210 , nasa-1277575 , and nasa-1285169 . lllllllllllll 1 & 08064389 - 4731532 & 19.21 @xmath20 0.02 & 17.65 @xmath20 0.01 & 14.98 @xmath20 0.01 & 1 & 12.39 @xmath20 0.03 & 12.38 @xmath20 0.03 & 12.33 @xmath20 0.04 & 12.32 @xmath20 0.04 & & 0 & ciii + 2 & 08064805 - 4728118 & 13.43 @xmath20 0.01 & 12.31 @xmath20 0.01 & 11.15 @xmath20 0.01 & 1 & 9.52 @xmath20 0.03 & 9.57 @xmath20 0.03 & 9.56 @xmath20 0.03 & 9.52 @xmath20 0.03 & & 0 & ciii + 3 & 08065006 - 4732219 & 19.65 @xmath20 0.03 & 18.22 @xmath20 0.01 & 15.50 @xmath20 0.03 & 1 & 12.75 @xmath20 0.03 & 12.66 @xmath20 0.03 & 12.66 @xmath20 0.06 & 12.71 @xmath20 0.07 & & 0 & ciii + 4 & 08065158 - 4728058 & 14.55 @xmath20 0.01 & 13.34 @xmath20 0.01 & 12.03 @xmath20 0.01 & 1 & 10.20 @xmath20 0.03 & 10.25 @xmath20 0.03 & 10.21 @xmath20 0.03 & 10.19 @xmath20 0.03 & & 0 & ciii + 5 & 08065419 - 4730210 & 14.54 @xmath20 0.01 & 13.34 @xmath20 0.01 & 12.05 @xmath20 0.01 & 1 & 10.23 @xmath20 0.03 & 10.25 @xmath20 0.03 & 10.20 @xmath20 0.03 & 10.17 @xmath20 0.03 & & 0 & ciii + 6 & 08070052 - 4732093 & 13.58 @xmath20 0.01 & 12.82 @xmath20 0.01 & 12.03 @xmath20 0.01 & 1 & 10.98 @xmath20 0.03 & 11.01 @xmath20 0.03 & 10.99 @xmath20 0.03 & 11.00 @xmath20 0.03 & & 0 & ciii + 7 & 08070716 - 4721462 & 12.99 @xmath20 0.01 & 11.90 @xmath20 0.01 & 10.74 @xmath20 0.01 & 1 & 9.16 @xmath20 0.03 & 9.18 @xmath20 0.03 & 9.16 @xmath20 0.03 & 9.12 @xmath20 0.03 & 1.59 @xmath20 0.61 & 0 & ciii + 8 & 08071155 - 4719512 & 16.09 @xmath20 0.01 & 14.77 @xmath20 0.01 & 13.13 @xmath20 0.01 & 1 & 11.10 @xmath20 0.03 & 11.10 @xmath20 0.03 & 11.07 @xmath20 0.03 & 11.05 @xmath20 0.03 & & 0 & ciii + 9 & 08071312 - 4721124 & 18.05 @xmath20 0.01 & 16.60 @xmath20 0.01 & 14.03 @xmath20 0.01 & 1 & 11.35 @xmath20 0.03 & 11.33 @xmath20 0.03 & 11.26 @xmath20 0.03 & 11.22 @xmath20 0.03 & & 0 & ciii + 10 & 08071363 - 4725155 & 13.39 @xmath20 0.01 & 12.73 @xmath20 0.01 & 11.94 @xmath20 0.01 & 1 & 10.86 @xmath20 0.03 & 10.84 @xmath20 0.03 & 10.81 @xmath20 0.03 & 10.83 @xmath20 0.03 & & 0 & ciii +

the chromosphere is in a highly dynamic state , varying on timescales of minutes or less . much of the dynamics in the magnetized regions associated with the magnetic network and plage are dominated by short - lived , jet - like features . a whole range of names has been applied to describe these chromospheric features . at the quiet sun limb , they are traditionally called spicules , where they are observed in as thin , elongated features that develop speeds of 10 - 30 km s@xmath1 and reach heights of on average 5 - 9 mm during their lifetimes of 3 to 15 minutes @xcite . their widths ( 0.2 - 1 mm ) and dynamics have , until recently , been very close to the resolution limits of observations . as a result of this limitation , the superposition inherent in limb observations , and the difficulty associated with interpreting h@xmath0 data , the properties of spicules have not been very well constrained , which has led to a multitude of theoretical models ( for a review , see , * ? ? ? the same issues have plagued a solid identification of a counterpart to spicules on the disk . observations on the disk in h@xmath0 reveal dark mottles in quiet sun regions . these dark mottles seem similar to spicules , and usually appear in close association with small flux concentrations ( and bright points ) in the magnetic network that outlines the supergranular cells . there has been a long - standing discussion on whether spicules and mottles are the same phenomena . this discussion has focused on the different velocity distributions of mottles and spicules to conclude that mottles are not the disk counterpart of spicules @xcite . however , several groups have used more recent data to argue for a close relationship between mottles and spicules ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? dynamic , jet - like features can also be seen in and around active region plage regions . these features are shorter ( 1 - 4 mm ) and shorter - lived ( 3 - 6 minutes ) and appear to form a subset of what have traditionally been called active region fibrils . there are also fibrils that do not show jet - like behavior ( fig . [ fig_context ] ) . these are apparently low - lying , heavily inclined and more static loop - like structures that connect plage regions with opposite polarity magnetic flux . in the remainder of this paper , we will not discuss these low - lying fibrils , but focus on the motions of the shorter dynamic fibrils ( hereafter abbreviated as dfs ) . only recently have dfs become the focus of detailed study , with several papers studying their dynamics , influence on the transition region and their oscillatory properties @xcite . in the past , several authors have reported periodicities and oscillations in spicule observations , usually with dominant periods of order 5 minutes . such periodicities in radial velocity , half - width and line intensity have been suggested based on older observations at the limb , but their interpretation has been difficult due to the complicated line formation of h@xmath0 , line - of - sight superposition and limited spatial resolution @xcite . more recently , @xcite have discovered significant oscillatory power in dfs in active regions using high - resolution data from the swedish 1 m solar telescope @xcite . not all dfs show oscillations or recurrence , but a significant number do . most of the power in especially more inclined dfs resides in periods between 4 and 6 minutes @xcite . similar periods have been found by @xcite in mottles in enhanced network or weak plage . @xcite and @xcite developed a new numerical model to explain the cause of these ( quasi)periodic mottles or dfs . they compared their model to data from the transition region and coronal explorer ( trace * ? ? ? * ) and suggested that inclined magnetic fields allow leakage of normally evanescent photospheric p - modes ( which have dominant periods of 5 minutes ) into the atmosphere where they form shocks that drive dfs . a recent paper by @xcite uses extremely high resolution observations obtained at the sst and significantly more advanced numerical modelling to show that dynamic fibrils are indeed caused by magneto - acoustic shock waves . in this paper , we directly expand on the work by @xcite . here we perform a more comprehensive analysis of the properties and motion of dfs using the same extremely high spatial and temporal resolution sst dataset of @xcite . in addition , we use advanced 2/3d numerical radiative mhd simulations that include the convection zone , photosphere , chromosphere and lower corona to interpret all of the measurements and correlations found in the sst data analysis , many of which have not been described in @xcite . in [ sec : instrumentation ] we discuss the details of the observations and instrumentation , including a description of a novel , advanced optical setup that allowed a diffraction limited 88 minute long timeseries in linecenter . we describe the typical temporal evolution and measure properties of dfs in [ section : dfdescription ] . this section also deals with intriguing regional differences and correlations in df properties that seem to be related to the oscillatory behavior of dfs . to interpret these findings , we use a numerical model that is described in [ section : discussion ] . this section also summarizes our observational findings and contains a more detailed description of the new model ( introduced by * ? ? ? * ) that explains our observations . our results strongly suggest that dfs naturally form as the result of chromospheric shock waves driven by convective flows and ( global ) oscillations in the photosphere as suggested by @xcite . in [ section : qs ] we discuss in more detail how this model and our observations of active region dfs relate to equivalent quiet sun phenomena . the dynamics of quiet sun mottles have been studied extensively by several groups recently @xcite . while the lack of high spatial and temporal resolution data has made it difficult to properly interpret the findings , this recent work seems to have resolved a long - standing issue @xcite regarding the paths mottles follow during their ascent and descent . @xcite and @xcite use data to establish that mottles follow parabolic paths . in fact , observations by @xcite indicate that spicules at the limb similarly trace parabolic paths . mottles initially appear in the blue wing of extending in the process to lengths of some 2 to 8 mm @xcite . after reaching maximum extent , they fade from the blue wing and appear in the red wing of to recede generally along the same path . throughout most of their lifetime they appear as dark , elongated features in linecenter . the measurements of the parabolic paths traced by mottles have been difficult to interpret . the projected decelerations of the top of the mottle are generally observed to be far less than solar gravity . attempts have been made to interpret such low decelerations as projection effects superposed on ballistic flights . however , the projection angles necessary are typically not compatible with measured doppler shifts @xcite , ruling out a purely ballistic interpretation . in addition , a purely ballistic flight would imply large initial velocities that have not been observed @xcite . these observational findings have been difficult to reconcile with the plethora of models for spicules or mottles @xcite . in [ section : qs ] we describe the similarities between mottles and dfs in more detail , and discuss the applicability of our model to quiet sun mottles . our results are summarized in [ section : summary ] . the observations were obtained with the swedish 1-m solar telesope on la palma . the solar optical universal polarimeter ( soup , * ? ? ? * ) was used to obtain narrow - band images in the linecenter of ( filter fwhm 12.8 pm ) . to compensate for seeing deformations in real - time , the observations were aided by the use of the sst adaptive optics system ( ao , * ? ? ? * ) consisting of a tip - tilt mirror , a bimorph mirror with 37 actuators , a correlation tracker , and a shack hartmann wavefront sensor . as a further measure to reduce seeing aberrations , the data were post - processed with the multi - object multi - frame blind deconvolution image restoration method ( momfbd , * ? ? ? * ) . this method is complementary to the use of ao since it reduces residual aberrations not corrected by the ao and it corrects the areas in the fov that are not used for the ao wavefront sensing . the science cameras consisted of 3 new , high speed cameras of type sarnoff cam1m100 , with a ccd readout time of only 10 ms and an effective exposure time of 15 ms , resulting in a frame rate of approximately 37 frames - per - second ( fps ) . the setup used was specially built for momfbd processing of the data and consisted of one camera behind the soup filter and 2 `` wideband '' cameras in front of the soup filter but after the soup pre - filter ( a 0.8 nm fwhm interference filter centered at 656.3 nm ) . the light for the wideband cameras was split off the main beam by a 90/10 beamsplitter in front of the soup filter and split again evenly for the 2 wideband cameras , which were set up as a phase diversity pair with one camera at 12 mm defocus on a 45 m focal length . a schematic drawing of the red beam optical set - up is shown in fig . [ fig_optical_setup ] . to ensure simultaneous exposure , a specially designed chopper with a useful aperture of 22@xmath222 mm and a variable duty cycle of @xmath3 - @xmath4 was used as a shutter . to guarantee correct labeling of all camera frames , a unique exposure number was generated by the shutter electronics and stored on the camera pcs . this scheme guarantees that simultaneously exposed frames can be uniquely identified , a feature that is required for momfbd processing but difficult to achieve using the internal clock of the observing pcs at the frame rates used . the large data rate produced by the cameras ( 72 mb@xmath5s@xmath1 per camera ) was reduced by a factor of 3 using real - time data compression , sent from the camera computers to disk - writing computers through ordinary gigabit network connections , and then passed through fiberchannel links to the raid controllers of an atabeast system . more details of the data acquisition , as well as first results from this unique dataset are described by @xcite . the observations were obtained on 04-oct-2005 and consist of a time series of about 78 minute duration , centered on one of the larger spots in the young active region ar10813 at s7,e37 . the observing angle @xmath6 between the line - of - sight and the local vertical is @xmath7 ( @xmath8 ) . the field - of - view ( fov ) was about 64@xmath264 at a pixel scale of 0063 . this corresponds to a fov of 46,400 x 46,400 km , with a pixel scale of 45.7 km . the diffraction limit of the sst at 6563 is 0.165 , i.e. , 120 km . the images from all cameras were jointly processed with the momfbd image restoration method , yielding 2 restored images for each restoration ( 1 wide band and 1 ) . since the common field - of - view of the cameras within an momfbd set is separately aligned to sub - pixel accuracy , the restored images have near - perfect alignment . sets from 12 exposures were used per restoration ( i.e. , 36 images in total ) resulting in a time series with 3 fps . for this study , we selected every third image , giving a time series with a fixed cadence of 1 s between images . the time series were corrected for diurnal field rotation , aligned and de - stretched . the ( local ) offsets were determined on the wide band images and subsequently applied to the images . the field of view of the time series contains two small sunspots of noaa active region ( ar ) 10813 . the larger of the spots is split into two parts by a lightbridge as seen in the left panel of fig . [ fig_context ] . the spots are surrounded by areas of plage , which includes some pores to the north of the larger spot . the chromosphere , as imaged in h@xmath0 linecenter ( right panel of fig . [ fig_context ] ) , shows many almost horizontal fibrils that connect the spots to surrounding opposite polarity pores and plage . many of these horizontal fibrils are associated with the superpenumbra of the larger spot . these fibrils are longer ( up to 20 - 40 ) and more static than the dynamic fibrils that are the focus of this paper . most of the dfs are associated with the plage in the region outlined by x=35 - 60and y=20 - 45 . the chromosphere in this region is dominated by relatively short jet - like features that appear to rise and fall within a matter of minutes . during their ascent and descent , individual dfs seem to follow the same , roughly straight , trajectory . at any given time , scores of dfs are visible in this region of interest . dfs often seem to occur in semi - coherent patches which contain several dfs at any given time . the location and orientation of these patches seems to be governed by the topology of the underlying magnetic plage region . within such a patch , it is sometimes difficult to uniquely identify individual dfs because of the semi - coherent behavior over a few arcseconds : some of these coherent fronts show significant substructure . the region which is dominated by dfs appears somewhat brighter in h@xmath0 linecenter than surrounding regions . this may be because hot coronal loops are connected to this region . such a correlation has previously been described by @xcite who find a correlation between the brightness in h@xmath0 linecenter and that of upper transition region moss emission in fe ix / x 171 ( using trace data ) . bright moss emission occurs at the footpoint regions of hot ( @xmath9 mk ) coronal loops . a full disk magnetogram taken by soho / mdi , centered in the middle of the time series , at 09:39 ut on 4-oct-2005 , shows that the two spots and plage in the region of interest share the same polarity ( see contours of fig . [ fig_context ] ) . most of the associated magnetic flux seems to connect to opposite polarity to the north and west of the spots , for example the pores to the north and the plage outside the field of view to the north . this can be deduced from the potential field extrapolation shown in the left panel of fig . [ fig_potext ] . the field extrapolation seems to correlate reasonably well with features outlined by h@xmath0 in some regions , including our region of interest . however , the extrapolation deviates from the h@xmath0 observations in several regions of the active region to the north and east of the spots , as well as close to the pores to the north of the large sunspot . some df properties vary significantly with position in the field of view . several regions were defined : region 1 is at the edge of a plage region where the magnetic field seems more inclined ( see discussion below ) . region 2 covers a dense plage region ( adjacent to region 1 ) where the fields seem more vertical . the average magnetic field strength , as derived from a full - disk mdi magnetogram ( with 2pixels ) , is of order 50 - 100 mx @xmath10 for region 1 , and 150 - 200 mx @xmath10 for region 2 . the other regions contain fewer dfs and their magnetic topology is not as well defined as in regions 1 and 2 . the right panel of fig . [ fig_potext ] illustrates the location of the two main regions ( blue is region 1 , red is region 2 ) . through visual inspection of a time lapse movie of the linecenter images , we found many fibrils in the plage region in proximity to the two spots near the center of the image . we analyzed dfs that occurred throughout the 78 minute long time series . to analyze the properties of the dfs , we manually choose the general direction of a df ( or set of dfs ) . a guideline is drawn in this direction , and a subset of the data , parallel to this guideline , is extracted for further analysis . this procedure results in detailed movies of individual dfs , as well as xt-plots that show the evolution of the extent of the df along the guideline as a function of time . a total of 257 dfs were analyzed in this fashion . in figs . [ fig_spiclife_1 ] , [ fig_spiclife_2 ] , and [ fig_spiclife_3 ] we present the evolution of three typical dfs . figure [ fig_spiclife_1 ] shows a df in region 2 . the six panels show that the df rises rapidly to a maximum length in roughly 130 s and recedes along the same path some 260 s after first becoming visible . it is darker than its surroundings , especially at maximum extension , and seems to follow a straight path in the direction that is shared by all surrounding features . the df has some internal structure , most clearly visible at @xmath11 s and @xmath12 s. the maximum extent is relatively modest at some 2 . despite the substructure , the df is thin with a width of some @xmath13 . it is also well isolated even though a number of other dfs are visible in the immediate vicinity . the length of the df as a function of time is well described by a parabola , as shown in the xt-plot in the top panel of fig . [ fig_spicxt ] . this is the case for the vast majority of the dfs that do not suffer from superposition and where the df guideline was chosen with a correct angle , i.e. where the df motion runs along the guideline . all 257 dfs used in the analysis below meet these conditions . each individual df is therefore fitted with a parabola and the deceleration , maximum velocity ( on ascent or descent ) , maximum extent and duration were derived . to determine the beginning and end points of the parabola , we use the following method . the steep parabolic ascent and descent of a typical df ( and the very high cadence of our data ) allows for a robust determination of the beginning and end time of each df . the intersection ( in the xt-plot ) between the parabolic path and a horizontal line at the beginning time or end time is used to determine the location of the `` root '' of the df at , respectively , the beginning and end of its lifetime . we then use these points in the `` xt''-plot to fit a parabola to the path followed by the df . the prevalence of parabolic paths is illustrated in all three panels of fig . [ fig_spicxt ] , where a wide variety of dfs are evident . the df studied in this case ( top panel ) follows its fitted parabola very well with a constant deceleration given by 216 m s@xmath14 , about two thirds of solar gravity ( 274 m s@xmath14 ) , a maximum velocity of 27 km s@xmath1 , and a maximum extent of some 1800 km . the properties for this df and all others discussed in this paper are measured along the projected path . an example of a much wider df in region 2 is shown in fig . [ fig_spiclife_2 ] . this df has extensive internal structure and rises over a front of about @xmath15 , a width that is largely maintained throughout the descending phase of its relatively short lifetime of 188 s. while rising , the df width is not constant along its axis but the df is wider at the bottom . it is typical that such wide dfs do not rise and fall as a rigid body but rather show phase and amplitude variations in velocity at various positions away from the axis . it is also worth noting that this df clearly becomes darker as it rises and brightens somewhat as it falls back along the same path it rose along . the xt-plot in the middle panel of fig . [ fig_spicxt ] reveals that this df also follows a path that is very close to a parabola with a deceleration of 286 m s@xmath14 , a maximum velocity of 27 km s@xmath1 , and a maximum extent of 1300 km . a longer lived df from region 1 with a lifetime of some 290 s is shown in fig . [ fig_spiclife_3 ] . this df is located in a region where images indicate that highly inclined magnetic field lines connect to the region of opposite polarity plage and pores to the west , in the top right corner of fig . [ fig_context ] . the df is wide , roughly 1 , and relatively long . it is also superposed on another df that is evident in the bottom panel of fig . [ fig_spicxt ] . detailed examination of the movie accompanying fig . [ fig_spiclife_3 ] shows the separate evolution of both dfs . the superposition makes it difficult to determine whether these dfs are formed at the exact same location . residual effects of image distortion from atmospheric seeing can cause the jagged appearance of these xt profiles . despite such distortions and the effects of superposition , a good fit for this df can still be obtained with a deceleration of 200 m s@xmath14 , a maximum velocity of 28 km s@xmath1 , and a maximum extent of some 2000 km . the properties of the parabolic fits to the xt-plots of 257 dfs were calculated . visual inspection of df time lapse movies confirmed the proper alignment of the dfs and quality of the parabolic fits . the upper left panel of fig . [ fig_hist ] shows that the dfs suffer a deceleration of on average 146 m s@xmath14 with a significant spread ranging from 40 to 320 m s@xmath14 and a standard deviation of 56 m s@xmath14 . this is significantly less than solar gravity . note that the median of all df properties is illustrated and described in the caption of fig . [ fig_hist ] , whereas we enumerate average values in the text . also note that these deceleration values are twice as high as the values reported in @xcite . the latter were mistakenly divided by two . the maximum df lengths are on average @xmath16 km , ranging from 400 to @xmath17 km with a standard deviation of 620 km , as shown in the upper right panel of fig . [ fig_hist ] . the lengths found for dfs are smaller than the widely reported values of @xmath18 to @xmath19 km for spicules at the limb @xcite , but fit well with the values reported in @xcite . on average , the maximum velocity ( on either ascent or descent ) is 18 km s@xmath1 ranging from 8 to 35 km s@xmath1 with a standard deviation of 6 km s@xmath1 , as shown in the middle left panel of fig . [ fig_hist ] . the maximum velocity is never observed to be lower than 8 km s@xmath1 . df durations are shown in the middle right panel of fig . [ fig_hist ] . the durations vary from 120 to 650 s , with an average of 290 s , and a standard deviation of 85 s. the median duration is lower at 250 s , because of the significant tail of longer duration dfs . in addition to the parabolic fits , we have also measured df widths . to determine the width of a df , we calculate xt-plots ( with x along the direction of the guideline ) for a range of locations that are each offset by one pixel in the direction perpendicular to the guideline . we then find the central axis of the df through visual inspection of the xt-plots : the location with the best - defined xt-plot is chosen as the central axis . starting at this location we find the edges of the df in the direction perpendicular to the df axis . we search for the two positions of maximum intensity gradient furthest from the axis where the df is still visible in the xt-plots . this is done for two different locations along the df , at 50% and at 80% of the maximum extent for a specific time . the width is measured in this fashion at three different times in the df lifetime , at 25% , 50% ( maximum extent ) , and 75% of the df duration . although there is a large spread , we find that the widths of dfs do not vary much with height nor time . the lower left panel of fig . [ fig_hist ] includes all the widths measured and shows a mean of 340 km with a standard deviation of 160 km . while the widest df was measured to be 1100 km , most dfs have widths between the difraction limit of 120 km and 380 km , with a significant tail of wider dfs extending to 700 km . based on the shape of the histogram it seems likely that we are not resolving all dfs . however , a significant fraction is clearly resolved . visual inspection shows that the tops of most dfs are very sharply delineated ( see , e.g. , fig . [ fig_spicxt ] ) . furthermore , inspection of xt-plots shows that dfs often are outlined by a bright ring , as shown in fig . [ fig_bright_ring ] . to determine how sharp the transition at the top of the df is , we calculated the distance along the df axis between the location of maximum intensity gradient and the location of maximum intensity in the bright ring above the df . we refer to this measure as the thickness of the bright ring . it is unclear what causes the bright ring . perhaps it is related to the steep transition from chromospheric to coronal temperatures ? detailed radiative transfer calculations will be necessary to determine what causes the bright ring . regardless of the uncertainty in interpretation , we measured the thickness of the bright ring for all 257 dfs and found an average thickness of 240 km with a standard deviation of 120 km . over half are less than 200 km and many are unresolved , as shown in the lower right panel of fig . [ fig_hist ] . despite a large spread in properties we find clear differences between df statistics in the two regions defined in the right panel of fig . [ fig_potext ] . on average , dfs in region 1 show lower decelerations , greater lengths , slightly higher maximum velocities , and longer durations than in region 2 ( fig . [ fig_hist ] ) . we also found a slight tendency for dfs to be wider in region 2 ( lower left panel of fig . [ fig_hist ] ) , but found no difference in transition region thickness between the two regions . visual inspection shows that region 2 does show more dfs that are wide with significant substructure , such as the df illustrated in fig . [ fig_spiclife_2 ] . to further illustrate the difference between region 1 and 2 , maps of the location of dfs , color coded to show various properties , are shown in fig . [ fig_regionalmaps ] . as can be seen , there are clear regional differences in deceleration , duration , and length . there is a prevalence of smaller decelerations in region 1 with typical values around 100 m s@xmath14 , versus roughly 200 m s@xmath14 in region 2 . even more clearly , the durations in region 1 are on average about 5 minutes , whereas we find 3 minutes in region 2 . similarly , df lengths in region 1 are typically around @xmath20 km , whereas shorter dfs with lengths around @xmath21 km dominate region 2 . the spatial pattern of maximum velocities is not as clear with a large spread of velocities in both regions . in principle some of these differences could be caused by projection effects due to the angle between the df axis and the line of sight vector . a df that propagates along a direction that is parallel to the direction of the line of sight would not be observed as a moving feature in these movies . to measure the absolute extent , velocity and deceleration along the path of the df , the line of sight needs to be perpendicular to the axis of the df . note that the df durations are independent of the line of sight as they do not suffer from these projection effects . to exclude the possibility that the regional differences in deceleration , velocity and length are caused by projection effects , we try to estimate the absolute direction of propagation of each df . assuming that dfs propagate along the magnetic field , their direction can be estimated from the potential field extrapolation shown in the left panel of fig . [ fig_potext ] . to estimate the effect of projection , we calculate the angle between the line of sight vector and the direction of the magnetic field at a reference height of @xmath22 km above the photosphere . this height was chosen to be very roughly the sum of the top height of the quiescent chromosphere and the typical df length . the angle between extrapolated field lines and the line of sight vector was used to correct the properties of each df for projection . histograms for these corrected values are shown in fig . [ fig_hist_los ] . not surprisingly , all corrected values are larger than the observed values . in addition , the corrected values show the same spatial differences between region 1 and 2 as the uncorrected values . to estimate how reliable the potential field extrapolation is , we measured the mismatch between the direction of the df axis and the direction of the magnetic field as projected onto the surface , and found it to be of order 20 30 degrees . a mismatch of this magnitude can lead to significant errors in calculating the projection angle if the df axis is nearly parallel to the projection of the line of sight vector onto the surface . this means that the corrected values should be considered with caution , especially since the calculations are also based on various assumptions whose validity is unclear . the poor resolution of the mdi magnetogram , the possible non - potentiality of the active region , and the uncertainty in the actual df height with respect to the chosen reference height , all could lead to significant errors in the estimated projection angle . for example , the potential field extrapolation clearly fails in the top part of region 1 where nearly horizontal fields are predicted . such horizontal fields are imcompatible with the visual appearance of linecenter images . in addition , since horizontal fields at this location are almost parallel to the line of sight vector , they lead to very large and obviously erroneous corrections of the local df parameters . despite these reservations , it is very difficult to conceive of a magnetic topology with an orientation to the line of sight that would remove the regional differences for all derived parameters . in addition , the regional differences in duration ( which are independent of projection effects ) as well as the correlations found between various parameters in the following subsection , strongly suggest that regional differences are not caused by projection effects . scatterplots of various df properties reveal intriguing correlations . the most striking correlation , also presented in @xcite , is that between the deceleration and the maximum velocity of the dfs , in which a linear relationship is evident ( upper left panel of fig . [ fig_scattergood ] ) . the larger the deceleration the df suffers from , the higher the maximum ( initial or final ) velocity it shows . note that there are , again , significant differences between region 1 and 2 : the dfs in region 2 typically have larger decelerations for a given maximum velocity . a linear fit to the observed relationship between the deceleration and velocity shows that both regions have almost identical slopes . another interesting correlation is that between the maximum length and the duration of a df ( upper right panel of fig . [ fig_scattergood ] ) . the longer dfs tend to have longer lifetimes . as expected from the histograms in the upper and middle right panels of fig . [ fig_hist ] , the dfs in region 1 gravitate towards the upper right of this scatterplot . we also find that the maximum velocity and maximum length of dfs is well correlated ( lower right panel of fig . [ fig_scattergood ] ) : dfs with higher maximum velocity tend to be longer . however , there is a difference in the slope in regions 1 and 2 . a linear fit for region 1 reveals a slope of 87 km per km s@xmath1 , and 60 km per km s@xmath1 for region 2 . dfs in region 1 are longer for the same maximum velocity . the deceleration of dfs shows a somewhat less clear correlation with the df duration ( lower left panel of fig . [ fig_scattergood ] ) . the longest - lived dfs typically suffer from the lowest deceleration . this correlation shows a large spread , and is not quite linear for the total population of dfs . it seems that each region may have its own linear relationship . dfs in region 1 show a more steep dependence of duration on deceleration , whereas those in region 2 show a smaller range of durations , for a range of decelerations that is similar in extent to that of region 1 . based on the correlations described in the above , one would perhaps also expect a correlation between the deceleration and maximum length , or between the maximum velocity and duration . however , such correlations are not clear from the observations , as demonstrated in fig . [ fig_scatterbad ] . the large scatter and different slopes that are evident in the plots that do show correlations may be partially responsible for the lack of clear correlation between deceleration and maximum length or the maximum velocity and duration . however , it is also clear that there are very significant regional differences in fig . [ fig_scatterbad ] , with each region occupying a different area in the scatterplots , and in some cases showing different slopes . it seems probable that such regional differences also are responsible for the lack of clear correlations . when we take the line - of - sight correction into account , none of these correlations , or the lack of correlation , change qualitatively . evidently , the correlation between deceleration and velocity , or that between maximum velocity and length can not change significantly because of projection effects , since both parameters would be corrected by the same factor . the fact that the other correlations , between duration on the one hand , and either deceleration , length or velocity , do not change qualitatively indicates that projection effects do not dominate our dataset . the same arguments also apply to any kind of selection effects that are caused by the fact that dfs with very large line - of - sight velocities could disappear from or be invisible in our relatively narrow ( 128 m ) linecenter bandpass . since dfs are much wider spectral features ( with widths of order 1000 m , see , e.g. * ? ? ? * ) than the width of the bandpass , this means that only features with line - of - sight velocities of order 20 or more km / s would be invisible or only partially visible in the linecenter timeseries . while in a few cases we can see some fading of the df during its descent , it is still possible to trace the top of the df to its end point . given these arguments , the passband effect can only change the derived correlations marginally . note that all of the measurements on which the correlations are based , depend on the identification of a single df rising above the chromospheric background . the latter can be quite noisy and sometimes less well defined . filtergrams taken in the wings of h@xmath0 might be helpful for a more exact determination of the root location of each df , but such filtergrams were unavailable in this dataset . this means that the measured length and maximum velocity could in fact be somewhat higher than reported here . however , given the rapid rise and fall of dfs , the measured duration is quite insensitive to this effect . since we are measuring the acceleration by fitting parabolas to the height profile , the acceleration is also insensitive to the position of the chromospheric background . overall , this implies that the relative uncertainty in determining the roots of dfs are unlikely to qualitatively change the correlations reported here . the interpretation of these correlations , or the lack thereof , will be discussed at length in [ section : discussion ] . the regional differences in df durations evident in the middle right panel of fig . [ fig_hist ] , with 3 minute lifetimes dominating region 2 and lifetimes around 5 minutes more dominant in region 1 , suggest a role for chromospheric waves or oscillations in the formation mechanism of dfs . oscillations in the chromosphere are typically not steady harmonic waves , but wave trains of finite duration . for this reason we perform a wavelet analysis of the data , since it allows for localization in time of periodic signals . the mother wavelet we use is the complex valued morlet wavelet ( @xmath23 ) which consists of a plane wave modulated by a gaussian , an appropriate shape for wave trains . to estimate the statistical significance of the wavelet power spectra ( 95 % confidence interval ) , we compare them to theoretical spectra for white noise , following @xcite and @xcite . taking into account the cone of influence @xcite , we calculate for each superpixel the number of wavepackets in the timeseries for which the wavelet power is significant . this is done for a range of wave periods from 150 to 600 seconds . a superpixel is defined as the average of 4x4 original 0.063 pixels . note that the original data is oversampled by a factor of almost 3 with respect to the diffraction limit ( 120 km ) of the sst at the wavelength of . the upper left and upper right panels of fig . [ fig_osc_panels ] show , for each location of the lower left panel , the number of significant wavepackets ( during the timeseries ) for waves with a period of 180 s and 300 s respectively . the lower right panel shows which wave period dominates , i.e. , has the most wavepackets , at each location of the field of view . from this figure , we find that many locations show significant oscillatory power throughout the timeseries . the dominant period of this power varies across the field of view . clearly visible are the dominant 3 minute oscillations in the sunspots around x=27 and y=37 and at x=45 and y=15 . in addition , the dense plage region which we have designated as region 2 , shows a preponderance of 3 minute oscillations . in contrast , region 1 , adjacent to the dense plage region , is dominated by oscillations with long periods around and above 5 minutes . in this region , the magnetic field is more inclined than in the dense plage region . the longer , less dynamic and more horizontal fibrils , such as those in the superpenumbra of both spots , seem to be dominated by oscillations with periods longer than 10 minutes . it is interesting to note that the dominant period in region 2 corresponds exactly to the df duration we observe in region 2 . apparently , dfs typically live for about 3 minutes in this region , which is also the dominant periodicity of the oscillations . likewise , region 1 is dominated by 5 minute oscillations , and contains dfs with lifetimes of order 5 minutes . since the oscillation measurements are based on the same data , it seems clear that some dfs are periodic or recurring . this was also described by @xcite and suggests a relationship between chromospheric waves and dfs , which will be further explored in [ section : discussion ] . most dfs follow parabolic paths with a symmetrical ascending and descending phase . these paths are characterized by a large initial velocity , usually of order 15 - 20 km s@xmath1 ( middle left panel of fig . [ fig_hist ] ) , that decreases linearly with time until the df has retreated completely . at the end of its life , a df s downward velocity is roughly identical to the initial upward velocity . the linear decrease with time of the velocity , i.e. , the deceleration , is usually between 120 and 280 m s@xmath14 ( upper left panel of fig . [ fig_hist ] ) . there is a clear linear relationship between the deceleration and the maximum ( down- or upward ) velocity of the df ( upper left panel of fig . [ fig_scattergood ] ) . dfs reach maximum lengths of order 1,000 to 2,000 km ( upper right panel of fig . [ fig_hist ] ) during their lifetimes of 3 to 8 minutes ( middle right panel of fig . [ fig_hist ] ) . the duration and maximum length are well correlated ( upper right panel of fig . [ fig_scattergood ] ) , which indicates that uncertainty about the line - of - sight correction may not play an important role in the regional differences we find . we resolve more than half of the dfs we observe . the average thickness is 240 km ( lower left panel of fig . [ fig_hist ] ) , with a range between the diffraction limit ( 120 km ) and 1000 km . some dfs occur over a wider spatial area , whereas others are very thin and stay thin throughout their lifetime . many of the wider dfs have variable widths as a function of time . these give the visual appearance of a wave front with varying phase speeds or amplitude in the direction perpendicular to the df axis . there is often a brightening above the df , which can be clearly seen in xt-plots ( fig . [ fig_bright_ring ] ) . this brightening might be a sign of increased temperatures at the top of the df . the thickness of this transition is not resolved and is less than 120 km in more than half of the dfs studied ( lower right panel of fig . [ fig_hist ] ) . dfs have significantly different properties in two regions within our field of view . region 2 contains dense plage , i.e. , a collection of strong magnetic field concentrations in which the magnetic field is generally more vertical . region 1 is located at the edge of this dense plage region , where the magnetic field is more inclined , as it connects to relatively nearby opposite polarity plage . the dfs in the dense plage region are shorter ( @xmath24 km ) , have higher decelerations ( @xmath25 m s@xmath14 ) , slightly lower velocities ( @xmath26 km s@xmath1 ) and shorter durations of about 3 minutes ( figs . [ fig_hist ] and [ fig_regionalmaps ] ) . in contrast , the dfs in the region where the field is more inclined are longer ( @xmath27 km ) , have lower decelerations ( @xmath28 m s@xmath14 ) , slightly higher velocities ( @xmath29 km s@xmath1 ) and longer durations of about 5 minutes ( figs . [ fig_hist ] and [ fig_regionalmaps ] ) . in addition , df lifetimes correlate well with the dominant periodicities observed in both regions : 3 minute oscillations dominate in the dense plage region , whereas 5 minute oscillations dominate in the region with more inclined field ( fig . [ fig_osc_panels ] ) . our measurements of dfs and oscillatory power strongly suggest that chromospheric waves drive dfs upward . these chromospheric waves are generated by convective flows and ( global ) oscillations in the photosphere and convection zone . the magnetic field concentrations provide a channel for these disturbances to propagate up through the chromosphere . during their path upwards , they shock and drive chromospheric plasma upwards , forming dfs in the process . this mechanism was originally suggested in rudimentary form by @xcite . some models have further studied the effects of the convective part of the photospheric driver on the chromosphere in the so - called rebound - shock model @xcite . more recently , @xcite included both the convective driver and global oscillations in varying magnetic topologies to directly describe their ( lower resolution ) observations of dynamic fibrils . the latter model s predictions agree well with the current findings . this scenario is confirmed by analysis of advanced numerical simulations ( see @xcite and next subsection ) in which dynamic fibril - like features with properties identical to those in our data occur . from the simulations , it is clear that the observed parabolic paths indicate that dfs are shock driven . when a shock hits the transition region ( i.e , the top of a df ) , the transition region is catapulted upwards at a velocity that exceeds the sound speed in the chromosphere ( @xmath30 km s@xmath1 ) . shock waves generally have velocity profiles in the form of n- or sawtooth shapes ( e.g. * ? ? ? thus , a plasma parcel passing through a shock wave will first experience a sudden kick in velocity - the shock itself and thereafter a gradual linear deceleration as the shock recedes . assuming a shock wave with amplitude or shock strength of @xmath31 and linear deceleration @xmath32 @xmath33 it is trivially verified that a plasma parcel passing through the shock will describe a parabolic path ; integrating gives a parcel position that moves as @xmath34 df formation from chromospheric shocks is compatible with the fact that we do not find any dfs with initial ( maximum ) velocities of less than 10 km s@xmath1 . waves with smaller amplitudes will not form shocks and hence dfs . the observed lower limit to the maximum velocities provides an estimate of the sound speed in the upper chromosphere ( 10 km s@xmath1 ) . its value is compatible with chromospheric temperatures from solar models . more generally , the observed maximum velocities provide a robust estimate of the mach number of chromospheric shocks . our observations indicate that shocks with mach numbers of 1 through at least 3.5 are ubiquitous in the magnetized chromosphere , and drive dfs . analysis of numerical simulations @xcite further shows that several different effects influence the shock formation and evolution so that the observed correlations ( e.g. , between deceleration and velocity ) and regional differences follow naturally . the oscillatory power spectrum of the photospheric driver is dominated by power at 5 minutes . the chromosphere acts as a filter to the input from below , filtering out waves with periods that are longer than the local acoustic cutoff period . in general , the acoustic cutoff period depends on the inclination of the magnetic field lines to the vertical @xcite . under conditions of vertical magnetic field ( e.g. , region 2 ) , this means that the chromosphere is dominated by oscillations and waves with periods at the acoustic cutoff of 3 minutes . as a result , the dfs formed in the dense plage of region 2 are driven by shocks with periods of 3 minutes and thus have lifetimes of order 3 minutes . in contrast , the inclined field , such as that found in region 1 , increases the acoustic cutoff period . this allows the photospheric waves with periods of 5 minutes to leak into the chromosphere , develop into shocks and form dfs with similar lifetimes . since most of the oscillatory power in the photosphere occurs for periods of 5 minutes , the driver in inclined field regions is generally stronger than in regions with more vertical field . this effect is enhanced by the fact that , in dense plage regions , the amplitude of convective flows and global oscillations is generally reduced at the photospheric level . dfs that form along inclined field lines also experience only the component of gravity along the magnetic field , so that decelerations are generally lower in the inclined field region . the observed correlation between deceleration and maximum velocity can be understood in terms of shock wave physics and the fact that dfs are driven by single shocks . for a given wave period , n-shaped shock waves will show a steeper decline with time in velocity for a greater shock strength . in other words , dfs with higher maximum velocity , i.e. , those that are driven by stronger shocks , will show greater deceleration , for a given df lifetime . it is the combination of these effects that explains the observed regional differences and correlations . region 2 contains more vertical field , so that only 3 minute power drives the chromospheric waves and shocks . given the photospheric power spectrum and the reduced power in dense plage regions , this leads to lower amplitudes , i.e. , lower maximum velocity . in addition , the short lifetimes and more vertical field lead to higher decelerations , and shorter lengths . conversely , region 1 contains more inclined field , allowing the full photospheric peak power at 5 minutes to leak into the chromosphere to drive dfs . as a result , the velocity amplitude is greater , and the longer lifetimes and less vertical field lead to lower decelerations and longer lengths . these findings are confirmed when comparing the observations with recent advanced numerical radiative mhd simulations that include the convection zone , photosphere , chromosphere , transition region and corona @xcite . the equations of mhd are solved using an extended version of the numerical code described in @xcite and in more detail by nordlund & galsgaard at http://www.astro.ku.dk/@xmath35kg . the code functions by using a sixth order accurate scheme to determine partial spatial derivatives . the equations are stepped forward in time using the explicit 3rd order predictor - corrector procedure by @xcite , modified for variable time steps . in order to supress numerical noise , high - order artificial diffusion is added both in the forms of a viscosity and in the form of a magnetic diffusivity . thermal conduction along the magnetic field in the corona and transition region is treated by operator splitting and using a multi - grid solver . the radiative flux divergence from the photosphere and lower chromosphere is obtained using the method of @xcite based on group mean opacities as modified by @xcite to account for scattering . for the upper chromosphere , transition region and corona we assume effectively thin radiative losses . for the middle and upper chromosphere we parametrize the optically thick radiative losses in strong lines and continua from hydrogen and singly ionized calcium by using escape probabilities calculated in 1d with the code of @xcite . further description of the 3d code may be found in @xcite . the 2d simulations described here are run on a grid of @xmath36 points spanning @xmath37 mm@xmath38 . in height , these models cover a region from 2 mm below to 10 mm above the photosphere . the grid has uneven spacing in the vertical , @xmath39 , direction with a grid size of 30 km in the photosphere and chromosphere becoming gradually larger in the corona . we have also run 2d calculations on coarser grids of @xmath40 and @xmath41 points , as well as 3d runs of @xmath42 points spanning @xmath43 mm@xmath44 , with results similar to those reported here . at the start of a given simulation , a potential magnetic field is added to a model of the solar atmosphere . the resultant model is then allowed to evolve in time until a quasi - steady state is achieved . this takes on the order of a half hour solar time . the models are convectively unstable due the radiative losses in the photosphere . this results in a temperature structure similar to that derived in semi - empirical solar models . the average temperature at the bottom boundary is maintained by setting the entropy of the fluid entering the computational domain . the bottom boundary , based on characteristic extrapolation , is otherwise open , allowing fluid to enter and leave the computational domain as required . the magnetic field at the lower boundary is advected with the fluid . to prevent coronal cooling the upper temperature boundary is set to evolve towards a temperature of @xmath45 k on a timescale of 10 100 s. ( in the 3d models coronal temperatures are maintained by self consistent magnetic dissipation . ) a snapshot of the temperature structure in the lower part of the model including the convection zone , photosphere , and chromosphere is shown in fig . [ sim_snapshot ] . waves generated by convection propagate through the chromosphere , steepen and form shocks roughly 1 mm above the photosphere . in regions above the level where plasma @xmath46 the propagation is modified by the magnetic field and the waves are subjected to reflection , refraction and mode coupling as described by @xcite . of interest to us here are the shocks that propagate along field lines that enter the coronal plasma as indicated in fig [ sim_snapshot ] . these shocks lift and accelerate the transition region as they propagate into the corona . plotting the motion of the transition region along one such field line as a function of time , e.g. a xt plot , reveals that the position of the transition region describes a parabola ( fig . [ sim_xt ] , compare with fig . [ fig_spicxt ] ) . in the left panel of fig . [ sim_xt ] the logarithm of the temperature is shown , the right panel shows the velocity . it is worth noting that except for a brief moment in the early ascending phase the simulated df shows both upflowing and downflowing velocities along its length at a given time . during descent , the df has only downflowing plasma . note also that near its base the simulated df shows downflows for a major portion of its lifetime . we have measured the transition region motion along several field lines in several ` df - like ' events and fitted the motion with parabolas . the resulting correlation between measured maximum velocities and decelerations ( fig . [ sim_dec - maxv ] ) show the same trend as found in the observed dfs ( upper left panel of fig . [ fig_scattergood ] ) : large decelerations are correlated with large maximum velocities . the range in maximal velocities and decelerations is approximately the same as that found in the observed dfs , and the slope in the correlation is about the same . inspection of fig . [ sim_xt ] also indicates that the simulated dfs have the same range of heights and durations that the observed dfs display . higher resolution 3d simulations with varying magnetic topologies will be necessary to explain the range of df widths , the internal df substructure , and what determines the sometimes sheet - like structure of some dfs . such simulations will also help to fully explain the range of periods and lifetimes that we observe . how do these findings relate to quiet sun phenomena such as mottles and spicules at the limb ? there are many similaties between mottles and dfs . both phenomena appear as highly dynamic , dark features in the wings and core of . they are associated with magnetic flux concentrations that can be observed as bright points in the blue wing of . more importantly , like dfs , mottles also follow parabolic paths . in an excellent analysis of the dynamics of inclined mottles ( which they refer to as disk spicules ) @xcite find that : * mottles seem to be generated several hundred kilometers above the photosphere , * mottles undergo real mass motion of order 10 - 30 km s@xmath1 , extending up to 2 - 10 mm during their lifetime of 2 - 15 minutes , * most mottles are observed to ascend and descend , * mottles in weak plage , or the interior of network cells are more vertical and shorter than the more inclined quiet sun mottles , * the appearance of mottles is varied , they can be thick , thin , tapered or not , curved or straight , * some mottles show downward motion close to their base during the ascending phase , * the decelerations observed in mottles are too small to be consistent with a purely ballistic flight ( at solar gravity ) and the observed doppler shifts , * the largest doppler signal appears at the beginning of the ascending phase , and at the end of receding phase , * the observed velocity profiles are compatible with impulsive acceleration and constant deceleration afterward . * the maximum upward velocity is usually identical in amplitude to the maximum downward velocity , * there is a linear relationship between the duration and the maximum length of mottles , * there is a correlation between the average velocity and maximum length , * nearby mottles are often found to be driven by a common disturbance . these findings agree strikingly well with our observations and modelling of dfs , with only a few discrepancies . the lifetimes and maximum lengths of the mottles @xcite analyzed are typically longer than those of the dfs we describe . we believe this is in part a selection effect , as we chose to focus only on highly dynamic and shorter features in our dataset . we note that our dataset also contains a large number of highly inclined fibrils that are longer lived and less dynamic than the dfs ( see , e.g. , fig . [ fig_osc_panels ] ) . in addition , preliminary analysis of high - quality sst data of quiet sun mottles indicates that the most inclined mottles are on average longer , but that many of the mottles actually show lengths ( 1,000 - 5,000 km ) that are similar to those of dfs . until further detailed analysis , it can however not be excluded that dfs associated with plage are on average somewhat shorter in length and lifetime than quiet sun mottles . preliminary analysis of our numerical simulations suggests that these differences could be related to large scale differences in magnetic topology . further work is necessary to establish the exact relationship between mottles and dfs . however , it should be noted that in addition to these striking similarities with mottles , recent observations of limb spicules @xcite show clear parabolic paths with decelerations and maximum velocities of the same order as those reported for mottles @xcite and those we find for dfs . older observations of limb spicules have also shown evidence for correlations between velocities and maximum lengths @xcite , and between duration and height @xcite . these correlations are similar to the ones we find for dfs . in addition , @xcite has reported on collective behavior of spicules or mottles over several thousand kilometers . as discussed also by @xcite , these similarities strongly suggest that chromospheric shock waves cause significant excursions of the upper chromosphere in both active region and quiet sun . such a scenario was proposed by @xcite using sst and trace data . further detailed analysis of our numerical simulations can help shed light on several unresolved issues , such as the role of reconnection in forming quiet sun jets , or whether the observed height of limb spicules can be explained by the mechanism proposed here . @xcite suggest that it is possible that limb spicules consist of two populations : jets that are driven by shocks ( as described in this paper ) , and jets caused by reconnection . the latter jets could form a subset that on average is taller than the shock driven jets , and perhaps be a part of a continuous spectrum of reconnection jets that includes surges , macrospicules and upflow events @xcite . these taller jets may be the features that have uv counterparts ( e.g , observed by sumer ) , which @xcite calls `` spicules '' . the combination of high - resolution chromospheric data and advanced numerical simulations strongly suggests that dfs are formed by upwardly propagating chromospheric oscillations / waves . these waves are generated in the convection zone / photosphere as a result of global p - mode oscillations and convective flows . the magnetic topology acts as a filter , so that only waves of certain periods can propagate into and through the chromosphere , forming shocks along the way , up to regions where h@xmath0 is formed . * dfs follow a parabolic path * dfs in dense plage regions are shorter , slower , undergo larger deceleration and live shorter * dfs in regions adjacent to dense plage regions are longer , faster , undergo less deceleration , and live longer * dfs in dense plage regions show periodicities around 180 s * dfs in regions adjacent to dense plage regions show periodicities around 300 s * the velocity and deceleration of all dfs is well correlated : the higher the velocity , the more deceleration * this correlation is reproduced well in numerical simulations where dfs are driven by shocks caused by upward propagating waves in the chromosphere . * dfs in the simulations follow parabolic paths as well . * the simulations reproduce the observed values of maximum velocities , decelerations and lengths well . from the results describe above and by @xcite , it seems clear that in active regions , most dynamic fibrils are formed by chromospheric shocks driven by convective flows and oscillations in the photosphere . , g. b. , bjelksj , k. , korhonen , t. k. , lindberg , b. , & petterson , b. 2003 , in innovative telescopes and instrumentation for solar astrophysics . keil & s.v . , vol . 4853 , 341350

shock waves have been observed in a wide range of environments outside the earth , from the solar wind to the hot gas in galaxy clusters . however , the mechanism whereby the gas in these environments is shocked has been poorly understood . while shock transitions in the earth s atmosphere are mediated by molecular viscosity ( and hence direct particle collisions ) , those in interstellar space and the solar wind are too dilute to form in this way . in non - relativistic shocks , the role of collisions is effectively played by collective interactions of the plasma with the magnetic field . this results in a multi - scale shock transition having sub - structure at ion kinetic length scales ( larmor radius or inertial length ) and potentially electron kinetic scales ( inertial lengths or whistler mode ) ( e.g. , schwartz et al . ( this volume ) , treumann 2009 ) . such plasmas are termed collisionless . the magnetic fields threading through the charged particle plasmas in space endow the plasmas with elastic properties , much like a fluid . the kinetic energy of the inflowing gas is dissipated within this fluid via collective interactions between the particles and magnetic field , transferring energy from the magnetic field to the particles . the collective processes are the result of the dc electromagnetic fields present in the shock transition layer , kinematic phase mixing , and also plasma instabilities ; the last give rise to a rich range of plasma waves and turbulent interactions . it has long been known that these processes may heat the electrons beyond the mass - proportional value predicted by the rankine - hugoniot jump conditions - ample evidence is found in spacecraft studies of solar wind shocks ( schwartz et al . 1988 ) and multi wavelength spectroscopy of supernova remnants ( ghavamian et al . 2001 , 2002 , 2003 , 2007 ; laming et al . 1996 ; rakowski et al . 2008 ) and galaxy cluster gas ( markevitch et al . 2005 ; markevitch & vikhlinin 2007 ; russell et al . 2012 ) understanding how this process depends on such shock parameters as shock speed , preshock magnetic field orientation and plasma beta has been slow . in collisionless plasmas , the downstream state of the plasma can not be uniquely determined from the upstream parameters because the rankine - hugoniot jump conditions only predict the _ total _ pressure downstream , not the individual contributions from the electrons and ions : @xmath2 . at the limit of a strong shock , @xmath3 and @xmath4 are each 4 times their preshock values , so the relative values of @xmath0and @xmath5immediately behind the shock are wholly dependent upon the nature of the collisionless heating processes occurring at the shock transition . although an mhd description can be used to describe the behavior of the gas far upstream and far downstream of the shock , a more detailed kinetic approach is required for understanding how the dissipation at the shock front transfers energy from plasma waves and turbulence to the electrons and ions . non - relativistic collisionless shocks can be broadly sorted into three categories : slow , intermediate and fast . the three types are defined according to the angle between the shock velocity and upstream magnetic field , as well as the relative value of the shock speed compared to the upstream sound speed ( @xmath6 ) and alfvn speed ( @xmath7 ) . most astrophysical shocks are quasi - perpendicular ( i.e. , they propagate at a nearly right angle to the preshock magnetic field ) , allowing only for fast - mode propagation . in that case , the relevant quantity is the magnetosonic mach number , @xmath8 ( @xmath9 ) . collisionless shocks have also been classified according to whether the flow speed exceeds the sound speed in the downstream plasma ( kennel 1985 ) . above the critical mach number where the flow is subsonic , the dissipation of flow energy into thermal energy can no longer be maintained by electrical resistivity , and plasma wave turbulence ( the cause of which are instabilities generated when the electron and ion distribution functions are distorted at the shock transition ) is required ( kennel 1985 ) . shocks above the critical mach number are termed supercritical , while those below are termed subcritical . note that even for subcritical shocks , observations suggests that kinetic processes other than resistivity and turbulence contribute to the shock dissipation ( greenstadt and mellott 1987).there are also other , higher critical mach numbers related to the formation of subshocks ( kennel et al . 1985 ) and non - steady cyclic shock reformation beyond the whistler critical mach number ( krasnoselskikh et al . 2002 ) . the critical mach number for quasi - perpendicular shocks is estimated to be rather low @xmath102.8 ( edmiston and kennel 1984 ) . the mach numbers of most snr shocks are expected to be well in excess of this value , meaning that they are both fast - mode and supercritical . as one approaches the high mach numbers expected for astrophysical shocks ( @xmath1120 - 100 for an ambient magnetic field of 3 @xmath12 ) , a greater and greater fraction of the incoming ions become reflected back upstream from the shock front , corresponding to an increasingly turbulent and and disordered shock transition . physical parameters such as @xmath13 , @xmath14 and @xmath15 no longer jump in an ordered manner ( i.e. , the transitions are no longer laminar ) . in addition to the reflected particles , the hot ions from downstream become hot enough to escape upstream , further enhancing the population of ions in front of the shock . these ions , which form a precursor , are now believed to play an essential role in the dissipation of high mach number collisionless shocks . aside from providing the seed population for the acceleration of cosmic rays , the precursor ions are likely generate a variety of plasma waves capable of selectively heating the electrons over the ions , thereby providing an important mechanism for raising @xmath16 above the mass - proportional value of @xmath17 . taken at face value , the rankine - hugoniot jump conditions predict that electrons and ions will be heated in proportion to their masses : @xmath18 in collisionless shocks , there are three relevant timescales to consider : the time required for coulomb collisions to isotropize a distribution of electrons , @xmath19 , the time required for coulomb collisions to isotropize distribution of ions , @xmath20 , and finally the time required for the electrons and ions to equilibrate to a common temperature , @xmath21 ( spitzer 1962 ) . after a time scale @xmath19 or @xmath20 , the particles in question attain a maxwellian velocity distribution . the self - collision time , @xmath22 for electrons of density @xmath3 and temperature @xmath0 is ( spitzer 1962 ) : @xmath23 where ln@xmath24 is the coulomb logarithm , equation [ massprop ] has been used to write the relaxation time in terms of the shock speed and @xmath25 is the shock speed in units of 1000 kms@xmath26 . for a young snr having @xmath271000 kms@xmath26 , postshock density n@xmath281 @xmath29 and ln @xmath3030 the time required to establish a maxwellian distribution at the electron temperature given by equation 1 is @xmath31 0.004 yrs . the electrons are isotropized by self - collisions first , then the protons , and finally over a longer timescale the electrons and protons equilibrate to a common temperature . for this reason , coulomb collisions alone are unable to establish equilibration electron and proton distributions at the shock front . this equilibration is described by the relation @xmath32 where @xmath33 the temperatures @xmath0 and @xmath5 equilibrate to a common density - weighted average temperature @xmath34 , given by @xmath35 , where @xmath36 is the mean mass per particle ( @xmath37 for cosmic abundances ) . for mass - proportional heating , the timescale given by equation [ tep ] is @xmath382000 yrs for @xmath39 @xmath401000 kms@xmath26 , of similar order but longer than the proton - proton isotropization timescale , @xmath41 . these are longer than the age of the snr , substantially so at higher @xmath39 indicating that for minimal heating ( i.e. , @xmath16@xmath421/1836 ) the electrons and ions will not equilibrate to @xmath34 during the lifetime of the snr ( itoh 1978 ; draine & mckee 1993 ) . the arguments above indicate that coulomb collisions are ineffective at both isotropizing the heavy ion particle distributions and equilibrating the electron and ion temperatures at the transtions of collisionless shocks . the emission spectra of non - radiative snrs ( dominated mostly by x - ray and ultraviolet emission ) should therefore be sensitive probes of the collisionless heating processes at the shock transition . we consider the observational constraints of these processes below . the most useful shocks for studying collisionless equilibration processes are those exhibiting detectable emission from the immediate postshock gas . to be diagnostically useful , the emission should arise from the region close to the shock front , where temperature disequilibrium between electrons , protons and heavy ions is substantial enough to affect both the relatives fluxes and relative velocity widths of emission lines . the shape and extent of the spatial profile for the different emission species behind the shock is also a useful diagnostic , especially for the uv resonance lines of he ii @xmath431640 , c iv @xmath441548 , 1550 , n v @xmath441238,1243 , and o vi @xmath431032 , 1038 . for a given shock speed , the distance behind the shock where the emission peaks depends strongly on the initial electron temperature , and electron temperature immediately behind the shock . the ubiquity of non - radiative snrs , as well as their relatively simple geometry and very high shock speeds , makes these objects the most important laboratories for investigating the efficiency and nature of electron - ion and ion - ion equilibration . other non - radiative shocks available for study are those occurring in stellar wind bubbles ( for example , wolf - rayet bubbles ) ( gosset et al . 2011 ) and in galaxy cluster shocks ( e.g. , the bullet cluster ( markevitch et al . 2005 ) and abell 2146 ( russell et al . however , these shocks are far less frequently observed than those in snrs , leaving the latter as the most important objects providing both a broad range of shock speeds and corresponding diagnostic information ( such as proper motions and spatially resolved structure ) . shocks in star forming regions , such as hh objects and their associated bow shocks , tend to be clumpy and complicated , and are usually located in dense environments ( @xmath45100 - 1000 @xmath29 ) with low enough shock speeds ( @xmath39@xmath46100 - 200 kms@xmath26 ) to be radiative . in those cases , the optical and uv emission ( where the most valuable shock diagnostic line emission arises ) is dominated by emission from the cooling and recombination zones far downstream from the shock . in those regions the cooling of the gas below 100,000k and the accompanying compression of the gas result in a collisional plasma with @xmath47 . this erases any memory of the initial electron - ion equilibration . furthermore , a significant fraction of the ly @xmath48 continuum produced in the recomination zone is expected to pass upstream and ionize the preshock gas . in strong shocks ( @xmath39@xmath4980 kms@xmath26 ) , this results in complete ionization of hydrogen ( shull & mckee 1979 ; cox & raymond 1985 ; dopita et al . 1993 ) , thus precluding the use of collisionally excited balmer line emission from neutral h ( described in the next section ) as a temperature equilibration diagnostic . in the late 1970s it was discovered that very fast ( @xmath102000 kms@xmath26 ) shocks in young snrs could generate detectable optical emission very close to the shock transition ( chevalier & raymond 1978 ; chevalier , kirshner & raymond 1980 ) , providing a valuable diagnostic of physical conditions at the shock before coulomb collisions or cooling could alter them . this emission is produced by collisional excitation of h i as it flows into the shock front . the cold neutral component does not interact directly with the plasma waves and turbulence at the shock , while the ionized component is strongly heated and compressed by a factor of four ( when the shock is strong ) . some of the cold h is destroyed by collisional ionization ; however , the rest of the cold h undergoes charge exchange with hot ions behind the shock , generating a separate population of hot h. approximately 1 in every 5 collisions results in collisional excitation to the n=3 level of h , producing h@xmath48 and ly @xmath52 emission . the h@xmath48 line from the cold neutrals is narrow and reflects the preshock temperature ( @xmath5330,000 k ) , while that from the hot neutrals is broad ( typically @xmath54500 kms@xmath26 ) , and reflects the postchock temperature ( and to a large extent , the velocity distribution ) of the protons ( chevalier & raymond 1978 ; chevalier , kirshner & raymond 1980 ; smith et al . 1991 ; ghavamian et al . 2001 ) ( figure [ fig : balmerprofile ] ) . in balmer - dominated shocks , the broad to narrow h@xmath48 flux ratio is proportional to the ratio of the charge exchange rate to the ionization rate , with the latter being highly sensitive to the electron and proton temperatures . this makes the broad to narrow ratio , @xmath55 , very sensitive to @xmath16 . there is only weak dependence of @xmath55 on the preshock h i fraction and preshock temperature , mainly due to differences in the amount of ly @xmath52 converted into h@xmath48 in the narrow component ( ghavamian et al . 2001 ; van adelsberg et al . 2008 ) . -0.2 in the first systematic attempt to use balmer - dominated snrs to infer @xmath16 for collisionless shocks was attempted by ghavamian et al . ( 2001 ) . using the broad h@xmath48 line widths measured from a sample of balmer - dominated shocks , they constrained the range of plausible shock speeds between the limits of minimal equilibration and full equilibration . they then predicted the broad - to - narrow ratios for a grid of shock models over this range of @xmath39 and @xmath16 , allowing @xmath16 to be constrained . the range of broad component h@xmath48 widths observed in snrs ranges from @xmath10250 kms@xmath26 for the slowest balmer - dominated shocks ( cgynus loop ) , to @xmath10 500 kms@xmath26 for intermediate - velocity shocks ( rcw 86 ) and finally @xmath102600 kms@xmath26 for the fastest shocks ( snr 0509@xmath5667.5 ) . this corresponds to a well - sampled range of shock speeds : nearly a factor of 10 . the primary uncertainty in measurement of the broad component width at low shock speeds ( @xmath57200 kms@xmath26 , as seen in the northeastern cygnus loop ; hester et al . 1994 ) is disentangling the broad and narrow components when they are of comparable width . at high shock speeds ( @xmath582000 kms@xmath26 ) the main difficulty is the baseline uncertainty of the surrounding continuum : if the peak of the broad line is low and the width very large , errors in ascertaining where the broad line merges into the background can lead to underestimates of the broad component width . the range of @xmath55 for this sample of balmer - dominated shocks lies between 0.4 and 1.2 ( kirshner , winkler & chevalier 1987 ; smith et al . 1991 ; ghavamian et al . 2001 ; 2003 , rakowski , ghavamian & laming 2009 ) . however , it does not not vary with broad component width in a monotonic fashion . recently there has been substantial improvement in the modeling of balmer - dominated shocks . earlier calculations of the broad h@xmath48 line profile assumed that it formed from a single charge exchange ( chevalier , kirshner & raymond 1980 ; smith et al . 1991 ; ghavamian et al . 2001 ) , and it treated the hot and cold neutrals as two separate , distinct populations , with a given neutral belonging to either one or the other . however , in reality an interaction tree is required to track the number of photons emitted by each neutral over multiple excitations and charge exchanges . these effects were first incorporated in the balmer - dominated shock models of heng & mccray ( 2007 ) , who also found that charge exchange results in a third population of neutrals having velocity widths intermediate between the hot and cold neutrals . further improvements in modeling of balmer - dominated shocks were included by van adelsberg et al . ( 2008 ) , who included the momentum transferred by charge exchange between the hot neutrals and protons . this allowed the bulk velocity of the postshock neutrals to be calculated separately from those of the protons . inclusion of the momentum transfer showed that for shock speeds @xmath571000 kms@xmath26 , charge exchange effectively couples the fast neutral and thermal proton distributions , while for high shock speeds ( @xmath585000 kms@xmath26 ) , it does so far less effectively . this results in a fast neutral distribution that is skewed relative to the protons in velocity space and an average velocity that is much higher for the fast neutrals than the protons ( van adelsberg et al . together , the inclusion of all these effects has enhanced the ability of the models to match the observed broad - to - narrow ratios ( and hence predict @xmath16 ) . in particular , the newer models can now match the low broad - to - narrow ratio ( @xmath55@xmath590.67 ) observed in knot g of tycho s snr , yielding @xmath16@xmath590.05 , @xmath39@xmath591600 kms@xmath26 . however , even after all the additional physics is included , the balmer - dominated shock models are still unable to reach the low broad - to - narrow ratios measured along the rims of dem l71 ( @xmath55@xmath590.2 - 0.7 ; ghavamian et al . 2003 ; rakowski , ghavamian & laming 2009 ) . the electron - ion equilibration in dem l 71 was instead determined via comparison of broad h@xmath48 line fwhm with postshock electron temperatures measured from _ chandra _ observations ( rakowski , ghavamian & hughes 2003 ) . the most promising explanation advanced for the @xmath55 discrepancy has been added narrow component flux from the shock precursor ( raymond et al . 2011 ; morlino et al . 2012 ) , hitherto not included in the earlier shock models ( ghavamian et al . 2003 , rakowski , ghavamian & laming 2009 ) . these developments are described in more detail in the next section . the plot of @xmath16 versus @xmath39for the available sample of balmer - dominated shocks shows a declining trend of equilibration with shock speed ( ghavamian et al . 2007 ; heng et al . 2007 ; van adelsberg et al . 2008 ) . the trend is described by ghavamian et al . ( 2007 ) as full equilibration for shock speeds up to and including 400 kms@xmath26 , and a declining equilibration proportional to the inverse square of the shock speed above 400 kms@xmath26 . this description can be characterized in the following way : @xmath60 where the functional form of the the @xmath16 relation is designed to asymptotically transition to @xmath16=@xmath17 at very high shock velocities . the most up to date plot of @xmath16 versus shock speed , reproduced from van adelsberg et al . ( 2008 ) , is shown in figure [ fig : plotequil ] ( note that the shock models used in producing these plots do not include contribution from the shock precursor ) . although the inverse correlation between @xmath16 and @xmath39was largely confirmed by van adelsberg et al ( 2008 ) , there may be some evidence of departure from the @xmath16@xmath50 @xmath61 relation at shock speeds exceeding 2000 kms@xmath26 . van adelsberg et al . find that when all three measured broad component widths and broad - to - narrow ratios from sn 1006 are included in the plot ( @xmath39@xmath102200 - 2500 kms@xmath26 ) , a slight upturn in the @xmath16-@xmath39 relation appears . the @xmath16 ratios for those cases are found to be @xmath100.03 , superficially similar to @xmath62 , rather than @xmath63 . the reason for the discrepancy is not clear , nor whether this indicates a breakdown in the @xmath61 dependence at shock speeds exceeding 2000 kms@xmath26 . some caveats to consider when interpreting the upturn in @xmath16 seen in figure [ fig : plotequil ] are as follows : for shock speeds @xmath582000 kms@xmath26 collisional ionization and excitation of h are primarily caused by proton ( and to a lesser degree , alpha particle ) impact ( laming et al . 1996 ; ghavamian 1999 ; ghavamian et al . 2001 ; tseliakhovich et al . experimentally measured cross sections for these interactions have still not been available to high precision ( uncertainties @xmath1020%-30% still exist ) , although more sophisticated theoretical calculations are now becoming available ( see , for example , tseliakhovich et al . in addition , as the broad component width increases , the h@xmath48 profiles are spread out over an increasing number of pixels , resulting in noisier spectra and greater measurement uncertainty in the broad component width . these larger error bars in turn result in a larger uncertainty in @xmath39 , especially at shock speeds of 2000 kms@xmath26 and higher . the best way to further constrain the equilibration - shock speed relation is to add new data points to the curve shown in figure [ fig : plotequil ] . to this end , we present balmer - dominated h@xmath48 profiles for two additional positions in tycho s snr ( different from those of knot g presented by kirshner , winkler & chevalier ( 1987 ) , smith et al . ( 1991 ) and ghavamian et al . ( 2001 ) ) . these profiles , shown in figure [ fig : tychocennwspec ] , were acquired with a moderate resolution spectrograph in 1998 ( for details on the observational setup , see ghavamian 1999 and ghavamian et al . the profile marked ne was obtained from a clump of h@xmath48 emission located along the northeastern edge of tycho s snr , approximately 1@xmath64 northward of knot g. the clump appears behind the main body of the balmer filaments and exhibits a broad component that is substantially doppler shifted to the red ( 10.7@xmath651.2 , or about 490 kms@xmath26 ) . the doppler shift reflects the bulk velocity of the hot posthock proton distribution , so the significant velocity shift of the broad component centroid indicates that the shock in the ne has a substantial velocity component into the plane of the sky , i.e. , that the ne shock is located on the far side of the blast wave shell . the broad h@xmath48 width of the ne shock is 1300@xmath6565 kms@xmath26 ( the smallest broad component width measured in tycho s snr so far ) , with @xmath55=0.85@xmath650.04 . the nw shock , on the other hand , has a broad h@xmath48 width of 2040@xmath6555 kms@xmath26 ( the largest broad component width measured in this snr so far ) , with @xmath55=0.45@xmath650.15 . neither of these broad - to - narrow ratios is strictly reproduced by the latest models of van adelsberg et al . ( 2008 ) , with the lowest predicted ratios being for case b ( the assumption of optically thick conditions for ly @xmath52 photons in the narrow component ) and for low equilibrations ( @xmath16@xmath570.1 ) . for these low equilibrations , the corresponding shock speeds for the ne and nw shocks in tycho are approximately 1400 kms@xmath26 and 2250 kms@xmath26 ( using figures 5 and 10 of van adelsberg et al . 2008 ) . in figure [ fig : plotequil ] we have added data from the ne and nw shocks in tycho s snr to the @xmath16-@xmath39 plot . the two data points help fill in a portion of the plot where the data are sparse : the region between approximately 1200 kms@xmath26 and 1500 kms@xmath26 , as well as the region beyond 2000 kms@xmath26 , where the existing data are taken entirely from sn 1006 . the added point near 1400 kms@xmath26 is fully consistent with the @xmath61 relation , while @xmath16 is not well constrained enough for the point at 2250 kms@xmath26 to contradict the appearance of an upturn at the highest shock velocities . it is clear that proper characterization of the equilibration - shock velocity curve above 2000 kms@xmath26 will require both higher signal - to - noise spectra on existing balmer - dominated shocks , as well as new data points beyond a broad component width of 2500 kms@xmath26 . although the inverse squared relation between equilibration and shock velocity appears has been the most salient result of the study of balmer - dominated shocks , there have been discrepant results reported in a small subset of cases . recently the broad h@xmath48 component in the lmc snr 0509@xmath5667.5 was detected for the first time ( helder , kosenko & vink 2010 ) . broad emission was observed along both the northeastern rim ( fwhm 3900@xmath65800 kms@xmath26 ) and southwestern rim ( fwhm 2680@xmath6570 kms@xmath26 ) , with the former being the fastest balmer - dominated shock detected to date having the characteristic broad and narrow component h@xmath48 emission . interestingly , the broad - to - narrow ratios for both shocks are exceptionally low , @xmath55=0.08@xmath650.02 and 0.29@xmath650.01 along the ne and sw rims , respectively . as noted by helder , kosenko & vink ( 2010 ) these ratios are nearly twice as large as the smallest ones predicted by the models of van adelsberg et al . ( 2008 ) , precluding measurement of @xmath16 from the balmer - dominated spectra and implicating excess narrow component h@xmath48 emission in a cosmic ray precursor once again . using rgs spectra of snr 0509@xmath5667.5 acquired with _ xmm - newton _ , they obtained a forward shock speed of approximately 5000 kms@xmath26 in the sw . this implies a broad h@xmath48 width of 3600 kms@xmath26 , substantially smaller than the measured width of 2680 kms@xmath26 . helder , kosenko & vink ( 2010 ) suggest that this indicates some thermal energy loss ( @xmath1020% ) to cosmic ray acceleration . this picture is supported by the presence of nonthermal x - ray emission in their fitted rgs spectra of snr 0509@xmath5667.5 . the result described above relies upon an accurate disentangling of the bulk doppler broadening from the thermal doppler broadening in the x - ray lines . the disentangling depends on parameters such as the ratio of reverse shock to forward shock velocities , as well as the ratio of the gradients in the reverse and forward shock velocities , which in turn had to be assumed from evolutionary snr models . the result , while intriguing , is still significantly uncertain . on the other hand , helder , kosenko & vink ( 2010 ) found that the x - ray shock velocities from the ne could only be reconciled with the observed broad h@xmath48 width there if @xmath16@xmath660.2 , which clearly which predicts @xmath16@xmath67 predicted from equation [ tetpeq ] . if this result were to be confimed by future observations , it would present a new challenge in understanding how electron - ion equilibration occurs in fast collisionless shocks . one possibility , given the presence of nonthermal x - ray emission in the spectra of 0509@xmath5667.5 ( warren & hughes 2004 ; helder , kosenko & vink 2010 ) is that the moderate loss of thermal energy to cosmic ray acceleration may have slightly increased the compression and reduced the temperature at the shock front compared to the case with no acceleration ( decourchelle , ellison & ballet 2000 ; ellison et al . 2007 ) . both of these effects would tend to render the plasma more collisional , possibly explaining the @xmath16@xmath660.2 result . however , it is also worth noting that the shock velocity used for obtaining @xmath16 in the ne is subject to the same model dependence and uncertainties as the sw measurements , so similar caution is required in its interpretation . a similar combined optical and x - ray study of rcw 86 was performed by helder et al . ( 2009,2011 ) . there , the broad component h@xmath48 widths were supplemented with electron temperatures measured from _ xmm - newton _ rgs spectra from the same projected locations along the rim . one of the main results of this study was that the slower shocks ( broad h@xmath48 fwhm @xmath10500 - 600 kms@xmath26 ) showed showed @xmath16@xmath101 , agreeing with earlier results from similar shocks observed both in rcw 86 and elsewhere ( ghavamian et al . ( 1999 , 2001 , 2007 ) ) . however , helder et al . ( 2011 ) found while the results were indeed consistent with low equilibration at the shock front for fast shocks ( @xmath16@xmath590.02 for broad fwhm @xmath101100 kms@xmath26 ) ) and higher equilibration for the slower shocks ( @xmath16@xmath591 for broad fwhm @xmath10650 kms@xmath26 ) , their x - ray derived electron temperatures were inconsistent with @xmath680.3 kev at the shock front contradicting the suggestion of ghavamian et al . ( 2007 ) that shocks above 400 kms@xmath26 may all heat electrons to roughly 0.3 kev . however , a major caveat of these results is that the forward shock in rcw 86 is believed to be impacting the walls of a wind - blown bubble ( williams et al . 2011 ) , resulting in substantial localized variations in shock speed around the rim . these variations occur as different parts of the forward shock impact the cavity wall at different times . while the broad component h@xmath48 widths closely trace the current position of the shock front , the x - ray emission behind that shock arises over a much more extended spatial scale , and is sensitive to the history of the forward shock interaction with the cavity wall . furthermore , narrowband h@xmath48 imagery of rcw 86 with the eso very large telescope ( helder et al . ( 2009 , 2011 ) ) shows a complex morphology of filaments , especially along the eastern side of the snr . the broad h@xmath48 components of these filaments exhibit substantial , localized variations in line width , ranging from @xmath10600 kms@xmath26 fwhm to 1100 kms@xmath26 ghavamian ( 1999 ) , helder et al . ( 2009 , 2011 ) ) . these variations reflect localized changes in density and viewing geometry along the line of sight . as such , uniquely mapping the observed balmer filaments to their corresponding x - ray emission in _ xmm - newton _ data ( especially given the somewhat coarse 10@xmath69 spatial resolution of that instrument ) is fraught with uncertainty . additional corroboration for these results would be desirable . perhaps the most important and exciting recent development in our understanding of balmer - dominated shocks has come with the development of new kinetic - based shock models . blasi et al . ( 2012 ) have introduced a kinetic model for following the momentum and energy exchange between neutrals and ions , along with the back - reaction of those neutrals when they pass back upstream and form a fast neutral precursor . rather than assume a maxwellian velocity distribution for the neutrals ( as had been done in previous models , despite the lack of thermal contact between neutrals needed to justify such an assumption ) , both the ion and neutral distributions are computed from their appropriate boltzmann equations . building on these models , morlino et al . ( 2012a , b ) have confirmed what had been suspected earlier ( smith et al . 1994 ; hester et al . 1994 ; sollerman et al . 2003 ) , namely that the broadening of the narrow component beyond the expected ism value ( @xmath1025 - 30 kms@xmath26 instead of 10 kms@xmath26 ) is most likey due to heating in a cosmic ray precursor . in particular , morlino et al . ( 2012 ) found that the characteristic charge exchange length of the incoming neutrals exceeds that of the neutrals crossing back upstream , so that the narrow component width is impacted not by the neutral return flux , but rather by heating in the cosmic ray precursor . raymond et al . ( 2011 ) predicted a similar broadening of the narrow component , though they focused mainly on the contribution of collisional excitation in the precursor to the flux in the narrow h@xmath48 component . spatially resolved line broadening of the narrow h@xmath48 component was detected in ground - based longslit spectroscopy of knot g in tycho ( lee et al . in addition , a small ramp - up in h@xmath48 emission was observed ahead of knot g in hst imagery tycho s snr ( lee et al . 2010 ) . the results , taken together , are strong evidence for the presence of cosmic ray precursors in balmer - dominated snrs . one prediction of the new balmer - dominated shock models is the existence of a third component of the h@xmath48 emission ( morlino et al . 2012a , b ) . when the hot neutrals escape upstream , they undergo charge exchange with the colder preshock protons . this results in fast protons with cold neutrals , with the former rapidly equilibrating with preshock protons and pre - heating the gas . the temperature of the equilibrated protons in the precursor lies between the temperature of the far upstream protons ( @xmath105000 k ) and the far downstream protons ( @xmath10 10@xmath70 - 10@xmath71 k ) , typically @xmath1010@xmath72 k. further charge exchange between these warm protons and the preshock neutrals gives rise to a third , warm neutral component ( neither fast nor slow ) having velocity widths of hundreds of kms@xmath26(morlino et al . interestingly , the presence of a third h@xmath48 component was first observationally reported by smith et al . ( 1994 ) in their high resolution echelle spectroscopy of balmer - dominated snrs in the large magellanic cloud ( an example of one of the spectra from smith et al . ( 1994 ) is reproduced in figure [ fig:0509_3rdhalphacomp ] ) . a third h@xmath48 component was also reported in high resolution spectra of knot g in tycho s snr ( ghavamian et al . 2000 ) . in their models of balmer - dominated shock emission , morlino et al . ( 2012a ) found that the importance of the third component relative to that of the broad and narrow components depends strongly on the preshock neutral fraction and @xmath16 , in line with earlier theoretical predictions on properties of a fast neutral precursor ( smith et al . the fact that the third component has been detected in tycho s snr ( width @xmath10150 kms@xmath26 ) is consistent with the high preshock neutral fraction ( @xmath730.9 ) inferred from the broad - to - narrow ratio of knot g by ghavamian et al . a similar third component may have been detected in high resolution spectra of snr 0509@xmath5667.5 , where measurement of the narrow component width required the inclusion of an additiona component of width 75 kms@xmath26 ( smith et al . 1994 ) . the new fast neutral precursor models predict that a substantial fraction of the h@xmath48 excitation in balmer - dominated shocks can arise ahead of the shock , where warm neutrals are excited by electron impact . interestingly , the relative contribution of the preshock h@xmath48 to the total ( upstream + downstream ) is sensitive to @xmath16 behind the shock . morlino et al . ( 2012a ) found that up to 40% of the total h@xmath48 flux from a balmer - dominated shock can arise from the fast neutral precursor when @xmath39@xmath102500 kms@xmath26 and @xmath16=1 both upstream and downstream . in these models the preshock contribution to the total flux drops substantially for lower downstream equilibrations for @xmath39@xmath581000 kms@xmath26 ( the slowest shocks considered by morlino et al . this is generally consistent with the fact that in most cases , shock models not including the precursor h@xmath48 emission have been able to match the observed broad - to - narrow ratios . in other words , if the postshock temperature equilibration were not low for such fast shocks , the agreement between the observed and predicted @xmath55would have been substantially worse for such remnants as tycho s snr and sn 1006 . sn 1006 is an example of a snr accessible to uv spectroscopy due to its galactic location , 450 pc up from the galactic plane , and therefore with relatively low extinction due to intervening dust and gas . the hopkins ultraviolet telescope ( hut ) observed the uv resonance lines of he ii @xmath431640 , c iv @xmath441548 , 1550 , n v @xmath441238,1243 , and o vi @xmath431032 , 1038 ( raymond et al . 1995 ) emitted from the balmer dominated filament in the nw quadrant . the line showed doppler broadening consistent with that of the h @xmath48 broad component observed in the optical , indicating insignificant ion - ion equilibration . laming et al . ( 1996 ) were also able to infer the degree of electron - ion equilibration at the shock . -0.2 in he ii @xmath431640 , by virtue of its relatively high excitation potential @xmath10 48 ev ) , is excited only by electrons , and its intensity is therefore directly related to the electron temperature . c iv @xmath441548 , 1550 , n v @xmath441238,1243 , and o vi @xmath431032 , 1038 , by contrast , have much lower excitation potentials of @xmath74 and 12 ev , so although these ionization states are established by electrons , the line emission in these transitions can also be excited by impacts with hot protons and @xmath48 particles , and the intensity ratio of he ii @xmath431640 to c iv @xmath441548 , 1550 , n v @xmath441238,1243 , and o vi @xmath431032 , 1038 can be sensitive to the post - shock electron - proton temperature equilibration . the spatial distribution of the uv resonance line emission , when spatially resolved , provides additional constraints on the degree of electron - ion equilibration at the shock front . for shocks slower than 1500 kms@xmath26 , @xmath75 at the shock front results in both a more rapid rise and higher maximum in emissivity of the uv resonance lines with distance behind the shock ( see figure [ fig : ovispatial ] ) . laming et al . ( 1996 ) calculated impact excitation cross sections for protons and @xmath48 particles colliding with li - like ions , using a partial wave expansion with the coulomb - bethe approximation , and applying a unitarization procedure following seaton ( 1964 ) . they found a degree of equilibration of order @xmath76 or less , which implied for a 2250 km s@xmath26 shock an electron temperature immediately postshock of @xmath77 k , in very good agreement ( and actually predating ) the optical results discussed above in subsection 4.1 . the ion - ion equilibration in sn 1006 was revisited by korreck et al.(2004 ) , using higher spectral resolution fuse data comprising o vi @xmath431032 , 1038 and the broad ly @xmath781025 emission lines . they found a slightly broader line profile in ly @xmath52 , implying less than mass - proportional heating and possibly a small degree of ion - ion equilibration . sn 1987a represents another sn / snr in a region of the sky accessible to uv observations . hst cos observed the he ii @xmath431640 , c iv @xmath441548 , 1550 , n v @xmath441238,1243 and n iv @xmath441486 lines emitted from the reverse shock ( france et al . when combined with optical spectroscopy of h @xmath48 , the @xmath16 ratio at the shock is determined to be in the range 0.14 0.35 , significantly higher than similar ratios coming from balmer dominated forward shocks . france et al . ( 2011 ) argued that a different equilibration mechanism is likely at work . considering the relative youth of sn 1987a , and the fact that the reverse shock is the origin of the emission , significant populations of cosmic rays and associated magnetic field amplification are unlikely . in fact , in the expanding ejecta the magnetic field is likely to be very weak , leading to a very high aflvn mach number shock . as will be discussed below in connection with shocks in galaxy clusters , electron heating in such a case is likely to be due to acceleration in the cross - shock potential . the cross - shock potential is effective at heating electrons , and so may explain the higher @xmath16 in sn 1987a . the forward shock of sn 1987a has also been observed in x - rays with the grating instruments on chandra ( e.g. zhekov et al . 2009 ) . in general electron heating well below complete equilibration is seen , though precise interpretation is difficult because one observation sees emission from shocks at a variety of different velocities , due to irregularities in the density of the surrounding medium . the inverse relationship between the temperature equilibration and shock speed is an interesting result from studies of balmer - dominated snrs . however , the applicability of this result to both fully ionized shocks and shocks undergoing efficient cr acceleration ( @xmath5850% of their energy transferred to crs ) remains unsettled . recently , vink et al . ( 2010 ) used a two - fluid - model for cosmic rays and thermal gas to simulate the effect of cosmic ray acceleration on the temperature and ionization structure of fast , non - relativistic shocks . they found that if 5% of the shock energy were to be channeled into cosmic rays ( the minimum needed if snrs are the dominant source of cosmic rays ) then approximately 30% of the postshock pressure must reside in cosmic rays ( corresponding to a ratio of cosmic ray to total postshock pressure , w , of 0.3 ) . for w=0.3 , vink et al . ( 2010 ) predicted a lowering of the average temperature of the postshock gas to @xmath1070% of the value when the cosmic ray contribution is ignored . this is a significant alteration of the postshock temperature profile , and should result in much more rapid equilibration of electrons and protons close to the shock . however , do the effects described above actually occur in snr shocks ? one of the principle lines of evidence cited by vink et al . ( 2010 ) in support of this picture was the result found in rcw 86 by helder et al . ( 2009 ) . in that snr , the broad h@xmath48 widths of balmer - dominated filaments were found to be nearly 50% smaller than the minimum allowed given their x - ray proper motions . filaments in the ne of the snr exhibited broad h@xmath48 widths of 1000 kms@xmath26 , but their apparent x - ray counterparts , which exhibited strong x - ray synchrotron ( nonthermal ) emission , exhibited proper motions indicating shock speeds of 3000 - 6000 kms@xmath26 . this result , along with the theoretical prediction that x - ray synchrotron emission requires shock speeds of at least 2000 kms@xmath26 ( aharonian et al . 1999 ) was taken by helder et al . ( 2009 ) and vink et al . ( 2010 ) as evidence for substantial energy loss ( w@xmath100.5 ) from the balmer - dominated shocks to cosmic rays . however , this association has now been refuted by subsequent multi - epoch optical imagery of the h@xmath48 filaments , which have failed to show the kind of high proper motions seen in the nonthermal x - ray filaments ( helder et al . 2012 , in preparation ) . instead , they show proper motions consistent with shock speeds predicted by the broad h@xmath48 widths without energy loss to cosmic rays(@xmath10600 - 1200 kms@xmath26 ) , implying that w@xmath790.2 . the association between the balmer - dominated shocks studied spectroscopically by helder et al . ( 2009 ) and the x - ray filaments was due either to coincidental spatial alignment , or due to sudden deceleration of the outer shock in rcw 86 during its encounter with the surrounding cavity wall ( williams et al . 2011 ; helder et al . 2012 , in preparation ) . -1.2 in the lack of association between the balmer - dominated filaments and the non - thermal x - ray filaments in rcw 86 raises some important questions about the feasibility of using balmer - dominated shocks to study electron - proton equilibration in cases where 50% or more of the thermal energy is diverted to cosmic rays . in snrs such as sn 1006 ( koyama et al . 1996 ; katsuda et al . 2010a ) , tycho s snr ( warren et al . 2005 ; katsuda et al . 2010b ) , cas a ( vink & laming 2003 ) and rx j1713.7@xmath563946 ( koyama et al . 1997 ; slane et al . 1999 ; tanaka et al . 2008 ) the presence of strong synchrotron x - ray filaments has been interpreted as evidence for highly efficient cosmic ray acceleration . the narrowness of the synchrotron filaments most likely reflects the short emitting lifetimes of the ultra high energy electrons ( energies @xmath1010 - 100 tev ) as they spiral in the postshock magnetic field ( vink & laming 2003 ) . the detection of @xmath80-ray emission from the shells of sn 1006 ( acero et al . 2010 ) and rx j1713.7@xmath563946 ( aharonian et al . 2006 ; abdo et al . 2011 ) has shown that cosmic rays are accelerated to energies as high as 100 tev in these snrs . in all cases thermal x - ray emission has been exceptionally faint due to the very low inferred preshock densities ( @xmath570.1 @xmath29 ) , making it more likely that the overall x - ray emission will be dominated by synchrotron radiation from the most energetic cosmic rays . snrs expanding into such low density media can propagate at the high shock speeds required for cosmic ray acceleration ( @xmath582000 kms@xmath26 ; aharonian et al . 1999 ) for a longer time , allowing their structure to become modified by the back pressure from the cosmic rays . investigating the temperature and ionization structure of such shocks with balmer line spectroscopy requires finding balmer - dominated shocks exhibiting x - ray synchrotron radiation . all of the known snrs exhibiting balmer - dominated shocks have also been imaged at x - ray wavelengths with _ chandra _ or xmm , allowing reasonably detailed searches for shocks emitting both h@xmath48 and synchrotron x - ray emission ( the latter producing hard continuum that is dominant at energies of 2 kev and higher ) . a detailed comparison for all balmer - dominated snrs has not yet been published . however , even a cursory comparison between the narrowband h@xmath48 and hard x - ray images of these snrs shows a distinct _ anticorrelation _ between shocks emitting in these two bands . for example , overlaying h@xmath48 and _ chandra _ images of tycho s snr acquired during the same epoch ( 2007 ) ( figure [ fig : tycho_balmer ] ) shows little or no correlation between the prominent balmer - dominated filaments on the eastern and northeastern edges and the non - thermal x - ray filaments ( e@xmath543 kev ) circling the remnant . the balmer - dominated filaments ( shown in red in figure [ fig : tycho_balmer ] ) on the eastern side of tycho s snr are seen projected 30@xmath69 - 1@xmath64 inside the edge of the nonthermal x - ray filaments ( marked in green ) , an indication that portion of the shell along the line of sight has significantly decelerated . the balmer filaments are seen at the outermost edge of the thermally emitting x - ray ejecta ( marked in blue ) , but only at locations where little or no nonthermal x - ray emission is present . the bright optical filament known as knot g ( at the far left edge of figure [ fig : tycho_balmer ] ) is the only location where balmer filaments and x - ray synchrotron emission appear coincident . however , upon closer inspection the anticorrelation between the balmer line and synchrotron emission can be seen in knot g as well : the upper half of the filament , where balmer line emission is strongest , exhibits minimal synchrotron emission , while the opposite is true in the lower half of the filament . the enhanced nonthermal emission inside of knot g may be due to the strong recent deceleration of knot g , where the snr is currently propagating into a strong density gradient at the outermost edge of an h i cloud ( ghavamian et al . the lack of optical / x - ray synchrotron correlation is especially striking given that the balmer - dominated filaments in tycho s snr have a high enough shock velocity ( @xmath101800 - 2100 kms@xmath26 ) to accelerate particles to tev energies . the anticorrelation between the optical and nonthermal x - ray emission can be observed in other snrs as well , including sn 1006 , where recently x - ray proper motions have been measured along the entire rim by katsuda et al . ( 2012 ) . as with tycho s snr , the locations of the balmer - dominated filaments and the nonthermal x - ray filaments along the nw rim of sn 1006 are mutually exclusive . instead , the balmer - dominated shocks are closely associated with thermal x - ray filaments having a proper motion consistent with a shock velocity of 3300@xmath65200@xmath65300 kms@xmath26 ( statistical and registrational uncertainties , respectively ) for a distance of 2.2 kpc . this result is in excellent agreement with the shock velocity of 2890@xmath65100 kms@xmath26 determined from the broad h@xmath48 width and broad - to - narrow ratio of the nw filament by ghavamian et al . such close agreement is a strong indication that little substantial energy has been lost from the thermal plasma to cosmic ray acceleration , similar to optical proper motion studies from rcw 86 ( helder et al . , in preparation ) . from the above discussion it appears that balmer - dominated snrs , while offering powerful diagnostics of @xmath16 and @xmath39 , are not useful for investigating equilibration the extreme cases of strongly cosmic - ray modified shocks . in fact , the very condition allowing for the detection of the balmer line emission - presence of neutral gas ahead of the shock - is also responsible for limiting the fraction of shock energy lost to cosmic ray acceleration . quantatitive evaluations of this effect by drury et al . ( 1996 ) and reville et al . ( 2007 ) show that when the preshock gas is significantly neutral , alfvn waves driven by the cosmic rays ahead of the shock are dissipated by ion - neutral damping . as long as the charge exchange frequency , @xmath81 ( @xmath82 ) is larger than the alfvn wave frequency , @xmath83 ( @xmath84 ) the ions and neutrals oscillate coherently , and ion - neutral damping is not important . however , when @xmath85 the neutrals are left behind by the ions in the alfvn wave motion , and during the incoherent oscillation between the two , charge exchange exerts a drag on the alfvn waves , damping them . the condition required for alfvn waves to not be strongly damped in the precursor can be written out as @xmath86 where @xmath87 is the alfvn speed of the ions ahead of the shock and @xmath88 is the preshock neutral density . given that cosmic rays resonantly scatter off alfvn waves having doppler shifted frequencies comparable to their gyrofrequency , and that the cosmic ray gyrofrequency is related to its energy via @xmath89 , the inequality above can be cast in terms of the energy , @xmath90 , below which a significant fraction of the cosmic ray flux out of the shock is reduced by ion - neutral damping : @xmath91 where we have set @xmath928.4@xmath9310@xmath94 @xmath95 s@xmath26 ( kulsrud & cesarsky 1971 ) , @xmath96 is the preshock magnetic field strength in units of 3 @xmath12 , @xmath15 and @xmath97 are the total preshock density and neutral fraction , and @xmath98 is the preshock temperature in units of 10@xmath99 k. for balmer - dominated snrs , where recent models have required moderate amplification of the preshock magnetic field ( @xmath1003 - 5 ; ghavamian et al . 2007 ) and where the preshock temperature may exceed 20,000 k ( raymond et al . 2011 ) , @xmath1014 tev for the typical case where @xmath1020.5 @xmath29 . snrs exhibiting nonthermal x - ray emission are believed to contain cosmic rays with energies of tens of tev , so @xmath1034 tev is certainly high enough reduce the effectiveness of balmer - dominated shocks in producing nonthermal x - ray emission . however , as noted earlier , a modest back pressure from cosmic rays is required to explain the width of the h@xmath48 narrow component line , as well as the low broad - to - narrow ratios seen in some snrs ( rakowski et al . 2009 ; raymond et al . in fact , one model for electron heating in fast collisionless shocks requires at least some feedback from the cosmic rays in order to explain the moderate heating of electrons in snrs , as well as the inverse squared relationship between @xmath16 and @xmath39 ( ghavamian et al . 2007 , described in the next section ) . furthermore , as pointed out by drury et al . ( 1996 ) , the ion - neutral damping of alfvn waves in the precursor is unimportant for cosmic rays which have already exceeded @xmath90 . since the acceleration time for cosmic rays shortens considerably with shock speed ( @xmath104 ; malkov & drury 2001 ) , the fastest balmer - dominated shocks are more likely to have accelerated particles beyond @xmath90 and hence will begin to exhibit nonthermal x - ray emission and cosmic - ray modified shock structure . a good example is the aforementioned snr 0509@xmath5667.5 , where the shock speeds exceed 5000 kms@xmath26 ( helder , kosenko & vink 2010 ) and nonthermal x - ray emission from cosmic ray accelerated electrons is detected from the forward shock . the forward shocks in more evolved balmer - dominated snrs ( such as sn 1006 and tycho s snr ) will have swept up more mass and slowed down to speeds @xmath572000 kms@xmath26 , by which point @xmath105 will have lengthened and the shocks will be less cosmic - ray dominated . given the lack of in situ measurements of the particle distributions in snrs , the electron heating mechanisms in these shocks have been studied primarily via numerical methods . on one hand , a number of studies have focused on electron heating in relativistic shocks , with the aim of modeling high energy emission from gamma - ray bursts ( e.g. , gedalin et al . 2008 ; sironi & spitkovsky 2011 ) . these shocks are in a different area of parameter space than the snr shocks discussed here , and the physics governing the electron heating in relativistic shocks is substantially different . at the very high alfvnic mach numbers characteristic of gamma - ray bursts , the shock transition becomes very thin ( less than an electron gyroradius ) . electrons in this case may be once again accelerated by the cross shock potential , similar to the very low mach number case . on the other hand , a number of other studies consider non - relativistic shocks relevant to snrs ( @xmath570.01c ) , where accelerated particles such as cosmic rays or solar energetic particles ( seps ) may play an important role in establishing their shock structure . these studies have sought to identify plasma waves capable of boosting electrons to mildly relativistic energies ( e.g. , amano & hoshino 2010 ; riquelme & spitkovsky 2011 ) , with the objective of understanding how electrons are injected into the cosmic ray acceleration process . this is a different ( though related ) question from what we consider here , namely how electrons are promptly heated to temperatures @xmath105@xmath9310@xmath70 k at the shock front ( ghavamian et al . 2007 ; rakowski et al . 2008 ) . this limits our consideration of the work done so far to two broad scenarios of electron heating in fast , non - relativistic collisionless shocks . one scenario is based on lower hybrid wave heating in the cosmic ray precursor ( laming 2000 ; ghavamian et al . 2007 ; rakowski et al . 2008 ) while the other is based on counterstreaming instabilities ahead of the shock ( e.g. , the buneman instability , cargill & papadapoulos ( 1988 ) , matsukiyo 2010 ; dieckmann et al . we discuss these mechanisms below in turn . the most significant result of the balmer - dominated shock studies , the inverse squared relation between @xmath16 and @xmath39 , places a useful constraint on the range of plausible equilibration mechanisms at the shock front . the simplest way to obtain @xmath16@xmath106 is to set @xmath107 at shock speeds of 400 kms@xmath26 and higher , while allowing @xmath1 to rise according to the rankine - hugoniot jump conditions , @xmath108 . the requirement that @xmath16=1 at @xmath39=400 kms@xmath26 gives @xmath1090.3 kev for @xmath39@xmath54400 kms@xmath26 , independent of shock velocity ( ghavamian et al . although there may be marginal evidence of a departure from this relation at shock speeds exceeding 2000 kms@xmath26 ( van adelsberg et al . 2008 ) , a velocity - independent heating of electrons in snr shocks is an important clue to the nature of plasma heating processes in fast collisionless shocks . it suggests that plasma processes ahead of the shock front are an important ( if not dominant ) source of electron heating in snrs ( ghavamian et al . 2007 ; rakowski et al . 2009 ) . as mentioned earlier , strong interstellar shocks are expected to form a precursor where cosmic rays crossing upstream give rise to alfvn waves and turbulence ( blandford & eichler 1987 ; jones & ellison 1991 ) , compressing and pre - heating the gas before it enters the shock . as long as the shock is strong ( @xmath1101/4 ) and cosmic ray pressure does not dominate the postshock pressure ( @xmath111b / b does not greatly exceed unity , with @xmath5720% of the postshock pressure provided by cosmic rays ) the thermal heating within the precursor does not depend strongly on shock velocity . the limited range of narrow component h@xmath48 widths observed in balmer - dominated snrs over a wide range in shock speeds ( sollerman et al . 2003 ; raymond et al . 2011 ) is consistent with the relative insensitivity of the preshock heating to shock speed . since the widening of the h@xmath48 narrow component line is now believed to arise in a precursor where the gas is heated by the damping of cosmic - ray driven waves ( wagner et al . 2008 ; raymond et al . 2011 ; morlino et al . 2012 ) , it stands to reason that perhaps the physical processes generating a constant electron heating with shock speed ( @xmath1090.3 kev ) also originate within the cosmic ray precursor . the above argument was used by ghavamian et al . ( 2007 ) and rakowski et al . ( 2008 ) to advocate for a heating model where lower hybrid waves within the cosmic ray precursor preheat electrons to a constant temperature before they enter the shock front . this model was based on the conception of mcclements et al . ( 1997 ) , who suggested that lower hybrid waves driven by the reflected population of nonthermal ions could generate lower hybrid waves ahead of the shock , pre - heating electrons and injecting them into the cosmic ray acceleration process . the condition for generating such waves is that the shock be quasi - perpendicular , and that the reflected ions form a beam - like ( gyrotropic ) distibution . a similar scenario was suggested by ghavamian et al . ( 2007 ) and rakowski et al . ( 2008 ) , but with one crucial difference : the reflected particles considered are ultra - relativistic cosmic rays rather than suprathermal ions . the lower hybrid waves are electrostatic ion waves which propagate perpendicular to the magnetic field and whose frequency is the geometric mean of the electron and ion geofrequencies , @xmath112 . the group velocity of these waves is directed primarily along the magnetic field lines ( @xmath113 ; laming 2001 ) and the waves can simultaneously resonate with ions moving across the field lines and electrons moving along the field lines . although the growth rate of lower hybrid waves is generally small ( rakowski et al . 2008 ) , their group velocity perpendicular to the magnetic field ( and hence the shock front ) can be on the order of the shock velocity ( @xmath114@xmath39 ) . this allows the lower hybrid waves to remain in contact with the shock for long periods of time , attaining high intensities capable of effectively heating the electrons ( mcclements et al . 1997 ; ghavamian et al . 2007 ) . in the case of cosmic rays , the time spent by the electrons in the precursor is @xmath115 . the kinetic energy acquired by the electrons in the precursor is @xmath116 , where @xmath117 is the momentum diffusion coefficient of electrons ( ghavamian et al . 2007 ) . for lower hybrid wave turbulence , @xmath118 ( karney 1978 ; ghavamian et al . 2007 ; rakowski et al . 2008 ) , so that @xmath119 , as needed to account for the inverse squared relationship between equilibration and shock speed . note that under the assumption that nonlinear amplification of the preshock magnetic field is not too strong ( @xmath120 ) , @xmath121 is that of bohm diffusion , which scales as 1/b , so that @xmath122 is also approximately independent of b. during the past decade more refined models of cosmic ray acceleration have shown that a non - resonant mode of alfvn waves , having a higher growth rate than the previously considered resonant mode ( skilling 1975 ) , can be excited by cosmic rays in the precursor ( bell & lucek 2001 , bell 2004 , 2005 ) . unlike for the resonant case , the non - resonant amplification allows for @xmath123 , driving preshock magnetic fields to values as high as 1 mg ( vink & laming 2003 ; berezhko et al . 2003 ; bamba et al . 2005 ; ballet 2006 ) . such magnetic fields are hundreds of times stronger than the canonical preshock magnetic field of 3 @xmath12 and high enough to account for the observed narrowness of x - ray synchrotron - emitting rims in such snrs as sn 1006 ( assuming the narrowness is due to rapid cooling of high energy electrons behind the shock ; see ballet 2006 and morlino et al . additional studies have suggested that non - resonant amplification may dominate early in the life of the snr , while resonant amplification may take over during the sedov - taylor stage of evolution ( amato & blasi 2009 ; schure et al . 2012 ) , though in either case , @xmath12410 is readily attained . such a strong magnetic field effectively reduces the acceleration time for particles , and is very well suited for explaining how cosmic rays can reach the knee in the cosmic ray spectrum near 10@xmath125 ev ( bell & lucek 2001 ; eriksen et al . 2011 ) . an important factor influencing the effectiveness of lower hybrid wave heating of electrons is the orientation of the preshock magnetic field relative to the shock front . lower hybrid wave heating is only effective in perpendicular shocks ( vink & laming 2003 , ghavamian et al . 2007 ; rakowski et al . given their spherical global geometry , snr blast waves generally propagate at a range of angles to the interstellar magnetic field . x - ray observations and models of such snrs as sn 1006 ( orlando et al . 2007 ; petruk et al . 2008 ) have indicated that perpendicular shocks are far more effective at accelerating cosmic rays than parallel shocks . although the detailed implications of such differences have not yet been worked out for the lower hybrid wave heating model , rakowski et al . ( 2008 ) argue that even for quasi - parallel shock geometries the cosmic ray current driving the nonresonant alfvn waves will generate a significant perpendicular magnetic field ahead of the shock ( such a possibility has also been inferred from numerical simulations ; riquelme & spitkovsky 2011 ) . this would allow lower hybrid wave growth to overtake modified alfvn wave growth for arbitrary orientations of the far upstream magnetic field , and allow for a more ubiquitous role for lower hybrid wave heating of electrons . the amplification of the preshock magnetic field well beyond its far upstream value introduces an interesting possibility : effective lowering of the alfvnic ( and hence magnetosonic ) mach number of the shock . for the balmer - dominated shocks , where analysis of the optical spectra has shown that at best only a moderate fraction ( @xmath5720% ) of the shock energy has likely been channeled into cosmic rays , the widening of the h@xmath48 narrow component has been interpreted as nonthermal broadening caused by the lowest frequency waves in the precursor ( ghavamian et al . 2007 ; though see raymond et al ( 2011 ) for a thermal intepretation ) . to explain the 30 - 50 kms@xmath26 widths of the h@xmath48 narrow component , the preshock magnetic field must be enhanced by a factor of a few . for the non - resonant alfvn waves in the bell ( 2004 ) mechanism , the magnetic field energy density immediately behind the shock is given by ( schure et al . 2012 ) @xmath126 where @xmath127 is the fraction of the shock ram pressure channeled into cosmic rays . solving this expression for b gives @xmath128 , where @xmath129 is the shock speed in units of 1000 kms@xmath26 . for @xmath1300.1 - 0.2 , @xmath451 @xmath29 , a postshock compression factor of 4 and balmer - dominated shock speeds @xmath102000 kms@xmath26 , this gives @xmath1314 - 10 ahead of the shock . correspondingly , @xmath132 can increase by nearly the same factor , so that @xmath133 can be reduced by as much as an order of magnitude . treumann & jaroschek ( 2008 ) describe the physical picture in this case as that of the shock having to prevent an increasing number of ions from crossing the shock jump by deflecting an increasing number of them at higher and higher mach numbers . this deflection is necessary so that the ability of the shock to dissipate the inflowing energy is not overwhelmed . by deflecting these ions back upstream into a precursor , the net inflow of momentum and energy density into the shock is reduced , reducing the net difference in velocity between the inflowing and outflowing ions . this effectively reduces the mach number in the frame of the upstream medium . note that the ions in the precursor are only mildly compressed ( wagner et al . 2008 ; morlino et al . 2012 ) , which only weakly counteracts the rise in b. in addition , @xmath132 only scales as @xmath134 , but scales directly as b. the result is that _ given the compelling evidence for enhanced preshock magnetic fields in snrs shocks , the mach numbers of these shocks may be overestimated by as much as an order of magnitude . _ as we describe in section [ sec:5 ] , a unified description of solar wind and snr shocks , where the physics of electron - ion temperature equilibration occurs over a similar range in mach numbers and involves a similar range of physical processes , may be possible . similar to the lower hybrid wave model , the buneman instability - driven wave model focuses on the region immediately ahead of a quasi - perpendicular shock . however , unlike the lower hybrid wave model , the buneman instability models consider the reflected nonthermal ion distribution , rather than ultrarelativistic cosmic rays . in the latter model , @xmath1020% of the ions are reflected backstream against the incoming electron and ion plasma ( papadapoulos 1988 ; cargill & papadapoulos 1988 ) . here the upstream plasma is not electrically neutral due to the positive charge of the reflected ion distribution . in such cases , a drift is induced between the electrons and ions . the microinstabilities excited by this configuration depend upon the relative size of the electron thermal speed relative to the electron - ion drift velocity . the buneman instability occurs when the drift velocity of the reflected ions relative to the upstream electrons exceeds the thermal speed of the upstream electrons ( @xmath135)(cargill & papadapoulos 1988 ) , a condition which occurs for very high mach number ( @xmath13650 ) shocks . if the reflected ion current upstream is strong enough , the electron current generated to counteract it may produce a large enough drift between the preshock ions and electrons to cause a secondary buneman instability when the ion speed exceeds the electron thermal speed ( dieckmann et al . the buneman instability generates electrostatic plasma waves which damp by rapidly heating the preshock electrons to @xmath137 kev , where @xmath138 is the shock speed in units of 1000 kms@xmath26 . until their thermal speed matches the electron - ion drift speed , at which point the instability saturates . the rapid heating of the electrons perpendicular to the magnetic field results in @xmath1391 and makes it possible for an ion acoustic instability to occur between the preshock electrons and either the reflected ions or preshock ions ( cargill & papdapoulos 1988 ) . the waves generated by the ion acoustic instability can then transfer a substantial fraction of the shock energy ( tens of percent ) into electron thermal energy . this process occurs over a length scale of @xmath140 ( as opposed to @xmath141 for the cosmic ray precursor ) , resulting in a @xmath16@xmath590.2 , independent of shock speed . _ this is in strong disagreement with the equilibrations obtained for the balmer - dominated shocks . _ a number of other electron heating mechanisms , such as the modified two - stream instability and electron - cyclotron drift instability , have been proposed for collisionless shocks based on results from particle in cell ( pic ) simulations ( umeda et al . 2012a , 2012b ) . a unified picture proposed by matsukiyo ( 2010 ) suggests that electrons can be strongly energized at low mach numbers ( @xmath14210 ) via a modified two - stream instability , where the velocities of the reflected / incoming ions and the electrons are lower than the thermal speed of the electrons and the electrons are able to damp out the buneman instability . in this case , obliquely propagating whistler mode waves are excited , having frequencies between the electron cyclotron frequency and the lower hybrid wave frequency . when the electron - ion drift velocity and electron thermal speed become nearly equal , the electron - cyclotron drift instability becomes important ( umeda et al . 2012a ) , exciting waves with frequencies that are multiples of the electron cyclotron frequency . at higher mach numbers the electron thermal speed is lower than that of the ions , and the buneman instability / ion acoustic wave process descrbied earlier is predicted to take over . the amount of electron heating predicted by the buneman instability / ion acoustic wave model scales as @xmath143 ( cargill & papadapoulos 1988 ; matsukiyo 2010 ) , so that for shocks in the 2000 kms@xmath26 - 10,000 kms@xmath26 range , @xmath1442 - 50 kev . this is clearly at odds with @xmath1450.3 kev observed between 400 kms@xmath26 and 2000 kms@xmath26 for balmer - dominated shocks . one explanation for this discrepancy is that growth of the buneman - like and two - stream instabilities described above requires that the reflected ions form a distribution function with a positive gradient at some velocity ( laming 2000 ) . this distribution forms when specularly reflected ions have a mostly monoenergetic , beamlike configuration . at the low mach numbers in the solar wind ( @xmath5710 ) , where the shock structure is laminar , the reflected ions closely resemble a monoenergetic beam . however , at the higher mach numbers , where the shock front is more turbulent and disordered , the reflected ions are likely to have a greater spread in energy and are probably less beamlike ( laming 2000 ) . this would lead to suppression of buneman - like instabilities . however , this line of reasoning is still speculative , and the real explanation for the lack of agreement between the observed @xmath16 and those predicted by models in this section remains to be explored . cosmic - ray driven processes may ultimately provide a better explanation for electron heating at snr shocks than those involving reflected suprathermal ions . the cross shock potential arises from the charge separation produced by the different gyroradii for ions and electrons as they cross the shock transition . it accelerates electrons into the post shock layer , and can be a means of electron heating at subcritical shocks with an approximately laminar structure . at supercritical shocks , the time dependence and non - locality introduced reduces the degree of electron heating . however at sufficiently alfvn mach number ( where the shock transition becomes thin ( on the order of the electron convective gyroradius or inertial length ) , significant electron heating may again occur . in the absence of magnetic field amplification by cosmic rays , this might be expected to happen at snr shocks . however it is much more likely in environments where the plasma beta is low , such as galaxy clusters . it may also occur in cases where a significant population of cosmic rays is unlikely due to the low age of the shock , as in gamma - ray bursts . from the beginning , a detailed study of the fastest collisionless shocks has been hampered by one inherent limitation : they occur in objects which are too remote for in situ study . although some collisionless shocks in our solar system reach alfvnic mach numbers as high as 30 ( such as those around saturn ; achilleos et al . 2006 ; masters et al . 2011 ) , there are no physical phenomena in our solar system energetic enough to produce the type of shocks seen in snrs ( mach numbers @xmath10100 or more if no preshock magnetic field enhancement in the cr is assumed ) . in addition , the range of plasma betas attainable in the solar system is larger than the range of plasma betas attained in the interstellar medium . another fundamental difference between solar wind and snr shocks is the fact that the former are short - lived phenomena confined to small spatial scales ( millions of km ) in a curved ( bow shock ) geometry , whereas the latter are sustained for thousands of years , on spatial scales of parsecs , often well - described by a planar geometry . this results in an irreducible difference between the two types of shocks : particles crossing back and forth between upstream and downstream can remain in contact with snr shocks for long periods of time , allowing accelerated crs to acquire much more energy in snr shocks than in solar wind shocks . this potentially allows the crs to create shock precursors with properties needed to heat electrons and influence @xmath16 . the heating of electrons at the earth s and other planetary bow shocks has been the subject of much theoretical and observational work . typical features of the electron temperature change , @xmath146 , observed at the bow shock as observed by @xcite include ( a ) an approximate relationship between heating and the incident solar wind energy @xmath147 , where @xmath148 is the component of the solar wind s velocity incident upon the shock , and ( b ) a relationship between the change in temperature normalized by the incident energy and the fast magnetosonic mach number , @xmath149 . a similar approximate relationship holds between the normalized electron temperature change and the alfvn mach number @xmath133 , especially for shocks with a low plasma @xmath52 , which is the ratio between thermal and magnetic pressures . recent work by @xcite shows that this relationship with @xmath133 holds well at saturn s bow shock . this is particularly interesting as @xmath133 at saturn is often much larger than at earth . in figure [ fig : combined ] we plot the ratio of electron and ion temperatures downstream of the shock against the magnetosonic , and alfvn mach numbers , as well as the upstream flow velocity relative to the shock . the data in these figures are taken from bow shock crossings of the isee spacecraft , and is a subset of those listed in @xcite , consisting of 61 crossings for which all the necessary data is available . it is well known that quantities other than those displayed here may be more appropriate , specifically the change in electron temperature over the change in total temperature @xmath150 or even @xmath151 are better correlated with inverse mach numbers than @xmath152 @xcite . nevertheless , we use the latter quantity here as it enables a comparison with data from extra - solar system and outer planetary shocks at where less data are available . furthermore , the approximate inverse dependence upon @xmath8 , @xmath133 , and @xmath39 is still quite apparent in this data . it is interesting to note that the relationship with the mach numbers is much more favourable than the dependence upon @xmath39 , indicating that the mach number is the more relevant quantity for organizing the relationship between @xmath16 and shock strength . is plotted versus shock speed ( left ) , alfvnic mach number ( center ) and magnetosonic mach number ( right ) . green symbols show data from crossings of earth s bow shock ( @xcite ) , while the black symbols show data from crossings of saturn s bow shock ( @xcite ) . shock speeds for the saturnian bow shock are based on a solar wind model and an assumed shock speed with respect to the spacecraft of 100 kms@xmath26 , and ion temperatures are based on electron distribution measurements and the application of the rankine - hugoniot conditions ( see masters et al . ( 2011 ) for a full discussion of shock parameter derivations at saturn s bow shock ) . red symbols show data acquired from balmer - dominated snr shocks ( van adelsberg et al . 2008 ) , and assume @xmath132=9 kms@xmath26 , @xmath153=11 kms@xmath26 . , title="fig:",width=480 ] -.5 in many mechanisms may be involved in the heating of electrons in solar system shocks . proposed mechanisms involve acceleration of electrons by a cross - shock potential @xcite , wave turbulence @xcite , microinstabilities @xcite , and electron trajectory scattering @xcite . the existence of a cross - shock potential may be deduced from the generalized ohm s law in which a gradient in electron thermal pressure gives rise to an electric field . examining the energetics of electrons crossing the shock may simplified by working in the de hoffmann - teller frame of reference @xcite , defined as the frame in which the shock is at rest , and in which the magnetic field and plasma flow velocity are ( anti-)parallel . in this case the the electric field is dominated by that generated by the electron pressure gradient , and the work done on electrons crossing the shock is determined by the cross - shock potential . additional mechanisms are required to scatter electrons to pitch angles that are more perpendicular , and to flatten the distribution so that empty regions of phase space are filled . this results in a distribution whose temperature is controlled to a large extent by the de hoffmann - teller frame cross - shock potential and downstream density . in addition to direct measurement of the electric fields within the shock , @xcite , comparison of upstream and downstream electron phase space distributions have shown that these are consistent with electron acceleration by a cross - shock potential in the de hoffmann - teller frame @xcite . figure [ fig : combined ] may indicate that similar mechanism(s ) heat the electrons in solar wind and in snr shocks . _ this is especially appealing when we remember that @xmath133 in snrs may be overestimated due to preshock amplification of magnetic field by cosmic rays . _ in their study of the terrestrial bow shock and interplanetary shocks , schwartz et al ( 1988 ) found that the electron - ion temperature equilibration organizes best by @xmath154 . given the difficulty in determining the mach numbers of snr shocks , the equilibration dependence on shock strength has been characterized via the shock speed instead , and found to obey @xmath155 1/@xmath39@xmath156 ( @xmath39 is much more accurately known than the mach numbers ) . subsection 4.1 outlined a model for snr electron heating , where the cosmic ray diffusion coefficient @xmath157 is assumed independent of @xmath39 . from quasi - linear theory ( blandford and eichler 1987 ) , @xmath158 where @xmath148 is the energy density of turbulence ( @xmath159 ) and @xmath160 is the cosmic ray velocity . for resonant amplification , we evaluate @xmath148 at @xmath161 , where @xmath162 is the gyroradius and @xmath163 the parallel component of the cosmic ray velocity . for relativistic cosmic rays , where @xmath164 , this results in @xmath165 , assumed constant with @xmath39 . however , for nonrelativistic suprathermal particles , @xmath160 will most likely be proportional to the shock velocity @xmath39 , which with the same assumptions leads to @xmath1661/@xmath39 for solar wind shocks ( as opposed to @xmath167 in snrs ) . this argument is admittedly loose , and should not be be viewed as much more than a hypothesis to motivate further work . in our arguments above we have made considerable assumptions about @xmath121 . the most obvious one is that @xmath157 as written above applies to parallel shocks , whereas we are most likely dealing with quasi - perpendicular cases . this may reduce the difference anticipated between solar wind and smr shocks , depending on the turbulence spectrum ( summarized in appendix a of rakowski et al . 2008 ) . the degree of cosmic ray magnetic field amplification at snr shocks required to bring snr data points in figure [ fig : combined ] into alignment with solar wind data points is approximately an order of magnitude or less at the highest velocities considered . given the degree of magnetic field amplification expected from cosmic ray acceleration , this appears to be highly plausible . in the case of saturated nonresonant instability ( bell 2004 , 2005 ) by resonant scattering ( luo & melrose 2009 ) , @xmath168 is expected . in the case of nonresonant saturation , higher values , but strongly dependent on @xmath169 , are predicted . saturation by electron heating ( i.e. the @xmath133 where growth of lower hybrid waves becomes greater than the growth rate of magnetic field ) leads to similar magnetic field enhancement , with @xmath170 ( rakowski et al . such magnetic field amplification is less likely at solar wind shocks . the suprathermal particle densities are lower in solar wind shocks , and the ambient magnetic fields are higher , much closer to where the nonresonant instability would saturate ( if not already beyond it ) . collisionless shocks occur over a vast range of length scales , with those in galaxy clusters being among the largest . while the shock speeds in the galaxy cluster shocks are similar to those in supernova remnants ( up to 4000 kms@xmath26 ; markevitch 2005 ; markevitch & vikhlinin 2007 ; russell et al . 2012 ) , they occur in environments that are substantially different from both the solar wind and the ism . these differences can be encapsulated via the plasma beta , defined as the ratio of the thermal pressure to the magnetic pressure of a plasma ( @xmath171 ) . utilizing solar wind parameters listed by bruno & carbone ( 2005 ) , this ratio ranges from around unity at 1 au under fast solar wind conditions ( wind velocity @xmath10900 kms@xmath26 ) to around 20 for the quiescent wind ( wind velocity @xmath10300 kms@xmath26 ) . the plasma @xmath52 of the ism is close to that of the fast solar wind , @xmath1721 - 4 ( assuming a b=3@xmath36 g , n= 1 @xmath29 and t=10@xmath99 k ) . on the other hand , the electron temperature of the intracluster medium ( icm ) ranges from 10@xmath173k - 10@xmath71k ( 1 - 10 kev ) , with number densities steeply declining from @xmath10 10@xmath174 @xmath175 near the cluster centers to @xmath1010@xmath176 @xmath175 at the outer edges . the corresponding sound speed is close to 1000 kms@xmath26 , nearly two orders of magnitude higher than in the general ism . the magnetic fields measured in galaxy clusters are actually close that of the ism , typically on the order of a microgauss ( carilli & taylor 2002 ) . therefore , @xmath17750 kms@xmath26 in galaxy clusters , so that @xmath1781 and the magnetic field pressure has negligible contribution to the dynamics of shocks in galaxy clusters . this puts galaxy cluster shocks in a different region of parameter space than solar wind and snr shocks . clusters such as 1e0657@xmath5656 ( the bullet cluster ) and a520 show strongly enhanced x - ray emission from collisionless shocks , formed during major mergers when gas from one cluster plunges through gas from the other ( markevitch et al . 2005 ; markevitch & vikhlinin 2007 ; russell et al . shocks moving mostly along the plane of the sky have a favorable viewing geometry and appear as giant curved structures hundreds of kiloparsecs in length . the large clumps of infalling gas drive bow shocks into the cluster gas , which has already been heated to at least 1 kev , and produces thermal bremsstrahlung emission peaking close to that energy . this is another major difference between collisionless shocks in the ism and those in the icm . while the alfvnic and magnetosonic mach numbers of snr shocks are very difficult to determine due to the lack of available observational constraints on magnetohydrodynamic quantities upstream ( such as @xmath14 , @xmath13 and @xmath15 ) , those in galaxy clusters can readily be measured by spectral analysis of x - ray emission upstream . comparison of this emission to that of the enhanced postshock region gives the density contrast between the downstream and upstream gas ( i.e. , @xmath179 ) . this density contrast yields the sonic mach number , @xmath180 , via solving the rankine - hugoniot jump conditions ( russell et al . 2012 ) : @xmath181 where @xmath80 is the ratio of specific heats of the cluster gas . measurement of these jumps from x - ray observations have yielded @xmath1821.5 - 3 for the bullet cluster ( markevitch et al . 2005 ) , abell 520 and abell 2146 ( russell et al . 2012 ) . using these estimated mach numbers , the shock velocity itself can be calculated via @xmath39=@xmath183 , where @xmath153 is the upstream sound speed as inferred from the x - ray spectra . this yields shock speeds ranging between 2500 kms@xmath26 and 4000 kms@xmath26 , similar to the fastest known balmer - dominated shocks . however , fits to the x - ray spectra behind these shocks show prompt electron - ion equilibration at the shock front , consistent with @xmath16=1 ( markevitch et al . 2005 ; markevitch & vikhlinin 2007 ) , despite the extremely high shock speeds involved . this result can only be reconciled with equilibration measurements from the solar wind and snrs if the equilibration depends on the mach number , rather than @xmath39 . furthermore , given the low mach numbers found in the galaxy clusters , it is plausible that the shock transitions in these cases are laminar ( as opposed to turbulent like the snr and fastest solar wind shocks ) , with electron heating occurring efficiently at the shock front via the same type of cross - shock potential as seen in the slowest solar wind and slowest snr ( @xmath39@xmath53 400 kms@xmath26 ) shocks . this is of course a speculation ; further insight into collisionless cluster shocks may be obtained via numerical simulations for the appropriate conditions . there have been exciting advances in the study of electron - ion temperature equilibration in collisionless shocks during the past few years . perhaps most notable has been the growing realization that temperature equilibration and cosmic ray acceleration may be intertwined processes . optical studies of collisionless shocks in partially neutral gas ( balmer - dominated shocks ) have shown that the electron - ion temperature equilibration is a declining function of shock speed , well characterized as @xmath16@xmath184 . this relationship most likely arises due to electron heating ahead of the shock that is nearly independent of shock speed above 400 kms@xmath26 . cosmic ray precursors , with moderately amplified preshock magnetic field and density , are the most logical sites for electron heating in snr collisionless shocks . the transition to fully equilibrated snr shocks at speeds below 400 kms@xmath26 may be due to a less turbulent and more laminar structure at low shock speeds and mach numbers . this allows the electrons to experience a more uniform cross - shock potential , and hence higher energization compared to higher shock speeds and mach numbers , where the shock jump is more turbulent and disordered . the magnetosonic mach numbers of snr shocks may match those in solar wind shocks if there is a approximately an order of magnitude increase in the alfvn speed of the preshock gas in snrs compared to the average ism value . this is possible if there is a moderately amplified preshock magnetic field ( @xmath1010 ) , and is readily provided by compression and heating in a cosmic ray precursor . in solar wind shocks , the precursor is due to suprathermal , non - relativistic ions , resulting in a @xmath16@xmath185 seen in the solar wind . while balmer - dominated shocks have allowed us to elucidate some of the physics of electron - ion temperature equilibration , ion - neutral damping limits most of those observed to cases where the shock structure has not become strongly modified by cosmic ray acceleration . given this limitation , _ electron - ion equilibration studies of fast , collisionless shocks in fully pre - ionized gas would be highly desirable_. such a sample would allow the equilibration to be studied over a range of speeds where shocks are increasingly affected by feedback from the accelerated cosmic rays . in such circumstances it is unclear what would happen to the @xmath16 versus @xmath39 relation . if , as predicted by amato & blasi ( 2009 ) , bell s non - resonant cosmic ray instability takes over from the resonant instability at the highest shock speeds , then additional electron heating may occur in the fastest shocks ( @xmath39@xmath585000 kms@xmath26 ) , resulting in substantial deviation from the @xmath16@xmath184 relation . such deviations may already have been seen in the fastest ( @xmath39@xmath582000 kms@xmath26 ) shocks , where there is evidence that @xmath16 does not settle down to @xmath17 , but rather @xmath100.03 . other deviations may have been detected in snr 0509@xmath5667.5 , where @xmath16 for a 5000 kms@xmath26 shock has been estimated to be @xmath100.2 , substantially higher than predicted by the inverse squared relation . however , the study of such shocks would be challenging . without an h@xmath48 broad component to constrain the range of @xmath39 , shock speeds would have to be determined by other means , such as proper motion studies . that would require x - ray and or uv imagery of snrs with well - constrained distances ( such as those in the lmc or smc ) , over multiple epochs . it would also require spectroscopy of the forward shocks in these snrs , in order to constrain both the electron temperature ( via x - ray continuum fits and and uv emission line ratios ) and the ion temperature ( e.g. , via he ii , c iv , n v and o vi resonance lines ) . an important test of our ideas concerning electron heating by cosmic ray generated waves in a shock precursor would be to measure electron temperatures at snr shocks exhibiting strong cosmic ray modification and substantial magnetic field amplification ( @xmath186100 ) . several snrs show x - ray filaments produced by synchrotron radiation from cosmic ray electrons , and are generally distinct from those shocks with strong balmer emission . the absence of neutral material ahead of these shock means that optical and uv emission is weak , and electron temperatures will have to be measured from x - ray spectra . difficulties arise in distinguishing thermal electron bremsstrahlung from cosmic ray electron synchrotron emission , requiring data of high signal to noise . further complications in some snrs ( e.g. cas a ) stem from scattering of bright x - ray emission from the ejecta such that it coincides spatially with emission from the forward shock . such scattering may either be local , due to snr dust , or instrumental , due to telescope imperfections . this leaves sn 1006 as the best likely target for such an observation , since due to its evolutionary state , only the outer layers of ejecta have encountered the reverse shock . the development of missions like solar orbiter and solar probe plus will allow _ in situ _ measurements of shocks in the solar wind , most likely associated with coronal mass ejections , much closer to the sun . these will probe a different parameter regime , where the magnetic field pressure dominates over the gas pressure ( low @xmath52 , similar to ism shocks ) . as such , measurements here might yield insights into the properties of similar plasma in the precursors of snr shocks where the magnetic field has been amplified by cosmic rays . g. acknowledges support by hst grant hst - go-11184.07-a to towson university . jml acknowledges support by grant nnh10a009i from the nasa astrophysics data analysis program , and by basic research funds of the office of naval research . k. a. eriksen , j. p. hughes , c. badenes , r. fesen , p. ghavamian , d. moffett , p. p. plucinsky , c. e. rakowski , e. m. reynoso , and p. slane . . _ astrophysical journal _ , 728:0 l28 , february 2011 . doi : 10.1088/2041 - 8205/728/2/l28 . k. france , r. mccray , s. v. penton , r. p. kirshner , p. challis , j. m. laming , p. bouchet , r. chevalier , p. m. garnavich , c. fransson , k. heng , j. larsson , s. lawrence , p. lundqvist , n. panagia , c. s. j. pun , n. smith , j. sollerman , g. sonneborn , b. sugerman , and j. c. wheeler . _ astrophysical journal _ , 743:0 186 , december 2011 . doi : 10.1088/0004 - 637x/743/2/186 . e. a. helder , j. vink , c. gh . bassa , a. bamba , j. a. m. bleeker , s. funk , p. ghavamian , k. j. van der heyden , f. verbunt , and r. yamazaki . . _ science _ , 325:0 719 , august 2009 . doi : 10.1126/science.1173383 . a. masters , s. j. schwartz , e. m. henley , m. f. thomsen , b. zieger , a. j. coates , n. achilleos , j. mitchell , k. c. hansen , and m. k. dougherty . . _ journal of geophysical research ( space physics ) _ , 116:0 a10107 , october 2011 . doi : 10.1029/2011ja016941 . h. r. russell , b. r. mcnamara , j. s. sanders , a. c. fabian , p. e. j. nulsen , r. e. a. canning , s. a. baum , m. donahue , a. edge , l. j. king and c. p. odea . . _ monthly notices of the royal astronomical society _ , 423:0 236255 , june 2012 . doi : 10.1111/j.1365 - 2966.2012.20808.x . t. tanaka , y. uchiyama , f. a. aharonian , t. takahashi , a. bamba , j. s. hiraka , j. kataoka , t. kishishita , m. kokubun , k. mori , k. nakazawa , r. petre , h. tajima , and s. watanabe . . _ astrophysical journal _ , 685:0 9881004 , october 2008 . doi : 10.1086/591020 . j. s. warren , j. p. hughes , c. badenes , p. ghavamian , c. f. mckee , d. moffett , p. plucinsky , c. e. rakowski , e. reynoso , and p. slane . . _ astrophysical journal _ , 634:0 376389 , november 2005 . doi : 10.1086/496941 . b. j. williams , w. p. blair , j. m. blondin , k. j. borkowski , p. ghavamian , k. s. long , j. c. raymond , s. p. reynolds , j. rho , and p. f. winkler . . _ astrophysical journal _ , 741:0 96 , november 2011 . doi : 10.1088/0004 - 637x/741/2/96 . c. s. wu , d. winske , m. tanaka , k. papadopoulos , k. akimoto , c. c. goodrich , y. m. zhou , s. t. tsai , p. rodriguez , and c. s. lin . . _ space science reviews _ , 37:0 63109 , january 1984 . doi : 10.1007/bf00213958 .

with the discovery of the standard model ( sm ) higgs - like scalar at the lhc @xcite , the sm particle content seems complete . in particular , the mass and couplings of the neutral higgs boson seem to disfavor an additional chiral generation of quarks and leptons @xcite . however , additional vector - like fermions , in which an sm generation is paired with another one of opposite chirality , and with identical couplings , are less constrained , as there is no quadratic contribution to their masses . these states appear as natural extensions of the sm particle content , in theories with warped or universal extra dimensions , as kaluza - klein excitations of the bulk field @xcite , in non - minimal supersymmetric extensions of the sm @xcite , in composite higgs models @xcite , in little higgs model @xcite and in gauged flavor groups @xcite . vector fermions have identical left- and right - handed couplings and can have masses which are not related to their couplings to the higgs bosons @xcite . depending on the dominant decay mode , the limits on new vector - like fermions range from @xmath2 gev @xcite , rendering them observable at the lhc . vector quarks can modify both the production and decays of the higgs boson at the lhc , while vector leptons do not carry @xmath3 charge and can only modify the decay patterns of the higgs . study of the latter would be a sensitive probe for new physics . the lepton states contribute to self - energy diagrams for electroweak gauge boson masses and precision observables , and consistent limits on their masses and mixings have been obtained @xcite . vector leptons have been studied in the context of the sm @xcite , but less so for models beyond the sm , where they can also significantly alter the phenomenology of the model . in the sm , introducing heavy fermions provides a contribution of the same magnitude and sign to the one of the top quark and interferes destructively with the dominant @xmath4-contribution , reducing the @xmath5 rate with respect to its sm value . recent studies indicate that cancellations between scalar and fermionic contributions allow a wide range of yukawa and mass mixings among vector states @xcite . an investigation of vector leptons in the two higgs doublet model @xcite showed that the presence of additional higgs bosons alleviates electroweak precision constraints . introducing vector leptons into supersymmetry @xcite can improve vacuum stability and enhance the di - photon rate by as much as 50% , while keeping new particle masses above 100 gev and preserving vacuum stability conditions . in the present work , we investigate vector leptons in the context of the higgs triplet model ( htm ) . we do not deal with lhc phenomenology ( pair production and decay ) of the extra leptons , which has been discussed extensively in the literature @xcite , choosing instead to focus on signature features of this model . we have previously shown that in the higgs triplet model , enhancement of the @xmath6 rate is possible only for the case where the doublet and triplet neutral higgs fields mix considerably @xcite . we extend our analysis to include additional vector - like leptons in the model and investigate how these affect the higgs di - photon decay rate , with or without significant mixing in the neutral higgs sector . originally , both cms and atlas experiments at lhc observed an enhancement of the higgs di - photon rate , while the di - boson rates ( @xmath7 ) have been roughly consistent with sm expectations . at present cms observes @xmath8 times the sm rate , while atlas observes @xmath9 times the sm rate @xcite . given these numbers , it is possible that either the sm value will be proven correct , or a modest enhancement will persist . a further test of the sm is the correlation of the decay @xmath1 with one for @xmath10 . we also include the prediction for the vector lepton effect on branching ratio of @xmath1 and comment on the relationship with the di - photon decay . we have an additional reason to investigate the effects of vector leptons in the higgs triplet model . the model includes doubly - charged higgs bosons , predicted by most models to be light . being pair - produced , the doubly - charged higgs bosons are assumed to decay into a pair of leptons with the same electric charge , through majorana - type interactions @xcite . assuming a branching fraction of 100% decays into leptons , _ i.e. _ , neglecting possible decays into @xmath4-boson pairs , the doubly - charged higgs mass has been constrained to be larger than about 450 gev , or more , depending on the decay channel . however , if the vector leptons are light enough , which they can be , the doubly - charged higgs bosons can decay into them and thus evade the present collider bounds on their masses . we investigate this possibility in the second part of this work . our work is organized as follows . we introduce the higgs triplet model with vector leptons in section [ sec : model ] . in section [ sec : neutral ] we analyze the effect of the vector leptons on the decays of the neutral higgs bosons , and discuss the constraints on the parameter space coming from requiring agreement with present lhc data . we include both loop - dominated decays , @xmath0 in [ subsec : gamgam ] , and @xmath11 in [ subsec : zgam ] . in section [ sec : doublycharged ] we analyze the effect of the vector leptons on the production and decay mechanisms of the doubly - charged higgs at the lhc . we summarize our findings and conclude in section [ sec : conclusion ] . the electroweak gauge symmetry is broken by the vevs of the neutral components of the doublet and triplet higgs fields , @xmath41 where @xmath42 and @xmath43 are the doublet higgs field and the triplet higgs field , with @xmath44 ( 246 gev)@xmath45 . higgs masses and coupling constants in the presence of non - trivial mixing in the neutral sector have been obtained previously @xcite . one can invoke new symmetries to restrict the interaction of the vector leptons with each other , or with the ordinary leptons , or disallow the presence of bare mass terms in the lagrangian . for instance , 1 . if an additional @xmath46 symmetry under which the primed , double primed and the ordinary leptons have different charges , this would forbid explicit masses @xmath47 , @xmath48 , @xmath49 and @xmath50 in the lagrangian . vector leptons would get masses only through couplings to the higgs doublet fields @xcite . 2 . if one imposes a symmetry under which all the new @xmath51 singlet fields are odd , while the new @xmath51 doublets are even , this forces all yukawa couplings involving new leptons to vanish , @xmath52 , and the masses arise only from explicit terms in the lagrangian @xcite . 3 . finally one can impose a new parity symmetry which disallows mixing between the ordinary leptons and the new vector lepton fields , under which all the new fields are odd , while the ordinary leptons are even @xcite , i.e. , such that @xmath53 ; alternatively one might choose these couplings to be very small . in this analysis the focus will be on higgs decays . we investigate the model subjected to symmetry conditions 1 and/or 2 ; and we neglect mixing between the ordinary and the new vector leptons . when allowed , stringent constraints exist on the masses and couplings with ordinary leptons . new vector leptons are ruled out when they mediate flavor - changing neutral currents processes , generate sm neutrino masses or contribute to neutrinoless double beta decay . recent studies of models which allow mixing between the ordinary leptons and the new ones exist @xcite , but there restrictions from lepton - flavor violating decays force the new leptons to be very heavy @xmath54 tev , or reduces the branching ratio for @xmath55 and of @xmath0 decay to 30 - 40% of the sm prediction , neither desirable features for our purpose here . in the higgs triplet model , distinguishing signals would be provided by lighter vector leptons . since imposing no mixing between ordinary and new leptons allows new lepton masses to be as light as @xmath56 gev , perhaps without a reduction in the higgs di - photon branching ratio , we investigate this scenario here . in the charged sector , the @xmath57 mass matrix @xmath58 is defined as @xcite @xmath59 with @xmath60 and @xmath61 , with @xmath62 the vev of the neutral component of the higgs doublet . the mass matrix can be diagonalized as follows : @xmath63 the mass eigenvalues are @xmath64 , \label{eq : eigenvalues}\ ] ] with @xmath65 by convention , @xmath66 . for simplicity we assume lepton yukawa couplings @xmath67 and @xmath68 are real so that the transformations that diagonalize the mass matrix are real orthogonal matrices : @xmath69 the angles @xmath70 are given by @xmath71 the eigenstates of the vector lepton mixing matrix enter in the evaluation of @xmath0 and @xmath1 in the next section . the presence of the vector leptons affects the loop - dominated decays of the neutral higgs , @xmath0 and @xmath1 , and possible relationships between them . in the higgs triplet model , singly- and doubly - charged bosons also enter in the loops , creating a different dynamic than in the sm . we analyze these decays in turn , and look for possible correlations between them . recently , the triplet higgs model has received renewed interest recently because of attempts to reconcile the excess of events in @xmath0 observed at the lhc over those predicted by the sm . such an enhancement hints at the presence of additional particles , singlets under @xmath3 , but charged under @xmath72 which affect only the loop - dominated decay branching ratio , while leaving the production cross section and tree level decays largely unchanged . vector leptons are prime candidates for such particles , so we study their contribution to the higgs decay branching . the decay width @xmath0 is @xmath73_{htm } & = & \frac{g_f\alpha^2 m_{h}^3 } { 128\sqrt{2}\pi^3 } \bigg| \sum_{f } n^f_c q_f^2 g_{h ff } a_{1/2 } ( \tau^h_f ) + g_{h ww } a_1 ( \tau^h_w ) + \tilde{g}_{h h^\pm\,h^\mp } a_0(\tau^h_{h^{\pm } } ) \nonumber \\ & + & 4 \tilde{g}_{h h^{\pm\pm}h^{\mp\mp } } a_0(\tau^h_{h^{\pm\pm } } ) + \frac{\mu_{e_1}g_{hff}}{m_{e_1 } } a_{1/2}(\tau^h_{e_1 } ) + \frac{\mu_{e_2}g_{hff}}{m_{e_2 } } a_{1/2}(\tau^h_{e_2 } ) \bigg|^2 \ , , \label{partial_width_htm}\end{aligned}\ ] ] with @xmath74 given in eq . ( [ eq : eigenvalues ] ) , and where the loop functions for spin @xmath75 , spin @xmath76 and spin @xmath77 are given by : @xmath78\ , \tau^{-2 } \ , , \label{eq : ascalar}\\ a_{1/2}(\tau)&=&-\tau^{-1}\left[1+\left(1-\tau^{-1}\right)f\left(\tau^{-1}\right)\right ] , \label{eq : afermion}\\ a_1(\tau)&= & 1+\frac32 \tau^{-1}+4 \tau^{-1}\left(1-\frac12 \tau^{-1}\right)f\left(\tau^{-1}\right ) , \label{eq : avector } \end{aligned}\ ] ] and the function @xmath79 is given by : @xmath80 ^ 2 \hspace{0.5 cm } & \tau>1 \ , , \end{array } \right . \label{eq : ftau } \end{aligned}\ ] ] with @xmath81 , and @xmath82 the mass of the particle running in the loop @xcite . in eq . ( [ eq : thm - h2gaga ] ) the first contribution is from the top quark , the second from the @xmath4 boson , the third from the singly - charged higgs boson , the fourth from the doubly - charged higgs boson , and the last two from the vector leptons . we used the following expressions for the couplings of the higgs bosons with charged vector leptons : @xmath83 the couplings of @xmath84 to the vector bosons and fermions are as follows : @xmath85 with @xmath86 , and the scalar trilinear couplings are parametrized as follows @xmath87 \ , , \nonumber \\ & & \tilde{g}_{h h^+h^-}= \frac{m_w}{2 g m_{h^{\pm}}^2 } \bigg\{\left[4v_\delta(\lambda_2 + \lambda_3 ) \cos^2{\beta_{\pm}}+2v_\delta\lambda_4\sin^2{\beta_{\pm}}- \sqrt{2}\lambda_5v_\phi \cos{\beta_{\pm}}\sin{\beta_{\pm}}\right]\sin\alpha \nonumber \\ & & + \left[4\lambda_1\,v_\phi \sin{\beta_{\pm}}^2+{(2\lambda_{4}+\lambda_{5 } ) } v_\phi \cos^2{\beta_{\pm}}+ ( 4\mu-\sqrt{2}\lambda_5v_\delta)\cos{\beta_{\pm}}\sin{\beta_{\pm}}\right]\cos\alpha\bigg\ } \ , . \label{eq : redgcallittlehhp}\end{aligned}\ ] ] since the new leptons do not affect the higgs production channels , the effect on the di - photon search channel at the lhc is expressed by the ratio @xmath88_{htm}}{[\sigma ( gg \to \phi ) \times br(\phi \to \gamma \gamma ) ] _ { sm } } \nonumber \\ & = & \frac{[\sigma ( gg \to h ) \times \gamma(h \to \gamma \gamma)]_{htm}}{[\sigma ( gg \to\phi ) \times \gamma(\phi \to \gamma \gamma)]_{sm } } \times \frac{[\gamma ( \phi)]_{sm}}{[\gamma(h)]_{htm } } , \end{aligned}\ ] ] where @xmath42 is the sm neutral higgs boson and where the ratio of the cross sections by gluon fusion is @xmath89 here @xmath90 is the mixing angle in the cp - even neutral sector : @xmath91 with @xmath92 in @xcite we investigated the higgs boson decay branching ratio into @xmath93 with respect to the sm considering the lightest higgs boson is the @xmath94 signal excess observed at the lep at 98 gev , while the heavier higgs boson is the boson observed at the lhc at 125 gev , in a higgs triplet model without vector leptons , and found that this is the only scenario which allows for enhancement of the @xmath0 branching fraction . we thus set the values 125 gev and 98 gev for the @xmath84 and @xmath95 masses respectively , and adjust the parameters @xmath96 accordingly . vector lepton masses and mixing parameters depend on @xmath47 and @xmath48 , the explicit mass parameters in the lagrangian ; and @xmath97 , the vector leptons yukawa parameters . in the limit of vanishing dirac mass terms @xmath47 and @xmath48 , the pre - factors @xmath98 in ( 11 ) go to one . it then follows that there is destructive interference between the dominant @xmath4- boson contribution and the charged leptons loops @xcite . in this limit , despite possible enhancement from singly- and doubly - charged higgs bosons in the loop , we find a large suppression of the di - photon rate . we present the plots for the relative signal strength of @xmath99 , defined in eq . ( [ eq : rgamgam ] ) as a function of @xmath100 ( or equivalently @xmath101 ) , for various values of doubly charged higgs bosons mass , in the left - side panel of fig . [ fig : hgg ] , for @xmath102 . clearly , for this case ( no mixing ) the decay of the @xmath84 is suppressed significantly with respect to the value in sm over the whole region of the parameter space in @xmath103 . -0.3 in @xmath104{fig1leftpanel.pdf } & \hspace*{-1.5 cm } \includegraphics[width=3.1in , height=3.1in]{fig1middlepanel.pdf } & \hspace*{-2.1 cm } \includegraphics[width=3.1in , height=3.1in]{fig1rightpanel.pdf } \end{array}$ ] allowing mixing in the neutral higgs sector changes the relative contributions of the charged higgs to the di - photon decay . we show decay rates for @xmath0 as a function of @xmath105 for different values of doubly - charged higgs boson mass considering @xmath106 gev ( and 200 gev ) in fig . [ fig : hgg ] middle ( and right ) panels respectively . considerations for relative branching ratios are affected by the fact that the total width of higgs boson in the htm for @xmath107 is not the same as in the sm . the relative widths factor is @xmath108_{htm}}{[\gamma(\phi)]_{sm}}=\frac{[\gamma ( h \to \sum\limits_f f { \bar f})+ \gamma ( h \to ww^ * ) + \gamma ( h \to zz^*)+ \gamma ( h \to \nu \nu)]_{htm}}{[\gamma ( \phi \to \sum\limits_f f { \bar f})+ \gamma ( \phi \to ww^ * ) + \gamma ( \phi \to zz^*)]_{sm } } . \end{aligned}\ ] ] the plots in fig . [ fig : hgg ] correspond to symmetry condition 1 in section [ sec : model ] , that is , @xmath109 . however , if mixing with sm leptons is forbidden , but the vector leptons are still allowed to mix with each other , the pre - factors @xmath110 in eq . ( [ eq : thm - h2gaga ] ) are not one , and can modify the higgs di - photon decay . in the next plots we investigate the effect of non - zero mass parameters @xmath47 and @xmath48 , for fixed values of the yukawa couplings . in fig . [ fig : hemlcont1 ] we present the contour plots of constant @xmath111 for @xmath112 in the plane of the explicit mass terms @xmath113 and @xmath48 , for various values of doubly - charged higgs bosons mass and @xmath114 . the contours are labeled by the value of @xmath111 . the vector lepton masses are restricted to values for which @xcite @xmath115 , where @xmath116 are given in eq . ( [ eq : eigenvalues ] ) . the plots indicate that it is difficult to obtain any significant enhancement of the ratio @xmath111 for @xmath117 , and this does not depend on the chosen values for the doubly - charged higgs mass ; while for @xmath107 , enhancements are possible for various values of @xmath118 . @xmath119{fig2upperleftpanel.pdf } & \hspace*{-0.4 cm } \includegraphics[width=2.3in , height=2.2in]{fig2uppermiddlepanel.pdf } & \hspace*{-0.4 cm } \includegraphics[width=2.3 in , height=2.2in]{fig2upperrightpanel.pdf}\\ \hspace*{-0.9 cm } \includegraphics[width=2.3in , height=2.2in]{fig2lowerleftpanel.pdf } & \hspace*{-0.4 cm } \includegraphics[width=2.3in , height=2.2in]{fig2lowermiddlepanel.pdf } & \hspace*{-0.4 cm } \includegraphics[width=2.3in , height=2.2in]{fig2lowerrightpanel.pdf } \end{array}$ ] in fig . [ fig : hemlcont2 ] we investigate the dependence of @xmath111 on the yukawa couplings and vector lepton masses . we show contour plots for fixed @xmath99 in a @xmath120 plane , with @xmath101 and @xmath121 , for various values of @xmath105 and doubly - charged higgs boson masses . enhancements are possible here for all values of @xmath105 , but while for @xmath117 the decay @xmath0 is enhanced for large vector lepton masses and yukawa couplings , for @xmath107 we observe enhancements for light vector lepton masses and small yukawa couplings . @xmath119{fig3upperleftpanel.pdf } & \hspace*{-0.4 cm } \includegraphics[width=2.3in , height=2.2in]{fig3uppermiddlepanel.pdf } & \hspace*{-0.4 cm } \includegraphics[width=2.3 in , height=2.2in]{fig3upperrightpanel.pdf}\\ \hspace*{-0.9 cm } \includegraphics[width=2.3in , height=2.2in]{fig3lowerleftpanel.pdf } & \hspace*{-0.4 cm } \includegraphics[width=2.3in , height=2.2in]{fig3lowermiddlepanel.pdf } & \hspace*{-0.4 cm } \includegraphics[width=2.3in , height=2.2in]{fig3lowerrightpanel.pdf } \end{array}$ ] if we wish to study the light vector leptons parameter space where @xmath0 is enhanced , @xmath107 is preferred . the enhancement is affected by mixing in the vector lepton sector , the various values for doubly - charged higgs bosons mass and values of @xmath105 . as the plots cover only a limited range of the parameter space , in the tables below we give the ranges for the values of the ratio @xmath122 for the various scenarios . in table [ tab : y1/6_me_ml ] , we fix the value of the yukawa coupling to @xmath123 , allow the vector lepton masses to vary in the ( 100 - 500 ) gev range , and show the values for @xmath99 for different @xmath105 and doubly - charged higgs masses . we note that the relative branching ratios are very sensitive to both doubly charged higgs mass values and values of @xmath105 . enhancements in the branching ratio of @xmath0 are possible for light @xmath124 gev , and are much more pronounced at large @xmath105 . note that for @xmath117 , the result is independent of @xmath118 , in agreement with the results obtained in @xcite . the reason is the following . in eq . ( [ eq : redgcallittlehhp ] ) , for @xmath117 , the coupling between neutral and doubly - charged higgs @xmath125 = \frac{m_w } { g m_{h^{\pm \pm}}^2}\bigg [ 2\frac { m^2_{h^{\pm \pm}}}{v_\phi^2 } v_\phi \bigg]=2 \frac{m_w } { g v_\phi } , \end{aligned}\ ] ] where we used the expression for @xmath126 from @xcite , is independent of @xmath118 . in table [ tab : y1/6_h_e ] we allow , in addition to mass variations , variations in the yukawa coupling @xmath127 . this means allowing both explicit ( dirac ) masses and additional contributions by electroweak symmetry breaking , @xmath128 . the dependence on the yukawa coupling @xmath67 is much weaker than on @xmath105 or on @xmath118 . however , one can see from the tables that modest enhancements of the ratio @xmath99 are possible for @xmath117 for large vector leptons yukawa couplings , unlike in the case of the triplet model without vector leptons . this would then be a clear distinguishing feature : enhancements of the decay @xmath0 in the absence of mixing in the neutral sector . the absence of mixing would manifest itself in observing tree - level decays ( @xmath129 and @xmath130 ) identical to those in the sm . there seems to be a minimum value of @xmath99 for @xmath131 , where the contribution from the doubly - charged higgs bosons is important for small doubly - charged masses and counters the contribution from the vector leptons . this is a suppression of the branching ratio for @xmath0 due to the fact that the vector lepton contribution interacts destructively with the dominant @xmath132 contribution . as a general feature , @xmath99 increases when we lower the doubly charged higgs mass and increase @xmath105 . this rules out part of the parameter space . for instance , for @xmath133 gev , the mixing can not be larger than @xmath134 , and for @xmath135 gev , mixings larger than @xmath136 are ruled out . if the value of @xmath137 is increased to @xmath138 gev , only modest enhancements are possible , and only for @xmath117 , for vector lepton explicit masses in the @xmath139 gev range and @xmath123 . increasing the vector leptons yukawa coupling increases the overall ratio @xmath99 . in most models , the @xmath140 and @xmath141 partial decay widths are correlated or anti - correlated , though usually the enhancement / suppression in the @xmath142 channel is much smaller compared to that in the @xmath143 channel . however , as in models with new loop - contributions to @xmath144 , sensitivity to both is expected , we study the correlation between the two here , in the presence of vector leptons . investigation of the branching ratio of @xmath1 is also further justified by the recent results from cms and atlas @xcite , which indicate branching fractions consistent with the sm expectation at 1@xmath145 in the higgs boson @xmath84 mass region at 95% c.l . . the decay width for @xmath1 is given by @xcite : @xmath146_{htm } & = & \frac{\alpha g_f^2 m_w^2 m_{h}^3 } { 64\pi^4 } \left ( 1-\frac{m_z^2}{m_h^2 } \right ) ^3 \bigg|\frac{1}{c_w } \sum_{f } 2 n^f_c q_f ( i_3^f-2q_f s_w^2 ) g_{h ff } a^h_{1/2 } ( \tau^h_f , \tau^z_f ) \nonumber \\ & + & \frac{(i_3^e-2q_e s_w^2)(2q_e)}{c_w } \bigg[\frac{\mu_{e_1}g_{hff}}{m_{e_1 } } a_{1/2}(\tau^h_{e_1 } , \tau^z_{e_1 } ) + \frac{\mu_{e_2}g_{hff}}{m_{e_2 } } a^h_{1/2}(\tau^h_{e_2 } , \tau^z_{e_2 } ) \bigg ] \nonumber \\ & + & c_w g_{h ww } a^h_1 ( \tau^h_w , \tau^z_w)- 2s_w \tilde{g}_{h h^\pm\,h^\mp}g_{zh^\pm h^\mp } a^h_0(\tau^h_{h^{\pm } } , \tau^z_{h^\pm } ) \nonumber \\ & - & 4s_w \tilde{g}_{h h^{\pm\pm}h^{\mp\mp } } { g}_{z h^{\pm\pm}h^{\mp\mp } } a^h_0(\tau^h_{h^{\pm\pm } } , \tau^z_{h^{\pm \pm } } ) \bigg|^2 \ , , \end{aligned}\ ] ] where @xmath147 @xmath148 ( with @xmath149 ) , and the loop - factors are given by @xmath150i_1(\tau^h,\tau^z).\nonumber\end{aligned}\ ] ] the functions @xmath151 and @xmath152 are given by @xmath153\nonumber \\ & + & \frac{\tau^{h\,2}\tau^z}{\left(\tau^h-\tau^z\right)^2 } \left[g\left(\tau^{h\,-1}\right)-g\left(\tau^{z\,-1}\right)\right],\nonumber\\ i_2(\tau^h,\tau^z ) & = & -\frac{\tau^h\tau^z}{2(\tau^h-\tau^z)}\left[f\left(\tau^{h\,-1}\right)-f\left(\tau^{z\,-1}\right)\right],\end{aligned}\ ] ] where the function @xmath79 is defined in eq . ( [ eq : ftau ] ) , and the function @xmath154 is defined as @xmath155 , \hspace{0.5 cm } & ( \tau\geq 1 ) \ , . \end{array } \right . \label{eq : gtau } \end{aligned}\ ] ] in eq . ( [ eq : hzg ] ) we list , in order , the ordinary leptons , vector leptons , @xmath4 boson , singly - charged higgs , and doubly - charged higgs contribution . the scalar couplings @xmath156 and @xmath157 are given in eq . ( [ littleh1tt ] ) , and the scalar trilinear couplings @xmath158 and @xmath159 are given in eq . ( [ eq : redgcallittlehhp ] ) . the remaining couplings in eq . ( [ eq : hzg ] ) are given by @xmath160 we proceed to perform a similar analysis as in sec . [ subsec : gamgam ] . we show first the variation of the branching ratio @xmath1 with the mass @xmath161 , for various values of the doubly - charged higgs masses , for the case of no mixing in the neutral sector , @xmath162 ( shown in fig . [ fig : hgzme ] , left - hand panel ) , and as a function of the mixing angle @xmath105 for @xmath163 gev , and @xmath164 gev , in the middle and right panels of fig . [ fig : hgzme ] , respectively . we have chosen the same parameter values as in fig . [ fig : hgg ] , for comparison . it is clear that the branching ratio into @xmath165 is fairly independent of both @xmath166 and @xmath118 , and always just below the sm expectations . note that the severe suppression seen in @xmath5 for @xmath117 ( fig . [ fig : hgg ] , left side panel ) does not occur here , and the results of the left side of fig . [ fig : hgzme ] are consistent with the data at lhc . but the variation with the mixing angle @xmath90 is pronounced , and the branching ratio can reach almost twice its sm value for @xmath167 . however , correlated with our predictions from sec . [ subsec : gamgam ] and lhc measurements for @xmath99 , the parameter space corresponding to an enhanced @xmath1 , for both @xmath168 gev and 200 gev , for doubly - charged higgs mass @xmath169 gev is ruled out . for all other values considered , the value for @xmath170 is close to , or below the sm expectations . this is general prediction of the model . -0.3 in @xmath171{fig4leftpanel.pdf } & \hspace*{-1.7 cm } \includegraphics[width=3.1in , height=3.0in]{fig4middlepanel.pdf } & \hspace*{-1.9 cm } \includegraphics[width=3.1 in , height=3.0in]{fig4rightpanel.pdf } \end{array}$ ] for a large range of parameter space , the decay @xmath1 can be suppressed significantly with respect to the sm . we plot decay rates for @xmath172 as a function of @xmath105 for different values of doubly charged higgs boson mass , considering @xmath173 gev and 200 gev in fig . [ fig : hgzme ] , middle and right panel respectively . again , considerations for relative branching ratios are affected by the fact that the total width of higgs boson in the htm is not the same as in the sm . the widths are the same as those in the sm for @xmath84 for @xmath117 , while for @xmath107 we take into account the relative widths factors , eq.([eq : width ] ) . in tables [ tab : yz1/6_me_ml ] and [ tab : yz1/6_h_e ] we present the explicit ranges of @xmath170 for varying @xmath174 and for a range of @xmath67 parameters . we choose a fixed value for @xmath123 in table [ tab : yz1/6_me_ml ] , as the preferred choice from other analyses @xcite , and for comparison with table [ tab : y1/6_me_ml ] . comparison of tables [ tab : y1/6_me_ml ] and [ tab : yz1/6_me_ml ] shows that the decay @xmath1 is far more stable against variations in masses and values for @xmath105 than @xmath0 , making it a less sensitive indicator for the presence of vector lepton states . in table [ tab : yz1/6_h_e ] we also allow variations in the yukawa coupling @xmath127 . as before this amounts to allowing both explicit and contributions by electroweak symmetry breaking , @xmath128 , to vector lepton masses . comparison of the tables [ tab : yz1/6_me_ml ] and [ tab : yz1/6_h_e ] indicates that the results are not very sensitive to variations in the yukawa coupling @xmath67 , or the vector lepton mass parameters @xmath174 . however , the relative branching ratios are very sensitive to values of @xmath105 . while the branching ratio into @xmath165 is almost always suppressed with respect to its sm value , there is a small region of the parameter space , light @xmath175 and @xmath176 where enhancement is possible ; but as discussed before , this region is ruled out by constraints from @xmath0 measurement , table [ tab : y1/6_h_e ] . note that for @xmath117 the branching ratio is as before , independent of the mass of @xmath175 and about the same as in the sm . the discovery of the doubly - charged higgs bosons would be one of the most striking signals of physics beyond sm , and a clear signature for the higgs triplet model . the decay modes of @xmath175 depend on the value of the vev of the neutral triplet higgs component , @xmath177 . when @xmath178 mev , the dominant decay mode of @xmath175 is into lepton pairs . if @xmath179 mev , the main decay modes of @xmath175 are into @xmath180 pairs are kinematically allowed . ] , and into @xmath181 , if kinematically allowed . searches for @xmath175 were performed at large electron positron collider ( lep ) @xcite , the hadron electron ring accelerator ( hera ) @xcite and the tevatron @xcite . the most up - to - date bounds have been more recently derived by atlas and cms collaborations at the lhc . assuming a drell - yan - like pair production , these collaborations have looked for long - lived doubly - charged states , and after analyzing 5 fb@xmath182 of lhc collisions at a center - of - mass energy @xmath183 tev and 18.8 fb@xmath182 of collisions at @xmath184 = 8 tev , they constrained the masses to lie above 685 gev @xcite . the assumption is that the doubly - charged higgs bosons decay 100% into a pair of leptons with the same electric charge through majorana - type interactions @xcite , thus neglecting possible decays into @xmath4-boson pairs , shown to alter the pattern of @xmath175 branching fractions @xcite . in this work , we allow both decays into @xmath185 bosons , and also include the effects of decays into vector leptons , which , if light enough , would modify the decays of the doubly - charged higgs bosons further . we take @xmath186 gev throughout our considerations is small enough to satisfy electroweak precision conditions , but large enough to allow decay into gauge boson and charged higgs @xcite . ] . the main production mode for @xmath175 is the pair production @xmath187 and the associated production @xmath188 . the production cross sections for both the vector boson fusion @xmath189 , and for weak boson associated production @xmath190 , are proportional to @xmath191 and much less significant for @xmath192 . at hadron colliders the partonic cross section for the leading order ( lo ) production cross section for doubly charged higgs boson pair is @xmath193,\ ] ] where we defined @xmath194 with @xmath195 the third component of the isospin for particle @xmath196 , @xmath197 the square of the partonic center of mass energy , @xmath198 , and @xmath90 the qed coupling constant evaluated at scale @xmath199 . the hadronic cross section is obtained by convolution with the partonic density functions of the proton @xmath200 where @xmath201 is the parton luminosity and @xmath202 ( @xmath203 is the total energy squared at the lhc ) . the cross section for pair - production , including nlo corrections , has been evaluated in @xcite . depending on mass parameters in the model , the doubly - charged higgs boson can decay into lepton pairs , including vector leptons , @xmath132 pairs , or @xmath204 states . in the higgs triplet model , the decay rate for @xmath175 into leptons is @xmath205 ^ 2 , \end{aligned}\ ] ] where @xmath82 is the mass of the @xmath206th lepton ( @xmath207 or @xmath208 ) and @xmath209 for @xmath210 , ( @xmath211 ) . similarly the decay rate of @xmath175 into fourth generation vector leptons is , if kinematically allowed @xmath212 \frac{m_{h^{\pm \pm}}}{4 \pi}\left ( 1-\frac{m_{e_i}^2}{m_{h^{\pm \pm}}^2}-\frac{m_{e_j}^2}{m_{h^{\pm \pm}}^2 } \right ) \left [ \lambda \left(\frac{m_{e_i}^2}{m_{h^{\pm \pm}}^2 } , \frac{m_{e_j}^2}{m_{h^{\pm \pm}}^2 } \right ) \right]^2 , \nonumber \\\end{aligned}\ ] ] with @xmath213 the mass eigenvalue from eq . ( [ eq : eigenvalues ] ) . in addition , we use the decay rates of @xmath175 into @xmath214 and @xmath215 : @xmath216^{3/2 } , \end{aligned}\ ] ] where @xmath217 is the mixing angle in the singly - charged higgs sector and @xmath218 we investigate the branching ratios of @xmath175 in two distinct parameter regions : * condition 1 : when @xmath219 is not kinematically allowed . then @xmath175 decays into leptons and @xmath132 pairs only : @xmath220 with the total width for condition 1 @xmath221 * condition 2 : when @xmath219 is kinematically allowed . then @xmath175 is able to decay into charged higgs and gauge bosons as well : @xmath222 with the total decay width for condition 2 @xmath223 we investigate the decay patterns of @xmath175 and present plots of the production cross section times the branching fractions under various conditions . to cover a wide range of parameter space , we distinguish two cases for each condition , depending on the vector lepton masses . we set the yukawa coupling of the vector leptons with the doublet higgs bosons to be @xmath224 for both cases . * case a corresponds to very light vector leptons : @xmath225 gev . for this case we obtain for the eigenvalues , @xmath226 gev , @xmath227 gev , the latter of which is close to the allowed minimum . * case b corresponds to intermediate mass vector leptons @xmath228 gev , @xmath229 gev . for this case we obtain for the eigenvalues , @xmath230 gev , @xmath231 gev . in fig . [ fig : condition1 ] we show the corresponding graphs for condition 1 ( when the decay @xmath219 is not kinematically allowed ) , case a on the top row and case b on the bottom row . we plot @xmath232 with @xmath233 as specified in the attached panels , as functions of the doubly - charged masses . on the left side of the figure the yukawa couplings of vector lepton with the triplet higgs boson are @xmath234 and on the right side of the figure , @xmath235 . we set @xmath236 throughout . for small vector lepton coupling @xmath237 , the decay into @xmath238 is dominant at low @xmath175 masses ( this is because we assumed the couplings @xmath32 with ordinary leptons to be all equal ; should we have chosen them diagonal , the branching ratio into @xmath239 would dominate ) . at high @xmath240 masses the branching ratio into @xmath214 dominates and can reach 40% , branching ratios into ordinary leptons reaching at most 40% , and those into vector leptons remaining at or below 20% level . if the triplet yukawa coupling to vector leptons is allowed to increase to @xmath241 , the decay into the two lightest vector leptons @xmath242 becomes dominant , and can reach 80 - 90% when kinematically allowed ( in the @xmath243 gev region for @xmath227 gev ) and overwhelms the other decay modes , which are now below 5% . the only difference between case a and case b in this figure are threshold effects . for case b , @xmath244 gev for decay into pairs of @xmath242 states , as @xmath231 gev ; otherwise the branching ratios are the same . -0.6 in @xmath245{fig5topleftcondition1case1yukawa001.pdf } & \hspace*{-0.8 cm } \includegraphics[width=3.5in , height=3.5in]{fig5toprightcondition1case1yukawa01.pdf}\\ \hspace*{-1.2 cm } \includegraphics[width=3.5 in , height=3.5in]{fig5bottomleftcondition1case2yukawa001.pdf } & \hspace*{-0.8 cm } \includegraphics[width=3.5in , height=3.5in]{fig5bottomrightcondition1case2yukawa01.pdf } \end{array}$ ] in fig . [ fig : condition2 ] we plot the same quantities for condition 2 ( when @xmath219 is kinematically allowed ) , also case a on the top row and case b on the bottom row , @xmath235 on the left hand side and @xmath241 on the right hand side . the decay pattern is very different here , and it is dominated by @xmath219 . for small vector lepton coupling @xmath246 the branching ratios for decays into vector and ordinary lepton pairs , and @xmath132 pairs are very small , and reach at most 1% . increasing the vector lepton coupling to @xmath241 , the decay into @xmath204 still dominates throughout the parameter space where it is kinematically allowed and can reach a branching fraction of over 90% , while the decay into vector leptons can have branching ratios of up to 25% . again , the decay rates into @xmath132 boson pairs and ordinary leptons are below 1% , and the only difference between case a and case b are , as in fig . [ fig : condition1 ] , threshold effects . -0.6 in @xmath247{fig6topleftcondition2case1yukawa001.pdf } & \hspace*{-0.6 cm } \includegraphics[width=3.5in , height=3.5in]{fig6toprightcondition2case1yukawa01.pdf}\\ \hspace*{-1.2 cm } \includegraphics[width=3.5 in , height=3.5in]{fig6bottomleftcondition2case2yukawa001.pdf } & \hspace*{-0.6 cm } \includegraphics[width=3.5in , height=3.5in]{fig6bottomrightcondition2case2yukawa01.pdf } \end{array}$ ] despite no new signals of physics beyond the sm at the lhc , the sm can not be the complete theory of particle interactions . an extension of the sm by additional vector - like leptons is not ruled out experimentally , and has been shown to provide a dark matter candidate . in models beyond the sm , the vector leptons can alter not only the phenomenology of the higgs , but also of other additional particle representations predicted by the models . we provide an example within the higgs triplet model , where previously we showed that , in the absence of triplet - doublet higgs mixing in the neutral sector ( @xmath117 ) , there is no enhancement of the rate of decay of @xmath0 in this model with respect to the sm expectation . introducing vector leptons does not affect any of the tree - level decays of the neutral higgs boson observed at the lhc , nor its production decay . however , loop decays into electroweak particles , such as @xmath1 and @xmath0 would be affected . we show that , for the no - mixing scenario ( @xmath117 ) the decays rates into @xmath93 and @xmath165 do not depend on the doubly charged higgs mass , and thus without the additional vector leptons , these decays would be unchanged from the sm expectations . with vector leptons , modest enhancements or suppressions are possible , more so for @xmath0 , where for large yukawa couplings , the rate of decays could even double . under the same circumstances , the decay width for @xmath1 remains practically unchanged from its sm value . the model thus presents a mechanism for enhancing one loop - decays and not the other , which seems to be consistent with the lhc data ( so far ) . if @xmath107 , the effect of the doubly - charged higgs bosons is felt for both @xmath248 and @xmath249 , most spectacularly so for very light @xmath169 gev , which is ruled out for @xmath250 . parameter space regions where light doubly - charged higgs masses ( 200 - 250 gev ) , _ and _ significant mixing in the neutral sector coexist , are disfavored . in general , there are many parameter combinations for which the decay @xmath248 is enhanced , but few for an enhanced @xmath251 , and these regions are ruled out by the branching ratio for @xmath0 . however , if the decay @xmath248 is ( modestly ) enhanced , while @xmath1 is the same as the sm to 1@xmath145 , small mixing angles and light doubly charged higgs bosons @xmath252 gev are preferred . the fact that there are no regions of the parameter space consistent with present measurements of the branching ratio for @xmath0 and an enhanced rate for @xmath1 is a feature of this model , valid over the whole explored range of the parameter space . other than that , there are no definite correlations or anti - correlations between there two loop - dominated decays . the intermediate mass doubly - charged higgs boson can decay into light vector leptons , which would alter its decay profile significantly . we explored this possibility and found that , if the singly charged higgs mass is such that the decay @xmath219 is not kinematically accessible , dominant branching ratios into vector leptons , if kinematically accessible , are expected for triplet yukawa couplings @xmath253 , whereas for small @xmath254 branching ratios into ordinary leptons , vector leptons and @xmath132 pairs are comparable for @xmath255 gev . if and where the decay @xmath219 is kinematically accessible , its corresponding branching ratio is the largest , while the branching fraction into vector leptons could reach 20 - 25% for @xmath253 . under both these circumstances , the decay patterns of the doubly charged higgs bosons are significantly altered , raising the hope that they can be found at masses around 300 - 500 gev . the work of s. b. and m.f . is supported in part by nserc under grant number sap105354 . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 716 * ( 2012 ) 1 ; s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 716 * ( 2012 ) 30 . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . lett . * 104 * , 091801 ( 2010 ) ; a. lister [ cdf collaboration ] , arxiv:0810.3349 [ hep - ex ] . k. agashe , g. perez and a. soni , phys . d * 71 * , 016002 ( 2005 ) ; k. agashe , g. perez and a. soni , phys . d * 75 * , 015002 ( 2007 ) ; a. carmona and j. santiago , jhep * 1201 * , 100 ( 2012 ) ; g. -y . huang , k. kong and s. c. park , jhep * 1206 * , 099 ( 2012 ) ; c. biggio , f. feruglio , i. masina and m. perez - victoria , nucl . b * 677 * , 451 ( 2004 ) ; d. e. kaplan and t. m. p. tait , jhep * 0006 * ( 2000 ) 020 ; h. -c . cheng , phys . d * 60 * , 075015 ( 1999 ) . t. moroi and y. okada , mod . phys . lett . a * 7 * ( 1992 ) 187 ; t. moroi and y. okada , phys . b * 295 * ( 1992 ) 73 ; k. s. babu , i. gogoladze , m. u. rehman and q. shafi , phys . d * 78 * ( 2008 ) 055017 ; s. p. martin , phys . d * 81 * ( 2010 ) 035004 ; p. w. graham , a. ismail , s. rajendran and p. saraswat , phys . d * 81 * ( 2010 ) 055016 ; s. p. martin , phys . d * 82 * ( 2010 ) 055019 ; j. kang , p. langacker and b. d. nelson , phys . d * 77 * ( 2008 ) 035003 . b. a. dobrescu and c. t. hill , phys . * 81 * ( 1998 ) 2634 ; r. s. chivukula , b. a. dobrescu , h. georgi and c. t. hill , phys . rev . d * 59 * ( 1999 ) 075003 ; h. collins , a. k. grant and h. georgi , phys . rev . d * 61 * ( 2000 ) 055002 ; h. -j . he , c. t. hill and t. m. p. tait , phys . d * 65 * ( 2002 ) 055006 ; c. t. hill and e. h. simmons , phys . * 381 * ( 2003 ) 235 [ erratum - ibid . * 390 * ( 2004 ) 553 ] ; r. contino , l. da rold and a. pomarol , phys . d * 75 * ( 2007 ) 055014 ; phys . d * 79 * ( 2009 ) 075003 ; k. kong , m. mccaskey and g. w. wilson , jhep * 1204 * ( 2012 ) 079 ; a. carmona , m. chala and j. santiago , jhep * 1207 * ( 2012 ) 049 ; m. gillioz , r. grober , c. grojean , m. muhlleitner and e. salvioni , jhep * 1210 * ( 2012 ) 004 . n. arkani - hamed , a. g. cohen , e. katz and a. e. nelson , jhep * 0207 * ( 2002 ) 034 ; t. han , h. e. logan , b. mcelrath and l. -t . wang , phys . d * 67 * ( 2003 ) 095004 ; m. perelstein , m. e. peskin and a. pierce , phys . d * 69 * ( 2004 ) 075002 ; m. schmaltz and d. tucker - smith , ann . nucl . part . * 55 * ( 2005 ) 229 ; m. s. carena , j. hubisz , m. perelstein and p. verdier , phys . rev . d * 75 * ( 2007 ) 091701 : s. matsumoto , t. moroi and k. tobe , phys . d * 78 * ( 2008 ) 055018 . a. davidson and k. c. wali , phys . * 60 * ( 1988 ) 1813 ; k. s. babu and r. n. mohapatra , phys . d * 41 * ( 1990 ) 1286 ; b. grinstein , m. redi and g. villadoro , jhep * 1011 * ( 2010 ) 067 ; d. guadagnoli , r. n. mohapatra and i. sung , jhep * 1104 * ( 2011 ) 093 . a. azatov , m. toharia and l. zhu , phys . d * 80 * , 035016 ( 2009 ) . atlas collaboration atlas - conf-2013 - 019 ; s. chatrchyan et al . [ cms collaboration ] , phys . b * 718 * , 348 ( 2012 ) . f. del aguila , j. de blas and m. perez - victoria , phys . d * 78 * , 013010 ( 2008 ) . a. joglekar , p. schwaller and c. e. m. wagner , jhep * 1212 * , 064 ( 2012 ) . j. kearney , a. pierce and n. weiner , phys . d * 86 * , 113005 ( 2012 ) ; h. davoudiasl , h. -s . lee and w. j. marciano , phys . d * 86 * , 095009 ( 2012 ) ; k. j. bae , t. h. jung and h. d. kim , phys . rev . d * 87 * , 015014 ( 2013 ) ; m. b. voloshin , phys . d * 86 * , 093016 ( 2012 ) ; d. mckeen , m. pospelov and a. ritz , phys . d * 86 * , 113004 ( 2012 ) ; h. m. lee , m. park and w. -i . park , jhep * 1212 * , 037 ( 2012 ) ; c. arina , r. n. mohapatra and n. sahu , arxiv:1211.0435 [ hep - ph ] ; b. batell , s. jung and h. m. lee , jhep * 1301 * , 135 ( 2013 ) ; h. davoudiasl , i. lewis and e. ponton , arxiv:1211.3449 [ hep - ph ] . ; j. fan and m. reece , arxiv:1301.2597 [ hep - ph ] ; a. carmona and f. goertz , arxiv:1301.5856 [ hep - ph ] ; c. cheung , s. d. mcdermott and k. m. zurek , arxiv:1302.0314 [ hep - ph ] ; w. -z . feng and p. nath , arxiv:1303.0289 [ hep - ph ] ; c. englert and m. mccullough , arxiv:1303.1526 [ hep - ph ] . k. ishiwata and m. b. wise , phys . d * 84 * , 055025 ( 2011 ) . s. k. garg and c. s. kim , arxiv:1305.4712 [ hep - ph ] . a. joglekar , p. schwaller and c. e. m. wagner , arxiv:1303.2969 [ hep - ph ] . g. f. giudice and o. lebedev , phys . b * 665 * , 79 ( 2008 ) ; e. del nobile , r. franceschini , d. pappadopulo and a. strumia , nucl . b * 826 * , 217 ( 2010 ) ; s. p. martin , phys . rev . d * 81 * , 035004 ( 2010 ) . f. arbabifar , s. bahrami and m. frank , phys . d * 87 * , 015020 ( 2013 ) . [ cms collaboration ] , cms - pas - hig-13 - 001 and cms - pas - hig-13 - 005 ; g. aad et al . [ atlas collaboration ] , [ arxiv:1307.1427 [ hep - ex ] ] and atlas - conf 2013 - 029 . a. g. akeroyd , mayumi aoki , and hiroaki sugiyama . d * 77 * , 075010 ( 2008 ) ; a. g. akeroyd , mayumi aoki , and hiroaki sugiyama . d * 79 * , 113010 ( 2009 ) ; takeshi fukuyama , hiroaki sugiyama , and koji tsumura . jhep * 03 * , 044 ( 2010 ) ; s. t. petcov , h. sugiyama , and y. takanishi . d * 80 * , 015005 ( 2009 ) ; takeshi fukuyama , hiroaki sugiyama , and koji tsumura . d * 82 * , 036004 ( 2010 ) ; a.g . akeroyd , cheng - wei chiang , and naveen gaur , jhep * 1011 * , 005 ( 2010 ) ; a. arhrib , r. benbrik , m. chabab , g. moultaka and l. rahili , jhep * 1204 * , 136 ( 2012 ) ; a. arhrib , r. benbrik , m. chabab , g. moultaka and l. rahili , [ arxiv:1202.6621 [ hep - ph ] ] ; a. arhrib , r. benbrik , m. chabab , g. moultaka , m. c. peyranere , l. rahili and j. ramadan , phys . d * 84 * , 095005 ( 2011 ) ; e. j. chun , h. m. lee and p. sharma , jhep * 1211 * , 106 ( 2012 ) ; a. melfo , m. nemevek , f. nesti , g. senjanovi and y. zhang , phys . rev . d 85 , 055018 ( 2012 ) ; m. aoki , s. kanemura and k. yagyu , phys . d * 85 * , 055007 ( 2012 ) ; s. kanemura and k. yagyu , phys . rev . d * 85 * , 115009 ( 2012 ) ; m. aoki , s. kanemura , m. kikuchi and k. yagyu , [ arxiv:1211.6029 [ hep - ph ] ] . k. ishiwata and m. b. wise , arxiv:1307.1112 [ hep - ph ] . r. dermisek and a. raval , arxiv:1305.3522 [ hep - ph ] . s. chatrchyan _ et al . _ [ cms collaboration ] , arxiv:1307.5515 [ hep - ex ] ; [ atlas collaboration ] , atlas - conf-2013 - 009 . c. -s . chen , c. -q . geng , d. huang and l. -h . tsai , phys . d87 , * 075019 * ( 2013 ) ; c. -s . chen , c. -q . geng , d. huang and l. -h . tsai , phys . b * 723 * , 156 ( 2013 ) ; p. s. bhupal dev , d. k. ghosh , n. okada and i. saha , jhep * 1303 * , 150 ( 2013 ) [ erratum - ibid . * 1305 * , 049 ( 2013 ) ] . g. abbiendi _ et al . _ [ opal collaboration ] , phys . b * 526 * ( 2002 ) 221 ; g. abbiendi _ et al . _ [ opal collaboration ] , phys . b * 577 * ( 2003 ) 93 ; p. achard _ et al . _ [ l3 collaboration ] , phys . b * 576 * ( 2003 ) 18 ; j. abdallah _ et al . _ [ delphi collaboration ] , phys . b * 552 * ( 2003 ) 127 . j. sztuk - dambietz [ h1 and zeus collaboration ] , j. phys . * 110 * ( 2008 ) 072042 ; a. aktas _ et al . _ [ h1 collaboration ] , phys . b * 638 * ( 2006 ) 432 ; v. m. abazov _ et al . _ [ d0 collaboration ] , phys . * 101 * ( 2008 ) 071803 ; v. m. abazov _ et al . _ [ d0 collaboration ] , phys . rev . * 108 * ( 2012 ) 021801 ; d. e. acosta _ et al . _ [ cdf collaboration ] , phys . * 93 * ( 2004 ) 221802 ; t. aaltonen _ et al . _ [ the cdf collaboration ] , phys . * 101 * ( 2008 ) 121801 . s. chatrchyan _ [ cms collaboration ] , arxiv:1305.0491 [ hep - ex ] ; g. aad _ et al . _ [ atlas collaboration ] , phys . b 722 ( 2013 ) 305 ; s. chatrchyan _ et al . _ [ cms collaboration ] , eur . j. c * 72 * ( 2012 ) 2189 ; g. aad _ et al . _ [ atlas collaboration ] , eur . j. c * 72 * ( 2012 ) 2244 . c. -w . chiang , t. nomura and k. tsumura , phys . d * 85 * , 095023 ( 2012 ) ; s. kanemura , k. yagyu and h. yokoya , arxiv:1305.2383 [ hep - ph ] . m. muhlleitner and m. spira , phys . rev . d * 68 * , 117701 ( 2003 ) .

edge localised modes ( elms ) are bursty , quasi - periodic expulsions of energy and particles from the plasma edge in the high - confinement state of toroidal magnetised plasmas @xcite . elms originate from the steep pressure gradient pedestal in the outer closed field line region of a tokamak . a substantial release of magnetic and potential energy , caused by a rapid rise of the rate of momentum transfer through the flux surfaces , leads to enhanced transport and loss of heat and particles into the open scrape - off layer ( sol ) field line region . the filamented peak heat fluxes during elms on the bounding divertor plates present serious restrictions on the performance of future fusion experiments like iter @xcite and necessitate the development of suitable control techniques @xcite . elms generically have been observed in all divertor tokamaks since the initial discovery of the high - confinement `` h - mode '' state in the early 1980s @xcite , while characteristics like frequency and intensity are found to vary widely depending on experimental conditions @xcite . stellarator experiments also are reported to show similar bursty quasi - periodic edge activity in the presence of edge transport barriers @xcite . indication that equilibrium drifts are part of the process has been found by experimental diagnosis of an asymmetry of the energy and particle fluxes on each of the divertor plates and evidence that this is sensitive to the directions of the toroidal current and magnetic field @xcite . phenomenologically , edge localised mode events in tokamaks have been compared to solar flare eruptions @xcite , and , in the cataclysmic variability of observed radiation emanating from the plasma , they may seem to bear also some remote resemblance to the outbursts of pressure - driven dwarf novae @xcite , although the specific instability mechanisms behind these phenomena are clearly of a substantially different nature . on the other hand , the nature of elms has also not yet been completely clarified , in the sense that there is no first - principles based predictive theory available . a number of characteristically different variations of elms have been observed in experiments ( e.g. type i , ii or iii elms ) which may actually be caused by different instability mechanisms @xcite . `` type iii '' elm observations show a remarkable similarity to global bursts found in computations of drift - wave turbulence , where the turbulence generated flows and fluxes are closely tied to the self - consistent equilibrium evolution in the plasma edge @xcite . `` type i '' elms in the experiment specifically show stronger magnetic activity during the burst and are usually associated with the onset of a magneto - hydrodynamic ( mhd ) ideal ballooning mode ( ibm ) instability when a threshold pressure gradient is reached in the steep edge pedestal of an h - mode plasma @xcite . the possible role of an additional current driven instability ( `` peeling - ballooning mode '' ) for `` type i '' elms has also received recent interest @xcite . the ideal ballooning instability with nonlinear phases or aftermath has provided a paradigm for recent and current study of the elm phenomenon theoretically . explosive mhd instability and associated critical phenomena have been advanced analytically @xcite . early braginskii fluid simulations were given in support of this @xcite . large , nonlinear mhd codes have been studying this in more detail , resolving several numerical problems and paying specific attention to the details of the magnetic geometry , with mixed results on the phenomenology of the instability beyond its linear phase @xcite . whether or not the ideal ballooning instability or its peeling / ballooning variant is actually responsible for elms occuring in experiments ( here noting the absence of a demonstrably well - resolved l - to - h mode transition in computations from first principles ) , the nonlinear phase of this `` ideal mhd blowout '' or `` ibm blowout '' phenomenon is of physical interest . the ability of the instability to destroy the edge pedestal layer is undisputed , and the term `` blowout '' is germane . typical physical parameters lead to a situation wherein the ideal mhd interchange growth rate is comparable to the parallel alfvn transit time , one and the same with the basic ideal ballooning criterion . this is fast with respect to mhd but not to microturbulence . provided they can simultaneously treat global mhd ( i.e. , the spectrum covers both global and ion gyroradius scales ) , edge - turbulence computations can also address the phenomenon and are even in a position to treat parts of it which fall outside the paradigm of the mhd model . here we present nonlinear gyrofluid computations of ibm blowout events localised to the edge / sol region of a tokamak plasma , resolved to below the ion gyroradius @xmath0 . as in the large mhd models the self - consistent evolution of the equilibrium is not only included but is also an integral part of the dependent variables . an electromagnetic 6-moment gyro - fluid model ( `` gemr '' ) for both electrons and ions is used , including an energy - conserving treatment of finite ion larmor radius effects and higher - moment terms which occur in the toroidal drift @xcite . the energetic consistency is a key feature since the reaction of parts of the system such as flows and currents which have low energy content can play a central role in the indirect nature of nonlinear dynamics by serving as transfer channels . although the gyrofluid model has a different formulation of nonlinear polarisation ( the generalisation of `` vorticity '' to a two - fluid setting ) than a fluid or mhd model , full correspondence in the regime of validity of the latter models has been shown @xcite . in this sense the model is a superset of global reduced mhd and electromagnetic microturbulence . the global dynamics is in the reduced mhd regime @xcite due to low absolute edge beta values ( @xmath1 ) but the small scales depend on treating the ion gyroradius to arbitrary order . the model does not assume instabilities to occur at any particular scale , but when they occur the resulting spread of the spectrum is found to reach @xmath0 within a few eddy turnover times of the onset of nonlinearity in every case computed . the work reported herein addressed the following points . the relationship of the mhd dynamics to the ideal threshold is much discussed but in the gyrofluid model several mechanisms of microinstability are also present . the threshold issue is obscured by the existence of ion temperature gradient ( itg ) drift instabilities at all beta values , with a smooth transition between them . nevertheless , varying character in the saturation phases is observed at different values . we also address the issue of possible `` rho - star '' dependence ( here the local rho - star is given by @xmath2 where @xmath3 is @xmath0 evaluated at @xmath4 and @xmath5 is the scale length of the @xmath6 profile , both evaluated at the mid - pedestal location in minor radius ) which would indicate a role for diamagnetic drifts . the gradient is sufficiently steep that @xmath7 so that none of these effects can be ordered small . we examine the necessary resolution to obtain converged cases , which gives the spectral range actively involved in nonlinear saturation . the main instability at experimentally relevant beta is the ideal ballooning instability but the main saturation process is energy exchange with broadband electromagnetic drift wave / itg turbulence which the instability itself generates . due mostly to this , the results turn out to be outside of both the magneto - hydrodynamic ( mhd ) and collisional braginskii paradigms , on which most previous approaches have been based . on the other hand , the gemr model still uses delta - f equations , so the actual profile phenomenology in the sol region is not well represented , and as in other work with gemr we concentrate on the properties of the nonlinear dynamics @xcite . following sections give the details of the gemr model , a discussion of the physical difference between itg and mhd dynamics and their relative roles , the methods used to pre - set the axisymmetric equilibrium state so that the computations are done in the absence of axisymmetric oscillations whose decay times can be slower than the ibm rise time , the computations of the blowout phase itself , and the possibility and limitations of quantitative comparison to experiment . the gyrofluid model used herein is given in ref . it is based on the original derivations given in refs.@xcite , with corrections given as motivated by free energy conservation . an alternative derivation using the conservation laws of the underlying gyrokinetic model is given in ref.@xcite . the equations are normalised to spatial scale @xmath8 and time scale @xmath9 where @xmath8 is the minor radius and @xmath10 is the reference sound speed given by @xmath11 , that is , in terms of the electron temperature ( in energy units ) and ion mass . the main parameters are the drift parameter , electron dynamical beta , and electron collisionality , respectively given by @xmath12 where @xmath13 and @xmath14 are the speed of light and fundamental charge , @xmath15 and @xmath16 are the pressure and braginskii collision time @xcite of the electrons , and @xmath17 is the drift scale . note that if @xmath4 then @xmath18 . the reference for @xmath19 is the deuterium mass @xmath20 . for the electrons the physical value of the mass @xmath21 is always used . with global geometry there is no single magnetic field line connection length , but one with a profile with @xmath22 a function of the minor radius coordinate , where @xmath23 is a constant giving the reference value of the major radius . each species @xmath24 has its own set of gyrofluid moment variable equations , one each for density @xmath25 , parallel velocity @xmath26 , parallel and perpendicular temperature @xmath27 and @xmath28 , and parallel and perpendicular energy components of the parallel heat flux @xmath29 and @xmath30 . they are coupled to the electrostatic and parallel magnetic potentials @xmath31 and @xmath32 through self consistent field equations . the tilde denotes a dependent variable to distinguish from constant parameters . the correspondence of these equations to the braginskii fluid model is given in ref . each species is characterised by a background charge density , temperature / charge ratio , and mass / charge ratio , given by the normalised parameters @xmath33 for electrons @xmath34 and @xmath35 . for the main ions @xmath36 while @xmath37 is kept as a parameter . for the trace ions @xmath38 is always taken . for hot trace ions @xmath39 is large while @xmath40 is still moderate . the species gyroradius @xmath41 is given by @xmath42 , which is always small for electrons , close to unity for main ions , and can be moderate or large for trace ions . for this study a simplified geometry is taken with @xmath43 except in curvature terms , so magnetic pumping and shaping effects @xcite are neglected . since dynamics at the electron gyroradius scale are neglected the gyroaveraging effects on @xmath32 are not treated . the moment equations are given by @xmath44 @xmath45 @xmath46 @xmath47 @xmath48 @xmath49 the linearised pressures are given by @xmath50 collisional dissipation is controlled by a collision parameter @xmath51 analogous to @xmath52 and a set of numerical constants for each species , @xmath53 which are the thermal force , thermal conductivity , and viscosity coefficients involved in parallel dissipation , with values given by the braginskii model @xcite , though that regime is never reached in core turbulence . for electrons these coefficients are @xmath54 and @xmath55 and @xmath56 , and for ions they are @xmath57 and @xmath58 and @xmath59 , respectively . the resistive dissipation is given by @xmath60 with parallel current given by @xmath61 noting that it is the electron heat fluxes that enter @xmath62 and as with @xmath63 the @xmath62 term enters the parallel motion in the same way for every species . the anisotropy dissipation is given by @xmath64 the thermal conduction components are given by @xmath65 @xmath66 @xmath67 the landau damping effects are modeled by @xmath68 with thermal velocity @xmath69 given by @xmath70 . the gyroaveraging is done through pad approximants to operators which would act in fourier space , @xmath71 and then the gyroaveraged potentials are @xmath72 noting that they are species dependent . the main exb advection and the flr correction pieces are given by @xmath73 - \nu_\perp\dpp^4 f + \nu_\parallel\ddpl f\ ] ] @xmath74\ ] ] where the @xmath75 $ ] symbols denote the nonlinear bracket defined below , and @xmath76 denote the artificial dissipation coefficients . the parallel derivative is given by @xmath77\ ] ] the curvature operator is given by @xmath78\ ] ] the species are coupled through two equations which give the self consistent response of the field potentials to the evolution of the gyrofluid moment variables . the electrostatic potential is governed by quasineutrality , which sets the total space charge density to zero . the density for each species is given by a part due to the gyrocenters and another due to polarisation , which is what sets @xmath79 . this polarisation equation is given by @xmath80 where @xmath81 gives the gyroscreening of the main ions . in this version of the model the electron and trace ion contributions to gyroscreening are neglected , due to small @xmath82 and zero @xmath83 , respectively . on the right hand side the electrons and main ions enter with oppositely signed @xmath83 values and again here the trace ions do not enter . this establishes them as a trace species . in a similar way the trace ions are left out of the induction equation due to the zero @xmath83 , so that @xmath84 determines @xmath32 noting the way that the normalisation scales it with respect to @xmath85 . more detail , including energy conservation and the relationship of these equations to it , is given in refs . @xcite . in these expressions the operators @xmath86 and @xmath87 and the nonlinear brackets are determined by the representation of the geometry that is used . the magnetic geometry is a simplified representation of an axisymmetric magnetic field using field aligned hamada coordinates . the field representation is @xmath88 where @xmath89 is the physical toroidal angle about the symmetry axis , @xmath90 is the magnetic flux function , and @xmath91 is given by the constraint of mhd equilibrium . without loss of generality one may define magnetic flux coordinates @xmath92 with the following properties , @xmath93 where @xmath94 is the volume enclosed by the surface with flux @xmath90 , then @xmath95 and @xmath96 are two further flux functions found , and the subscript denotes the partial derivative with respect to @xmath97 . the ratio @xmath98 is another flux function which gives the pitch of the magnetic field . the angles are cyclic on @xmath99 $ ] which determines the coordinate jacobian to be unity . hence the metric tensor @xmath100 satisfies @xmath101 . the function @xmath102 is found by the constraint that @xmath103 is cyclic on the unit interval @xmath104 where @xmath105 is a simple cyclic coordinate on the flux surface in the poloidal plane ( e.g. , path length on the curve , or the physical angle position about the magnetic axis with respect to any branch cut ) . then @xmath106 is found @xmath107 where the angle brackets give the flux surface average defined as @xmath108 now , @xmath109 is also defined and the toroidal angle coordinate is given by @xmath110 which completes the prescription . the field aligning is a one - to - one and onto coordinate transform from @xmath92 to @xmath111 given by @xmath112 where @xmath8 is a reference minor radius and @xmath113 is an arbitrary function of @xmath114 which is chosen to make the off diagonal perpendicular metric element @xmath115 vanish at a particular location . this is called the shifted metric procedure @xcite . the point is that at any position in @xmath116 where perpendicular drifts or @xmath86 is evaluated , the coordinate elements are rectangular , and the information contained in magnetic shear enters only in derivatives in the parallel coordinate . the magnetic field components now satisfy @xmath117 at all locations in space for any choice of @xmath113 . hence although the perpendicular coordinate elements are rectangular only for one particular location in @xmath116 the field aligning is exactly satisfied everywhere . the coordinate metric elements @xmath118 are found in the @xmath92 representation . then the grid locations @xmath119 are all given their own members of the family of these field aligned coordinates via @xmath120 then we have @xmath121 at @xmath119 . the drift tensor elements are given by @xmath122 that is , @xmath123 is the only independent , nonvanishing component . we have @xmath124 which defines the bracket @xmath125 $ ] for any scalar field quantities @xmath79 and @xmath126 . with @xmath79 the electrostatic potential this gives the exb advection term , for electrons for which flr corrections are neglected by taking @xmath127 . for ions this is generalised by the flr corrections as given above . for local equations as those used here ( nonlinearities kept only in advection and magnetic flutter ) the geometry model must be modified to retain free energy conservation by the equations . in particular , any function of @xmath114 such as a radially varying normalised parameter ( such as the temperature / charge ration @xmath39 ) multiplying any of the curvature terms will cause them to fail to conserve free energy . the compressibility is already split away from advection and placed into the curvature terms , so the brackets must retain the properties of incompressible advection . this means that the model for @xmath128 which is used must satisfy @xmath129 and to avoid any confusion the curvature terms are written with @xmath130 , which can be given any spatial dependence . correspondingly , the mhd version of the continuity equation ( neglecting diamagnetic fluxes ) has already been manipulated according to @xmath131 @xmath132 and then terms such as @xmath133 are neglected due to the ordering . the result is the same as the local continuity equation @xmath134 where @xmath135 is a normalising constant , if we identify the dependent variable @xmath136 with @xmath137 . it is important to note that with the self consistent background being evolved this also includes the profile : @xmath138 . the simplified geometry is now defined by its operators . the laplacian and gyroaverging operators neglect @xmath87 . the form in the ampere s law ( which does not involve the gyroradius ) is given by @xmath139 the form in the gyroaverging operations and polarisation is given by @xmath140 the brackets are defined by @xmath141=q[f , g]_{xy}+[f , g]_{xs}-q_{,x}(s - s_k)[f , g]_{ys}\ ] ] for any two scalar fields @xmath142 with each bracket piece defined by @xmath141_{ij } = \lp{\partial f\over\partial x^i}{\partial g\over\partial x^j } - { \partial g\over\partial x^i}{\partial f\over\partial x^j}\rp\ ] ] with the third piece not used since these are always evaluated at @xmath119 . nonlinear computations on ideal or peeling ballooning mode elm scenarios in tokamaks have previously been based on magnetohydrodynamic ( mhd ) models @xcite , and the respective codes had largely been tailored towards mode structures expected by linear analysis . restriction to single - fluid mhd equations including two - fluid correction terms allows computations of large to mesoscale dynamics in realistically shaped 2d tokamak equilibria including x - point geometry crossing the separatrix , due to the lack of a need to use field - aligned coordinates . recent mhd simulations are able to reproduce spatio - temporal mode structures , divertor footprints and other characteristics with good agreement to experimental diagnostics , in particular for the growth and immediate blowout phase of the finger - like instability @xcite . however , as soon as the dynamics becomes nonlinear , the spectrum broadens to include scales normally associated with drift wave dynamics . this turbulence is basically of a drift - alfvn type with strong edge itg character ( see ref . @xcite for the signatures of the various mode structure types ) , made more powerful than otherwise by the energetic access to the long - wavelength mhd component . the itg character results from the steeper logarithmic gradients in both temperature profiles , and the lack of involvement of ion temperature fluctuations in the parallel nonadiabatic response of the electrons . although edge turbulence of this type ( cf . @xcite ) is strongly suppressed in the h - mode phase and is not initially involved in the instability phase , the experience of edge turbulence becomes relevant as these physical components become involved in all the nonlinear phases of the blowout . the saturation and aftermath of the blowout , which in an experiment carries a large part of the actual transport losses that lead to a degradation of the pedestal , should be expected to involve physics not contained in the mhd model . ion temperature gradient ( itg ) driven modes and ideal ballooning modes are both essentially caused by the gradient and curvature driven interchange instability and show similar character in their initial linear growth phases . instability is achieved above their respective critical gradients , determined by the ratio @xmath143 between density gradient to ion temperature gradient scale lengths for itg , and by the ideal mhd ballooning parameter @xmath144 for ibm , where @xmath145 is the magnetic shear parameter and @xmath146 is a critical threshold which depends on beta , collisionality , and toroidicity . the principal difference between these modes is the relative role of the parallel alfvn responses which tend towards enforcement of an adiabatic response in the electrons ( parallel force balance , with the electron pressure gradient ) . this is nonexistent in an mhd model and subdominant generally for an mhd instabililty . but it constrains the electrons if the @xmath147 parameter is below threshold , still allowing the itg instability because the adiabatic response does not involve the ions . the itg instability exists at all @xmath147 values , and furthermore for the edge situation the itg instability and mode structure transitions smoothly to a drift wave one in a nonlinear setting for @xmath148 . hence there is actually no finite threshold in values of @xmath149 or @xmath147 for the nonlinear edge situation . therefore , the existence of a threshold in the experiment is the same as the mechanism which maintains the h - mode , which is not yet well understood . it follows that a gyrofluid model computation of an elm scenario with enough resolution to allow for ion - gyroradius based dynamics has to face the lack of a simple linear threshold in the parameter space . the transition from initial ( micro-)instabilities to generic edge turbulence was studied in detail in ref . as the most unstable linear modes crystallise out of an initial random bath of small - amplitude perturbations , the linear growth rate rises and becomes steady . the maximum value of the instantaneous growth rate of total fluctuation free energy @xmath150 given by @xmath151 may be taken as the maximal linear growth rate . the curve of @xmath152 then falls very sharply to zero ( over about @xmath153 ) as saturation occurs . there is some structural adjustment over the next few @xmath154 as the spectrum fills out , and then the turbulence is fully developed . but over the adjustment phase the value of @xmath155 is well below its previous maximum . we will use the same diagnostic herein , except that the total ion exb heat flux @xmath156 averaged over the computational domain is used instead of @xmath150 because most of @xmath150 is represented by the self consistent profiles while @xmath156 is entirely due to fluctuations . visualisation of the tokamak elm burst : the edge region of a circular poloidal tokamak cross - section is shown on the left with computational perpendicular @xmath116-domains ( @xmath157 2 , 5 , 8 , 11 , 14 of @xmath158 total ) mapped onto the circle during the maximum growth phase ( @xmath159 in units of @xmath9 ) of an ideal ballooning mode . on the right , the outer midplane area ( @xmath160 with size @xmath161 and @xmath162 ) is magnified and the radial domain stretched by a factor of 2 for better visibility at times @xmath159 ( max . flux ) and @xmath163 ( turbulent aftermath).,width=188 ] the ibm elm blowout scenario is similar to this , initially , except that the instability is not a microinstability . nevertheless , the scale differs by less than an order of magnitude : the toroidal mode numbers for linear itg instabilities are in the range of @xmath164 while the main ideal ballooning mode is near mode number @xmath165 , on the entire flux surface , for these typical @xmath166 values ( recall @xmath167 for @xmath168 , and the ratio @xmath169 is not far from unity in experimental cases ) . the itg mode numbers are determined by the ion gyroradius and the dimensionless scale ratio @xmath170 , while the mhd values are determined by the width of the pedestal . since @xmath5 is not larger than about @xmath171 in an h - mode pedestal , these scales are not disparate . due to the closeness of the native scales , the ibm instability very quickly transfers energy to the ion gyroradius scale , which is only about two or three steps away in a turbulence cascade which acts at a factor of two at each step . this brings the itg and drift alfvn physics into play , and in any nonlinear stages the latter involves stable shear alfvn component with substantial energy content . the moment of saturation is defined as the time at which the ion exb heat flux averaged over the computational domain is maximum ; we refer to this as the `` peak - flux '' time . the results shown below indicate that coincides with the establishment of the fully turbulent regime . representation of this phase requires resolution of all scales and inclusion of the appropriate drift wave physics in the model . how much resolution is actually needed depends on the strength of the blowout , which is situation dependent . we therefore include a resolution scan on the toroidal mode number spectrum . the result is that convergence is reached in the temporal phases just after peak flux only if the spectrum reaches down to @xmath172 . the self - consistent equilibrium coupling in our gyrofluid model only allows treatment of shifted circular @xmath173 geometry ( the difference to mhd codes is that the high - resolution flr - gyrofluid dynamics necessitates field - aligned coordinates ) . effects of flux surface shaping are postponed to later studies . our focus here is on basic physics issues regarding the ability of the model to capture mhd phenomena ( see also ref . @xcite concerning global alfvn oscillations with self consistent profiles ) , resolution , turbulent character , scalings and tests which are also accessible by @xmath173 geometry . in particular , the theory of explosive instability has been formulated for circular geometry @xcite and is therefore accessible by the present model . first - principles based local drift wave edge turbulence simulations are not able to obtain a realistic h - mode edge state with the known experimental characteristics : correct density and temperature pedestal profiles shapes or strength of flow shear are not obtained by self - consistent evolution by specifying core sources only , nor has a threshold transition character been found in any verified edge turbulence simulation @xcite . therefore some kind of `` modelling '' has to take place when the ibm instability ( as an h - mode phenomenon ) and its subsequent nonlinear evolution is simulated with a nonlinear gyrofluid turbulence code : although the realistic development of an edge transport barrier ( and thus a full elm cycle ) can not be directly obtained , one still may prescribe the h - mode pedestal profile before the onset of an elm , known from experimental data , as an initial state for the simulation . as a base case for the prescribed pedestal profiles the well diagnosed edge characteristics of asdex upgrade h - mode shot # 17151 is used here @xcite . the local parameters , taken as mid pedestal values , correspond to electron and ion temperatures @xmath174 ev , @xmath175 ev , densities @xmath176 m@xmath177 , magnetic field strength @xmath178 t , major torus radius @xmath179 m , aspect ratio @xmath180 , perpendicular temperature gradient length @xmath181 cm , density gradient length @xmath182 cm . the profile of the safety factor @xmath183 is parabolic yielding local values at the last closed flux surface ( lcfs , @xmath184 ) of @xmath185 and @xmath186 . the radial domain of the simulations covers a range of @xmath187 on either side of the lcfs . the nominal pressure values are @xmath188 and @xmath189 . all times are given in units of @xmath9 . the computational grid is given in terms of the @xmath111 coordinate domains . the spacing is equidistant . for each grid point in the third coordinate @xmath119 the @xmath190-coordinate is @xmath191 as defined in eqs.([eqcoords],[eqalphak ] ) so that for each grid point @xmath192 the coordinate system @xmath193 is the one which has @xmath194 at @xmath119 . for the nominal case @xmath162 perpendicular and @xmath195 parallel mesh points are used . the radial domain with @xmath196 spans the plasma edge region between the h - mode pedestal top , with plasma core parameters as inner boundary values , and the outer bounded scrape - off layer region ( @xmath197 ) . this represents a spatial range from the global scale to smaller than the ion gyroradius scale , with a ratio @xmath198 between ion gyroradius @xmath199 and minor torus radius @xmath8 . in the sol region ( @xmath200 ) the parallel boundary condition is replaced by a debye sheath model whose treatment is given in ref . @xcite . radial tokamak edge pedestal profiles of the plasma pressure before and after an elm blowout event across the separatrix into the scrape - off layer ( sol ) . the area shaded in grey shows the present radial numerical simulation domain.,width=283 ] the initial conditions are thus prescribed and are based on experimentally diagnosed radial temperature and density pedestal profiles @xmath201 and @xmath202 for each species ( electrons and ions ) . a consistent electrostatic potential @xmath203 is derived by numerically solving the neoclassical balances ( parallel dynamics , toroidal drifts , and collisional dissipation , but not transport ) in a pre - processing step with a modified ( zonally frozen ) gemr setup , resulting in a time - steady 2d dissipative solution . the @xmath173 mhd equilibrium in turn is set internally in gemr by computing the shafranov shift , and in the present setup would not be consistently described by prescribed in terms of external or coupled shaped equilibrium solvers @xcite . the parallel and perpendicular electron and ion temperatures , @xmath204 , @xmath205 , @xmath206 and @xmath207 , are directly adopted and fixed from experimentally derived values by filtering the zonal component out of the total time derivative @xmath208 . their values are zonally frozen by setting @xmath209 where @xmath210 is the zonal ( parallel and perpendicular ) average of all right - hand - side terms @xmath211 in the temperature evolution ( eqs . 5,6 ) including numerical dissipation . on the other hand , gyrocenter densities @xmath212 and @xmath213 have to be set to obey relaxation relations that allow the vorticity to freely evolve into equilibrium . this is achieved by freezing the zonal component of the sum @xmath214 , where @xmath215 , as the density part of the pressure during the equilibration phase , but allowing the difference ( i.e. , vorticity ) to evolve freely . this ensures that the contribution of densities to the total pressure is zonally frozen through the relation @xmath216 while the densities relax regarding to @xmath217 and @xmath218 the numerical solution of the equilibration phase , starting directly from realistically steep pedestal profiles @xmath219 , into steady state is delayed by long , weakly damped global geodesic alfvn oscillations . convergence is expedited by ramping up all of the gradients gradually from zero to prescribed value over the first @xmath220 of the run by @xmath221 this pre - processing equilibration phase is run until convergence with reduced perpendicular resolution ( @xmath222)=(@xmath223 ) , and without the exb and magnetic flutter nonlinearities , which allows establishment of the 2d structure in a smooth manner . then , the resolution is increased to the nominal values ( @xmath224 ) , and a random turbulent bath with relative amplitude @xmath225 is added to the background pedestal profiles inside the closed flux surface region . this procedure reduces transient alfvnic and geodesic acoustic ringing and prepares a reproducible initial state . however , the following sudden release of the nonlinearities also leads to transient oscillations . depending on parameters , these may still be present at the onset of the instabilities , and can obscure a clean view on the nonlinear growth rates , which will be relevant for the discussion below on diagnosing linear or explosive instability . a schematic sketch of the equilibrium pedestal profiles , representing an idealised asdex upgrade h - mode scenario @xcite , is shown in fig . [ f : profiles ] together with the final relaxed state after the elm blowout phase ( which is discussed in the next section ) . when this initialised pedestal pressure profile is ideal ballooning unstable , the ibm instability in gemr simulations is observed to be linearly growing in the pedestal region and at the onset of nonlinearity further overshoots and saturates , representing the beginning of the turbulent blowout phase during which a substantial fraction of the pedestal energy is thrown onto the sol . subsequent evolution involves turbulence in both the pedestal and sol regions as the original pedestal energy is dissipated . previous nonlinear approaches on elm ideal ballooning mode burst computations have treated only the initial growth and nonlinear phases , focused on low - wavelength modes and resolved only the mhd - relevant scales , excluding treatment or discussion of the ion gyroradius scales . this however precludes the development of fully developed microturbulence which is caused by the onset of nonlinearity ( robust transfer of free energy to smaller scales ) . ultimately , this microturbulence decides both the transition to nonlinearity and the eventual saturation ; that is , most of the post - peak transport curve . this affects the mhd scenario of nonlinear explosive growth which most previous work advances . however , we find that the exclusion of the ion gyroradius scales produces an under - resolved situation . we measure the quantitative character of the growth in both linear and nonlinear phases with a growth curve @xmath226 defined in terms of the heat flux @xmath156 instead of the fluctuation free energy as explained above . the heat flux is computed as a zonal ( flux surface ) average , @xmath227 @xmath228 and then the time trace @xmath229 , normalised in terms of @xmath230 at nominal parameters , is computed as a volume average over the central half of the radial domain . the ibm instability , which follows the random seeding of the pre - processed pedestal profile in our computations , is very violent , growing for the nominal case at a rate @xmath231 , just below the ideal interchange rate . on the other hand the growth curve @xmath152 appears qualitatively the same as in basic turbulence cases . at all time points in the nonlinear phase @xmath152 is well below its previous maximum . just after initial saturation there is some nonlinear evolution in which @xmath152 crosses zero before settling down into long - term statistical saturation . at late times the initial blowout no longer imprints the results : with a fixed source one merely finds bursty turbulence thereafter . hence , there is no evidence for explosive instability in this nominal case , which , if present , would be visible in a finite time singularity in the fluctuating free energy @xmath232 rising with a power @xmath233 . the burst and the resulting decaying turbulence act to transport the pedestal plasma across the lcfs , where energy and particles are lost within short times by parallel boundary outflow to the scrape - off layer ( sol ) limiter . without maintenance of a heat and particle source at the core boundary , this leads to decay of the initial gradient with a rapid onset at the initial transition to nonlinearity at about @xmath234 . this time point also corresponds to the peak - flux time . total energy @xmath235 ( top figure ) and ion heat flux @xmath229 ( bottom figure ) during the elm blowout : without source maintenance ( dashed curve ) ; with the source term switched off at the time of maximum instability ( here : at @xmath236 ) ; and with the source maintained for the whole simulation time.,title="fig:",width=283 ] total energy @xmath235 ( top figure ) and ion heat flux @xmath229 ( bottom figure ) during the elm blowout : without source maintenance ( dashed curve ) ; with the source term switched off at the time of maximum instability ( here : at @xmath236 ) ; and with the source maintained for the whole simulation time.,title="fig:",width=283 ] however , the ion temperature profile already starts decaying during the low - amplitude linear growth phase , due to neoclassical ( finite larmor radius ) transport ( for detail on how this works in a fluid model see ref.@xcite ) . this situation is shown in fig . [ f : energy ] ( left frame , dashed line ) , where the total pedestal energy is reduced by a quarter of its initial content before the peak flux phase . therefore we test sensitivity against sources . maintenance of the ion temperature profile by a fixed ion heat source localised to the inner boundary ( following the method for driven cases in ref . @xcite ) ensures nearly unchanged gradients during the linear growth phase . at the end of the simulation time ( @xmath237 ) after the elm burst , the energy content is reduced to around 10% of the initial content when the source is absent or is switched off at the peak flux time ( @xmath234 ) , and to around 20% when the source is maintained for the whole simulation time . the influence of either switching off the source or maintaining the zonal profiles on the nonlinear ion heat flux @xmath229 is shown in fig . [ f : energy ] ( right ) : the linear growth phase and the peak flux is not significantly changed by the source . as expected , the heat flux saturates for late times with maintained core boundary inflow and is otherwise decaying . consistency check for different perpendicular resolutions by varying the grid point number @xmath238 between 64 and 512 . convergence is seen for @xmath239 and higher.,width=283 ] consistency checks on the perpendicular resolution are made by varying @xmath238 between 64 , 128 , 256 and 512 . the evolution of @xmath229 for these cases is shown in fig . [ f : nyres ] . the @xmath240 case is clearly under - resolved and overestimates the total energy decay by a factor 2 . for @xmath241 the initial part of the post - peak decay phase disagrees with the @xmath162 case . convergence is found for @xmath239 and higher . the nonlinear saturation phase thus requires resolving the ion gyroradius scale . this is the clearest indication that turbulence for toroidal mode numbers beyond about @xmath242 extending down to the @xmath243 range is involved in the blowout saturation process . perpendicular mode number @xmath244 spectra for @xmath159 ( top ) and @xmath163 ( bottom ) around the peak flux phase . the vorticity spectrum ( bold lines ) is already flattening down to the ion gyroradius scale for @xmath163 only a few @xmath245 after the maximum linear growth phase.,title="fig:",width=283 ] perpendicular mode number @xmath244 spectra for @xmath159 ( top ) and @xmath163 ( bottom ) around the peak flux phase . the vorticity spectrum ( bold lines ) is already flattening down to the ion gyroradius scale for @xmath163 only a few @xmath245 after the maximum linear growth phase.,title="fig:",width=283 ] the transition from linear instability to turbulence is studied during the peak flux phase around @xmath234 . [ f : kyspecs ] shows perpendicular mode number @xmath244 spectra of the squared amplitude of various fluctuating plasma quantities ( density @xmath246 , ion temperature @xmath247 , electrostatic potential @xmath79 and vorticity @xmath248 ) for @xmath159 and @xmath163 . this time difference corresponds to @xmath249 in physical units and around @xmath250 in local drift units , which is only slightly faster than the overshoot and saturation times known from edge microinstability cases @xcite . initially , the ion temperature gradient ( itg ) driven microinstability and the ideal ballooning mode ( ibm ) compete in growth out of the random low - amplitude bath . the itg mode grows strongest near the separatrix due to radially local steepening by parallel sol diffusion , which in our simulations may be overestimated by using the standard fluid bohm outflow boundary conditions . for @xmath251 the linear ibm is clearly dominant near a toroidal mode number of 9 - 10 for the nominal aug parameters , consistent with experimental observations @xcite . around @xmath252 rapid formation of a turbulent cascade range in the spectra is observed and the vorticity spectrum is already spread out to the ion gyroradius scale ( @xmath253 ) . this is a manifestation of the role of self generated drift wave turbulence in the saturation process . because of the way both the diffusive mixing and vorticity scattering nonlinearities enter the drift wave physics @xcite the spectrum is held together as a unit , all scales down to @xmath0 are involved in the saturation phase of the overall elm blowout transport . the involvement of @xmath0 at unit order , in turn , necessitates a nonlinear gyrofluid model ( and at some later date , a nonlinear gyrokinetic model ) . the characteristics of the blowout at various plasma beta is studied for @xmath254 , with the absolute @xmath255 a factor of @xmath256 higher due to the @xmath257 value . a competition between unstable growth of itg high - n modes and ibm low - n modes is observed : for low @xmath255 an early onset of itg turbulence degenerates the profile and influences the strength of the ibm . for large @xmath255 the nonlinear saturation is acting faster and stronger . the two effects lead to a rather nondistinctive result concerning the peak elm flux as shown in fig . [ f : beta ] , with a continuous transition between itg and ibm triggered onset of the turbulent aftermath . competing growth of itg , microtearing and ibm instabilities determine onset times and intensities of the blowout time traces @xmath229 when scaling with the local plasma beta : for @xmath258 only an itg micro - instability is growing around @xmath259 and saturating on a low h - mode transport level . for @xmath260 the ibm instability dominates and leads to a clean elm signal around @xmath261 . in the nominal case ( @xmath262 ) itg and ibm mhd growth compete initially with similar growth rates but the ideal ballooning mode takes over around @xmath263 and further determines the blowout . for @xmath264 a microtearing instability is saturating on a gyro - bohm transport level around @xmath261 and is transformed into more violent itg mhd turbulence after deterioration of the initial electron temperature gradient ( @xmath265 ) . , width=283 ] the presence or absence of background current gradient terms , which in this model are set by replacing @xmath266 everywhere the electron parallel velocity @xmath267 appears in the equations , with @xmath268 given by the @xmath106 profile , was found to have no discernible effect on the result . indeed self - consistent inclusion of @xmath269 effects with possible role in the l - to - h transition was among the motivations of the gemr model in the first place @xcite . however , the magnitude of even an impressive pedestal current peak is only one to three times the nominal saturation current given by @xmath270 , while values closer to @xmath271 are required to enter the energetics effectively . since @xmath272 for the nominal case , current gradient effects are very weak . the main conclusion of this study is that the qualitative nature of the saturation and aftermath of the initial ibm blowout is the same as for generic edge turbulence given a small - amplitude start . the transition from linear mode structure and energetics to turbulence found for these blowout cases is the same as in ref . only the nature of the linear mode itself differs . the blowout saturates upon its own self generated drift - alfvn turbulence , with a strong ion temperature component given the gradients . the vorticity spectrum reaches quickly to the ion gyroradius ( @xmath0 ) scale , requiring the gyrofluid model and explaining the numerical difficulties seen with braginskii models it can be argued that the cases given in ref . @xcite crash on entry to the nonlinear stage . convergence in the aftermath requires resolving at least @xmath0 . unfortunately , due to the lack of a self consistent h - mode state in a well resolved computation , no threshold is found . at lower beta values one simply finds generic edge turbulence driven by the temperature gradients . it is not clear that this scenario really describes an actual elm , although the energetic growth and decay curves are not unrealistic . however , with a well resolved transition to nonlinearity in both the energetic peak and aftermath phases , our studies find no evidence of an explosive mhd phenomenology . indeed it can be argued that nonlinear mhd processes are pre - empted by the efficient transfer to smaller scales through the two - fluid drift wave physics more commonly associated with microturbulence . this work was supported by the austrian science fund fwf under contract y398 , by a junior research grant ( `` nachwuchsfrderung '' ) from university of innsbruck , and by the european communities under the contract of associations between euratom and the austrian academy of sciences , carried out within the framework of the european fusion development agreement . the views and opinions herein do not necessarily reflect those of the european commission . h. strauss , l. sugiyama l. , c.s . chang , et al . 21st iaea fusion energy conf . ( chengdu , china , 2006 ) ( vienna : iaea ) cd - rom file th / p8 - 6 and http://www-naweb.iaea.org/napc/physics/fec/fec2006/html/index.htm

correlated quantum many - body systems exhibit a wide range of unconventional phenomena . their rich physics emerges from the intricate entanglement structure of those systems , which however at the same time renders their theoretical study a challenging task . some of the most important insights into the physics of those systems have thus been obtained through wavefunction ansatzes , such as the bcs or the laughlin state . in recent years , ideas from quantum information have led to classes of ansatz wavefunctions constructed to capture the entanglement structure present in interacting quantum systems @xcite . in particular , projected entangled pair states ( peps ) @xcite describe correlated many - body systems by associating a local tensor to each site which builds up the entanglement in the wavefunction through auxiliary indices . peps provide a faithful approximation for low - energy states of systems with local interactions @xcite , making them the basis of powerful variational algorithms @xcite . at the same time , they form a framework for the construction of models : from the local tensor , one can construct parent hamiltonians which inherit its symmetry structure and have a global wavefunction built from the tensor as their ground state @xcite . thus , they can be used to construct `` peps models '' ( fig . [ fig : peps]a ) , where the physics is encoded in a tensor @xmath1 which gives rise to a local hamiltonian @xmath2 and a ground state wavefunction @xmath3 on the same footing , forming a versatile framework for the study of correlated quantum systems @xcite . within this framework , the key problem is to understand how the local properties of the tensor ( and thus the hamiltonian ) determine the global properties , i.e. , quantum order , of the wavefunction . in the last years , considerable progress has been made in the study of topological order @xcite in peps models : it has been understood how topological order is related to the structure of the local tensor , and how this allows to construct _ all _ topological ground states from a _ single _ tensor @xcite . more recently , the mechanism behind topological phase transitions within this framework has been clarified @xcite , and it has been found that it can be related to symmetry breaking in the so - called transfer operator @xcite . finally , peps models have also been shown to provide a natural framework to study the entanglement properties of correlated quantum systems through entanglement hamiltonians @xcite associated to the boundary of the system @xcite . * ( a ) * peps models use a tensor @xmath1 to provide a description of a wavefunction @xmath3 and an associated parent hamiltonian @xmath2 on the same footing . * ( b ) * properties of the tensor such as on - site symmetries are inherited by @xmath2 . * ( c ) * the transfer operator @xmath4 encodes all correlations of the wavefunction . ] the aim of this paper is to initiate the development of an analogous framework for studying conventional long - range order and spontaneous symmetry breaking within peps models , thereby complementing the understanding of topological phases . we consider peps models where the local tensor carries a @xmath0 on - site symmetry , and investigate _ i ) _ under which conditions long - range order emerges , and _ ii ) _ how the different symmetry broken states can be described in terms of the single peps tensor . we first show that long - range order originates in a degeneracy of the transfer operator , and how the latter in turn leads to spontaneous breaking of the symmetry under fluctuations . we subsequently characterize the symmetry broken states in terms of the peps tensor , and show that they correspond to the extreme points of the degenerate fixed point space of the transfer operator . we demonstrate that the symmetry broken peps wavefunctions can be used to determine the spontaneous magnetization , and that they give rise to a local entanglement hamiltonian , in contrast to previous findings @xcite , which establishes that all gapped phases can be assigned a short - ranged entanglement hamiltonian . let us start by introducing peps . w.l.o.g . , we restrict to a square lattice with @xmath5 sites . a peps model is described by a @xmath6-index tensor @xmath7 , @xmath8 , @xmath9 , with @xmath10 the physical dimension at each site and @xmath11 the _ bond dimension_. the peps wavefunction @xmath12 is obtained by arranging the tensors on the lattice and contracting the _ virtual indices _ @xmath13 of adjacent tensors as indicated by lines in fig . [ fig : peps]a , yielding @xmath14 . we will consider systems on a long cylinder ( or torus ) of circumference @xmath15 , such that the dependence on the boundaries becomes negligible . to any peps , one can construct so - called _ parent hamiltonians _ which have the corresponding peps as its ground state . parent hamiltonians ensure that the wavefunction on a small ( e.g. , @xmath16 ) patch looks correct , i.e. , as being built from the tensor @xmath1 , which is e.g.accomplished by a projector onto the span of the tensors on that patch ; under generic conditions , these hamiltonians have a unique finite - volume ground state @xcite . by blocking all @xmath15 tensors in a column , a peps can equally be considered as a quasi-1d tensor network ( a matrix product state , mps ) with tensors @xmath17 and bond dimension @xmath18 . in mps , a central role is played by the transfer operator @xmath19 ; in particular , @xmath20 appears in correlation functions at distance @xmath21 , whose decay is therefore governed by the spectrum of @xmath4 . for peps , @xmath4 has itself a 1d structure , cf . [ fig : peps]c . this leads to a more complex behavior , since in the thermodynamic limit the system grows simultaneously in both directions , and thus , the dimension of the space on which @xmath22 acts grows exponentially with the system size . let us now turn towards systems with on - site symmetries . in peps models , symmetries are encoded locally in the tensor : a symmetry action @xmath23 on the physical level translates to an action @xmath24 on the virtual system in a way where it cancels out when contracting tensors , see fig . [ fig : peps]c , giving rise to an invariant wavefunction @xmath25 @xcite . this induces a symmetry @xmath26=0 $ ] of @xmath4 , which is therefore block - diagonal in a basis of irreducible representations ( irreps ) of @xmath24 . at the same time , the parent hamiltonian @xmath2 enjoys by construction the same symmetry , i.e. , we obtain a peps model with symmetry @xmath27=0 $ ] . an instructive example is the `` ising peps '' @xcite with @xmath28 , where the ket ( bra ) corresponds to the physical ( virtual ) indices , and @xmath29 , @xmath30 . this model has a @xmath31 symmetry with non - trivial action @xmath32 . the wavefunction is of the form @xmath33 with @xmath34 the classical 2d ising model ; it thus has the same @xmath35 correlation functions and therefore undergoes a second - order phase transition at @xmath36\approx 0.349596\ .\ ] ] as we have seen , peps form a natural framework to model systems with on - site symmetries : by encoding the symmetry locally into the tensor , we obtain a local hamiltonian with the same symmetry which depends smoothly on the parameters of the tensor , and whose ground state wavefunction can be constructed from the very same tensor . however , we have also found that by changing parameters , such a model can undergo a phase transition to a symmetry broken phase . nevertheless , the peps wavefunction @xmath37 , which is the unique ground state of the system , remains invariant under the symmetry throughout the phase diagram , @xmath38 ; e.g. , for the ising peps at zero temperature ( @xmath39 ) , we have @xmath40 . this leads to the central question of this work : how can we characterize symmetry breaking in peps models , and how can we construct the distinct symmetry broken states starting from a single _ symmetric _ tensor @xmath1 ? let us for a moment leave aside peps models and consider the analogous question for general hamiltonians with a symmetry : generically , these systems have a unique finite volume ground state which therefore carries the full symmetry of the hamiltonian . since the average magnetization @xmath41 in those states is zero , the ordered phase is characterized through non - vanishing long - range order @xmath42 . in order to obtain the symmetry broken states , one couples the system to a small external field @xmath43 and considers the ground state @xmath44 , taking first the limit @xmath45 and subsequently @xmath46 ; if in the limit @xmath47 , one says that the system exhibits spontaneous symmetry breaking . how are these two notions of long - range order and spontaneous symmetry breaking related , and can one construct an approximate symmetry broken state for a finite system ? given a unique finite - volume ground state with long - range order such as @xmath48 and order parameter @xmath49 , one can construct an orthogonal state @xmath50 whose energy approaches the ground state as @xmath51 @xcite . a generic local perturbation will select a unique ground state from the space spanned by @xmath37 and @xmath52 , assuming a gap above . in fact , it can be proven that the only states which do not exhibit long - range order for any observable @xmath53 with @xmath54 , i.e. , which are stable under arbitrary perturbations and thus form the symmetry broken states , are @xmath55 @xcite ; this is , the symmetry broken states can be constucted from the symmetric ground state @xmath37 alone . it is straightforward to see from this argument that the value of the long - range order is just the spontaneous magnetization squared . it should be pointed out , however , that the above argument relies on the validity of the perturbative treatment which can not be justified rigorously in the limit @xmath45 , as the _ total _ perturbation diverges , and in fact , the equality between the two notions of symmetry breaking and the values of the corresponding order parameters has not been proven rigorously except for very few cases such as the ising model @xcite . in the following , we will follow a very similar reasoning in order to relate long - range order and spontaneous symmetry breaking in peps , and in particular to show how to construct the set of symmetry - broken states ( i.e. , those which are stable under arbitrary local perturbations ) starting from a single object , namely the unique symmetric peps wavefunction @xmath37 and its associated symmetric tensor @xmath1 . here , a central role will be played by the one - dimensional transfer operator and in particular its spectral properties and eigenstates , partly taking the role of the hamiltonian in the preceding discussion . we will start by showing that long - range order in the wavefunction is closely related to an approximate degeneracy in the spectrum of the transfer operator . subsequently , we study the behavior of the transfer operator under perturbations , and we show that _ arbitrary _ perturbations induce a splitting of the degenerate subspace in a _ fixed _ basis independent of the perturbation which therefore describes the symmetry broken states of the model . in the following discussion , we will restrict to the symmetry group @xmath31 , and denote the physical ( virtual ) symmetry action by @xmath56 ( @xmath57 ) . let us first see how long - range order in a peps model with @xmath0 on - site symmetry is related to the properties of the underlying tensor and in particular the transfer operator . we start by defining long - range order ( in analogy to ref . @xcite ) : we say that @xmath37 has long - range order if there exists a local operator @xmath58 with @xmath59 and @xmath60 ( the _ order parameter _ ) , and @xmath61 s.th . for sufficiently large @xmath15 , @xmath62 where the sum runs over all sites i.e . , on an infinite cylinder , the spins are correlated at least over a distance proportional to the circumference . [ away from fixed point wavefunctions , we indeed can not expect correlations along the cylinder over arbitrary distances ; on the other hand , e.g. in the ising peps correlations only break down after a distance @xmath63 . ] note that @xmath49 can depend on the lattice site , such as in the case of an antiferromagnet . in such a case , we will also have to restrict the possible values of @xmath15 accordingly , e.g. to even numbers ( or we have to block tensors ) , and we will tacitly assume this in the following . ( corresponding restrictions on @xmath64 will not be necessary in the limit @xmath65 away from fixed point wavefunctions . ) in 1d , i.e. , mps , long - range order is in one - to - one correspondence with a degeneracy of the transfer operator @xmath4roughly speaking , the only way to build long - range correlations over arbitrary distances using the constant virtual dimension is an exact degeneracy of the leading eigenvalue of @xmath22 . in 2d , however , the situation is much less clear due to the exponentially growing dimension of the virtual space on which @xmath4 acts . however , as we will show , under certain conditions eq . ( [ eq : lro ] ) still implies the existence of an almost degenerate second eigenvalue . to this end , define a `` dressed '' transfer operator @xmath66 , where we insert @xmath49 on the physical index at position @xmath67 , and correspondingly @xmath68 ; such operators arise in expectation values such as eq . ( [ eq : lro ] ) . assume that @xmath4 is diagonalizable , @xmath69 with @xmath70 . here , we use @xmath71 to denote ( eigen-)vectors on the level of the transfer operator , i.e. , on the virtual indices ; due to the ket - bra structure of the transfer operator , these vectors can themselves be regarded as operators , in which case we will write them without brackets . since @xmath72=0 $ ] , the eigenvectors of @xmath22 transform even or odd under the symmetry ( i.e. , as different irreps ) , and we will denote the largest eigenvalue in the even ( odd ) symmetry sector by @xmath73 ( @xmath74 ) , with corresponding eigenvectors @xmath75 . we thus have that @xmath76 , or ( if interpreting the eigenvectors as operators ) @xmath77 note that this implies that @xmath78 can not be positive semidefinite , and since @xmath22 must have a positive semidefinite fixed point ( as it is a completely positive map ) , it follows that @xmath79 . our goal is to show that @xmath73 and @xmath74 , as a function of @xmath15 , become degenerate as @xmath80 . we can thus assume that @xmath81 ( otherwise there is nothing to show ) . we will additionally restrict to the case where the largest eigenvalue @xmath73 is non - degenerate also within the even parity sector , since an exact degeneracy at _ finite _ @xmath15 hints an additional symmetry in the wavefunction which we would have to incorporate for a full description . this implies @xmath82 @xcite and we can w.l.o.g . normalize our tensors such that @xmath83 . we can now re - express eq . ( [ eq : lro ] ) as @xmath84 } { \operatorname{tr}[\mathbb t^{n_h } ] } + 1 \ , \ ] ] where the additive term @xmath85 stems from the case where both @xmath49 s are in the same column , and using @xmath86 . in order to simplify the r.h.s . , we split the sum at @xmath87 and use cyclicity of the trace , together with @xmath88 , to obtain two identical sums , @xmath89 } { \operatorname{tr}[\mathbb t^{n_h } ] } + 1\ .\ ] ] ( if @xmath64 is even , the term @xmath87 appears only once , but this bears no relevance . ) since the largest eigenvalue of @xmath22 is non - degenerate , we have that @xmath90 where @xmath91 upper bounds the second largest eigenvalue of @xmath22 @xcite . ( note though that @xmath92 can heavily depend on @xmath15 and properties of @xmath22 . ) using a sequence of triangle inequalities and taking the limit @xmath65 ( see appendix for details ) , we can then replace @xmath93 and obtain @xmath94 + 1 \label{eq : lro - top - fpt } \\ & = 2\sum_i \sum_{p=0}^\infty \lambda_i^p \underbrace{\rbra{l_+ } \mathbb t_{\hat z}\rket{r_i } \rbra{l_i } \mathbb t_{\hat z}\rket{r_+}}_{=:m_i}+1\ , \nonumber\end{aligned}\ ] ] where in the last step , we have expanded @xmath22 in its eigenbasis . as @xmath95 anticommutes with the symmetry , @xmath96 for eigenvectors @xmath97 from the even sector . thus , only eigenvectors from the odd sector contribute to the sum . for those eigenvectors , @xmath98 , and therefore @xmath99 if now @xmath100 , we have latexmath:[$\sum_i ( since @xmath86 ) and thus @xmath102 i.e. , @xmath103 : we find that long - range order implies that the gap of @xmath4 closes in the thermodynamic limit at least as @xmath104 . the required condition @xmath100 is automatically satisfied if @xmath4 and @xmath95 are hermitian , which in particular holds if the tensor and thus the peps model is invariant under combined reflection and time reversal , but turns out to be true also for a range of other models on the r.h.s . of eq . ( [ eq : lro ] ) might not be strong enough to bound the gap without extra assumptions : the convergence of @xmath105 to its fixed point can be roughly bounded by @xmath106 , where @xmath107 and @xmath108 is the dimension of the space @xmath4 acts on @xcite . correlations are thus lost at a scale set by @xmath109 ; in order for a lower bound @xmath110 to imply a bound on @xmath111 we thus need @xmath112 . ] . we thus find that under certain conditions , long - range order implies that the transfer operator has approximately degenerate eigenvalues @xmath113 in the even and odd sector , together with non - vanishing matrix elements @xmath114 . conversely , an approximately degenerate eigenvalue in the odd sector , together with non - vanishing matrix elements , will clearly give rise to long - range order . let us now investigate spontaneous symmetry breaking in peps models with long - range order , and in particular how to construct the symmetry broken states . the symmetry broken states are those states which remain ground states under generic perturbations , where for a given perturbation , we first need to take the thermodynamic limit and subsequently take the strength of the perturbation to zero . in the context of peps models , the natural perturbations to consider are perturbations of the tensor which can be realized by acting solely on the physical index , @xmath115 with @xmath116 , as these both correspond to small perturbations of the parent hamiltonian and leave us in the peps manifold @xcite . such a perturbation gives in turn rise to a perturbation of the transfer operator , @xmath117}$ ] , and thus the symmetry broken states will exactly correspond to boundary conditions given by the stable fixed points of @xmath118}$ ] in the appropriate limit . how will the transfer operator of a system with long - range order respond to such perturbations ? to this end , consider the limit @xmath80 where @xmath119 , and let us assume that @xmath4 has a gap below the two degenerate eigenvalues , @xmath120 . a sufficiently small perturbation will then induce a splitting within this subspace without mixing it with lower - lying eigenvectors ; the basis in which this splitting occurs is independent of the strength of @xmath121 and thus leaves us with a single fixed point @xmath118}^{n_h}\sim \rket{r_\uparrow}\rbra{l_\uparrow}$ ] in the limit @xmath65 , even if later @xmath122 . this fixed point is therefore the boundary condition for the symmetry broken state under perturbation @xmath121 . naively , one might expect that the symmetry broken fixed point sensitively depends on the exact form of the perturbation chosen . however , as we will show in the following , the symmetry broken fixed point @xmath123 is independent of @xmath121 . we will first consider the case of a hermitian transfer operator , i.e. , @xmath124 and @xmath125 , and a hermiticity - preserving perturbation . a generic such perturbation to @xmath22 will induce a splitting of the two degenerate eigenvalues in an orthogonal basis @xmath126 since @xmath127 are non - degenerate eigenvectors of @xmath128}$ ] , which is a completely positive ( cp ) map , they must be hermitian , and thus @xmath129 . since @xmath118}$ ] is a completely positive map , its leading ( non - degenerate ) eigenvector must be a positive operator , @xmath130 using eq . ( [ eq : eigvecsym ] ) and the fact that conjugation preserves positivity , we find that also @xmath131 since for a pair of positive operators @xmath132 , it holds that @xmath133\ge0 $ ] , it follows that @xmath134\\ & = ( r_+\rket{r_+ } - \gamma^2 ( r_-\rket{r_-}\ , \end{aligned}\ ] ] and thus @xmath135 . by changing the sign of @xmath121 , we can exchange the two leading eigenvectors , i.e. , @xmath136 becomes the leading eigenvector , and thus @xmath137 . following the same line of reasoning as before , this yields @xmath138 , and thus , @xmath139 . we thus find that regardless of the perturbation , the transfer operator always acquires the same pair of fixed points @xmath140 solely as a consequence of its complete positivity . note that @xmath127 are the extremal positive states @xmath141 , i.e. , the stable symmetry broken states are those where the symmetry is maximally broken , as intuitively expected . let us now consider the non - hermitian case . since @xmath142 , it follows that @xmath143 where hermiticity of the eigenvectors implies @xmath144 , and we w.l.o.g . choose @xmath145 . again , complete positivity of @xmath146}$ ] and @xmath147}$ ] , together with the possibility to change the ordering of eigenvectors by changing the sign of @xmath121 , implies that @xmath148 . to determine @xmath149 , consider @xmath150 and let @xmath151 be the orthogonal projector onto @xmath152 . complete positivity of @xmath22 implies that @xmath153 s.th . @xmath154 , but @xmath155 . applying @xmath22 repeatedly to @xmath156 would therefore converge to the non - positive @xmath157 , @xmath158 , which would be in contradiction to @xmath159 being completely positive , i.e. , mapping positive operators to positive operators . ] , and therefore @xmath160 , i.e. , @xmath161 . define @xmath162 since @xmath163 = \operatorname{tr}[\pi_{\mathcal s } l^\dagger_{\uparrow\downarrow}\pi_{\mathcal s}r_{\uparrow\downarrow}]= \operatorname{tr}[l^\dagger_{\uparrow\downarrow } r_{\uparrow\downarrow}]= ( l_{\uparrow\downarrow}|r_{\uparrow\downarrow})=1 $ ] , we have that @xmath164 as @xmath165 , and @xmath166 , @xmath167 , it follows that @xmath168 ( @xmath169 ) is contained in the negative ( positive ) eigenspace of @xmath78 , which implies that @xmath170 moreover , since @xmath171= \operatorname{tr}[l_{\uparrow\downarrow}r_{\downarrow\uparrow}]= ( l_{\uparrow\downarrow}|r_{\downarrow\uparrow})=0 $ ] , we have @xmath172 , and thus @xmath173 ( as @xmath174 ) . eq . ( [ eq : supp - ker - s ] ) has two implications : first , together with eq . ( [ eq : ltilde - ne-0 ] ) , @xmath175 , it shows that @xmath176 . since @xmath177 [ eq . ( [ eq : s - def ] ) ] , this implies that @xmath127 are again the extremal points of the positive cone @xmath178 . second , by combining eq . ( [ eq : supp - ker - s ] ) with eq . ( [ eq : ker - rup - rdn - orth ] ) , we find that @xmath179 , and thus @xmath180 = \operatorname{tr}[\pi_\mathcal s(l_+ + \lambda l_-)\pi_\mathcal s(l_+-\lambda l_-)]\ , \ ] ] which allows us to determine the corresponding value of @xmath149 as @xmath181/ \operatorname{tr}[(\pi_{\mathcal s}l_-)^2])\big]^{1/2}$ ] . together , this proves that also in the non - hermitian case , the fixed points of the transfer operator under perturbations , i.e. , the symmetry broken states , are uniquely determined independent of @xmath121 , and correspond to the extremal symmetry - broken states . having understood the structure of the symmetry broken states in the thermodynamic limit , let us now investigate the accuracy of our result for finite systems , and use our findings to re - examine the entanglement spectra and hamiltonians of models with symmetry breaking . to this end , we have performed numerical simulations for two models . first , we have considered the ising peps introduced in sec . [ sec : peps - models ] . as its transfer operator is isometric to the transfer operator of the 2d classical ising model and it exhibits the same @xmath35 correlation functions , it allows us to benchmark our numerical findings against exact results . as a second model , we have studied the square lattice aklt model with nematic field @xcite . the aklt model @xcite is constructed by placing spin-@xmath182 singlets on the links and subsequently projecting them onto the spin @xmath183 subspace , i.e. , @xmath1 is of the form @xmath184 , where @xmath185 projects onto the spin-@xmath183 space ; the model can be mapped to a ferromagnetic model through a sublattice rotation . subsequently , a `` nematic field '' @xmath186 , @xmath187 favoring large values of @xmath188 is applied to the aklt tensor which yields a model with @xmath189 symmetry and eventually leads to breaking of the @xmath31 symmetry @xcite . numerical results for the ising peps . * ( a , b ) * error analysis for @xmath190 , @xmath191 : a ) scaling of the error @xmath192 between the fixed point of @xmath118}$ ] , @xmath193 , and of its projection onto @xmath194 , @xmath195 ; we find @xmath196 . ( the data for different @xmath15 almost exactly coincide . ) b ) the deviation of @xmath197 from @xmath198 vanishes exponentially in @xmath15 . * ( c ) * magnetization , computed from the symmetry broken states @xmath199 , for @xmath200 ( blue ) , extrapolated curve ( red ) , and exact solution from the correspondence to the classical 2d ising model ( dashed green ) . ] in all cases , the simulations have been carried out through an exact diagonalization of the transfer operator ( cf . , e.g. , ref . @xcite ) , where the transfer operator @xmath22 is applied to a vector by sequential contraction of indices rather than by building its matrix representation . thus , the largest objects which need to be stored are on the order of the size of an eigenvector , @xmath201 , rather than that of the transfer operator , @xmath202 ; at the same time , one obtains results with machine precision , allowing for an accurate scaling . our derivation involved a number of assumptions , in particular neglecting the coupling to eigenstates other than @xmath194 in second and higher order perturbation theory even in the limit @xmath203 , and neglecting the splitting between @xmath204 . in order to assess the validity of neglecting the coupling to the other levels , we have compared the fixed point @xmath193 of the perturbed transfer operator @xmath118}$ ] , with @xmath121 a general perturbation , to the fixed point @xmath195 obtained after projecting @xmath118}$ ] onto @xmath205 and @xmath206 . [ fig : ising - conv - mag]a shows the result for the ising peps , where we find that the error @xmath207 scales as @xmath208 . this scaling is consistent with a second order perturbation treatment , where one assumes that the local perturbation @xmath121 has only a local effect on @xmath209 , and thus can only be undone by a term in its vicinity , leading to @xmath210 contributions , rather than @xmath211 , of amplitude @xmath212 each . the same behavior is found for the nematical aklt model , cf . [ fig : aklt - top - mag]a . we have subsequently also verified the validity of neglecting the splitting between @xmath204 : fig . [ fig : ising - conv - mag]b shows that the exact solution for the two - dimensional degenerate subspace , @xmath195 , in turn converges exponentially to @xmath213 as @xmath203 , consistent with an exponentially vanishing splitting @xmath214 . nematic aklt model . * ( a ) * error analysis , cf . [ fig : ising - conv - mag]b . * ( b ) * critical exponents for the correlation length , determined as @xmath215 in the disordered phase and @xmath216 from the magnetization fit ( cf . [ fig : ising - conv - mag]c ) in the symmetry - broken phase . * ( c ) * magnetization @xmath217 for @xmath200 and extrapolated value ( cf . [ fig : ising - conv - mag]c ) ; the inset shows the critical exponent of @xmath217 . ] the knowledge of the boundary conditions @xmath213 and @xmath218 of the symmetry broken states allows us to compute the approximate spontaneous magnetization for finite @xmath15 . for the ising peps , we find that the finite @xmath15 data reproduces the analytic expression for the infinite system very well away from the critical point , with the error vanishing exponentially in both @xmath15 and the distance from the critical point ; this is consistent with the expectation that the error should vanish as @xmath219 with @xmath220 the correlation length . on the other hand , there are strong finite size effects as we approach the critial point , as shown in fig . [ fig : ising - conv - mag]c ; indeed , it is natural to expect that finite size effects will dominate as soon as @xmath221 . nevertheless , it is still possible to accurately extrapolate the magnetization curve to the thermodynamic limit even very close to the critical point . to this end , we first determine the behavior of the correlation length in the proximity of the critical point by fitting @xmath222 as a function of @xmath15 , and subsequently fitting @xmath223 quadratically ( shown in the inset ; note that this makes an implicit assumption about the critical exponent of @xmath220 ) ; the critical point from the fit is @xmath224 , as compared to the exact value @xmath225 . the resulting fitting function @xmath226 is then used in the scaling ansatz @xmath227 $ ] , where the algebraic factor @xmath228 accounts for short - range effects independently in a first step is that in the relevant regime where @xmath229 , the second fit does not allow for an accurate determination of @xmath226 and might even return negative values . ] . altogether , this extrapolation yields a very accurate approximation to the analytical curve even in the regime very close to the critical point where finite size effects seemingly dominate , as shown in fig . [ fig : ising - conv - mag]c . we have subsequently applied the same analysis to determine the magnetization curve of the nematic aklt model . the data for @xmath200 and the extrapolated magnetization curve are shown in fig . [ fig : aklt - top - mag]c , and we find a value of @xmath230 for the critical point . using this value of @xmath231 , we have subsequently studied the critical scaling of the model . [ fig : aklt - top - mag]b shows the critical scaling of two different correlation lengths@xmath232 , obtained by fitting the magnetization as a function of @xmath15 as @xmath233 , and @xmath234 , obtained by an exponential fit @xmath235 of the correlation length @xmath236 extracted from the gap of the transfer operator both of which yield a critical exponent of @xmath237 for the correlation length . the inset of inset of fig . [ fig : aklt - top - mag]c shows the corresponding analysis for the magnetization , which is compatible with a critical exponent @xmath238 , therefore suggesting that the nematic aklt model is in the 2d ising universality class . interaction strength vs. interaction range of @xmath239 @xcite for * ( a ) * the ising peps ( @xmath240 ) and * ( b ) * the nematic aklt model ( @xmath241 ) , where blue = trivial phase , green = symmetry broken phase , and red = critical point . the insets show @xmath239 computed from @xmath242 , where the interaction becomes long - ranged after the phase transition . the main panels show @xmath239 derived from the symmetry broken states ( gray lines indicate the data from @xmath242 ) , and we find that in both cases , @xmath239 becomes more local again after the phase transition . ] let us now turn towards entanglement spectra and the entanglement hamiltonian . the entanglement spectrum of a state on e.g. an infinite cylinder is given by the spectrum of the reduced density operator @xmath243 of the half - infinite cylinder , and its low - energy part can be interpreted as the spectrum of a gibbs state @xmath244 of an `` entanglement hamiltonian '' @xmath239 @xcite . in the context of peps models , it has been shown @xcite that the spectrum of @xmath243 is equal to the spectrum of the symmetrized fixed point of @xmath4 , @xmath245 where @xmath246 denotes a possible label of the fixed point , such as in the case of multiple sectors . in turn , @xmath247 allows to define the entanglement hamiltonian through @xmath248 . in ref . @xcite , @xmath239 has been studied for both the ising peps and the nematic aklt model based on the unique fixed point of @xmath22 for finite @xmath15 , i.e. , @xmath249 , and it has been found that it is ( quasi-)local ( i.e. , the interactions decay exponentially with distance ) in the trivial phase , the interaction length diverges at the phase transition , and @xmath239 remains long - ranged in the symmetry - broken phase ; the corresponding results for both models are shown in the insets of figs . [ fig : enthams]a , b . however , as we have argued , in the symmetry broken phase the physically meaningful states , i.e. , those which are stable under arbitrary perturbations , are the symmetry broken ones , and we should therefore rather use the symmetry broken fixed points @xmath250 when determining @xmath239 . in fig . [ fig : enthams ] , we compare the interaction range of @xmath239 derived from @xmath250 , @xmath251 is not positive , we have used @xmath252 , with @xmath253 . ideally , @xmath254 should be choosen as the most negative eigenvector of @xmath251 , but we have used @xmath255 which is a very good approximation to it . ] with the one obtained from @xmath249 ( insets ) for the ising peps ( fig . [ fig : enthams]a ) and the nematic aklt model ( fig . [ fig : enthams]b ) : we find that by considering the latter , the locality of @xmath239 in the symmetry broken phase is restored , and @xmath239 diverges only at the phase transition , in accordance with the intuition that the interaction length of the entanglement hamiltonian should reflect the characteristic length scale of the system , and thus should be finite away from the critical point . together with previous findings on how to recover locality of @xmath239 in topological phases @xcite , this shows that by considering the physically meaningful ( this is , stable ) fixed points , a local entanglement hamiltonian can be obtained for _ all _ gapped phases , as intuitively expected . we have studied the occurence of long - range order and spontaneous symmetry breaking in peps models . we have shown that long - range order is closely related to a degeneracy in the transfer operator , and have characterized the symmetry broken ( i.e. , stable ) states in terms of the fixed points @xmath256 of the transfer operator , which we found to be the extremal positive semi - definite states @xmath257 irrespective of the model and the perturbation . these fixed points do not only yield the physically relevant boundary conditions in the symmetry broken phase , but each one by itself already carries the full information about _ both _ fixed points , as @xmath258 . moreover , they give rise to local entanglement hamiltonians , thereby establishing locality of the entanglement hamiltonian for all gapped phases . we acknowledge helpful conversations with ignacio cirac , david perez - garcia , volkher scholz , and frank verstraete . this work has been supported by the alexander von humboldt foundation , the erc grant wascosys , and by jara - hpc through grants jara0092 and jara0111 . 44ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1016/j.aop.2010.09.012 [ * * , ( ) ] , @noop ( ) , @noop * * , ( ) , @noop * * , ( ) , @noop * * , ( ) , @noop * * , ( ) , @noop * * , ( ) , @noop ( ) , @noop * * , ( ) , @noop * * , ( ) , @noop * * , ( ) , link:\doibase 10.1016/j.aop.2010.05.008 [ * * , ( ) ] , link:\doibase 10.1103/physrevlett.59.799 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop @noop * * , ( ) , @noop * * , ( ) , @noop * * , ( ) , @noop @noop ( ) , @noop _ _ ( , ) @noop ( ) , @noop ( ) , @noop * * , ( ) , @noop ( ) , link:\doibase 10.1103/physrevlett.101.010504 [ * * , ( ) ] , @noop * * , ( ) , @noop * * , ( ) , @noop * * , ( ) , @noop * * , ( ) link:\doibase 10.1007/bf02188685 [ * * , ( ) ] , @noop _ _ , vol . ( , ) @noop * * , ( ) , @noop * * , ( ) link:\doibase 10.1007/s00220 - 014 - 2188 - 5 [ * * , ( ) ] , link:\doibase 10.1103/physrevb.88.115108 [ * * , ( ) ] , @noop * * , ( ) , ' '' '' in this appendix , we give a detailed derivation of the equality of the r.h.s . of eqs . ( [ eq : lro - top ] ) and ( [ eq : lro - top - fpt ] ) . we start by splitting @xmath259 } { \operatorname{tr}[\mathbb t^{n_h } ] } = s(0,n_\mathrm{cut } ) + s(n_\mathrm{cut}+1,n_h-2)\ , \ ] ] where @xmath260 , and @xmath261 } { \operatorname{tr}[\mathbb t^{n_h } ] } \ .\ ] ] due to cyclicity of the trace , @xmath262 , where @xmath263 and thus @xmath264 , depending whether @xmath64 is even or odd . we will now show that @xmath265\ .\ ] ] to this end , we will make use of eq . ( [ eq : top - exp - conv ] ) , @xmath266 where @xmath91 , as well as @xmath267\big|\le \|\mathbb t_{\hat z}\|_\mathrm{op}\ , \|\mathbb t^p\|_\mathrm{op}\ , \|\mathbb t_{\hat z}\|_\mathrm{op}\ , \|x\|_\mathrm{tr } \le \zeta \|x\|_\mathrm{tr}\ ] ] with @xmath268 , which can be shown using hlder s inequality , the submultiplicativity of the operator norm , and @xmath269 and finally @xmath270\big|\ge \big|\operatorname{tr}[|r_+)(l_+|]\big|- \big|\operatorname{tr}[\mathbb t^{n_h}-|r_+)(l_+|]\big| \ge 1 - \big\|\mathbb t^{n_h}-|r_+)(l_+|\big\|_\mathrm{tr } \ge 1 - c\gamma^{n_h}\ .\ ] ] we now have @xmath271 } { \operatorname{tr}[\mathbb t^{n_h } ] } - \operatorname{tr}[\mathbb{t}_{\hat z}\mathbb{t}^p \mathbb t_{\hat z } |r_+)(l_+| ] \right\rvert \\[1ex ] & \le \left\lvert \frac{\operatorname{tr}[\mathbb{t}_{\hat z}\mathbb{t}^p \mathbb t_{\hat z } \mathbb t^{n_h - p-2 } ] } { \operatorname{tr}[\mathbb t^{n_h } ] } - \frac{\operatorname{tr}[\mathbb{t}_{\hat z}\mathbb{t}^p \mathbb t_{\hat z } |r_+)(l_+| ] } { \operatorname{tr}[\mathbb t^{n_h } ] } \right\rvert + \left\lvert \frac{\operatorname{tr}[\mathbb{t}_{\hat z}\mathbb{t}^p \mathbb t_{\hat z } |r_+)(l_+| ] } { \operatorname{tr}[\mathbb t^{n_h } ] } - \frac{\operatorname{tr}[\mathbb{t}_{\hat z}\mathbb{t}^p \mathbb t_{\hat z } |r_+)(l_+| ] } { \operatorname{tr}[|r_+)(l_+| ] } \right\rvert \\[1ex ] & \stackrel{\mathclap{(\ref{eq : app : tztptzx } , \ref{eq : app : tn - lowerbnd})}}{\le } \hspace*{1.3em } \frac{\zeta\ , \big\|\mathbb t^{n_h - p-2}-|r_+)(l_+|\big\|_\mathrm{tr}}{1-c\gamma^{n_h } } + \zeta \big\||r_+)(l_+|\big\|_\mathrm{tr}\ , \left\lvert \frac{\operatorname{tr}[|r_+)(l_+|]-\operatorname{tr}[\mathbb t^{n_h}]}{\operatorname{tr}[\mathbb t^{n_h}]\operatorname{tr}[|r_+)(l_+| ] } \right\rvert \\[1ex ] & \stackrel{\mathclap{(\ref{eq : top - exp - conv},\ref{eq : app : tn - lowerbnd})}}{\le } \hspace*{1.3em } \frac{\zeta\ , c\gamma^{n_h - p-2}}{1-c\gamma^{n_h } } + \zeta \big\||r_+)(l_+|\big\|_\mathrm{tr}\ , \frac{c\gamma^{n_h}}{1-c\gamma^{n_h } } \\[1ex ] & \le 2 \zeta c\nu\gamma^{n_h - p-2}\ , \end{aligned}\ ] ] where in the last step we have assumed that @xmath64 is sufficiently large such that @xmath272 , and have introduced @xmath273 . it follows that @xmath274 \right\rvert \le \sum_{p=0}^{n_\mathrm{cut}+\kappa } \delta_p \le \tfrac{n_h}{2}\times 2\zeta c\nu \gamma^{n_h/2 - 1}\ ] ] where we have used @xmath275 and @xmath276 . clearly , the r.h.s . goes to zero as @xmath277 , and thus , @xmath265\ ] ] as claimed .

early radio surveys found that classical radio galaxies ( @xmath2 jy ) evolve strongly with time in both density and luminosity , and are relatively rare in the local universe @xcite . more recent surveys which probe to fainter flux densities have shown that the normalised source count flattens below a few mjy @xcite . this flattening corresponds to a rapid increase in the number of faint radio sources , and has been interpreted as being due to a new population of sources which do not show up at higher flux density levels - the so called sub - mjy population . the existing sub - mjy samples are generally small due to the large amount of observing time required to reach these faint levels . several models , involving different classes of objects , have been developed to explain the observed sub - mjy source counts . these include a strongly evolving population of normal spiral galaxies @xcite or actively star forming galaxies @xcite . a non evolving population of local ( @xmath3 ) low luminosity radio galaxies has also been proposed to explain the sub - mjy population @xcite . it is now thought that the sub - mjy population is comprised of low luminosity agn , normal spirals and ellipticals , as well as starbursts , but the exact mix is unknown . the nature of the sub - mjy population is still not clear because of the difficulty of obtaining optical photometry and spectra for complete radio samples . the follow up optical identification and spectroscopy of sub - mjy radio samples require a lot of telescope time because these sources generally have very faint optical counterparts . conclusions about the sub - mjy sample have been inferred from surveys of varying radio sensitivity and different depths of optical follow up . approximately 41% of sub - mjy ( @xmath4 mjy ) radio sources are identified with optical counterparts to @xmath5 and these are mostly faint blue galaxies with optical - infrared colours indicative of active star formation @xcite . this result is confirmed by a spectroscopic study of optical counterparts ( @xmath6 ) to faint ( @xmath7mjy ) radio sources , which found that most of the sub - mjy sources have spectra similar to star forming _ iras _ galaxies @xcite . however , this result is obtained from spectroscopy of only 10% of their whole radio sample . a more spectroscopically complete sample is the marano field , where spectra were obtained for 50% of the 68 faint ( @xmath8 mjy ) radio sources @xcite . contrary to previous results of @xcite , the majority of the spectra are identified with early type galaxies . gruppioni et al . attribute this discrepancy to the fainter magnitude limit reached in their spectroscopic identifications , since the fraction of radio sources identified with early type galaxies increases abruptly at @xmath9 , which is the limit of the benn et al . sample . about 60% ( 219/386 ) of the faint ( @xmath10 mjy ) radio sources in the australia telescope eso slice project ( atesp , @xcite ) are identified to i = 22.5 mag . the spectra obtained for 70 sources with i @xmath11 19 are dominated by early type galaxies ( 60% ) . starbursts and post - starburst galaxies become important in the sub - mjy ( @xmath12 mjy ) regime , where they make up 39% of the i @xmath11 19 sample @xcite . nevertheless , prandoni et al . find that early type galaxies still constitute a significant fraction ( 25% ) of the sub - mjy radio sources , and sub - mjy samples with fainter spectroscopic limits become increasingly sensitive to early type galaxies . the phoenix deep field survey covers a 4.5 deg@xmath13 at 1.4 ghz and has a rms noise of 12 @xmath1jy over the most sensitive region @xcite . a subregion covering about a degree has been imaged to r @xmath14 24.5 and optical counterparts found for 76% of radio sources within the imaged area @xcite . spectra of the radio sources revealed that the sub - mjy population is made up of at least 32% starforming galaxies and 21% are quiescent galaxies @xcite . however , @xcite only have spectra for 17% of their radio sources , and the spectral completeness is biased to the optically brighter radio sources ( r @xmath15 ) . the most complete result in photometry is in the hubble deep field north ( hdfn ) where the vla was used to follow up observations of a patch of sky observed by the hst to depths of about 30th magnitude in the four hst broadband filters , f300w , f450w , f606w , and f814w ( approximately u , b , v , and i ) . eighty percent of the 79 faint ( @xmath16 mjy ) radio sources in the hdfn region were identified by hst and ground based images reaching @xmath17 @xcite . using optical and radio morphologies along with radio spectral index arguments , richards et al . concluded that 60% of the hdfn radio sources with optical counterparts were star forming galaxies , 20% were classed as agn , and the remainder were ambiguous . in 1998 , the hst observed a region of sky in the southern continuous viewing zone ( cvz ) as a complementary observation to the northern hubble deep field @xcite . this field is known as the hubble deep field south ( hdfs ) @xcite . a region containing a quasar suitable for the search of lyman alpha absorption systems was chosen . the main wfpc field of the hdfs observations reach a magnitude limit of @xmath1430th magnitude in the four hst broadband filters . simultaneous observations with the other hst instruments , nicmos and stis , reach similarly deep levels in the near - infrared and ultraviolet . in addition to the main hst deep fields , a mosaic of nine flanking fields were imaged to shallower levels of @xmath1425th in i ( f814w ) . from the ground , deep wide field images have been obtained , reaching depths of @xmath1425th magnitude in ubvri in a 44 @xmath18 44 arcmin region centered on the hdfs @xcite . spectroscopically , 194 galaxies in the main hdfs and flanking fields were targeted on the very large telescope ( vlt ) and reliable redshifts obtained for 97 targets @xcite . in addition , optical spectroscopy is available for 225 bright ( r@xmath19 ) sources in a @xmath20 arcmin region containing the stis and wfpc fields ( glazebrook , in preparation , but see http://www.aao.gov.au/hdfs/redshifts/ ) . the hdfs was selected as a region for deep radio follow up to take advantage of the wealth of publicly available optical / near - infrared photometric and spectroscopic data . observations over 1998 to 2001 were performed on the australia telescope compact array ( atca ) at all four available frequency bands . approximately 100 hours observing at each band yielded images at 1.4 , 2.5 , 5.2 , and 8.7 ghz with sensitivities of approximately 10 @xmath1jy rms . a detailed description of observations , catalogues and analysis of these australia telescope hubble deep field south ( athdfs ) images is described in norris et al . ( 2004 , hereafter paper i ) . the radio source counts at the faintest flux density levels ( @xmath12 mjy ) are not well determined due to the small number of surveys reaching this sensitivity regime . the sensitivity of our radio survey allows us to determine the source counts down to @xmath140.05 mjy . only the phoenix deep field @xcite and hubble deep field north @xcite reach similarly deep levels . this paper describes the image analysis and source extraction technique of the full 1.4 ghz athdfs survey . this paper is organised as follows . in section 2 we summarise the 1.4 ghz observations and data reduction . various image properties are discussed in section 3 and the source extraction and catalogue is detailed in section 4 . section 5 contains an analysis of the accuracy of our source parameters . the survey completeness is investigated in section 6 and the athdfs 1.4 ghz differential radio source count is presented in section 7 . paper i discussed in detail the observations and data reduction steps used to obtain the athdfs images . here we provide a brief summary . the 1.4 ghz observations were carried out over four years from 1998 to 2001 . they consist of single pointings centred on ra = 22h 33 m 25.96s and dec = @xmath2160@xmath22 38 09.0 ( j2000 ) . a wide variety of atca configurations was employed to ensure maximum _ uv _ coverage . the correlator was set to continuum mode ( @xmath23 mhz bandwidth ) , with each 128 mhz bandwidth divided into @xmath24 mhz channels . the primary flux density calibrator used was pks b1934 - 638 , while secondary gain and phase calibrations were performed through observations of either pks b2205 - 636 or pks b2333 - 528 . the data were inspected using the _ miriad _ interactive task _ tvflag_. after flagging , the 1.4 ghz data were split into the three observing bands ( 1.344 , 1.384 and 1.432 ghz ) before imaging and cleaning . this allowed the cleaning process to do a much better job of removing all the flux of a source . two iterations of both phase and amplitude self - calibration were then performed to improve the image quality . in this section we discuss two systematic effects which have to be taken into account in deriving source flux densities , the clean bias and bandwidth smearing . we use methods similar to those outlined in @xcite to investigate clean bias and bandwidth smearing . a master equation to derive flux densities corrected for these effects is given in appendix a. when the _ uv _ coverage is incomplete the cleaning process can redistribute flux from real sources to noise peaks , producing spurious sources . this effect is generally only a problem for snapshot observations where _ uv _ coverage is poor . although our _ uv _ coverage is very good , tests were performed to check the magnitude of the clean bias in our image . we performed a clean bias check by injecting point sources into the _ uv _ data at random positions . the _ uv _ data were then cleaned to the same rms level as our original image , and source peak fluxes after cleaning were compared to input values . the flux density distribution of injected sources was chosen to be similar to that of the real sources in the image : 40 at @xmath25 , 15 at @xmath26 , 15 at @xmath27 , 15 at @xmath28 , 15 at @xmath29 , 10 at @xmath30 , 3 at @xmath31 , 2 at @xmath32 , 1 at @xmath33 and 1 at @xmath34 . this simulation was repeated 50 times to get reliable number statistics . the results from this process are shown in figure [ fig : clnbias ] . the average source flux measured after the cleaning process ( s@xmath35 ) normalised to the true source flux ( s@xmath36 ) is shown for the various values of input source signal - to - noise ( s@xmath37 ) . it is evident that clean bias only affects the faintest sources . there is a sudden increase in the clean bias at a flux density of about @xmath30 , but even at the faintest bin ( @xmath25 ) the clean bias is only 4% . an analytical form of the clean bias effect was obtained by a least squares fit to the function @xmath38 the best fit values for @xmath39 and @xmath40 are 1.00 and @xmath210.23 , respectively . the resulting best fit curve is plotted in figure [ fig : clnbias ] . bandwidth smearing , the radio analogue of optical chromatic aberration , is a well - known effect caused by the finite width of the receiver channels . it reduces the peak flux density of a source , while correspondingly increasing the source size in the radial direction such that the total integrated flux density is conserved . the amount of smearing is proportional to the distance from the phase centre , and the bandwidth of the observations . in the simplest case of a gaussian beam and passband , the bandwidth smearing can be described by @xcite : @xmath41 where the ratio @xmath42 represents the attenuation of the source peak flux density with respect to that of an unsmeared source , @xmath43 is the passband width , @xmath44 is the observing frequency , @xmath45 is the distance from the phase centre , and @xmath46 is the synthesized beam fwhm . although the observational bandwidth is 4 mhz , the _ miriad _ task _ atlod _ effectively doubles the channel bandwidth . so with the athdfs channel width ( @xmath43 ) of 8 mhz and observing frequency ( @xmath44 ) of 1.4 ghz , equation [ eq : bandsmear ] simplifies to : @xmath47 the bandwidth smearing as a function of distance from the centre is plotted in figure [ fig : bandsmear ] , calculated with the appropriate athdfs beamsize parameter . the majority of our radio sources lie within the inner 10 @xmath18 10 arcmin region , where the bandwidth smearing is less than 5% . the attenuation increases rapidly however , and is @xmath48% at a radial distance of 20 arcmin from the phase centre . to investigate the noise characteristics of our 1.4 ghz image we constructed a noise map . the noise map which contains the pixel by pixel root mean square ( rms ) noise distribution was made using the sextractor package @xcite . sextractor initially estimates the local background in each mesh from the pixel data , and the local background histogram is clipped iteratively until convergence at @xmath49 around its median . the choice of mesh size is very important . when it is too small the background will be overestimated due to the presence of real sources . when it is too large , any small scale variations of the background are washed out . we ran sextractor with a mesh size set to 30 @xmath18 30 pixels , approximately 8 @xmath18 8 beams . a grey scale image of the sextractor noise map is shown in figure [ noisemap ] . as expected , the noise is lowest in the centre and increases with radial distance . the left panel of figure [ noiseprops ] shows the average noise as a function of distance from the centre . this increase in noise is roughly a parabolic shape due to the primary beam attenuation . a histogram of the noise map pixel values is shown in the right panel of figure [ noiseprops ] . this distribution peaks at a value of 10 @xmath1jy , but has a large wing at high noise values ( @xmath50jy ) due to the radially increasing noise . there are also regions of increased noise ( white pixels ) around bright sources which are clearly evident in the grey scale image ( figure [ noisemap ] ) . the increased noise around these sources is due to dynamic range effects , as discussed in paper i. the most affected region is noticeable as a high noise bump at approximately 6 arcmin from the image centre in the left panel of figure [ noiseprops ] . this corresponds to the area surrounding the brightest source ( @xmath51 = 155 mjy ) . the noise in this region is up to 3 times that of the surrounding unaffected region . although sextractor was developed for the analysis of optical data , the noise image obtained with it has been found to be reliable for radio images @xcite . we confirmed the accuracy of the sextractor noise map by comparing pixel values against rms values directly measured from our radio image . the directly measured values were obtained by selecting a set of random positions in the 1.4 ghz image and calculating the rms in the surrounding @xmath52 pixel box . each box was examined for any sources above 3@xmath53 before inclusion in the analysis to ensure that flux from real sources does not effect the statistics . figure [ noisecompare ] shows the mean and standard deviation of the difference between the sextractor pixel values and the directly measured values as a function of distance from the centre . the difference is very close to zero , and there is no systematic effect as a function of radial distance . we therefore concluded that the sextractor noise map is reliable and used it to perform the source extraction . we determined a sensible maximum radial distance for the source cataloging taking into account the following factors : + i ) the radial noise distribution , + ii ) the primary beam attenuation , and + iii ) the bandwidth smearing effect . + these three factors determine the efficacy of detecting a source with a particular flux density at different radial distances from the centre . the primary beam response at a radial distance of 20 arcmin is approximately 39% , resulting in an increased noise of @xmath1420 @xmath1jy , compared to @xmath1410 @xmath1jy at the image centre . furthermore , the bandwidth smearing effect is 18% at this distance and rapidly degrading , as discussed in section [ bandsmear ] . on this basis we choose to catalogue to a maximum distance of 20 arcmin from the image centre . to begin the source extraction we divided the original 1.4 ghz map by the noise map generated by sextractor , obtaining a ` signal - to - noise ' map . sextractor is useful for obtaining signal - to - noise maps , but it has not been reliably used for extracting sources from radio images . the commonly used routines are sfind @xciteand _ imsad_. although sfind may result in fewer false detections @xcite , there are no systematic differences in the source parameters found by either routine @xcite . we used the _ miriad _ task _ imsad _ to derive a preliminary list of source ` islands ' . this task searches for ` islands ' of pixels above a specified cutoff , and attempts to fit gaussian components to the ` islands ' . we executed _ imsad _ on the ` signal - to - noise ' map to detect all islands with a peak flux density level of @xmath54 . this resulted in a list of 688 source ` islands ' for further investigation . each source ` island ' found by _ imsad _ was examined and refitted with an elliptical gaussian to derive source flux densities and sizes . this refitting procedure was performed on the ` real ' image ( not the ` signal - to - noise ' map ) . all sources and fit parameters were visually inspected to check for obvious failures and poor fits that need further analysis . following @xcite , a reference peak value was derived from a second degree interpolation of the source ( _ miriad _ task _ maxfit _ ) . if the difference between the fitted peak and reference peak was less than 20% of the reference value and the fitted position was inside the 0.9@xmath55 flux density contour , then these sources were assigned a fit quality flag of 1 . gaussian integrated fluxes were also compared to the ones directly measured by summing the pixels greater than @xmath56 in the source area . in most cases the gaussian fit provided good values for the position and peak flux densities but not integrated flux densities . these sources are given a quality flag of 2 in the catalogue . there were some problematic cases which were classified as follows : + 1 ) sources fit by _ imsad _ with a single gaussian , but better described by two or more gaussians ( @xmath143% ) , and + 2 ) non gaussian sources not well fitted by a single gaussian , i.e. fit exceeds tolerance criteria described above ( @xmath140.3% ) , and + 3 ) obviously spurious sources which correspond to artefacts or noise peaks ( @xmath141% ) . in the first case above , the _ imsad _ islands were split into two or three gaussian components . the number of of successfully split islands are 19 in total ( 18 with two components , 1 with three components ) . the individual components of these sources are listed in the catalogue . the reference positions and peak flux densities from _ maxfit _ were adopted for the two non gaussian sources . the source position angle was determined by the direction along which it is most extended . the fwhm size of the axes was defined by the distance between two points on the @xmath56 contour parallel to ( major ) and perpendicular to ( minor ) this direction . these sources are flagged `` n '' in the catalogue . after accounting for the multiple components , non - gaussian sources and noise artefacts , we have a list of 693 sources , or source components . figure [ shist ] shows the peak flux density and signal - to - noise distributions of the @xmath54 sample . the spurious sources were low sn , and make up about 4% of the 4 5 @xmath57 sources . so to ensure we have a source catalogue of high reliability we apply a cut of @xmath58 , which includes 12 multiple component sources . there are clear signs of physical association between the individual components of these sources , as shown by the postage stamps in figure [ multi_stamps ] . the final 5@xmath57 catalogue has 466 sources . the ratio of integrated flux to peak flux is a direct measure of the extension of a radio source : @xmath59 where @xmath60 and @xmath61 are fwhm of the source axes , and @xmath62 and @xmath63 the fwhm of the synthesised beam axes . we therefore use this relation to determine whether our sources are resolved or unresolved ( see e.g. @xcite ) . in figure [ fig : deconvplot ] we plot the ratio of the total integrated flux density ( @xmath64 ) and peak flux density ( @xmath55 ) as a function of signal - to - noise ( @xmath65 ) for all the sources in the catalogue . if the values of @xmath66 are purely due to errors introduced by noise in our image , we can derive a criterion for extension by assuming that these statistical errors are also present for @xmath67 sources . we determined the lower envelope in figure [ fig : deconvplot ] which contains 90% of the sources @xmath66 and mirrored it above the @xmath68 line . the upper envelope is characterised by the equation : @xmath69 \ ; . \label{eq : deconv}\ ] ] only sources lying above this upper envelope and successfully deconvolved by _ miriad _ are considered resolved . from this analysis we conclude that 221/466 ( 47% ) of sources are resolved . deconvolved sizes are only given in the catalogue for resolved sources , and other sources have deconvolved sizes set to zero . radio sources can be made up of a nucleus with hot spots along , and at the end of , one or two jets . the individual components of a single source are often catalogued separately by gaussian fitting routines . this will skew number statistics , so a method must be devised to identify multiple components as belonging to a single source . similar to the technique of @xcite , we plot the sum of the fluxes of the components of each nearest neighbour pair versus their separation , as shown in figure [ nearest ] . the high density of points to the bottom right of the @xmath70-flux density plane is made up of the general population of single component sources . the pairs with a separation of about 10 arcsec are most likely subcomponents of single sources . this is evident from figure [ nearest ] where the sources which were successfully split into multiple components in section [ srcext ] , shown as circled large dots , cluster around this separation . we apply a maximum separation that increases with the summed flux density , @xmath71 : @xmath72 this maximum separation is shown as a dashed line in figure [ nearest ] . by varying the allowed separation with summed flux , bright components may be considered a single source even at large separations , while faint sources are kept as single sources . as a further constraint we apply a flux ratio cut to the nearest neighbour pairs . since the flux densities of real double sources are correlated this gives us another criterion to restrict the matched pairs to physically associated sources . we combine pairs only if their flux densities differ by a factor less than 4 . the sources which meet this further requirement are marked in figure [ nearest ] by a large dot . for sources which lie to the left of equation [ summedfluxsep ] , 20/23 pairs with separations less than 20 arcsec pass the flux ratio cut , while only 3/21 pairs with larger separations are passed . so this procedure seems to be successful in that almost all pairs with small separation are considered single sources , while very few distant pairs are passed . from this analysis we identify an additional 11 possible multiple sources . all of these sources were visually inspected and only 5 pairs appear to be real double sources with lobe - lobe or core - lobe morphology . the postage stamps of these pairs are shown in figure [ newmulti ] . these sources are accepted as double sources and flagged in the catalogue . the other 6 pairs are shown in figure [ notmulti ] . they are rejected as double sources as they show no sign of physical association . the full @xmath58 catalogue is presented in table [ catalogue ] . a description of table [ catalogue ] is as follows . _ column ( 1 ) _ source i d . _ column ( 2 ) _ source name . _ column ( 3 ) _ right ascension in j2000 . _ column ( 4 ) _ one sigma uncertainty of right ascension , in arcsec . this is calculated following @xcite and @xcite . _ column ( 5 ) _ declination in j2000 . _ column ( 6 ) _ one sigma uncertainty of declination , in arcsec . this is calculated following @xcite and @xcite . _ column ( 7 ) _ source 1.4 ghz peak flux density , in mjy . the values given here are not corrected for the systematic effects described in section [ imageanalysis ] appendix a describes how to obtain source peak flux densities corrected for the systematic effects . _ column ( 8) _ source 1.4 ghz integrated flux density , in mjy . _ column ( 9 ) and ( 10 ) _ the _ deconvolved _ major and minor axes ( fwhm ) of the source , @xmath73 , in arcsec . zero values refer to unresolved sources ( see section [ sec : decon ] ) . _ column ( 11 ) _ the _ deconvolved _ position angle ( pa , measured from n through e ) of the source , in degrees . zero values refer to unresolved sources ( see section [ sec : decon ] ) . _ column ( 12 ) _ the signal - to - noise ratio of the detection , calculated as _ imsad _ fitted peak/@xmath74 . _ column ( 13 ) _ gaussian fit quality flag : `` 1 '' refers to very good fits , `` 2 '' refers to poor integrated flux density , and `` n '' refers to non - gaussian sources . see section [ srcext ] for more details . _ column ( 14 ) _ notes on associated sources : components of a multiple source are indicated with the name of the other source(s ) with which they are associated . see section [ sec : multisources ] for more details . we performed monte carlo simulations to check the internal accuracy of our gaussian fits to the athdfs radio sources . firstly , a residual map was produced by removing all the sources detected above 5@xmath57 . gaussian sources were injected into the residual image at random positions and the _ miriad _ gaussian fitting task _ imfit _ was used to extract fitted parameters . the injected point sources had peak flux densities between @xmath25 and @xmath75 . fifty sources were injected at each peak flux density level per simulation ( 500 in total ) , and the simulation was repeated 20 times to get a sample size of 1000 sources per flux density level . by comparing the input to fitted output parameters we test the reliability of our source parameters , and obtain an estimate of the fit errors . in figure [ fig : fitcheck1 ] the ratio of the fitted peak flux density ( s@xmath35 ) to input peak flux density ( s@xmath36 ) is shown as a function of the source signal - to - noise . there is an evident systematic effect ( known as `` noise bias '' ) in the fitted peak flux density for sources with signal - to - noise ratios less than about @xmath76 . here the ratio of fitted to input peak flux density deviates from unity due to incompleteness at the lower signal - to - noise bins . if an injected source is coincident with a noise dip , its peak flux density is either under - estimated or it falls below the source detection threshold . this biases the fitted flux density to higher values because only sources that fall on noise peaks , and therefore have extra flux , are detected and measured . figure [ fig : fitcheck23 ] shows the ratio of output to input source sizes for both major and minor axes . there maybe a marginal systematic over - estimation of the source major axis for sources with signal - to - noise less than about @xmath31 . although the output major axis sizes are increasingly over - estimated from @xmath31 to @xmath25 , the over - estimation is less than the standard deviation in the measurements . in contrast there is no systematic effect present on the measurement of the minor axis . the accuracy of source positions was also checked in the simulations . the mean difference between output and input positions ( @xmath77 and @xmath78 ) are plotted in figure [ fig : fitcheck45 ] as a function of source signal - to - noise . no systematics are present , and we notice that rms values for the two axes ( @xmath79 and @xmath80 ) are similar . the positional accuracy of our catalogue is therefore derived as 1.1 arcsec at the limit of our survey ( 5@xmath57 ) , improving to 0.6 arcsec at 10@xmath57 and 0.1 arcsec at 100@xmath57 . ideally , our survey would be checked against an independent radio survey to determine the absolute accuracy of the source parameters . unfortunately there are no radio surveys in the region which reach the required depth . the parkes - mit - nrao ( pmn ) survey @xcite has a flux density limit of only 30 mjy at 4.85 ghz , while the sydney university molonglo sky survey ( sumss , @xcite ) reaches a peak flux density limit of 6 mjy at 843 mhz . the sumss source positions are not useful as a comparison because the sumss resolution is about 45 arcsec and source positional uncertainties are up to 10 arcsec , which are much greater than the athdfs internal positional uncertainties . flux density comparisons are not possible since these two surveys are at different frequencies and there is no apriori knowledge of the source seds . the absolute positional accuracy of our survey will depend on how well the positions of the phase calibrators are known . calibrator pks b2333 - 528 has been observed by vlbi and its position in the fk5 reference system is known to within 0.005 arcsec in each coordinate @xcite . the position of the second calibrator , pks b2205 - 636 , is known to within 0.1 0.25 arcsec ( atca calibrator manual , http://www.narrabri.atnf.csiro.au/calibrators/ ) . thus we expect the absolute accuracy of our source positions to be 0.25 arcsec . this absolute uncertainty in position is not included in the catalogue ( table [ catalogue ] ) , and needs to be added in quadrature to the source positional uncertainty . this affects mainly the brighter sources , as the positional uncertainty of sources with signal - to - noise less than 10 are dominated by the internal accuracy of our survey . in the athdfs , sources were included in the catalogue if the measured peak flux density exceeded five times the local noise . this can lead to incompleteness at faint flux density levels , which has to be taken into account when deriving source counts . in this section we discuss the various possible sources of incompleteness in the source catalogue . any source extraction method that relies on the ratio of peak flux density to the local noise will be affected by the variation of noise in the image . the area over which a source of a given flux density can be detected , also known as the visibility area , depends on the source peak flux density and the homogeneity of the noise distribution . the noise in our image is 11.0 @xmath1jy at the centre and is a function of the primary beam response , increasing towards the edges of the field , as quantified in section [ sec : noiseim ] . local variation in image noise is also caused by bright sources , which can increase the noise in its vicinity by up to 3 times that of an unaffected region ( see section [ sec : noiseim ] for details ) . the visibility area of the athdfs 1.4 ghz image is shown in figure [ visfig ] . as expected , the fraction of the total area over which a source of a given peak flux density can be detected increases between 0.05 and 0.1 mjy , and reaches 0.96 at 0.1 mjy . the rise in the visibility function is not as rapid as in the atesp @xcite or virmos @xcite surveys . this is because these surveys used a mosaicing technique , which results in a more uniform noise distribution than single - pointing surveys such as the athdfs . the visibility area also shows that sources with a peak flux density greater than 0.1 mjy can be detected over more than 95% of the whole image . the visibility area reaches 100% of the image at a peak flux density of 0.23 mjy . two additional effects which could be responsible for partial incompleteness in the source catalogue are bandwidth smearing and clean bias . these two effects have been extensively discussed in section [ imageanalysis ] . bandwidth smearing can lead to an underestimation of peak flux densities by up to @xmath1418% , although the amount is 5% or less for the majority of our sources . the clean bias affects both peak and integrated flux densities . it leads to an underestimation of flux densities by @xmath145% for the faintest sources ( 5@xmath53 ) , but has no effect on sources brighter than 10@xmath53 . in section [ imageanalysis ] we proposed a formula ( equation a1 ) to correct the measured peak and integrated flux densities for these two systematic effects . the correction depends on the source signal - to - noise ratio as well as the source distance from the image centre . however , peak flux densities are underestimated by @xmath1423% _ at most _ due to the combined effects of both clean bias and bandwidth smearing . since the highest level of noise is @xmath1450 @xmath1jy , it follows that clean bias and bandwidth smearing do not affect sources brighter than @xmath81 mjy . resolution bias is an effect in which weak extended sources may have peak flux densities that fall below the catalogued 5@xmath57 limit , yet still have total integrated flux densities above the survey limit . to derive a source count that is complete in terms of total flux density , the number of `` missing '' sources has to be estimated . given a maximum detectable angular size and knowledge of the intrinsic source size distribution as a function of flux density , this so called resolution bias can be corrected . here we follow the procedures of @xcite to correct the resolution bias . assuming a gaussian beam , the maximum size ( @xmath82 ) a source of total flux density @xmath71 can have before it drops below the @xmath83 detection limit can be calculated from the equation : @xmath84 in figure [ thetalims ] we plot the angular size ( @xmath70 ) of the athdfs sources as a function of the measured total flux density . the angular sizes are defined as the geometric mean of the major and minor deconvolved axes of the sources . we assume @xmath85 for point sources . we calculate @xmath82 from equation [ thetamaxfn ] with a full range of noise values from 5 to 30 @xmath1jy . the distribution of pixel values in the noise image of section [ sec : noiseim ] is used to determine the relative weight of each noise value and a weighted average @xmath82 is calculated . figure [ thetalims ] shows that the angular sizes of the largest athdfs sources are in good agreement with this weighted average @xmath86 function . as discussed in [ sec : decon ] , the deconvolution efficiency depends on the signal - to - noise of a source . an estimate of the minimum angular size ( @xmath87 ) that a source can have is derived from equations [ eq : fluxtheta ] and [ eq : deconv ] : @xmath88 = \frac{\theta_{\rm min}^2}{b_{\rm maj}b_{\rm min}}\ ] ] similar to the derivation of @xmath82 , a weighted average @xmath89 is calculated using the noise distribution of our image . the resulting average @xmath89 is plotted in figure [ thetalims ] as a dashed line . we note that the @xmath89 constraint is important at low flux density levels where @xmath82 becomes unphysical ( @xmath900 ) . also , the minimum deconvolved size is about 10 at the detection limit , whereas sources as small as @xmath141 are reliably deconvolved at higher signal - to - noise . using the two constraints described above , an overall angular size upper limit , @xmath91 , as a function of flux density @xmath71 is defined as : @xmath92 the incompleteness of the athdfs survey due to @xmath91 can be estimated with knowledge of the true source angular size distribution as a function of flux density . we assume an exponential form for the integral angular size distribution , @xmath93 @xcite : @xmath94 with @xmath95 where @xmath96 is in mjy , and @xmath97 is the estimated median source size at a given flux density . equations [ windeq1 ] and [ windeq2 ] together with equation [ thetalimeq ] allow us to estimate the fraction of sources larger than the maximum detectable size and therefore missed by our survey . the correction factor , @xmath98 , is then simply calculated as : @xmath99 in figure [ rescorrfig ] we show both @xmath93 ( left panel ) and @xmath98 ( right panel ) . the resolution bias peaks at about 0.1 mjy , but decreases at lower flux densities due to the @xmath89 constraint . in this section , we construct the differential radio source counts from the @xmath58 athdfs catalogue . in computing the source counts we have used the integrated flux density for resolved sources and peak flux densities for point sources . the counts were calculated by dividing the number of sources in each flux density bin by the total survey area and the bin width , and then normalising by the euclidean slope of @xmath100 . the reason for the normalisation is historical , relating to the early use of radio source counts to test various cosmological models against the number of sources that would be expected in a steady state euclidean universe @xcite . table [ countstable ] lists the final athdfs source counts , corrected for the visibility area , clean bias , bandwidth smearing and resolution bias as discussed above . for each flux density bin , the mean flux density ( @xmath11@xmath101@xmath102 ) , the number of detected sources ( @xmath103 ) , the number of effective sources after applying the corrections ( @xmath104 ) , and the corrected normalised differential radio source count ( @xmath105 ) , are given . we show the counts before and after corrections in figure [ athdfscounts1 ] . we notice that the corrections are most important in the flux density range 0.09 0.3 mjy , where they can increase the source count by up to approximately 30% . sources classed as `` multiple - component '' ( see section 4.4 ) only make one contribution to the source count . at these flux density levels , the difference between incorporating `` multiple - component '' sources as single objects and deriving counts from the `` full component '' catalogue is negligible . if all components are used , then the most significant change is in the @xmath11@xmath101@xmath102=0.119 mjy bin , where there are 4 additional sources . for this bin the difference is @xmath14 7% , which is less than the poisson error . the athdfs source counts are compared to those derived from other 1.4 ghz surveys in figure [ athdfscounts2 ] . the source counts from the faint images of the radio sky at twenty centimetres survey ( first , @xcite ) are shown , as well as counts from the atesp survey @xcite , phoenix deep field ( pdf , @xcite ) and hubble deep field north ( hdfn , @xcite ) . the solid line in figure [ athdfscounts2 ] is the linear least squares sixth order polynomial fit to this compilation of source counts . first counts below 2.5 mjy , were removed from the fit since first is incomplete at this limit . pdf counts above 2.5 mjy and atesp counts above 76 mjy were also excluded from the fit since the sampling is sparse in the higher flux density bins of these surveys . the resulting polynomial fit is given by : @xmath106 = \sum_{i=0}^{6 } a_i(\log[s/{\rm mjy}])^i,\ ] ] with @xmath107 , @xmath108 , @xmath109 , @xmath110 , @xmath111 , @xmath112 and @xmath113 . for comparison , a third order fit to source counts from 0.1 mjy to 10000 mjy by @xcite and a sixth order fit by @xcite to the pdf and first source counts between 0.06 mjy and 1 mjy are shown as dashed and dotted lines , respectively . a third order polynomial is insufficient to account for the curvature below 0.1 mjy suggested by both the hdfn and pdf surveys . a fourth order polynomial does not adequately model the point of inflection around 0.5 mjy @xcite , while fifth order polynomials do not reproduce the concave curvature seen at about 1 jy ( e.g. @xcite ) . hence the sixth order polynomial is necessary . the fitted polynomial has no physical basis , but is useful in quantifying the differential radio source count . there is a high level of consistency between the source counts derived from both small and large area surveys at flux densities @xmath114 0.5 mjy ( figures 19 and 20 ) . the differential source counts diverge at fainter flux density levels ( @xmath115 0.5 mjy ) the source count from the hdfn has been previously noted as being particularly low because of inadequate completeness corrections , as well as cosmic variance @xcite . in contrast , the count from the athdfs lies _ higher _ than all other counts at @xmath116 0.4 mjy , although within statistical agreement for @xmath115 0.3 mjy . the excess is significant for the athdfs at 0.37 mjy @xmath117 0.47 mjy compared to the latest ( 2003 ) pdf results and the counts from the virmos survey . over the 0.35 deg@xmath13 area catalogued by the athdfs survey , we have 22 sources in this flux density range , compared to 44 in the virmos field ( 1 deg@xmath13 ) and 150 in the latest pdf catalogue ( 4.5 deg@xmath13 ) . we attribute the variation in the athdfs as real given the small area of the athdfs , with a probable over density of @xmath14 3 8 sources compared to the average field . at the very faintest limits ( @xmath115 0.1 mjy ) it has been suggested that the differential source count steepens towards lower flux density limits @xcite . although there is little real confirmation , as in every case it relies on the faintest source count bin , this steepening is expected as the flat region of the count between 0.08 and 0.3 mjy can not continue infinitely because the implied surface density of radio sources would exceed the number of host galaxies @xcite . the differential count from the athdfs is in agreement with the general trend around @xmath101 = 0.08 mjy , although the last data point at 0.059 mjy lies well above those from the hdfn and pdf . thus we can not confirm the down - turn in the differential radio source count , and will have to wait for a deeper source catalogue to supply this data . source counts are important because they can be used to constrain the evolution of the radio population . the source count above 10 mjy is dominated by giant radio galaxies and qsos , which we group together in this discussion as agn . at sub - mjy levels ( 0.1 1 mjy ) there is an increasing number of blue galaxies with starforming spectral signatures . @xcite , @xcite , @xcite and others have concluded that the source count at these faintest levels requires two populations , agn and starforming galaxies , both of which undergo strong evolution . to derive a local luminosity function of starforming galaxies , @xcite start with the iras 60@xmath118 luminosity function @xcite and convert that to radio using the well - known fir - radio correlation @xcite . assuming that luminosity evolves as @xmath119 , @xcite found @xmath120 best matches the radio source counts . we perform a similar test to determine the rate of evolution of the local starforming galaxies required to match the radio source counts . we use the local radio luminosity function of starforming galaxies determined by @xcite for the 4583 uppsala galaxy catalogue ( ugc , @xcite ) galaxies crossmatched with the nrao vla sky survey ( nvss , @xcite ) . the ugc - nvss local rlf is the most comprehensive to date , very local ( @xmath121 ) , and probes the low luminosity end of starforming galaxies to @xmath122 w hz@xmath123 . the ugc - nvss local rlf , @xmath124 , is described by the local visibility function , @xmath125 , where @xmath126 + \frac{3}{2 } \log[l ] + 28.43 \ ; .\ ] ] fitting the visibility function directly , instead of the rlf , results in a better fit because the @xmath127 weighting prevents the visibility function from steepening as fast as the luminosity function at high luminosities . @xcite approximate the visibility function with a hyperbolic function : @xmath128 ^ 2 \right\}^{1/2 } \ ; , \ ] ] with y = 3.06 , b = 1.9 , x = 22.35 , and w = 0.67 , for starforming galaxies . for the agn component we use the agn counts from @xcite . the observed source count is shown in figure [ count_evolve ] along with several models for the counts of the local starforming population . we have explored a variety of evolution scenarios : pure luminosity evolution going as @xmath129 , pure density evolution going as @xmath130 , and a no evolution model . we confirm that two distinct populations , the starforming galaxies and agn , are required to account for the faint source counts . the counts are best fit by starforming galaxies undergoing @xmath131 luminosity evolution . this amount of evolution means the number of starforming galaxies exceeds the number of agn at about 0.25 mjy . the counts rule out stronger luminosity evolution of @xmath132 . a modest amount of density evolution is also allowed , but if it is large ( e.g. @xmath133 ) the source counts are too high at the @xmath1jy level . a downturn in radio source counts is needed at these levels so that the number of radio sources do nt exceed the number of host galaxies @xcite . thus , the radio source counts rule out large amounts of density evolution in the starforming galaxy population . similar values for the evolution of star forming galaxies have been found by @xcite and @xcite , who find @xmath134 and @xmath135 , respectively . these two studies use the local radio luminosity function of starforming galaxies from @xcite . despite different assumptions for the luminosity functions of both starforming and agn galaxies , there is good agreement between our work and these previous studies . redshift information is required to break the degeneracy between luminosity and density evolution . little density evolution is expected however . for example , @xcite jointly constrain the evolution parameters @xmath136 and @xmath137 to find @xmath134 and @xmath138 . we have obtained spectra of our radio sources which will allow us to explore galaxy evolution in detail . this further work will be discussed in a future paper . we have presented the extended observations of the hubble deep field south with the australia telescope compact array at 1.4 ghz . the new 1.4 ghz image reaches a maximum sensitivity of @xmath1411 @xmath1jy rms at a resolution of @xmath146.5 arcsec . the clean bias and bandwidth smearing effects have been investigated and quantified for our 1.4 ghz image . a catalogue of all sources brighter than @xmath139 within 20 arcmin of the image centre has been compiled , comprising 466 sources . the 1.4 differential radio source counts from the athdfs survey have been presented . we have corrected the counts for survey completeness and systematic effects , including visibility area , clean bias , bandwidth smearing and resolution bias . we find that the source counts in the athdfs are in general consistent with that from previous surveys , although the athdfs counts are greater than that found in the the hdfn by a factor of 1.2 2 . other deep surveys have also found the hdfn counts to be low ( e.g. phoenix , @xcite ; virmos , @xcite ) . thus , the hdfn counts are probably underestimated , or the hdfn samples a relatively under - dense region of radio sources . we find the radio source counts are best fit by the local starforming population with luminosity evolution of the form @xmath131 , although small amounts of density evolution are not ruled out . the athdfs survey has obtained radio images at 2.5 , 5.2 and 8.7 ghz in addition to 1.4 ghz . the next paper in the series will provide radio spectral indices and a detailed analysis of the athdfs sources across this radio spectrum . the hubble deep field south has also been the target of deep multi - colour optical photometry and spectroscopy . future papers in this series will present the optical identifications and spectroscopy of the athdfs radio sources . as discussed in section [ imageanalysis ] , the two systematics which affect the peak flux densities of the athdfs sources have been analysed . these two effects are the clean bias and bandwidth smearing . the flux densities given in the athdfs catalogue ( table 2 ) are not corrected for these effects . corrected flux densities can be obtained using the following formula : @xmath140 where @xmath141 is the flux density actually measured and catalogued , @xmath142 is the catalogued source signal - to - noise , @xmath45 is the source distance from the pointing centre ( ra = 22h 33 m 37.6s , dec = @xmath2160@xmath22 33 29 ) , and @xmath46 is the synthesized beam fwhm ( 6.6 ) . the term in the numerator of equation [ correction ] corrects for the bandwidth smearing , while the term in the denominator represents the clean bias correction . for integrated flux densities , the term in the numerator of equation [ correction ] should be set to 1 . , , uppsala general catalogue of galaxies ( acta universitatis upsaliensis . nova acta regiae societatis scientiarum upsaliensis - uppsala astronomiska observatoriums annaler , uppsala : astronomiska observatorium , 1973 ) , e. m. , jackson , c. a. , cannon , r. d. , mcintyre , v. j. , murphy , t. , bland - hawthorn , j. , bridges , t. , cole , s. , colless , m. , collins , c. , couch , w. , dalton , g. , de propris , r. , driver , s. p. , efstathiou , g. , ellis , r. s. , frenk , c. s. , glazebrook , k. , lahav , o. , lewis , i. , lumsden , s. , maddox , s. , madgwick , d. , norberg , p. , peacock , j. a. , peterson , b. a. , sutherland , w. , & taylor , k. 2002 , , 329 , 227 , r. , mathis , d. , & neuschaefer , l. 1990 , in evolution of the universe of galaxies ; proceedings of the edwin hubble centennial symposium , berkeley , ca , june 21 - 23 , 1989 ( a91 - 55951 24 - 90 ) . san francisco , ca , astronomical society of the pacific , 1990 , p. 389 , 389 1.8 in size , of the 5 source pairs which pass the single source test described in section [ sec : multisources ] and appear to be physically associated . the contour levels for each source are set at 3 , 5 , 10 , 20 , 50 and 100 @xmath57.,width=151 ] 1.8 in size , of the 5 source pairs which pass the single source test described in section [ sec : multisources ] and appear to be physically associated . the contour levels for each source are set at 3 , 5 , 10 , 20 , 50 and 100 @xmath57.,width=151 ] 1.8 in size , of the 6 source pairs which pass the single source test described in section [ sec : multisources ] but are not taken to be physically associated . the contour levels for each source are set at 3 , 5 , 10 , 20 , 50 and 100 @xmath57.,width=151 ] 1.8 in size , of the 6 source pairs which pass the single source test described in section [ sec : multisources ] but are not taken to be physically associated . the contour levels for each source are set at 3 , 5 , 10 , 20 , 50 and 100 @xmath57.,width=151 ] ccrrc 0.050 0.064 & 0.059 & 28 & 226.1 & 4.02 @xmath143 0.76 + 0.064 0.082 & 0.073 & 53 & 149.8 & 3.51 @xmath143 0.48 + 0.082 0.106 & 0.094 & 70 & 143.0 & 4.90 @xmath143 0.59 + 0.106 0.136 & 0.119 & 54 & 87.06 & 4.25 @xmath143 0.58 + 0.136 0.174 & 0.154 & 52 & 79.85 & 5.73 @xmath143 0.79 + 0.174 0.224 & 0.197 & 43 & 59.43 & 6.15 @xmath143 0.94 + 0.224 0.287 & 0.253 & 26 & 33.49 & 5.08 @xmath143 1.00 + 0.287 0.368 & 0.328 & 24 & 29.18 & 6.58 @xmath143 1.34 + 0.368 0.473 & 0.417 & 22 & 25.18 & 8.06 @xmath143 1.72 + 0.473 0.607 & 0.541 & 14 & 15.10 & 7.22 @xmath143 1.93 + 0.607 0.779 & 0.667 & 10 & 10.75 & 6.76 @xmath143 2.14 + 0.779 1.000 & 0.905 & 7 & 7.56 & 7.93 @xmath143 3.00 +

galactic black hole candidates ( bhcs ) are known to exhibit transitions between different x - ray spectral states . typically , a transient outburst begins in the low / hard state at a low luminosity . the x - ray spectrum in the low / hard state is roughly described by a hard power law with a high energy cutoff at @xmath6 . as the luminosity increases , these systems undergo a transition to the high / soft state ( so - called a hard - to - soft transition ) . the x - ray spectrum in the high / soft state is dominated by the disk emission of characteristic temperature @xmath7 . recently , two distinct types of hard - to - soft transitions , the bright / slow transition and the dark / fast transition , are reported ( e.g. , * ? ? ? * ; * ? ? ? the bright / slow transition occurs at @xmath8 and takes more than @xmath9 days . the system undergoes a transition from the low / hard state to the high / soft state via the `` bright / hard '' state and the very high / steep power law ( vh / spl ) state during the bright / slow transition . the x - ray spectrum in the bright / hard state is described by a hard power - law with a ( moderately ) high energy cutoff at @xmath10 and the luminosity is `` brighter '' than that in the low / hard state . the dark / fast transition occurs at less than @xmath11 and takes less than @xmath12 days . the system immediately switches from the low / hard state to the high / soft state during the dark / fast transition . here , @xmath13 is the eddington luminosity , @xmath14 is the black hole mass , and @xmath15 is the electron scattering opacity . @xcite analyzed the results of _ rxte _ observations of the bhc gx 339 - 4 in the rising phases of the transient outbursts ( this object showed the bright / slow transition in the 2002/2003 outburst and the dark / fast transition in the 2004 outburst ) . they found that the cutoff energy strongly anti - correlates with the luminosity and decreases from @xmath16 to @xmath17 in the bright / hard state , while the cutoff energy is roughly constant at @xmath16 in the low / hard state . this suggests that the electron temperature of an accretion disk emitting hard x - rays decreases as the luminosity increases in the bright / hard state . furthermore , the bright / hard state has been observed from @xmath18 up to @xmath8 . they concluded that such anti - correlation is explained by the scenario that the heating rate from protons to electrons via the coulomb collision balances the radiative cooling rate of inverse compton scattering . the main purpose of this paper is to present a model explaining the bright / hard state . in the conventional theory of accretion disks , the concept of phenomenological @xmath19-viscosity is introduced . in this framework , the @xmath20-component of the stress tensor , which appears in the angular momentum equation and the viscous heating term , is assumed to be proportional to the gas pressure , @xmath21 ( we ignore the radiation pressure in this paper because we focus on optically thin disks ) . here @xmath22 is the viscosity parameter introduced by @xcite . the magnetic field can be an origin of the @xmath19-viscosity because the maxwell stress generated by the magneto - rotational instability ( mri ) efficiently transports angular momentum in accretion disks ( e.g. , * ? ? ? * ) and the dissipation of magnetic energy contributes the disk heating ( e.g. , * ? ? ? * ) . optically thin , hot accretion disks have been studied to explain hard x - rays from bhcs . @xcite proposed that hard x - rays from cyg x-1 are produced in an inner optically thin hot disk . @xcite studied the structure and stability of optically thin hot accretion disks . we have to consider two - temperature plasma in such disks because the electron temperature is expected to become lower than the ion temperature in such a low density , high temperature region . the energy equations for ions and electrons are written in the form @xmath23 @xmath24 where @xmath25 is the viscous heating rate , @xmath26 is the energy transfer rate from ions to electrons via coulomb collisions , @xmath27 is the radiative cooling rate , and @xmath28 is the fraction of heating to electrons . the left hand - sides represent the heat advection terms for ions ( @xmath29 ) and electrons ( @xmath30 ) . we note that early works on optically thin , hot , two - temperature accretion disks assumed that the viscous heating acts primarily on ions ( @xmath31 ) . @xcite and ( * ? ? ? * hereafter sle ) constructed a model for optically thin two - temperature accretion disks . in the sle solutions , the dissipated energy is transferred from ions to electrons via coulomb collisions ( @xmath32 ) and radiated away by electrons ( @xmath33 ) . although the electron temperature is high enough to explain the x - ray spectrum in the low / hard state , the sle solutions are thermally unstable in the framework of the @xmath19-prescription of viscosity . @xcite pointed out the importance of heat advection in hot , magnetized accretion flows , and obtained steady solutions of optically thin disks . such geometrically thick , optically thin , advection - dominated accretion flows ( adafs ) or radiatively inefficient accretion flows ( riafs ) have been studied extensively by ( * ? ? ? * @xcite ) and @xcite . in the adaf / riaf solutions , a substantial fraction of the dissipated energy is stored in the gas as entropy and advected into the central object ( @xmath34 ) . only a small fraction of the dissipated energy is transferred to electrons and radiated away . the adaf / riaf solutions are thermally stable . @xcite found that the maximum mass accretion rate of the adaf / riaf solutions is @xmath35 which corresponds to @xmath36 , where @xmath37 is the eddington mass accretion rate . @xcite showed that the electron temperature in the adaf / riaf solutions weakly anti - correlates with the luminosity and decreases to @xmath38 . these features are consistent with the facts that the energy cutoff weakly anti - correlates with the luminosity around @xmath16 in the low / hard state , and that these systems undergo a transition from the low / hard state to other x - ray spectral states ( i.e. , the high / soft state during the dark / fast transition , and the bright / hard state during the bright / slow transition ) about at this maximum luminosity of the adaf / riaf solutions . however , the adaf / riaf solutions can not explain the strong anti - correlation in the range of @xmath39 and @xmath40 observed in the bright / hard state . the heat advection works as an effective cooling in the adaf / riaf solutions . @xcite presented a luminous hot accretion flow ( lhaf ) in which the heat advection for ions works as an effective heating . above the maximum mass accretion rate of the adaf / riaf solutions , the heat advection overwhelms the viscous heating and balances the energy transfer from ions to electrons ( @xmath41 ) . the lhafs are thermally unstable . however , @xcite concluded that the thermal instability will have no effect on the dynamics of the lhafs because the accretion timescale is shorter than the timescale of growth of the local perturbation at such high mass accretion rate . the lhaf solutions also can not explain the bright / hard state because the electron temperature is high and roughly constant at @xmath42 . in these models mentioned above , magnetic fields are not considered explicitly , and the ratio of the gas pressure to the magnetic pressure is assumed to be constant ( typically , @xmath43 ) . @xcite suggested that an accretion disk evolves toward two types of disks , a high-@xmath1 disk and a low-@xmath1 disk , by carrying out two - dimensional magnetohydrodynamic ( mhd ) simulations of the buoyant escape of the magnetic flux owing to the parker instability @xcite . in the high-@xmath1 disk , the magnetic flux escapes from the disk owing to the parker instability and @xmath1 inside the disk is maintained at a high value . global three - dimensional mhd simulations of optically thin , radiatively inefficient accretion disks also indicate that the amplification of magnetic fields becomes saturated when @xmath44 in a quasi - steady state except in the plunging region very close to the black hole ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? on the other hand , once a disk is dominated by the magnetic pressure , it can stay in the low-@xmath1 state because the strong magnetic tension suppresses the growth of the parker instability . @xcite demonstrated that an optically thin , radiatively inefficient , hot , high-@xmath1 ( adaf / riaf - like ) disk undergoes transition to an optically thin , radiatively efficient , cool , low-@xmath1 disk except in the plunging region when the mass accretion rate exceeds the threshold for the onset of a cooling instability . during this transition , the magnetic flux @xmath45 is almost conserved at each radius because the cooling timescale is shorter than that of the buoyant escape of the magnetic flux , where @xmath46 is the mean azimuthal magnetic field and @xmath47 is the half thickness of the disk . in this way , the magnetic pressure becomes dominant and supports the disk as the gas pressure decreases owing to the cooling instability . eventually , the disk stays in a quasi - steady , cool , low-@xmath1 state . because the mri is not yet stabilized in this quasi - steady state , the magnetic field still remains turbulent and dominated by the azimuthal component . as a result , the heating owing to the dissipation of the turbulent magnetic field balances the radiative cooling . @xcite performed local three - dimensional mhd simulations of strongly magnetized , vertically stratified accretion disks in a keplerian potential . they showed that strongly magnetized state is maintained near the equatorial plane because the buoyantly escaping magnetic flux is replenished by stretching of a radial field . the mri feeds off both vertical and azimuthal fields and drives turbulence . the maxwell and reynolds stresses generated by the turbulence become significant . therefore , they indicated that highly magnetized disks are astrophysically viable . we note that such low-@xmath1 disks are quite different from magnetically dominated accretion flows ( mdafs ; * ? ? ? * ) observed in the plunging region of optically thin accretion disks in global mhd and general relativistic mhd simulations ( e.g. , * ? ? ? * ) in terms of their energy balance and configuration of magnetic fields . outside the plunging region , the magnetic field becomes turbulent and dominated by the azimuthal component because the growth timescale of the mri is shorter than the inflow timescale . as a result , the released gravitational energy is efficiently converted into the thermal energy via the dissipation of the turbulent magnetic field . on the other hand in the plunging region , the ratio of the timescales is reversed because the inflow velocity increases with decreasing the radius in such disks . therefore , magnetic field lines are stretched out in the radial direction before turbulence is generated by the mri and is dissipated . as a result , a substantial fraction of the gravitational energy is converted into the radial infall kinetic energy . since the heating owing to the dissipation of turbulent magnetic fields becomes inefficient , the gas pressure becomes low , and the flow becomes magnetically dominated . although both of the low-@xmath1 disk and the mdaf are cool and magnetic pressure dominant , they are essentially different . we focus on the low-@xmath1 disk in this paper . @xcite constructed a steady model of optically thin , one - temperature accretion disks incorporating magnetic fields on the basis of these results of three - dimensional mhd simulations of accretion disks . @xcite extended it to the optically thick regime . they assumed that the @xmath20-component of the stress tensor is proportional to the total pressure . in order to complete the set of basic equations , they specified the advection rate of the azimuthal magnetic flux instead of @xmath1 . they found a new thermally stable solution , a low-@xmath1 solution , which can explain the results by @xcite . in the low-@xmath1 solutions , the magnetic heating enhanced by the strong magnetic pressure balances the radiative cooling . the disk temperature is lower than that in the adaf / riaf solutions and strongly anti - correlates with the mass accretion rate they also found that the low-@xmath1 solutions exist above the maximum mass accretion rate of the adaf / riaf solutions . therefore , they concluded that the optically thin low-@xmath1 disk can qualitatively explain the bright / hard state . however , they considered one - temperature plasma and bremsstrahlung emission as a radiative cooling mechanism . it is expected that the electron temperature becomes lower than the ion temperature and that synchrotron cooling and/or inverse compton scattering become effective in such disks . in this paper , we extend the model of optically thin , one - temperature disk to that of optically thin , two - temperature disks . we consider synchrotron emission and inverse compton scattering as a cooling mechanism in addition to bremsstrahlung emission . we obtained the thermal equilibrium curves and found that the optically thin low-@xmath1 solutions can quantitatively explain the bright / hard state . the basic equations are presented in section [ basic_eq ] . in section [ results ] , we present the thermal equilibrium curves . section [ discussion ] is devoted to a discussion . we summarize the paper in section [ summary ] . we extended the basic equations for one - dimensional steady , optically thin , two - temperature black hole accretion flows ( e.g. , * ? ? ? * ) incorporating magnetic fields . we adopt cylindrical coordinates @xmath48 . general relativistic effects are simulated using the pseudo - newtonian potential @xmath49 @xcite , where @xmath50 is the gravitational constant , @xmath14 is the black hole mass ( we assume @xmath51 in this paper ) , @xmath52 , and @xmath53 is the schwarzschild radius . for simplicity , the gas is assumed to consist of protons ( ions ) and electrons . the number density of ions and electrons are equal by charge neutrality , @xmath54 . we start with the resistive mhd equations @xmath55 @xmath56 = - \rho \nabla \psi - \nabla p_{\rm gas } + \frac{\mbox{\boldmath{$j$ } } \times \mbox{\boldmath{$b$}}}{c } ~ , \label{eq : vec_mom}\end{aligned}\ ] ] @xmath57 - \left ( \mbox{\boldmath{$v$ } } \cdot \nabla \right ) p_{\rm i } \nonumber \\ = \left(1 - \delta_{\rm heat } \right ) q^{+ } - q^{\rm ie } ~ , \label{eq : vec_ene_i}\end{aligned}\ ] ] @xmath58 - \left ( \mbox{\boldmath{$v$ } } \cdot \nabla \right ) p_{\rm e } \nonumber \\ = \delta_{\rm heat } q^{+ } + q^{\rm ie } - q_{\rm rad}^{- } ~ , \label{eq : vec_ene_e}\end{aligned}\ ] ] @xmath59 where @xmath60 is the density , @xmath61 and @xmath62 are the ion and electron densities , @xmath63 and @xmath64 are the ion and electron masses , @xmath65 is the velocity , @xmath66 is the magnetic field , @xmath67 is the current density , @xmath68 is the gas pressure , @xmath69 and @xmath70 are the ion and electron gas pressure , @xmath71 and @xmath72 are the ion and electron temperature , @xmath73 is the boltzmann constant , @xmath74 and @xmath75 are the internal energy of ions and electrons . here , @xmath76 and @xmath77 are the specific heat ratio for ions and electrons . in the energy equations for ions ( [ eq : vec_ene_i ] ) and electrons ( [ eq : vec_ene_e ] ) , @xmath25 is the heating rate , @xmath27 is the radiative cooling rate , and @xmath26 is the energy transfer rate from ions to electrons via coulomb collisions . here , @xmath28 represents the fraction of heating to electrons . in the induction equation ( [ eq : vec_ind ] ) , @xmath78 is the magnetic diffusivity , where @xmath79 is the electric conductivity . three - dimensional global mhd and local radiation - mhd simulations of black hole accretion disks showed that magnetic fields inside the disk are turbulent and dominated by the azimuthal component in a quasi - steady state ( e.g. , * ? ? ? * ; * ? ? ? on the basis of results of the simulations , we decomposed the magnetic fields into the mean fields @xmath80 and fluctuating fields @xmath81 and also decomposed the velocity into the mean velocity @xmath82 and the fluctuating velocity @xmath83 . we assume that the fluctuating components vanish when azimuthally averaged , @xmath84 , and that the radial and vertical components of the magnetic fields are negligible compared with that of the azimuthal component , @xmath85 , @xmath86 . here @xmath87 denotes the azimuthal average . let us derive the azimuthally averaged equations . we assume that the disk is in a steady state and in hydrostatic balance in the vertical direction . by azimuthally averaging equations ( [ eq : vec_con ] ) - ( [ eq : vec_ind ] ) and ignoring the second order terms of @xmath88 , @xmath89 , and @xmath90 , we obtain @xmath91 @xmath92 @xmath93 + \frac{\partial}{\partial z } \left ( \frac{\langle b_{\varphi } b_{z } \rangle}{4 \pi}\right)~,\end{aligned}\ ] ] @xmath94 @xmath95 + \frac{v_{\varpi}}{\varpi } \left ( \rho_{\rm i } \epsilon_{\rm i } + p_{\rm i } \right ) + \frac{\partial}{\partial z } \left [ \left ( \rho_{\rm i } \epsilon_{\rm i } + p_{\rm i } \right ) v_{z}\right ] \nonumber \\ - v_{\varpi } \frac{\partial}{\partial \varpi } p_{\rm i } - v_{z } \frac{\partial}{\partial z } p_{\rm i } = \left ( 1 - \delta_{\rm heat } \right ) q^{+ } - q^{\rm ie } ~,\end{aligned}\ ] ] @xmath96 + \frac{v_{\varpi}}{\varpi } \left ( \rho_{\rm e } \epsilon_{\rm e } + p_{\rm e } \right ) + \frac{\partial}{\partial z } \left [ \left ( \rho_{\rm e } \epsilon_{\rm e } + p_{\rm e } \right ) v_{z}\right ] \nonumber \\ - v_{\varpi } \frac{\partial}{\partial \varpi } p_{\rm e } - v_{z } \frac{\partial}{\partial z } p_{\rm e } = \delta_{\rm heat } q^{+ } + q^{\rm ie } - q^{-}_{\rm rad } ~,\end{aligned}\ ] ] @xmath97 -\frac{\partial}{\partial \varpi } \left [ v_{\varpi } \langle b_{\varphi } \rangle \right ] \nonumber \\ + \ { \nabla \times \langle \delta \mbox{\boldmath{$v$ } } \times \delta \mbox{\boldmath{$b$ } } \rangle \}_{\varphi } - \ { \eta_{\rm m } \nabla \times \left ( \nabla \times \mbox{\boldmath{$\bar{b}$ } } \right ) \}_{\varphi } ~,\end{aligned}\ ] ] where @xmath98 is the total pressure and @xmath99 is the azimuthally averaged magnetic pressure . the third and fourth terms on the right - hand side of equation ( [ eq : ind_phi ] ) represent the dynamo term and the magnetic diffusion term which we approximate later on the basis of the results of the numerical simulations . we assume that the radial velocity @xmath100 , the specific angular momentum @xmath101 , and @xmath102 are independent of @xmath103 , and that the disks are isothermal in the vertical direction . under these assumptions , the surface density @xmath104 , the vertically integrated total pressure @xmath105 , and the half thickness of the disk @xmath47 are defined as @xmath106 @xmath107 @xmath108 where @xmath109 is the keplerian angular velocity . here the subscript @xmath110 refers to quantities in the equatorial plane . using the equation of state for the ideal gas , the vertically integrated total pressure is expressed as @xmath111 now we integrate the other basic equations in the vertical direction . we obtain @xmath112 @xmath113 @xmath114 @xmath115 @xmath116 @xmath117 d\varpi dz + \mbox{const.}\end{aligned}\ ] ] where @xmath118 is the mass accretion rate , @xmath119 is the keplerian angular momentum and @xmath120 is the specific angular momentum swallowed by the black hole . in the energy equations , @xmath121 and @xmath122 are the vertically integrated heat advection terms for ions and electrons , @xmath123 , @xmath124 , and @xmath125 are the vertically integrated heating rate , radiative cooling rate , and energy transfer rate from ions to electrons via coulomb collisions . in equation ( [ eq : ind_int ] ) , @xmath126 is the radial advection rate of the azimuthal magnetic flux ( hereafter we call it the magnetic flux advection rate ) . in this paper , we assume that @xmath127 instead of equation ( [ eq : mom_pi_int ] ) because we focus on local thermal equilibrium solutions . global mhd simulations of radiatively inefficient , accretion flows ( e.g. , * ? ? ? * ; * ? ? ? * ) showed that the ratio of the azimuthally averaged maxwell stress to the sum of the azimuthally averaged gas pressure and magnetic pressure is nearly constant ( @xmath128 ) except in the plunging region . following the simulation results , we assume that the azimuthally averaged @xmath129-component of the maxwell stress inside a disk is proportional to the total ( gas and magnetic ) pressure @xmath130 integrating in the vertical direction , we obtain @xmath131 this is one of the key assumptions in this paper . when the magnetic pressure is high , the stress level can be high even though the gas pressure is low . we can rewrite this relation in terms of the kinematic viscosity , @xmath132 , as @xmath133 where @xmath134 @xmath135 is the sound speed , and @xmath136 is the alfvn speed . we complete the set of basic equations by specifying the radial distribution of the magnetic flux advection rate . if we perform the integration in the second term of the induction equation ( [ eq : ind_int ] ) , we obtain @xmath137 + \mbox{const . } \nonumber\end{aligned}\ ] ] where @xmath138 is the mean azimuthal magnetic field in the equatorial plane . according to the result of the global three - dimensional mhd simulation by @xcite , the magnetic flux advection rate at a radius is roughly unchanged before and after the transition from the adaf / riaf - like disk to the low-@xmath1 disk . hence , we adopt the magnetic flux advection rate as the parameter in order to complete the set of the basic equations . the magnetic flux advection rate depends on various mechanisms such as the escape of magnetic fluxes due to the magnetic buoyancy , the regeneration of azimuthal magnetic fields by the shear motion , the generation of magnetic turbulence through the mri , dissipation of magnetic fields due to the magnetic diffusivity , and magnetic reconnection . if the sum of the dynamo term and the magnetic diffusion term is zero in the whole region , the magnetic flux advection rate is constant in the radial direction . the global three - dimensional mhd simulation performed by @xcite indicated that the magnetic advection rate increases with decreasing radius , specifically , @xmath139 , in the quasi steady state as a result of magnetic dynamo and diffusivity processes . because it is hard to compute the dynamo term and the magnetic diffusion term explicitly from the local quantities , we parameterize the dependence of @xmath126 on @xmath140 by introducing a parameter , @xmath141 , as follows . @xmath142 where @xmath143 is the magnetic flux advection rate at the outer boundary @xmath144 . when @xmath145 , the magnetic flux advection rate is constant in the radial direction . when @xmath146 , the magnetic flux advection rate increases with decreasing radius ( see also figure 1 in * ? ? ? we determine the parameter @xmath143 by imposing the outer boundary condition , @xmath147 \left(\varpi_{\rm out } / r_{\rm s } \right)^{-1}$ ] and @xmath148 at @xmath149 . this leads that @xmath150 . in the local model presented in this paper , the value of @xmath141 just means the amount of the magnetic flux advection rate at a radius ( e.g , @xmath151 , @xmath152 , @xmath153 for @xmath145 , @xmath154 , @xmath155 , respectively ) . equation ( [ eq : phidot ] ) is the second key assumption in this paper . specifying the magnetic flux advection rate enables the magnetic pressure to increase when the disk temperature decreases . by contrast , if we specified the plasma @xmath1 at each radius instead of the magnetic flux advection rate , the decrease in the temperature results in a decrease in magnetic pressure . this is inconsistent with the results of three - dimensional mhd simulations ( e.g. , * ? ? ? we address the similarity between the concepts of the mass accretion rate and the magnetic flux advection rate in order to facilitate understanding of the concept of the magnetic flux advection rate . the right - hand side of equation ( [ eq : con_int ] ) represents the mass flux crossing at @xmath140 per unit time and we have denoted it by @xmath118 . if there is no mass loss or gain ( e.g , due to inflows and/or outflows ) , @xmath118 is constant in the radial direction . otherwise @xmath118 is a function of @xmath140 ( e.g. , @xmath156 , @xmath157 is a parameter ) . the right - hand side of equation ( [ eq : ind_int_ignore ] ) represents the azimuthal magnetic flux crossing at @xmath140 per unit time and we have denoted it by @xmath126 . if there is no loss or gain of azimuthal magnetic fields ( e.g , due to the dynamo and/or the magnetic diffusion ) , @xmath126 is constant in the radial direction . otherwise @xmath126 is a function of @xmath140 and we prescribed it as equation ( [ eq : phidot ] ) . that is , when @xmath158 , the magnetic flux is conserved in the radial direction , and when @xmath159 ( or @xmath160 ) , the magnetic flux increases ( or decreases ) with a decreasing radius ( see figure 1(b ) in * ? ? ? we note that the azimuthal magnetic flux inside a disk can increase when the azimuthal flux of opposite polarity buoyantly escapes from the disk ( e.g. , * ? ? ? * ) . in the conventional theory , the viscous heating was expressed as @xmath161 and assumed to heat primarily ions ( @xmath162 ) , where @xmath163 is the @xmath129-component of the total stress and @xmath164 is the angular velocity . the results of three - dimensional mhd simulations indicate that the dissipation of the magnetic energy dominates the total dissipative heating rate throughout a disk and is expressed as @xmath165 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? hereafter , we refer to it as the magnetic heating rate . following these simulation results , we employ magnetic heating as the heating mechanism inside a disk , and set the vertically integrated heating rate as follows : @xmath166 dz = - \alpha w_{\rm tot } \varpi \frac{d \omega}{d \varpi } ~ , \end{aligned}\ ] ] where we have used equation ( [ eq : al ] ) . we note that if the magnetic pressure is high , the heating rate can also be high even when the gas pressure is low . the eventual expression of the heating term is not at all unusual ( e.g. , * ? ? ? * ; * ? ? ? * ) except that we consider much lower values of @xmath1 . we will discuss in section [ discussion_main ] a lower limit of @xmath1 below which the mri is stabilized ( e.g. , * ? ? ? * ) so that this expression is no longer valid . we note that the magnetic heating does not always heat primarily ions . @xcite suggested that @xmath167 is required to fit the spectrum of sgr @xmath168 with riaf models . @xcite carried out local shearing box simulations of the nonlinear evolution of the mri in a collisionless plasma considering pressure anisotropy and showed that @xmath169 ( note that the definition of @xmath170 is different from that of @xmath171 in * ? ? ? the exact value we choose for this parameter is not so important , in particular , for the low-@xmath1 solutions . we find that the low-@xmath1 solutions presented in this paper practically unchanged for any value of @xmath172 . therefore , we adopt @xmath173 as a fiducial value . if the ion temperature is higher than the electron temperature , coulomb collisions transfer energy from ions to electrons . the energy transfer rate from ions to electrons per unit volume via coulomb collisions is given by @xcite @xmath174 , \end{aligned}\ ] ] where @xmath175 is the thomson scattering cross section and @xmath176 is the coulomb logarithm ( roughly @xmath177 ) . @xmath178 are modified bessel function of the second kind of the order @xmath179 , respectively . the dimensionless electron and ion temperatures are defined by @xmath180 for technical reason , we use the following formula which uses no special functions , and is accurate to within a factor of 2 when @xmath181 @xcite @xmath182 integrating in the vertical direction , we obtain @xmath183 & ( \theta_{\rm i } > 0.2)\\ \\ \displaystyle ( k t_{\rm e } - k t_{\rm i } ) \frac{(2 \pi)^{1/2 } + ( \theta_{\rm e } + \theta_{\rm i})^{1/2}}{(\theta_{\rm e } + \theta_{\rm i } ) } & ( \theta_{\rm i } < 0.2 ) \nonumber \end{array}\right . ~ .\end{aligned}\ ] ] we note that @xmath184 in almost all solutions presented in this paper . we assume that the radiative cooling occurs through electrons and consider bremsstrahlung , synchrotron , and compton cooling by bremsstrahlung and synchrotron photons as cooling processes . the vertically integrated radiative cooling rate is expressed as @xmath185 following @xcite ( see also * ? ? ? * ; * ? ? ? * ) , bremsstrahlung cooling rate per unit volume is @xmath186~,\end{aligned}\ ] ] where the subscripts @xmath187 and @xmath188 denote the electron - ion and electron - electron bremsstrahlung cooling rates , @xmath189 is fine - structure constant , the function @xmath190 and @xmath191 have the approximate form @xmath192 & ( \theta_{\rm e } > 1 ) \\ \\ \displaystyle 4 \left ( \frac{2 \theta_{\rm e}}{\pi^3}\right)^{1/2 } \left [ 1 + 1.781 \theta_{\rm e}^{1.34}\right ] & ( \theta_{\rm e } < 1 ) \end{array } \right . ~,\end{aligned}\ ] ] @xmath193 @xmath194 and @xmath195 is euler s number . integrating in the vertical direction , we obtain @xmath196 ~.\end{aligned}\ ] ] the synchrotron emissivity of a relativistic maxwellian distribution of electrons in the optically thin limit is given by @xcite : @xmath197 where @xmath198 with the fitting function @xmath199 given by @xcite @xmath200 integrating in the vertical direction , we obtain @xmath201 we assumed that the emission below a critical frequency , @xmath202 , is completely self - absorbed so that the emissivity can be approximated by the blackbody emission from the surface of the disk . following @xcite , we estimate @xmath202 as the frequency at which the synchrotron emission from the region @xmath203 is equal to the blackbody emission ( in the rayleigh - jeans limit ) from the upper and lower surfaces of the region . this condition gives the equation @xmath204 we obtain @xmath202 by solving this equation numerically . integrating over frequency , we obtain @xmath205 , \end{aligned}\ ] ] where the parameters @xmath206 , @xmath207 , @xmath208 , and @xmath209 are defined as @xmath210 and @xmath211 is the incomplete gamma function @xmath212 we adopted the prescription for the compton energy enhancement factor @xmath213 described by @xcite , which is defined to be the average change in energy of a photon between injection and escape : @xmath214 \nonumber \\ & \equiv & 1 + \eta_1 -\eta_2 \left ( \frac{x}{\theta_{\rm e}}\right)^{\eta_3 } ~,\end{aligned}\ ] ] where @xmath215 here , @xmath216 is the probability that an escaping photon is scattered , @xmath217 is the mean amplification factor in the energy of a scattered photon when the scattering electrons have a maxwellian velocity distribution of temperature @xmath218 . following @xcite , the vertically integrated compton cooling by bremsstrahlung and synchrotron photons are given respectively by @xmath219\right\ } ~,\end{aligned}\ ] ] @xmath220 ~ , \end{aligned}\ ] ] where @xmath221 . the vertically integrated heat advection terms for ions and electrons are expressed as @xmath222 @xmath223 where @xmath224 @xmath225 \frac{\partial \ln t_{\rm e}}{\partial \ln \varpi } \nonumber \\ + \frac{\partial \ln \sigma}{\partial \ln \varpi } - \frac{\partial \ln h}{\partial \ln \varpi}~ , \end{aligned}\ ] ] are dimensionless quantities of the order of unity ( hereafter we call them the entropy gradient parameter ) , @xmath226 , @xmath227 , respectively . the positivity ( negativity ) of the entropy gradient parameter means that the heat advection term works as an effective cooling ( heating ) . the entropy gradient parameter for ions , @xmath228 , has a positive value in adaf / riaf solutions , while it has a negative value in lhaf solutions . the global three - dimensional mhd simulations ( e.g. , * ? ? ? * ) and steady , vertically integrated , one - dimensional transonic solutions ( e.g. , * ? ? ? * ; * ? ? ? * ) of optically thin black hole accretion flows indicate that @xmath229 . following these results , we adopt @xmath230 as a fiducial value in this paper . we note that a value of @xmath228 is not important in sle and low-@xmath1 solutions because the heat advection term is negligible compared to the other terms . according to @xcite , @xmath231 can have a positive and negative value ( @xmath232 ) . however , they assumed that no viscous heat goes into electrons ( @xmath233 ) . in the energy equation for electrons , when @xmath172 , the dissipated magnetic energy term is typically greater than or comparable to the heat advection term in the inner region of the disk . therefore the exact value we choose for this parameter is not so important , in particular , for sle and low-@xmath1 solutions in which @xmath234 . we find that the results presented in this paper are practically unchanged for any value of @xmath231 between @xmath235 and @xmath236 when @xmath237 . we show the results for @xmath238 in most part of this paper . we solved the above basic equations at @xmath239 for given parameters @xmath118 , @xmath19 , @xmath141 , @xmath28 , @xmath228 , and @xmath231 . we obtained new thermal equilibrium solutions , low-@xmath1 solutions , in addition to the adaf / riaf ( for positive @xmath228 ) , sle , lhaf ( for negative @xmath228 ) solutions in the optically thin regime . figure [ al05ds3xp05sifoo ] shows the sequences of each thermal equilibrium solution in the @xmath104 versus @xmath240 , @xmath241(thin line ) , @xmath72(thick line ) , and @xmath1 plane . the disk parameters we adopted are @xmath242 , @xmath230 , @xmath238 , @xmath173 , @xmath243 ( solid ) , @xmath154 ( dashed ) , and @xmath110 ( dotted ) , respectively . here @xmath244 is the eddington mass accretion rate , @xmath245 is the energy conversion efficiency , and @xmath246 is the electron scattering opacity . we obtain three types of solutions , adaf / riaf ( for @xmath247 at this radius ) , sle ( for @xmath248 ) , and low-@xmath1 solutions ( for @xmath249 ) . we find that the low-@xmath1 solutions exist above the maximum mass accretion rate of the adaf / riaf solutions , @xmath250 . this indicates that the disk initially staying in the adaf / riaf state undergoes transition to the low-@xmath1 state when the mass accretion rate exceeds @xmath251 . furthermore , the electron temperature in the low-@xmath1 solutions is lower ( @xmath252 ) than that in the adaf / riaf solutions . the energy balance for ions and electrons is illustrated in figure [ al05ds3xp05sieie ] . the upper panel shows the ratio of the heat advection to the magnetic heating for ions , @xmath253 ( this quantity is referred to as the advection factor @xmath254 , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the lower panel shows the ratio of the heat advection to the total heating for electrons ( thin line ) and the fraction of the energy transfer via coulomb collisions to the total heating ( thick line ) . the electrons receive the total amount of the magnetic heating in the low-@xmath1 solutions as well as in the sle solutions even though we introduced the parameter @xmath28 which represents the fraction of heating to electrons . the fraction @xmath255 of the magnetic heating goes into ions and the fraction @xmath28 of the magnetic heating goes into electrons . however , almost all the magnetic heating going into ions is transferred to electrons via coulomb collisions in the low-@xmath1 solutions ( @xmath256 ) . eventually , the total amount of the magnetic heating goes into electrons . this means that the parameter @xmath28 does not appear practically in the energy balance for electrons . the radiative cooling overwhelms the heat advection in the low-@xmath1 solutions . therefore , the magnetic heating balances the radiative cooling in the low-@xmath1 solutions ( @xmath257 ) . the energy balance is essentially the same in both the sle solutions and the low-@xmath1 solutions ( @xmath258 ) the difference is which pressure dominates the magnetic heating . the magnetic heating , which is proportional to the total pressure in our model , is dominated by the gas pressure in the sle solutions and the magnetic pressure in the low-@xmath1 solutions . next we describe the main cooling mechanism in each solution . figure [ al05ds3xp05sirr ] shows the vertically integrated bremsstrahlung ( solid ) , bremsstrahlung - compton ( dotted ) , synchrotron ( dashed ) , and synchrotron - compton ( dash - dotted ) cooling rate for @xmath145 ( bottom ) , @xmath154 ( middle ) , and @xmath155 ( top ) . when @xmath145 , the synchrotron - compton cooling is dominant in the low-@xmath1 solutions for lower mass accretion rate while the bremsstrahlung - compton cooling is dominant for high mass accretion rates ( @xmath259 , @xmath260 ) even though @xmath261 . the synchrotron cooling is relatively ineffective for higher mass accretion rates because of the lower electron temperature . as @xmath141 increases , the electron temperature become high because the large magnetic flux enhances not only the synchrotron cooling but also the magnetic heating . as a result , the synchrotron and synchrotron - compton cooling become efficient . when @xmath243 , the synchrotron - compton is dominant in whole low-@xmath1 solutions . now we show that @xmath262 , @xmath263 , @xmath264 on the low-@xmath1 branch ( see also * ? ? ? * ) , where @xmath265 is the mean temperature . these relations depend on the dependence of @xmath143 on @xmath118 ( our outer boundary condition leads @xmath266 ) . here we introduce the parameter @xmath157 , @xmath267 , in order to leave this dependence explicitly . first , we derive the relations between @xmath118 and @xmath104 . since @xmath268 on the low-@xmath1 branch , equations ( [ eq : h2 ] ) , ( [ eq : mom_phi_int ] ) , and ( [ eq : al_int ] ) yield @xmath269 and @xmath270 . using equations ( [ eq : con_int ] ) , ( [ eq : ind_int_ignore ] ) , and ( [ eq : phidot ] ) , we find that @xmath271 , @xmath272 , and @xmath273 . next , we derive the relation between @xmath274 and @xmath104 from the energy balance of the low-@xmath1 solutions ( @xmath257 ) . here we roughly approximate the radiative cooling rate by @xmath275 for simplicity ( this is the same dependence as non - relativistic bremsstrahlung cooling for single temperature plasma ) . since @xmath276 , we find that @xmath277 . therefore @xmath278 . we find that @xmath279 , @xmath263 , and @xmath280 when @xmath281 . we investigate the relation between the mass accretion rate and the electron temperature in order to explain the anti - correlation between the luminosity and the cutoff energy observed in the bright / hard state . in the low-@xmath1 solutions , we expect that the luminosity is roughly proportional to the mass accretion rate since @xmath282 , and that the electron temperature roughly represents the cutoff energy since the inverse - compton scattering is the dominant radiative cooling mechanism . therefore , the relation between the electron temperature and the mass accretion rate is useful for comparison with the observational data . figure [ al05ds3xp05temd ] shows the relations between the electron temperature and the mass accretion rate for the same parameters as in figure [ al05ds3xp05sifoo ] . we find that the electron temperature is typically @xmath283 in the low-@xmath1 solutions while it is @xmath284 in the adaf / riaf solutions . furthermore , the electron temperature in the low-@xmath1 solutions strongly anti - correlates with the mass accretion rate . since @xmath285 and @xmath286 in the low-@xmath1 solutions , we find that @xmath287 for @xmath281 . according to @xcite , the mri is stabilized for @xmath288 ( we plot this critical point ( filled circle in figure [ al05ds3xp05temd ] ) at which @xmath289 ) . therefore , the low-@xmath1 solutions may not exist under the condition that @xmath290 . we discuss this issue in section [ discussion ] . we also show the results for @xmath291 in figure [ al20ds3xp05sifoo ] and [ al20ds3xp05temd ] . the maximum mass accretion rate of the adaf / riaf solution is around @xmath292 . this maximum mass accretion rate is higher than that for @xmath293 . here we investigate the dependence of @xmath294 on @xmath19 . since @xmath295 , @xmath296 and @xmath297 at @xmath298 , the energy equations yield ( i ) @xmath299 , ( ii ) @xmath300 , and ( iii ) @xmath301 q^{+ } \sim q_{\rm rad}^{-}$ ] . here we assume @xmath28 to be constant for simplicity . we find from equation ( i ) that @xmath302 . if @xmath124 can be written in a simple form of @xmath303 like @xmath304 , we find from equations ( ii ) and ( iii ) that @xmath72 at @xmath305 is independent of @xmath19 and @xmath251 is proportional to @xmath306 exactly . although @xmath124 has a complicated form in our model , our numerical results indicate that the @xmath72 is roughly independent of @xmath19 ( @xmath307 ) . using this value of @xmath72 instead of solving equation ( iii ) , we find from equation ( ii ) that roughly @xmath308 . this result is consistent with @xmath309 by @xcite . we investigated the dependence of the results on @xmath28 because @xmath28 is a poorly constrained parameter . we considered three cases : @xmath310 as an example of model in which ions and electrons receive equal amounts of the dissipated magnetic energy , @xmath311 ( @xmath312 ) as an example of a conventional model in which ions receive a substantial amount of the energy , and @xmath313 as an intermediate example . we note that @xmath314 lies between @xmath315 and @xmath316 in the solutions presented in this paper . figure [ al05xp05sifoo ] shows the thermal equilibrium curves for the same parameters as in figure [ al05ds3xp05sifoo ] but for @xmath317 ( solid ) , @xmath236 ( dashed ) , @xmath318 ( dotted ) , and @xmath319 ( dot - dashed ) , respectively . the energy balance for ions and electrons is illustrated in figure [ al05xp05sieie ] . the ion temperature decreases as @xmath28 increases because the magnetic heating for ions ( @xmath320 ) which is the only heating source for ions decreases . @xmath305 also decreases because of the decrease in the magnetic heating for ions . the electron temperature increases as @xmath28 increases in the adaf / riaf solutions . by contrast , in the low-@xmath1 solutions , the electron temperature is almost independent of the value of @xmath28 because @xmath321 does not appear practically in the energy equation for electrons ( @xmath322 ) . therefore , we find that the exact value of @xmath321 is not so important for the low-@xmath1 solutions . figure [ al05ds3xp05z000sifoo ] shows the thermal equilibrium curves for the same parameters as in figure [ al05ds3xp05sifoo ] but for @xmath323 ( dashed ) . we also plotted the thermal equilibrium curves for @xmath324 ( solid ) for comparison . we obtained the lhaf solutions in the high mass accretion rate and high surface density region . the heat advection works as a heating for ions and balances the energy transfer via the coulomb collisions in the lhaf solutions . above @xmath305 , the heat advection term overwhelms the magnetic heating . hence , the ion temperature becomes higher than that in the adaf / riaf solutions . in the lhaf solutions , the electrons receive a larger amount of heat from ions than that in the adaf / riaf solutions . nonetheless , the electron temperature becomes lower than that in the adaf / riaf solutions ( yet higher than that in the low-@xmath1 solutions ) because the radiative cooling becomes more effective in such higher surface density region . we find that our results are practically unchanged for any value of @xmath231 between @xmath235 and @xmath236 because the heat advection term is negligible compared to the other terms in energy equation for electrons except when @xmath311 . furthermore , even when @xmath325 , the other quantities except the electron temperature in the adaf / riaf solutions does not change practically . we show the thermal equilibrium curves plotted on the @xmath104 - @xmath71 ( thin ) and @xmath72 ( thick ) plane for @xmath311 , @xmath238 ( solid ) , @xmath110 ( dashed ) , @xmath235 ( dotted ) , @xmath243 ( left panel ) , and @xmath110 ( right panel ) in figure [ al05dd3sitt ] . we obtained thermal equilibrium solutions for an optically thin , two - temperature accretion disk incorporating magnetic fields . we included bremsstrahlung emission , synchrotron emission , and inverse compton scattering as the radiative cooling mechanisms , and introduced the parameter @xmath28 which represents the fraction of the magnetic heating to electrons . we prescribed the @xmath129-component of the azimuthally averaged maxwell stress is proportional to the total pressure . in order to complete the set of basic equations , we specified the radial distribution of the magnetic flux advection rate by introducing a parameter @xmath141 . we found a branch of low-@xmath1 solutions in addition to the usual adaf / riaf ( for @xmath326 ) , sle , and lhaf ( for @xmath327 ) solutions . here we remark why we can obtain the low-@xmath1 solutions . first , we prescribed that the @xmath129-component of the maxwell stress is proportional to the total pressure . therefore , if the magnetic pressure is high , we can obtain the magnetic heating rate balancing the radiative cooling rate even in high surface density and low temperature region . second , we specified the magnetic flux advection rate , @xmath328 , in order to complete the set of the basic equations . in the conventional theory , @xmath1 is assumed to be constant ( typically , @xmath329 ) , which means that the magnetic pressure is proportional to the gas pressure . this implies that @xmath330 is just multiplied by a constant ( @xmath331 ) . as a result , we can not obtain the sequence of the low-@xmath1 solutions . on the other hand in our model , a decrease in temperature results in an increase in magnetic pressure ( therefore an increase in the magnetic heating ) under the conservation of the magnetic flux in the vertical direction . this is the reason why we can obtain the sequence of the low-@xmath1 solutions . we note that the exact values of @xmath228 , @xmath231 , and @xmath28 are not so important in the low-@xmath1 solutions . the magnetic heating enhanced by the high magnetic pressure balances the radiative cooling in the low-@xmath1 solutions . the heat advection terms including @xmath228 and @xmath332 are negligible for both ions and electrons . furthermore , @xmath28 does not appear in the energy balance for electrons practically . let us discuss the lower limit of @xmath1 . since @xmath280 and @xmath333 under our outer boundary condition ( @xmath281 ) , we find that @xmath334 , that is , @xmath1 decreases as the mass accretion rate increases in the low-@xmath1 solutions . it has been confirmed by global mhd simulations ( e.g. , * ? ? ? * ) that the mri is not stabilized and hence the magnetic heating rate is expressed in the form of equation ( [ eq : qmag ] ) , at least , when @xmath335 . local mhd simulations ( e.g. , * ? ? ? * ) also indicated that the maxwell and reynolds stresses generated by magnetic turbulence are significant and yield an effective @xmath19-viscosity ( @xmath336 ) in highly magnetized disks ( @xmath329 ) . however , the expression of the magnetic heating employed in our paper may no longer be valid for much lower values of @xmath1 because strong magnetic fields suppress the growth of the mri . we implicitly assumed that the dissipation energy of the turbulent magnetic fields generated by the mri is converted into the thermal energy of the disk gas . therefore , such heating mechanism becomes ineffective if the mri is stabilized . @xcite studied the evolution of the mri in differentially rotating , magnetized flows beyond the weak - field limit , and showed that the mri is stabilized for toroidal alfvn speeds , @xmath337 , exceeding the geometric mean of the sound speed , @xmath338 , and the rotational speed , @xmath339 , ( i.e. , @xmath288 , or equivalently @xmath340 ) . our results satisfy this condition when @xmath341 , @xmath342 , @xmath343 for @xmath242 , and @xmath344 , @xmath345 , and @xmath346 for @xmath347 , respectively . these critical points are denoted by filled circles in figure [ al05ds3xp05temd ] and [ al20ds3xp05temd ] ( see also table [ tbl-1 ] ) . when @xmath1 falls below this critical value , the low-@xmath1 solutions may not exist because there is no heating source balancing the radiative cooling . as a result , the disk may undergo a transition to other states ( e.g , an mdaf - like disk or an optically thick disk ) . let us discuss the thermal stability in this subsection . a general criterion concerning the thermal instability of disks can be expressed as ( see * ? ? ? * ; * ? ? ? * ) @xmath348_{\sigma } > 0 ~ .\end{aligned}\ ] ] in the low-@xmath1 solutions , we ignore the heat advection term because @xmath349 . once again we approximate the radiative cooling rate by @xmath350 for simplicity . we find from @xmath351 and @xmath352 ( see section [ results_main ] ) that the low-@xmath1 solutions do not satisfy the criterion ( [ eq : criterion ] ) , that is , are thermally stable . we note that the thermal stability is independent of @xmath157 . we also remark the thermal stability of the other solutions ( adaf / riaf , sle , and lhaf ) in which the gas pressure is dominant ( @xmath353 ) . equations ( [ eq : h2 ] ) and ( [ eq : con_int ] ) yield @xmath354 and @xmath355 . we find from @xmath356 , @xmath357 , @xmath358 that the adaf / riaf solutions ( @xmath359 ) are thermally stable but the sle ( @xmath258 ) and lhaf ( @xmath360 ) solutions are thermally unstable . the main purpose of this paper is to explain the bright / hard state observed during the bright / slow transition in the rising phases of the transient outbursts of bhcs . in the low / hard state , the x - ray spectrum is described by a hard power law with a high energy cutoff at @xmath16 . when their luminosity exceed @xmath361 , these systems undergo a transition from the low / hard state to the bright / hard state . in the bright / hard state , the cutoff energy decreases from @xmath16 to @xmath362 as the luminosity increases from @xmath361 to @xmath363 ( e.g. , * ? ? ? * ) . beyond the bright / hard state , these systems undergo a transition to the high / soft state going through the vh / spl state . the adaf / riaf solution explains the x - ray spectrum in the low / hard state because the electron temperature is high ( @xmath364 ) . however , this solution does not exist at the high mass accretion rates and does not show the strong anti - correlation between the electron temperature and the mass accretion rate observed in the bright / hard state . the low-@xmath1 solution extends to such high mass accretion rates . therefore , the disk initially staying in the adaf / riaf state undergoes transition to the low-@xmath1 state when the mass accretion rate exceeds @xmath294 . on the low-@xmath1 branches , the electron temperature is low ( @xmath365 @xmath366 ) and strongly anti - correlates with the mass accretion rate . these features are consistent with the anti - correlation between the luminosity and the energy cutoff observed in the bright / hard state . therefore , the low-@xmath1 solution can explain the bright / hard state . in section [ discussion_main ] , we have discussed the lower limit of @xmath1 below which the mri is stabilized therefore the low-@xmath1 solutions may not exist . since @xmath367 in the low-@xmath1 solutions , this means that the low-@xmath1 solution has a maximum mass accretion rate , @xmath368 . the mass accretion rate and the electron temperature at the lower limit of @xmath1 are depicted by filled circles in figure [ al05ds3xp05temd ] and [ al20ds3xp05temd ] ( see also table [ tbl-1 ] ) . we found that when @xmath369 ( or equivalently , @xmath370 ) @xmath371 . this indicates that the disk staying in the adaf / riaf state undergoes transition to the low-@xmath1 disk when the mass accretion rate exceeds @xmath372 , after that , the disk undergoes transition to the optically thick disk when the mass accretion rate exceeds @xmath373 . in this case , the hard - to - soft transition occurs at @xmath373 ( not at @xmath251 ) . this can correspond to the bright / slow transition . here we also remark on the dark / fast transition during which the system undergoes transition from the low / hard state to the high / soft state without going through the bright / hard state and the vh / spl state . we found that when @xmath374 ( or equivalently , @xmath375 ) @xmath376 , that is , there is no optically thin , thermally stable solution above @xmath372 . therefore , the disk staying in the adaf / riaf state immediately undergoes transition to an optically thick disk without through the low-@xmath1 disk . this can correspond to the dark / fast transition . accordingly , we conclude that the bright / slow transition occurs when @xmath126 has a large value and the dark / fast transition occurs when @xmath126 has a small value . we have obtained the low-@xmath1 solutions for optically thin , two - temperature accretion disks incorporating the mean azimuthal magnetic fields , and concluded that the low-@xmath1 solutions explain the bright / hard state observed during the bright / slow transition of bhcs . we assumed that the energy transfer from ions to electrons occurs through coulomb collisions , and considered bremsstrahlung emission , synchrotron emission , and inverse compton scattering as the radiative cooling processes . we prescribed that the maxwell stress is proportional to the total ( gas and magnetic ) pressure . in order to complete the set of basic equations , we specified the radial distribution of the magnetic flux advection rate . accordingly , a decrease in temperature can result in an increase in magnetic pressure under the conservation of the magnetic flux in the vertical direction . in the low-@xmath1 solutions , the magnetic heating is enhanced by the high magnetic pressure . the fraction @xmath377 of the magnetic heating goes into ions and is transferred to electrons via coulomb collisions . the fraction @xmath28 of the magnetic heating goes into electrons . eventually , the total amount of the magnetic heating goes into electrons and balances the radiative cooling ( compton cooling by bremsstrahlung and/or synchrotron photons ) . the electron temperature is lower than that in the adaf / riaf solutions ( @xmath378 ) and strongly anti - correlates with the mass accretion rate in the low-@xmath1 solutions . these features are consistent with the x - ray spectrum observed in the bright / hard state . according to @xcite , the mri is stabilized for @xmath379 . this indicates that the low-@xmath1 solutions disappear below this critical point . when the magnetic flux advection rate is high ( @xmath369 ) , the critical mass accretion rate of the low-@xmath1 solutions , @xmath373 , is above the maximum mass accretion rate of the adaf / riaf solutions , @xmath294 . therefore , the disk initially staying the adaf / riaf state undergoes transition to the low-@xmath1 state when the mass accretion rate exceeds @xmath251 , after that , the disk undergoes transition to the optically thick disk when the mass accretion rate exceeds @xmath373 . this corresponds to the bright / slow transition . on the other hand , when the magnetic flux advection rate is low ( @xmath380 ) , @xmath381 is below @xmath294 . therefore , the disk initially staying the adaf / riaf state immediately undergoes transition to the optically thick disk when the mass accretion rate exceeds @xmath251 . this can correspond to the dark / fast transition . we are grateful to r. narayan for valuable discussion and closely examining my draft . this work is supported by the grant - in - aid for science research of the ministry of education , culture , sports , science and technology ( rm : 20340040 ) and grant - in - aid for jsps fellows ( 20.1842 ) . llccc @xmath382 & @xmath155 & @xmath383 & @xmath384 & @xmath385 + & @xmath154 & @xmath386 & @xmath387 & @xmath388 + & @xmath389 & @xmath390 & @xmath391 & @xmath392 + @xmath318 & @xmath155 & @xmath393 & @xmath394 & @xmath395 + & @xmath154 & @xmath396 & @xmath397 & @xmath398 + & @xmath110 & @xmath399 & @xmath400 & @xmath392 +

a century ago , the study of variable stars was a leading area of astronomy . with the increasing availability of large format optical detectors and inexpensive high speed computing this sub - field , as witnessed by projects such as asas , ogle , catalina sky survey , the palomar transient factory , panstarss and skymapper , is making a come - back . radio astronomy appears to be poised for a similar growth . at meter wavelengths , commercially available signal processing chips make it feasible to image the entire primary beam of a dipole or a cluster of dipoles . these technological innovations lie at the heart of lofar @xcite , mwa @xcite and lwa @xcite . at centimeter wavelengths the `` large number , small diameter '' ( lnsd ) array approach ( made possible by inexpensive signal processing , advances in commercial rf technology , innovative ideas in the design of small diameter telescopes and phased array focal planes ) has now been demonstrated to be a cost effective method of building high speed mapping machines @xcite . the lnsd approach has motivated a new generation of radio facilities : apertif / wsrt @xcite ; meerkat @xcite ; askap @xcite . lllllll[hbtp ] type ii sne & 0.1 - 1 & 10 & 100mpc & 16 & 0.04 & @xcite + type ib / c sne & 0.1 & 0.3 & 50mpc & 14.5 & @xmath1 & @xcite + sn1998bw - like & 0.1 & 0.1 & 300mpc & 18.4 & @xmath2 & @xcite + swj1644 + 57-like & 0.1 & 1 & @xmath3 & 21.7 & 0.1 & @xcite + orphan afterglows & 1 & 1 & 1gpc & 21.0 & @xmath4 & @xcite + ns - ns mergers & 0.11 & 0.13 & 800mpc & 20.5 & @xmath5 & @xcite + [ tab : listoftrans ] we divide radio transients into four categories based on two attributes . the first is the duration of the basic phenomenon ( shorter than or greater than a few seconds ) . the second is their location ( within the galaxy or extra - galactic ) . roughly speaking the duration maps to coherent versus incoherent emission and the location to repeated versus cataclysmic events . pulsars and related phenomenon ( giant pulses , nulling pulsars , erratic pulsars , rotating radio transients , and magnetars ) are the dominant category of short duration radio transients at meter and centimeter wavelengths . there are no secure examples of short duration radio transients that are located beyond the local group . flare stars and associated phenomena are prime examples of long duration radio transients of galactic origin . the focus of this paper is long duration transients of extra - galactic origin . known examples in this group are supernovae @xcite and gamma - ray burst afterglows @xcite . in both cases , the radio emission arises as the fast moving debris interacts with the circumstellar matter . in table [ tab : listoftrans ] we summarize the areal density of radio - emitting supernovae ( including the sub - classes ) and grb afterglows . note the areal density of `` live transients '' ( transients present at any given instant of time ) of both supernovae and grb afterglows is less than 0.05 per square degree . in 2007 , bower et al . [ hereafter , b07 ] reported on the analysis of a single field observed every week as a part of the very large array ( vla ) calibration program . the observations were conducted at 4.8ghz and 8.4ghz and lasted 22 years . the 944 epochs and the weekly cadence makes this data set a most valuable set to probe the decimeter band for long duration transients at the sub - millijansky level . these authors reported the discovery of eight transients found in only one epoch ( hereafter `` single - epoch '' ; duration , 20minutes @xmath6 1week ) and two transients found in rolling 2-month searches ( hereafter , `` multi - epoch '' sources ) . deep observations towards these sources were undertaken at optical , near - ir and x - ray bands . the most remarkable feature of the b07 sources is an absence of optical and near - ir counterparts , despite deep searches ( b07 ; @xcite ) . as noted by @xcite all extra - galactic transients ( regardless of the band at which the transient was discovered ) have detectable optical counterparts , namely , their host galaxies . furthermore , remarkably the areal density of transients live at any given time was estimated to be 1.5deg@xmath7 ( @xmath8mjy ) . this density exceeds that of all other known radio transient source populations by an order of magnitude ( or more ) ; see table [ tab : listoftrans ] . @xcite thus argued that the absence of an optical counterpart means that b07 transients have to be repeating sources of galactic origin , and proposed that b07 transients are old neutron stars ( which naturally satisfy the requirement of being optically almost invisible ) . given that the search for transient and strong variables is one of the primary motivators for the next generation radio facilities ( described earlier ) it is important to critically investigate the b07 transients since this class nominally dominates over all other known classes of radio transients ( see table [ tab : listoftrans ] ) . to this end , here we report on a re - analysis of the original data of b07 ( [ sec : obsreana ] , [ sec : findings ] ) and revisit the transient reported by @xcite ( [ sec : othersurveys ] ) . in [ sec : lognlogs ] we present an update of the expected rate of radio transients . in [ sec : futuresurveys ] we discuss these rates in relation to future synoptic radio imaging surveys and conclude . the data used by b07 arose from a calibrator program carried out during the period 19832005 . all observations were made in the standard continuum mode with 100mhz of total bandwidth in each of two adjacent 50-mhz bands ( ifs ) at center frequencies of 5ghz and 8.4ghz and in both hands of circular polarization . see b07 for more details about the full data - set . for the re - analysis we selected , from the archive , only the raw data relevant to the transients reported in b07 . this means the eight epochs from which the single - epoch transients were first found and the 3 + 8 data sets from which the two multi - epoch transients were found . data were taken at other radio frequencies in about half of the cases . some details of the single epoch and two - month transients can be found in table [ tab : listoffields ] ) . for the re - analysis we used aips @xcite . the data reduction and imaging followed the same path used by b07 with a small exception . b07 employed aips for the flagging and analysis of the single - epoch transients , and used the _ miriad _ package @xcite for imaging the two - month averages . we endeavored to make the calibration and the flagging of uv data ( aips task ` tvflg ` ) data for each epoch as uniform way . following these steps we ran each raw visibility data set through the vla pipeline ( ` vlarun ` ) . no flux density calibrator was observed during any epoch of these test observations . following b07 , the flux density of the phase calibrator ( b1803 + 784 ) was fixed to be 2.2jy ( 5ghz ) and 2.8jy ( 8.4ghz ) . for those epochs with 22ghz and 1.4ghz observations the flux density of the phase calibrator was taken to be 3jy and 2jy , respectively . it is evident from the strong variations in the radio light curves for b1803 + 784 that these mean values are only approximate . our reinvestigation confirm that at least during the period 19811999 the variation was less than 15% . fortunately , an accurate flux density scale is not crucial for our analysis since we report results in terms of the signal - to - noise . following b07 we applied a gaussian weighting to the visibility data in order to limit the effects of bandwidth smearing . this was done by applying a 150-k@xmath9 taper to all visibility data prior to imaging ( ` imagr ` ) . for each field we required that a source be present in both frequency bands ( ifs ) with similar flux densities and with similar positions . images were made with extra large fields - of - view . the wide field - of - view is necessary to reduce the effect of side - lobes that can mimic sources in narrow fields . these final analysis images had a size of about 40-arcmin at 5ghz data and 27-arcmin at 8.4ghz . for guidance , the full - width at half power for vla antennas is 45-arcmin/@xmath10(ghz ) , or 9.3-arcmin at 5ghz and 4.3-arcmin for 8.4ghz . measurements of the vla beam power response beyond the first null are given in @xcite , while the polynomial coefficients needed to correct for the primary beam attenuation can be found in the aips task ` pbcor ` . we do note that these corrections are uncertain for large angular distances from the beam center . all data were taken in the b1950 coordinate system . we stayed in the b1950 system throughout calibration and imaging . lllllll rt19840502 & se & 4.9 & 4.6 & 0.22 & 6.0 & phase center artifact . + rt19840613 & se & 4.9 & 4.6 & 3.3 & 5.7 & side - lobe of bright source . + rt19860115 & se & 4.9 & 4.6 & 1.3 & 14.8 & side - lobe of bright source . + rt19860122 & se & 4.9 & 4.6 & 4.6 & 14.5 & artifact . lower if is bad . + rt19870422 & 2 m & 4.9 & 4.6 & 8.0 & 12.8 & artifact . bad pointing . + rt20010331 & 2 m & 8.5 & 2.6 & 4.4 & 1.5 & no detection ( see [ sec : rt20010331 ] ) . + + rt19920826 & se & 4.9 & 4.6 & 2.0 & 21.2 & * d * snr : _ 6.4 _ ; ` 5.8 ` . + rt19970205 & se & 8.5 & 2.6 & 4.4 & 1.4 & * b * snr : _ 7.7 _ ; ` 5.7 ` ( ca ) + rt19970528 & se & 4.9 & 4.6 & 6.8 & 3.9 & * cnb * snr : _ 7.5 _ ; ` 5.6 ` . ( ca ) + rt19990504 & se & 4.9 & 4.6 & 8.9 & 18.9 & * d * snr : _ 7.3 _ ; ` 5.7 ` . + + rt19970528 & se & 8.5 & 2.6 & 6.8 & 1.3 & * cnb * no detection ( ba ) + rt19990504 & se & 8.5 & 2.6 & 8.9 & 8.3 & * d * no detection ( ba ) [ tab : listoffields ] below we offer a detailed report for each of the ten sources reported in b07 . summarizing our results for the impatient reader we find four of the eight single - epoch transients [ rt19840502 ( [ sec : rt19840502 ] ) , rt19860122 ( [ sec : rt19860122 ] ) , rt19840613 ( [ sec : rt19840613 ] ) , rt19860115 ( [ sec : rt19860115 ] ) ] and one of the two - month transients ( rt19870422 ; [ sec : rt19870422 ] ) to be artifacts . our re - analysis finds that the remaining long - duration transient , rt20010331 ( [ sec : rt20010331 ] ) did not have a significant detection when imaged with aips ( snr of 3.2 ) but the _ miriad _ imaging yields an snr of 7.4 . for the remaining single epoch sources [ rt19920826 , rt19970205 , rt19970528 , rt19990504 ] our analysis ( undertaken by daf ; see [ sec : obsreana ] for a summary of our data reduction ) finds reduced snr . the discrepancy with respect to the results reported in b07 concerned us and so another author ( gb ) reanalyzed these fields . some of the discrepancies arise from uv data flagging but even when the same flagging is used different snr algorithms yield measures which differ by 1 to 1.5@xmath11 . since the sources lie close to threshold of detection even a small shift of a single @xmath11 has an exponential effect in their confidence . in effect , the reality of this group ( taken as a whole ) apparently depends on details of algorithms in aips and _ miriad _ the investigation of which is beyond the scope of this paper . [ sec : rt19920826][sec : rt19990504 ] we detail the analyses and a summary of the snrs of these sources can be found in table [ tab : listoffields ] . we also re - investigated the lone transient candidate reported in @xcite . our re - examination which now correctly includes the number of independent beams that were searched for shows that there is a four percent probability that this candidate is due to noise ( see [ sec : othersurveys ] ) . in the next section ( [ sec : revisedb07 ] ) we synthesize these findings and present our conclusions about the b07 transients . b07 report finding a transient close to the pointing center ( 13@xmath12 ) with a primary beam - corrected flux density of @xmath13jy , or a snr=6.1 . we imaged the calibrated data set and confirmed the presence of emission at this level at the reported position . however , we identified a phase center artifact in the visibility data for the left hand polarization of the upper if band . the effect of this artifact in the image plane is to create strong positive and negative slidelobes , with the positive feature identified as a transient source ( figure [ fig:19840502 ] ) . when the upper if band data are removed the resulting image is noise - like and the peak flux density at the nominal position of the reported transient is @xmath14jy . additional observations were taken during the same epoch at 15ghz . the peak flux density at the same position is @xmath15jy . , @xmath16 , @xmath170.5 , 0.5 , 1.0 , 1.5mjy , with negative ( positive ) contours given by dashed ( sold ) lines . the synthesized beam size is shown in the lower left . ] b07 report finding a transient coincident with a possible host galaxy ( @xmath18 ) and with a primary beam - corrected flux density of @xmath19jy , or a snr=7.0 . our deconvolved image shows a source at that location . gaussian fitting suggests that the source is resolved and this conclusion is supported with an integrated flux density ( @xmath20jy ) being clearly larger than the estimate of the peak flux density of @xmath21jy . imaging the lower and upper ifs separately we find another discrepancy . the peak flux density in the lower if is four times weaker than the upper if band . the likely source of the problem can be seen by comparing the dirty image with the dirty beam ( figure [ fig:19840613 ] ) . rt19840613 appears to be an uncleaned side - lobe of j150123 + 781806 , one of the brighter persistent sources detected in 452 images made by b07 at 5ghz . the putative transient is 56@xmath12 away from j150123 + 781806 to the northwest , close to a local maximum ( 10% of peak ) in the dirty beam at this location . this side - lobe artifact is stronger in the upper band but it is still present in the lower band . deconvolution does not fully remove the side - lobe from the image and the effect is to produce a false transient . we investigated whether the artifact is due to short - timescale ( @xmath22min ) variability of j150123 + 781806 but after dividing the visibility data in half ( by time ) and re - imaging , we found no evidence for variability . the artifact may be due to some low level interference picked up by some antennas baselines but we were not able to identify the bad data . it is possible that rt19840613 is a real transient that unfortunately lies on the side - lobe of the dirty beam , but its gaussian fit parameters and its variation in the lower and upper sidebands suggest that it is not a real source . additional observations were made during this epoch at 1.5ghz . the data quality is good and the images have no obvious artifacts . no source is visible at the transient position . the peak flux density is @xmath23jy . . note that the transient candidate rt19840613 lies at the same angle and position as a side - lobe from the bright source j150123 + 781806 located in the bottom right corner of this image . , title="fig : " ] + . note that the transient candidate rt19840613 lies at the same angle and position as a side - lobe from the bright source j150123 + 781806 located in the bottom right corner of this image . , title="fig : " ] b07 report a transient with a primary beam - corrected flux density of @xmath24jy , or a snr=5.5 . some of the same issues with the image of the previously discussed rt19840613 were also seen for rt19860115 . in figure [ fig:19860115 ] we show the dirty image along with the synthesized beam . the rt19860115 appears to be an uncleaned side - lobe of j150123 + 781806 and hence not a real transient . rt19860115 lies at the same angular distance ( 3.3@xmath25 ) and position angle ( 100@xmath26 ccw cc ) of a local maximum in the side - lobe structure ( 25% of the peak beam ) . we were unable to identify the source of these strong side - lobes . as in the case of rt19840613 , we were able to rule out that the strong side - lobes originated from short - term variability of j150123 + 781806 . . note that the transient candidate rt19860115 lies at the same angle and position as a side - lobe from the bright source j150123 + 781806 located in the upper right corner of this image . , title="fig : " ] + . note that the transient candidate rt19860115 lies at the same angle and position as a side - lobe from the bright source j150123 + 781806 located in the upper right corner of this image . , title="fig : " ] b07 report a transient with a primary beam - corrected flux density of @xmath27jy , or a snr=6.4 . images made with all the visibility data do show emission at the position of rt19860122 with the reported peak flux density . however , there does appear to be some erroneous visibility data resulting in low - level rings in the image centered at the phase center . the bad data were traced to the lower if . if the data for the lower if band are removed there is no source at the reported transient position ( peak flux density of @xmath28jy ) . this is one of two transients found by binning individual epochs into two - month averages . b07 report a 5ghz transient with a primary beam - corrected flux density of @xmath29jy , or a snr of 6.1 . the source j150050 + 780945.5 is positionally coincident with a blue host galaxy ( @xmath30 ) and has been suggested by @xcite as a candidate afterglow from a binary coalescence event . three single epochs were used to form the average . they are in yyyymmdd format : 19870414 , 19870422 and 19870429 . a close inspection of the data reveals a problem . the last two of the three epochs used to find rt19870422 were pointed to a different part of the sky , an error that was made when the original observing files were written . the source is actually j210133.67 + 373528 , and appears only as a transient when added inappropriately to the data from the standard on - the - sky test field . in hindsight we recognize these to be the same type of errors as seen earlier ( see * ? ? ? this is the second of the transients which were found by binning individual epochs into two - month averages . b07 report a 5ghz transient with a primary beam - corrected flux density of @xmath31jy , or a snr=7.4 . separately , @xcite reported a marginal x - ray source at this position . eight single epochs were used to form the average . they are in yyyymmdd format : 20010306 , 20010314 , 20010321 , 20010328 , 20010403 , 20010411 , 20010418 and 20010425 . we calibrated and imaged the eight epochs of observations from 2001 march 6 to 2001 april 25 ( all in b - configuration ) . while the four quiescent sources from b07 can be seen in this deep image , we do not identify a significant source at the position of rt20010331 . the brightest peak within the synthesized beam ( natural weighting ) is @xmath32jy ( or 3.2@xmath11 ) . an independent reduction using the aips package by gb confirms the absence of this source . the first problem lies with the primary beam corrections reported in b07 . rt20010331 lies 4.4@xmath33 from the phase center , close to the 10% response radius of the primary beam . our estimate for the rms noise of @xmath34jy lies within a few percent of the theoretical value . the beam - corrected rms noise in this case ( at 10% response of the primary beam ) would then be about @xmath35jy , not the @xmath36jy given in b07 . rather than correcting at the 10% radius , it appears that the flux density and noise in b07 were mistakenly corrected at the 14% power level . this multiplicative error has no impact on the signal - to - noise . in order to investigate this signal - to - noise discrepancy between our image and b07 , we split the data in various ways ( separate epochs , months of march and april epochs , adjacent if bands ) and re - imaged , looking for a bright peak . none were found . the discrepancy between b07 and the work reported apparently can be traced back to differences in the _ miriad _ and aips imaging packages . our calibrated visibility data , when processed through _ miriad _ using nearly identical imaging parameters as those in aips , gives a flux density of @xmath37jy ( 7@xmath11 ) . we have no explanation for the discrepancy between the two results obtained from aips and _ miriad_. it is worth noting that the peak flux densities of the persistent sources identified by b07 agree in these images to @xmath38 . for this paper we accept the analysis given here . there was one epoch ( 2001 march 6 ) in which data were also taken at 5ghz . the peak flux density at the position of rt20010331 is @xmath39jy . in summary , we find no evidence to support that rt20010331 is a significant detection . b07 report a transient with a primary beam - corrected flux density of @xmath40jy , or a snr=6.4 . we confirm a source at this position but with a slightly reduced snr . using natural weighting of the gridded visibilities , our measured peak flux density is @xmath41jy , or 5.8@xmath11 . despite the lower significance , there is some confidence that rt19920826 is real since it appears in both ifs with comparable flux densities . further investigation by gb shows that the difference in the snr between b07 and the analysis here can be traced to differences in flagging of the uv data ( of two specific antennas ) . b07 report a transient with a primary beam - corrected flux density of @xmath42jy , or a snr=7.5 . these observations were taken during a time when some antennas were being moved to the b configuration and so we applied antenna position corrections ( via aips task ` vlant ` ) before calibration . rt19970528 is 6.8@xmath25 from the center of the image , where the response of the antennas is only 16% of their peak . the uncorrected flux density is @xmath43 . a point source search ( aips task ` sad ` ) of the four million pixels enclosed interior to a radius around this candidate , shows six other un - cataloged candidates with similar signal - to - noise . the primary - beam corrected flux density @xmath44mjy is similar that from to b07 . this estimate does not include a correction for temporal smearing due to the sources distance from the phase center , nor for the added uncertainty in the magnitude of the primary beam correction beyond the 20% point . correcting for these terms we get @xmath45mjy . gb re - investigated this source and found that with the b07 analysis the snr varies with the approach ( peak / rms , ` sad ` , ` jmfit ` ) between 6.2 , 7.4 and 7.1 . ( without any corrections discussed above ) . the corresponding snrs for the image discussed above is 6.1 , 5.1 and 6.8 . the source is a bit `` ratty '' and this may explain the variation in snr . we therefore find this source to be a weak detection . additional observations were made during this epoch at 8.5ghz . the source lies close to the first null ( @xmath46 from the phase center ) of the beam at this frequency . the attenuation by the primary beam is severe and hence the sensitivity is not sufficient to provide any strong spectral index constraints . b07 report a transient with a primary beam - corrected flux density of @xmath47jy , or a snr=7.3 . @xcite report an x - ray source in the vicinity of the radio source . deep multi - wavelength data is consistent with the x - ray source arising from a qso but located five arcseconds from the putative radio transient . we find the ( uncorrected ) peak flux density is @xmath48jy ( or an snr of 5.7 ) . rt19990504 lies 8.9@xmath33 from the phase center , close to the first null ( @xmath49 from the phase center ) where we would not expect to find sources . the polynomial expressions used in aips to correct for the beam attenuation are increasingly inaccurate outside the 20% response radius , and they are not applicable close to the null . the azimuthally averaged measured value of the beam response is 1.8% , implying a flux density correction factor of 55@xmath50 @xcite or @xmath51mjy . as with rt19970528 , the analysis of b07 data by gb finds snr of 5.6 ( peak / rms ) , 7.3 ( ` sad ` ) and 6.8 ( ` jmfit ` ) and a similar variation with the image reported here . it is worth noting that the source may be extended and also that a visual inspection of the annular region reveal a number of similar sources . we conclude that rt19990504 is not a robust detection . additional observations were made during this epoch at 8.5ghz . an image was made from these higher frequency data but , as expected , no source was found . this is the only single - epoch transient identified at 8.5ghz by b07 . they report a primary beam - corrected flux density of @xmath52jy , or a snr=7.8 . rt19970205 is 4.4@xmath25 from the center of the image , where the response of the antennas is only 9.8% of their peak . there is an elongated source at this position with a peak flux density of @xmath53jy ( using natural weighting ) . there is some indication that the source is extended since the integrated flux density is about twice the peak flux density . the peak flux density was likely underestimated by b07 since they did not fully correct for bandwidth or temporal smearing effects . these data were taken in the bna array with an integration time of 3 1/3 s and a 50-mhz bandwidth . in this observing configuration , the effects of temporal smearing with this dump time will likely reduce the peak flux density by only a few percent . however , chromatic aberration is expected to be larger with the effect of smearing a point source along the radial direction @xcite . accounting for the bandwidth effect , we measure a peak flux density of @xmath54mjy ( snr=5.7 ) . next , due to the uncertainty about the form of the primary response function beyond the 20% point ( aips task pbcor ) , we only know that the peak flux of rt19970205 lies in the range of 2.94.4mjy . as noted above the source is likely extended and this may account for the variation in snr ( from @xmath55 to 7.6 ) . @xcite presented a survey for 5ghz radio transients at low galactic latitudes . most of the data were reduced in near - real time ( 2hr delay ) and transient candidates were followed up with radio , visible light and x - ray instruments . the authors reported a single transient candidate , j@xmath56 with a peak specific flux of @xmath57mjy ( 5.8@xmath11 ) and with no obvious optical counterpart . we have re - analyzed this data set and confirm that the snr is 5.8 . considering that the search resulted from inspecting @xmath58 independent beams ( [ sec : numberindependentbeams ] ) and assuming gaussian statistics , we find that the probability of the highest event found in these many independent beams is attributable to chance or noise is 3.6% ( see [ sec : a ] ) . therefore , we advocate that this candidate is not a real event . as noted in [ sec : introduction ] the ten transients reported in b07 were remarkable for their apparent abundance ( 1.5 per square degree , at any given epoch ) _ and _ the lack of quiescent optical counterparts . in [ sec : findings ] we found that five of the b07 transients are artifacts arising out of ( rare ) data acquisition problems or imaging artifacts ( see table [ tab : listoffields ] ) . we find that the 2-month transient rt20010331 is not well detected . we argue that the single transient reported by @xcite is also not a robust detection . in [ sec : findings ] we present our analysis for the remaining four transients ( all of which are single - epoch transients ) : rt19920826 , rt19970205 , rt19970508 and rt19990504 . we find snrs which are typically 1-@xmath11 below that reported in b07 . the revised and b07 snrs are given in table [ tab : listoffields ] . we note that an independent pipeline written in parseltongue @xcite , a python interface to aips , and run on three of these four sources confirms the lower snr values found here @xcite : rt 19920826 ( snr=6.0 ) , rt 19970528 ( snr=4.4 ) , and rt 19990504 ( snr=4.0 ) . the lowering of snr ( from between 7 and 8 to between 5 and 6 ) has a pernicious effect when the number of independent beams which were search is included . in [ sec : numberindependentbeams ] we estimate this number to be @xmath59 . in [ sec : a ] we derive the probability density function for the highest @xmath60 values of @xmath61 gaussian random numbers . in figure [ fig : pdistpoints ] we plot the density function for the highest value ( @xmath62 ) and the fourth highest value ( @xmath63 ) . as can be seen from this figure [ fig : pdistpoints ] , if the snrs reported here are accepted then the global case for the remaining b07 transients is entirely weakened . if , on the other hand , the snrs reported in b07 are accepted then the four transients reported in b07 do argue for a new class of radio transients . ) and the fourth highest value ( @xmath63 ) of a population of @xmath64 gaussianly distributed random numbers with zero mean and unit variance . the probability density function plotted here is discussed in [ sec : a ] and the justification for @xmath61 can be found in [ sec : numberindependentbeams ] . the dots represent the snrs of the following single - epoch transients discussed in table [ tab : listoffields ] . ( from left to right ) : rt19905054 , rt19920826 , rt19970205 and rt19970228 . ( top ) : snrs as reported in this paper . ( bottom ) : snrs as reported in b07 . note that the vertical location of the dots is arbitrary . for details of these density distributions see [ sec : a ] . , title="fig:",width=259 ] ) and the fourth highest value ( @xmath63 ) of a population of @xmath64 gaussianly distributed random numbers with zero mean and unit variance . the probability density function plotted here is discussed in [ sec : a ] and the justification for @xmath61 can be found in [ sec : numberindependentbeams ] . the dots represent the snrs of the following single - epoch transients discussed in table [ tab : listoffields ] . ( from left to right ) : rt19905054 , rt19920826 , rt19970205 and rt19970228 . ( top ) : snrs as reported in this paper . ( bottom ) : snrs as reported in b07 . note that the vertical location of the dots is arbitrary . for details of these density distributions see [ sec : a ] . , title="fig:",width=259 ] the above approach of using a fixed threshold for all epochs does not result in optimal detection . in particular , the threshold for a low resolution survey is lower than that for a higher resolution survey ( since the latter has a correspondingly larger number of synthesized beams ) . b07 addressed this problem by requiring that the probability of a false detection ( pfd ) in an individual epoch was constant and less than @xmath65 where @xmath65 is the total number of images . with this approach , the expectation number of false detections is 1 for the entire survey . applying the b07 method we find the following pfds : rt19920826 ( log(pfd)=@xmath66 ) ; rt19970205 ( @xmath67 ) ; rt19970528 ( @xmath68 ) ; rt19990504 ( @xmath69 ) . with this more refined approach only rt19920826 and 19990504 survive . however , for reasons discussed in [ sec : rt19990504 ] we have misgivings about rt1990504 . an entirely different approach ( and in some ways orthogonal to the above snr based approach ) is to look at the angular distribution of the transient sources with respect to the primary axis . basic interferometry theory informs us that the dirty image is simply the fourier transform of the visibility data . as such the radiometric noise in the dirty image should be independent of the angular offset from the phase center . in contrast , the point source sensitivity decreases as one goes away from the pointing center and this is governed by the primary beam response ( assuming that the spectral resolution of the survey is high enough that the delay beam is larger than the primary beam ) . thus , once the minimum snr for detection is fixed , cosmic sources should be concentrated towards the pointing direction whereas noise spikes ( masquerading as threshold point sources ) should be uniformly distributed . in [ sec : primarybeam ] we derive the expected distribution of cosmic sources as a function of the angular offset . in figure [ fig : primarybeampoints ] we plot the expected cumulative distribution and also the angular offset of the four sources which are not artifacts but whose snr seems to be under dispute , namely rt19920826 ( [ sec : rt19920826 ] ) , rt19970205 ( [ sec : rt19970205 ] ) , rt19970528 ( [ sec : rt19970528 ] ) and rt19990504 ( [ sec : rt19990504 ] ) . from this figure one can see that only rt19920826 lies in the expected region whereas the remaining three are collectively improbable . . the sources are represented by dots and are ( from left to right ) : rt19920826 , rt19970528 , rt19970205 , and rt19990504 . the points are deliberately placed at cumulative probability of 1% . rt19920826 is firmly within the region where one naturally expects cosmic sources.,width=259 ] in summary , two different statistical tests , one based on snr and the other making use of the spatial signature provided by the primary beam , suggest that of the remaining four sources detected at threshold , only one , namely , rt19920826 is a good detection . thus a simple interpretation of our re - analysis is that the rate of b07 transients is considerably lower than that reported by b07 , perhaps an order of magnitude smaller . we acknowledge that the discrepancy between the analysis presented in b07 and the analysis presented here ( and re - investigated ) is disturbing . in the previous section ( [ sec : remainingfourtransients ] ) we investigate the reasons for the discrepancy in the snrs and variously find possible and plausible causes : flagging of data , the choice of data reduction package ( aips versus _ miriad _ ) and the specific method used to compute snrs . however , none of these explanations are satisfactory . we are continuing this investigation but at the present time we consider this topic to beyond the scope of the paper . in contrast , the sources which we find as artifacts have ready explanations ( see table [ tab : listoffields ] ) . one source is a result of the file header containing a pointing direction of the previous pointing . another is due to a systematic associated with local signals ( rfi ) . these signals do not have the natural fringe rate of cosmic sources and appear as candidates close to the phase center . two are side - lobes of a stronger sources . suffice to say that such rare errors will be found if one inspects sufficiently large number of beams ! in figure [ fig : rates ] we plot the areal densities of three known transients ( sn1998bw - like , type - ii rsn and swift j1644 + 57-like ) and that of two expected classes ( ns - ns mergers and orphan afterglows of long duration grbs ) of transients . the areal density of these five classes is also summarized in table [ tab : listoftrans ] . we briefly discuss each of these five classes of transients below . the traditional extra - galactic radio transients are type ii radio supernovae . the rate that we present here is based on the single radio sn detected in a blind radio survey by @xcite and it agrees with an independent estimate by @xcite . as illustrated by the example presented in @xcite , the main advantage in the search for radio sne in a blind survey is the unique view that it provides to the otherwise hidden population of heavily obscured sne . the observed rate of other types of radio sne ( e.g. ordinary type ib / c sne ) are considerably lower than that of type ii supernovae and are not discussed further . radio emission is expected from both classical long duration gamma - ray bursts ( grbs ) as well as the more abundant but less luminous grbs exemplified by grb980425 associated with the energetic sn ic supernova , sn1998bw @xcite . the radio emission is far brighter than that of ordinary core collapse sn ( ii , ib , ic ) and the increased volume makes up for the intrinsically smaller birth rate . the recent discovery of the energetic supernova sn2009bb @xcite demonstrates that radio surveys can find such sources without resorting to high energy ( gamma - ray ) missions . @xcite estimated the number of afterglows from classical grbs and whose explosion axis is directed away from us ( `` orphan '' afterglows ) . the expected rate depends strongly on the poorly constrained @xmath70-ray beaming . on one hand this makes any rate prediction uncertain . on the other hand , even a non - detection by the kind of survey that we discuss below will provide an independent and unique constraint on the average opening angle of long gamma - ray bursts , their true rate and total energy output @xcite . the areal density in figure [ fig : rates ] is derived using a typical opening angle of 10 degrees . a surprising and apparently an important development in the field of radio transients took place just this year with the discovery of a radio transient associated with the nucleus of a modest size galaxy . the source , swiftj1644 + 57 was initially detected as a hard x - ray transient @xcite . subsequent follow up found a bright , compact self - absorbed radio counterpart , localized at the center of a normal galaxy at @xmath71 @xcite . the current view is that swiftj1644 + 57 arose from a relativistic jet produced when a star was tidally disrupted as it passed too close to an otherwise dormant super - massive nuclear black hole ( e.g. * ? ? ? shortly , thereafter , a second candidate non - thermal tidal disruption event ( tde ) was recently proposed @xcite . events such as these give us an opportunity to study the activity of 10@xmath72 - 10@xmath73 m@xmath74 supermassive black holes in otherwise normal galaxies . the areal density in figure [ fig : rates ] is calculated assuming an observed rate of 0.2 @xmath75 swiftj1644 + 57-like events and a gamma - ray beaming factor of @xmath76 @xcite . nominally , swiftj1644 + 57-like sources appear to be the most frequent extra - galactic transients that will be found in radio transient searches . we acknowledge that the uncertainty of both the observed rate and the gamma - ray beaming is high and the true rate may be significantly different . now we come to the most uncertain as well as potentially the most important extra - galactic radio transient the merger of two neutron stars ( or a black hole and a neutron star ) . it is generally accepted ( or expected ) that short hard bursts are on - axis explosions of these mergers @xcite . as in long duration grbs , radio emission is expected by afterglow ( on - axis or orphan ) . the rates are highly uncertain because there are very few observations of short hard grbs . thus there still continues to be a debate about the geometry of these explosions ( `` jetted '' or not ) . next , while the expected radio emission is straightforward to estimate ( subject to the usual parametric uncertainties of the energy fractions of relativistic electrons and magnetic field ) an additional uncertainty is the density of the ambient gas ( which is necessary for the production of the afterglow emission ) . regardless of the uncertainty whether neutron star mergers are the sources of short grbs or not , a substantial sub- and mildly relativistic outflow is expected to be ejected during the merger . @xcite estimate radio emission from these outflows . the areal density in figure [ fig : rates ] is calculated based on their estimates for ns - ns mergers @xcite ] , assuming a ns - ns merger rate of 300 @xmath77 and that any merger ejects @xmath78 erg of a mildly relativistic outflow . we note that @xcite suggested that rt19870422 was the radio emission from the remains of a neutron star merger . however , as noted in [ sec : findings ] this source is an artifact . there are sound reasons to continue the exploration of the dynamic radio sky . radio searches are an ideal way to discover core - collapse supernovae embedded in or behind dusty regions . the discovery of sn2009bb shows that large radio searches can find urgently needed additional examples of nearby low luminosity grbs . next , the many rewards of radio follow up observations of swift1644 + 57 ( accurate localization , energetics , beaming , outflow velocity ) show the tremendous diagnostic power of radio observations of this entirely new class of extragalactic transients . as exciting as these developments are , the search for new classes of radio transients has involved several false starts . the euphoria that followed the discovery of a highly dispersed ( and therefore argued to be of extragalactic origin ) millisecond burst @xcite was rapidly diminished by the discovery of many such bursts , presumably of terrestrial origin @xcite ; but see @xcite . similarly , a long - duration transient found by @xcite and @xcite was later traced to a glitch in the vla on - line data taking system @xcite , apparently affecting 0.29% of all first survey pointings @xcite . finally , our re - analysis ( see [ sec : rt19870422 ] ) shows that the claim of late time radio emission from neutron star coalescence @xcite is premature . here we focused on the potentially new class of long duration radio transients reported in b07 . we rule out six of the ten transients and cast doubts on some of the remaining ones . however , even with one or two survivors the long duration transients of b07 remain of great interest . first , even with a diminished number of transients the implied areal rate of b07 transients would be comparable to the recently established class of swiftj1644 + 57 transients . however , _ unlike any other long duration radio transient _ ( to wit , supernovae ; active stars ; tidal disruption events ; gamma - ray burst afterglows , beamed or otherwise ) the b07 transients are remarkable for the absence of a quiescent optical counterpart . the event rt19920826 survives two independent tests . as such it useful to speculate on the origin of this transient . the absence of an optical ( b07 ) and near ir quiescent source could mean one of two origins . the event is extra - galactic in origin but the host galaxy is faint enough not to have been detected ( as does happen for a few percent of long duration grb host galaxies ) or that it is offset from a host galaxy ( as in the case for a few short hard bursts ; @xcite ) . alternatively , the event is a galactic neutron star and we have to then appeal to an optically invisible galactic ( neutron star ) population @xcite . clearly , a new survey which can net a dozen of such sources ( but brighter ! ) would help resolve the origin . for the galactic hypothesis one expect no quiescent counterpart even at hst sensitivity whereas detectable galaxies , in the majority , are expected for the extra - galactic hypothesis . there are great gains in the discovery of new classes of radio transients but at the same time the path to true success is littered with false starts or easy speculations . the way forward must incorporate the lessons learnt from the false starts . we elaborate on this conclusion below . to start with we believe that the field of radio transients ( at least in the decimeter band ) is sufficiently mature that any new survey which just sets an upper limit relative to the known population is of marginal value . future surveys have to be sufficiently deep and cover large enough sky that success ( i.e. detection of a few to many transients ) is assured . in our opinion , this means that a survey should be designed to find at least one or more swiftj1644 + 57-like transient ( figure [ fig : rates ] ) . next , timely and multi - wavelength follow - up is essential . for example , the transients reported in @xcite , @xcite and @xcite have plausible origins as supernovae and swiftj1644 + 57-like sources ( see figure [ fig : rates ] ) . however , the lack of timely follow - up or deep multi - wavelength followup of these events preclude us from coming to a definitive conclusion . finally , the search should be restricted to sources with a high level of significance . this certainly means paying attention to the large number of beams searched . however , at low thresholds and with large number of beams ( cf [ sec : findings ] ) it would be prudent to set thresholds beyond mere statistical considerations . a threshold of 9 or even @xmath79 may be appropriate . alternatively , an immediate verification of a transient by deeper observation or a confirmation by observations at other wavelengths would allow detection of transients closer to threshold . we start with a discussion of two recent developments . @xcite undertook an ambitious program similar in spirit to b07 , namely the investigation of fields surrounding vla calibrators . sources with duration between 4 and 40d were searched for . the total integration time was 435hr . no transient source in the ghz range with flux greater than 8mjy was found . the authors place an upper limit to the areal density of 0.032deg@xmath7 . assuming @xmath80 scaling this areal density is 4.4deg@xmath7 and is well above the b07 rate . the first survey imaged the sky in an hexagonal grid , in which each position in the survey footprint was observed on average 34 times @xcite . @xcite used this fact to construct light curves of sources detected in individual first survey snapshots . they identified 1627 variable candidates with variability exceeding @xmath81 . this effort is probably the largest variable and transient survey ever carried out . one disadvantage of such a survey for transient identification is that the co - added images are not much deeper than a single epoch image . this make it hard to tell if an apparent transient source is really a transient or just a variable source that exceeded the detection threshold in one of the epochs . the limitations discussed above lead us to suggest a new evla survey specifically tailored to systematically explore the radio sky . following that we review a far more ambitious survey the vast key project on askap . for the discussion below we will adopt the rates summarized in table [ tab : listoftrans ] . the rates are specified to a flux density of 0.3mjy and are extrapolated to higher flux densities as n(@xmath82 . a moderately ambitious survey with the evla can result in great progress . this survey has two virtues . one , the evla offers excellent spatial resolution . next , the evla is fully commissioned and is working to specifications . specifically consider a 100 square degree survey undertaken in the 24ghz band . an integration time of 85s results in a sensitivity of 0.3mjy ( 10-@xmath11 ) . a single epoch covering 100 square degrees would require 50 hours . as can be seen from figure [ fig : phase ] ( taken from table [ tab : listoftrans ] ) such a survey would have to explore a variety of time scales to probe the emerging classes of transient . fifteen epochs could reasonably cover the range of a week to years . noting the great importance of multi - wavelength imaging data , such a survey would sensibly focus on regions of sky where considerable multi - wavelength data ( including radio ) exists . one such region is , for example , the sdss equatorial stripe @xcite . furthermore , the high instantaneous sensitivity of the evla makes rapid follow up of newly transients possible . after the first four epochs the reference images will be twice as deep as the survey field . a single new epoch would then yield about ten swiftj1644 + 57-like sources and four supernovae . ten such images may find a new example of an sn1998bw - like event , several clear examples of orphan afterglows and have an excellent chance of finding the first examples of neutron star mergers . we note that these different classes of objects have different characteristics , both in duration ( figure [ fig : phase ] ) and also in host magnitudes and location with respect to host galaxy . therefore , it is possible to distinguish between different types of objects . as note earlier , rapid verification of a transient ( either by additional and deeper radio observations or observations at other wavelengths ) can reduce the requirement for a high detection threshold . this would then require close rapid reduction well within the reach of modern computers . the variable & slow transient ( vast ) is an approved key project of askap . the vast - wide survey aims to survey in the 1.2ghz band about 10,000 square degrees every day for 2 years . with 40-s integrations the expected single - epoch rms is 0.5mjy ( vast memo#1 ) . since the survey is planned to be undertaken daily the reference image will be built up quite rapidly . a comparison of data obtained at a new epoch when compared with the built - up reference image can be expected to detect about six swiftj1644 + 57-like sources . type ii sn and swiftj1644 + 57-like sources are relatively long lived and so one could consider averaging the daily images ( to say 10 days ) . the resulting summed data has a sensitivity of 0.16mjy . the expected number of transients per such summed image is nominally 32 ( type ii sn ) , 82 ( swiftj1644 + 57 ) , 8 ( orphan afterglows ) and 4 ( neutron star coalescences ) . to sum up , the dynamic radio sky remains a rich area for exploration . based on what we know about the areal rates for the _ known _ transient sources , future synoptic radio imaging surveys are expected to yield substantial numbers of exotic transients . such surveys will also provide the definitive test for the b07 population . daf thanks jim condon and alicia soderberg for important discussions early on in this project . we thank steve croft for a most careful reading of the paper and j. condon for making several insightful suggestions . the national radio astronomy observatory is a facility of the national science foundation operated under cooperative agreement by associated universities , inc . srk thanks the department of astronomy , university of wisconsin at madison for their hospitality . eoo is supported by an einstein fellowship and nasa grants . srk s research in part is supported by nasa and nsf . this research has made use of data from the university of michigan radio astronomy observatory which has been supported by the university of michigan and by a series of grants from the national science foundation , most recently ast-0607523 . this research has made use of nasa s astrophysics data system . , j. , et al . 2010 , classical and quantum gravity , 27 , 173001 bannister , k. w. , murphy , t. , gaensler , b. m. , hunstead , r. w. , & chatterjee , s. 2011 , , 412 , 634 bannister , k. w. , murphy , t. , gaensler , b. m. , hunstead , r. w. , & chatterjee , s. 2011 , erratum , july 2011 becker , r. h. , white , r. l. , & helfand , d. j. 1995 , , 450 , 559 bell , m. e. 2011 , `` the low frequency array and the transient and variable radio sky '' , phd thesis , southampton university . bell , m. e. , fender , r. p. , swinbank , j. , et al . 2011 , , 415 , 2 berger , e. 2009 , , 690 , 231 berger , e. , kulkarni , s. r. , frail , d. a. , & soderberg , a. m. 2003 , , 599 , 408 bloom , j. s. , et al . 2009 , arxiv:0902.1527 bloom , j. s. , et al . 2011 , science , 333 , 203 booth , r. s. , de blok , w. j. g. , jonas , j. l. , & fanaroff , b. 2009 , arxiv:0910.2935 bower , g. c. , saul , d. , bloom , j. s. , bolatto , a. , filippenko , a. v. , foley , r. j. , & perley , d. 2007 , apj , 666 , 346 bower , g. c. 2011 , , 732 , l12 bower , g. c. , whysong , d. , blair , s. , croft , s. , keating , g. , law , c. , williams , p. k. g. , & wright , m. c. h. 2011 , arxiv:1107.1517 burke - spolaor , s. , bailes , m. , ekers , r. , macquart , j .- p . , & crawford , f. , iii 2011 , , 727 , 18 burrows , d. n. , kennea , j. a. , ghisellini , g. , et al . 2011 , , 476 , 421 cenko , s. b. , fox , d. b. , moon , d .- s . , et al . 2006 , , 118 , 1396 cenko , s. b. , et al . 2011 , arxiv:1107.5307 chatterjee , s. , murphy , t. , & vast collaboration 2010 , bulletin of the american astronomical society , 42 , # 470.12 cotton , w. d. & perley , r. 2010 , obit development memo series no . 17 croft , s. , bower , g. c. , keating , g. , law , c. , whysong , d. , williams , p. k. g. , & wright , m. 2011 , , 731 , 34 croft , s. , tomsick , j. a. , & bower , g. c. 2011 , arxiv:1107.5039 dewdney , p. e. , hall , p. j. , schilizzi , r. t. , & lazio , t. j. l. w. 2009 , ieee proceedings , 97 , 1482 ellingson , s. w. , clarke , t. e. , cohen , a. , et al . 2009 , ieee proceedings , 97 , 1421 galama , t. j. , vreeswijk , p. m. , van paradijs , j. , et al . 1998 , , 395 , 670 gal - yam , a. , et al . 2006 , , 639 , 331 gehrels , n. 1986 , , 303 , 336 gehrels , n. , ramirez - ruiz , e. , & fox , d. b. 2009 , , 47 , 567 giannios , d. , & metzger , b. d. 2011 , , 1137 gregory , p. c. , & taylor , a. r. 1986 , aj , 92 , 371 gregory , p. c. , capak , p. , gasson , d. , & scott , w. k. 2001 , galaxies and their constituents at the highest angular resolutions , 205 , 98 greisen , e. w. 2003 , information handling in astronomy - historical vistas , 285 , 109 hodge , j. a. , becker , r. h. , white , r. l. , richards , g. t. , & zeimann , g. r. 2011 , , 142 , 3 johnston , s. , taylor , r. , bailes , m. , et al . 2008 , experimental astronomy , 22 , 151 jonas , j. l. 2009 , ieee proceedings , 97 , 1522 , v. , et al . 2004 , , 601 , l179 keane , e. f. , kramer , m. , lyne , a. g. , stappers , b. w. , & mclaughlin , m. a. 2011 , , 415 , 3065 kettenis , m. , van langevelde , h. j. , reynolds , c. , & cotton , b. 2006 , astronomical data analysis software and systems xv , 351 , 497 krding , e. , rupen , m. , knigge , c. , fender , r. , dhawan , v. , templeton , m. , & muxlow , t. 2008 , science , 320 , 1318 kulkarni , s. r. , frail , d. a. , wieringa , m. h. , et al . 1998 , , 395 , 663 law , n. m. , kulkarni , s. r. , dekany , r. g. , et al . 2009 , , 121 , 1395 lazio , j. 2008 , astronomische nachrichten , 329 , 330 levan , a. j. , et al . 2011 , science , 333 , 199 levinson , a. , ofek , e. o. , waxman , e. , & gal - yam , a. 2002 , apj , 576 , 923 lien , a. , chakraborty , n. , fields , b. d. , & kemball , a. 2011 , arxiv:1107.0775 lonsdale , c. j. , cappallo , r. j. , morales , m. f. , et al . 2009 , ieee proceedings , 97 , 1497 lorimer , d. r. , bailes , m. , mclaughlin , m. a. , narkevic , d. j. , & crawford , f. 2007 , science , 318 , 777 metzger , b. d. , et al.2010 , , 406 , 2650 metzger , b. d. , & berger , e. 2011 , arxiv:1108.6056 nakar , e. 2007 , , 442 , 166 nakar , e. , piran , t. , & granot , j. 2002 , , 579 , 699 nakar , e. , & piran , t. 2011 , , 478 , 82 , r. , piran , t. , & shemi , a. 1991 , , 379 , l17 nissanke , s. m. , sievers , j. l. , dalal , n. , & holz , d. e. 2011 , arxiv:1105.3184 sault , r. j. , teuben , p. j. , & wright , m. c. h. 1995 , astronomical data analysis software and systems iv , 77 , 433 thyagarajan , n. , helfand , d. j. , white , r. l. , & becker , r. h. 2011 , arxiv:1107.5901 ofek , e. o. , breslauer , b. , gal - yam , a. , frail , d. , kasliwal , m. m. , kulkarni , s. r. , & waxman , e. 2010 , , 711 , 517 ofek , e. o. , & frail , d. a. 2011 , , 737 , 45 ofek , e. o. , frail , d. a. , breslauer , b. , kulkarni , s. r. , chandra , p. , gal - yam , a. , kasliwal , m. m. , & gehrels , n. 2011 , arxiv:1103.3010 oosterloo , t. , verheijen , m. a. w. , van cappellen , w. , et al . 2009 , proceedings of wide field astronomy technology for the square kilometre array ( skads 2009 ) . november 2009 . chateau de limelette , belgium . , e. s. 1991 , , 380 , l17 perley , r. a. , schwab , f. r. , & bridle , a. h. 1989 , synthesis imaging in radio astronomy , 6 , rau , a. , et al . 2009 , , 121 , 1334 rossi , e. m. , perna , r. , & daigne , f. 2008 , , 390 , 675 rttgering , h. , de bruyn , a. g. , fender , r. p. , et al . 2003 , texas in tuscany . xxi symposium on relativistic astrophysics , 69 scott , w. k. 1996 , phd thesis , _ compact source variability at centimeter wavelengths _ , university of british columbia soderberg , a. m. , et al . 2006 , , 442 , 1014 soderberg , a. m. , et al . 2010 , , 463 , 513 sylvestre , j. 2003 , , 591 , 1152 thyagarajan , n. , helfand , d. j. , white , r. l. , & becker , r. h. 2011 , arxiv:1107.5901 totani , t. , & panaitescu , a. 2002 , , 576 , 120 de vries , w. h. , becker , r. h. , white , r. l. , & helfand , d. j. 2004 , , 127 , 2565 van velzen , s. , koerding , e. , & falcke , h. 2011 , arxiv:1104.4105 weiler , k. w. , panagia , n. , sramek , r. a. , et al . 2010 , , 81 , 374 welch , j. , backer , d. , blitz , l. , et al . 2009 , ieee proceedings , 97 , 1438 zauderer , b. a. , et al . 2011 , arxiv:1106.3568 de zotti , g. , massardi , m. , negrello , m. , & wall , j. 2010 , , 18 , 1 in this section , our goal is to compute the probability density function of the @xmath60th highest value of @xmath83 . let @xmath84 be the probability density function of @xmath85 with @xmath86 being the cumulative function . we denote the @xmath87 highest value by @xmath88 . thus the maximum of the series of measurements is @xmath89 and the minimum value is @xmath90 . let @xmath91 be the probability density function of @xmath92 . this means that at least one of the measurements lies is in the range @xmath93 $ ] . the probability density for this event is @xmath94 . next , then @xmath95 measurements must lie below this range and @xmath96 above this range . the probability for any value to be smaller than @xmath88 is @xmath97 and the probability for a value to higher than @xmath88 is is @xmath98 . this is now a binomial distribution with @xmath99 total values . thus the probability density function for @xmath88 is @xmath100^{m-1}.\ \label{eq : rho}\ ] ] the first combinatorial factor of @xmath61 accounts for the possibility that @xmath88 can occupy any position in the sequence . the second combinatorial factor , @xmath101 account for the combinations satisfying the condition that @xmath95 values lie below @xmath88 and @xmath96 lie above @xmath88 . for both the maximum ( @xmath62 ) and minimum ( @xmath102 ) equation [ eq : rho ] simplifies to that expected from basic considerations . now let us consider the specific case where @xmath103 follows gaussian statistics : @xmath104 where @xmath103 is normalized in units of @xmath11 . the probability that an event is extreme or lies within the range @xmath105 is @xmath106 the probability that an event is extreme in only one sense , maximum or minimum , and lies outside the range @xmath107 $ ] ( say ) is thus @xmath108= 1-\frac{1}{2}{\rm erfc}\big(\frac{h}{\sqrt{2}}\big ) \end{aligned}\ ] ] where @xmath109 . consider the case where @xmath110 ( say @xmath111 or more ) and @xmath60 is small , say 10 . then we can approximate @xmath112 , @xmath113 , ... , @xmath114 . furthermore using the approximation , @xmath115 , we find in the limiting case where @xmath88 is greater than a few [ so that @xmath116 : @xmath117 \big[\frac{n}{2}{\rm erfc}\big(\frac{h_m}{\sqrt{2}}\big)\big]^{m-1}. \end{aligned}\ ] ] provided that there is sufficient spectral resolution , the response of an interferometer to a source follows the antenna response function ( `` primary beam '' ) . we will assume that the response function is azimuthally symmetric and specified by @xmath118 where @xmath119 is the angular offset from the pointing axis . let the areal density of sources with flux density greater than @xmath120 be a power law , say , @xmath121 . for euclidean geometry and most reasonable luminosity functions , @xmath122 . next , we note that detection is really finding sources at a given snr and above . fortunately the noise distribution for an interferometric image is uniform . thus a source with a given flux density will have an snr , @xmath123 , that scales with the primary beam response , @xmath124 . the number of sources above a certain snr and contained outside an angular radius of @xmath125 is @xmath126 for the specific case of a gaussian beam , @xmath127 \label{eq : gaussian}\ ] ] where the traditional `` full width at half maximum '' ( fwhm ) is @xmath128 . substituting equation [ eq : gaussian ] into equation [ eq : ns ] we obtain @xmath129 half the sources will be detected outside the radius @xmath130 . for @xmath122 we obtain @xmath131 . the expression @xmath132 is more useful . a plot of @xmath133 can be found in figure [ fig : primarybeamdistribution ] . from the pointing aixs and assuming @xmath122 . the response curve is normalized by insisting that the integral of the curve ( from @xmath134 to @xmath135 ) is unity . fifty percent of the sources are within @xmath136 ( marked by a pentagram ) and 97% within @xmath137 ( marked by a cross ; at this radius the primary beam gain is only 0.1 relative to that on - axis ) . , width=288 ] it is not unusual to find no source after undertaking a survey . we wish to determine an upper limit to the number of sources that were being searched . the number detected is given by poisson statistics . the probability of finding @xmath138 sources is then given by @xmath139 where @xmath9 is the poisson parameter and equal to @xmath140 . our goal is to determine the maximum value of @xmath9 given a non - detection . as the value of @xmath9 is increased the probability of detection , by which we mean the probability of detecting one or more events , also increases . this probability is @xmath141 which we note is @xmath142 . this probability can be set to the desired confidence level , @xmath143 and thence @xmath144 the reader with a stronger physical bent may find the complement , @xmath145 , more appealing . regardless , we find @xmath146 $ ] at a confidence level of [ 95% , 99% , 99.9% ] . here we compute @xmath61 , the number of independent beams for the vla data that went into the analysis of b07 and @xcite . for b07 , a circular region with a radius of two times the half - power radius was searched for each epoch . the number of independent beams per epoch is the ratio of that area to the area of the synthesized beam in that particular epoch . for individual epochs , this value ranged from as small as @xmath147 to as large as @xmath111 . the total numbers of independent beams for the 5 and 8.4 ghz data are @xmath148 and @xmath149 , respectively . the smaller number of independent beams for 8.4ghz are due to the tapering of the visibility data , increasing the typical synthesized beam size at 8.4ghz . for @xcite the search was made for transients in 4.65radius circular region . the fwhm of the synthesized beam is @xmath150arc - seconds . though the goal was 16 epochs per pointing we only achieved an average of 15.7 images . with 141 epochs we derive @xmath151 .

the spatiotemporal dynamics of semiconductor lasers can be simulated successfully by semiclassical maxwell - bloch equations without including any quantum effects in the light field ( for an overview of the theory and modeling see @xcite and references therein ) . the classical treatment of the light field is justified by the high intensity of the laser light well above threshold . however , the incoherent noise required by the uncertainty principle in both the electrical dipole of the semiconductor medium and the light field itself is of significant importance when several cavity modes compete or when the laser is close to threshold . photon rate equations for multi - mode operation of semiconductor lasers show that the spontaneous emission terms may contribute significantly to the spectral characteristics of the light field emitted by the laser @xcite . such models assume a fixed mode structure determined entirely by the empty cavity . this assumption does not apply to gain guided lasers and to unstable resonators , however @xcite . in these cases it is therefore desirable to explicitly describe the spatial coherence of spontaneous emission . the spatial coherence of spontaneous emission and amplified spontaneous emission is even more important in devices close to threshold or devices with a light field output dominated by spontaneous emission such as superluminescent diodes and ultra low threshold semiconductor lasers @xcite . ultra - low threshold lasers may actually operate in a regime of negative gain where spontaneous emission is the only source of radiation @xcite . a description of the light field emitted by such devices therefore requires an explicit description of the spatial coherence in spontaneous emission as well . an approach to the consistent inclusion of the quantum noise properties of the light field in the dynamics of semiconductor laser diodes using nonequilibrium green s functions has been presented in @xcite . in this approach , the linear optical response of the medium is varied as a function of the time - dependent electron - hole distributions . although the non - equilibrium green s function presents an elegant solution for the description of many - body effects@xcite , the representation of the interband dipole dynamics by green s functions causes a non - markovian memory effect which is difficult to handle and is therefore usually neglected @xcite . moreover , the need to determine the green s function corresponding to the dynamically varying carrier distribution requires a computational effort far greater than that required for the integration of the corresponding maxwell - bloch equations . therefore , as stated in @xcite , an exact analytical investigation of the spatial mode structures in realistic cavities using non - equilibrium green s functions is out of reach . in order to simulate the spatiotemporal dynamics of multi mode operation , of lasers near threshold , low threshold lasers or superluminescent diodes , it is therefore desireable to formulate an alternative approach to the problem of spontaneous emission and amplified spontaneous emission in such devices which is based on maxwell - bloch equations . by including the spatiotemporal dynamics of the interband dipole in such equations , non - markovian terms are avoided and the quantum mechanical equations may be integrated in a straightforward manner . the starting point for our description of quantum noise effects is the dynamics of quantum mechanical operators of the field and carrier system . since the operator dynamics of the carrier system have been investigated in the context of maxwell - bloch equations before @xcite and the light field equations correspond exactly to the classical maxwell s equations , it is possible to focus only on the local light - matter interaction . once the properties of this interaction are formulated in terms of the expectation values of field - field correlations , dipole - field correlations , carrier densities , fields and dipoles , the dynamics of the carrier system and the light field propagation may be added . in section [ sec : lci ] , the quantum dynamics of the interaction between the light field and the carrier system is formulated in terms of wigner distributions for the carriers and of spatially continuous amplitudes for the light field . the equations are formulated for both bulk material and for quantum wells including the effects of anisotropic coupling to the polarization components of the light field . section [ sec : cdyn ] summarizes the effects of the dynamics of the electron - hole system in the semiconductor material . the light field dynamics are introduced in section [ sec : max ] . by quantizing maxwell s equation , the coupling constant @xmath1 introduced in section [ sec : lci ] is expressed in terms of the interband dipole matrix element . the complete set of quantum maxwell - bloch equations is presented in section [ sec : qmbe ] . based on this general formulation , specific approximate versions for quantum well edge emitting and vertical cavity surface emitting lasers are derived . the possibility of including two time correlations in the quantum maxwell - bloch equations is discussed and equations are given for the case of vertical cavity surface emitting lasers . in section [ sec : ase ] , analytical results for the spectra of gain and spontaneous emission in quantum wells as well as the spontaneous emission factor @xmath0 and the far field pattern of amplified spontaneous emission in broad area quantum well lasers are presented . section [ sec : concl ] concludes the article . in following , we will describe the active semiconductor medium in terms of an isotropic two - band model where , for the case of the holes , a suitably averaged effective mass is taken @xcite . a generalization to more bands is straightforward . in terms of the local annihilation operators for photons ( @xmath2 ) , electrons ( @xmath3 ) , and holes ( @xmath4 ) , the hamiltonian of the light - carrier interaction can be written as @xmath5 the operator dynamics associated with this hamiltonian are then given by @xmath6 the discrete positions @xmath7 correspond to the lattice sites of the bravais lattice describing the semiconductor crystal . each lattice site actually represents the spatial volume @xmath8 of the wigner - seitz cell of the lattice . for zincblende crystal structure , this volume is equal to one quarter of the cubed lattice constant . in the case of algaas structures the lattice constant is about @xmath9 m and @xmath10 @xcite . the photon annihilation operator @xmath2 therefore describes the annihilation of a photon within a volume @xmath8 . here , we will focus on the light - carrier interaction and the quantum noise contributions responsible for spontaneous emission . for that purpose , we extend the semiclassical description by including not only the expectation values of the field and dipole operators , @xmath11 and @xmath12 , respectively , but also the field - field correlations @xmath13 and the field - dipole correlation @xmath14 . the factorized equations of motion then read @xmath15 note that this set of equations already represents a closed description of the field dynamics . if , as in many experimental configurations , the absolute phase of the light field and dipole operators may be considered unknown , these equations are sufficient for a description of the light - carrier interaction . however , when two time correlations are of interest or in the case of coherent excitation by injection of an external laser it may also be necessary to additionally consider the dynamics of the field and dipole expectation values , i.e. , @xmath16 in the following , we will briefly discuss the implications of the factorization performed in the derivation of ( [ eq : factdyn ] ) . in order to formulate the dynamics of the light - matter interaction without including higher order correlations , three terms have been factorized in the time derivative of the field - dipole correlation ( [ eq : factdynbcd ] ) . the factorizations are @xmath17 no additional factorizations are necessary in the density dynamics of photons , electrons and holes . the three factorizations are based on the assumption of statistical independence between the densities of photons , electrons and holes . since we are not considering fluctuations in the particle densities , this is a necessary assumption . equations ( [ eq : fact1 ] ) and ( [ eq : fact2 ] ) separate the photon density from the carrier densities . these terms represent stimulated emission processes . therefore , a correlation of the photon density with the carrier densities would lead to a modified stimulated emission rate . below threshold , this effect will be small because the amplified spontaneous emission is distributed over many modes such that the local correlations between photon and carrier densities are weak . above threshold , the photon number fluctuations in the lasing mode cause relaxation oscillations . the photon number fluctuations are nearly ninety degrees out of phase with the carrier number fluctuations . therefore , the time averaged correlation is still negligible . equation ( [ eq : fact3 ] ) separates the electron and hole densities . this term represents the spontaneous emission caused by the simultaneous presence of electrons and holes in the same location . although it is reasonable to assume that the high rate of scattering at high carrier densities effectively reduce all electron - hole correlations to zero , it is important to note that the interband dipole @xmath18 implies a phase correlation between the electrons and the holes . in fact , the spontaneous emission term factorized according to equation ( [ eq : fact3 ] ) originates from the dipole - dipole correlation @xmath19 . note that this term could also be factorized into the product of dipole operators @xmath20 . the dynamics of the field - field and the field - dipole correlations are then identical to the dynamics of the products of the fields and dipoles . therefore , that factorization corresponds to the approximations of the conventional maxwell - bloch equations such as described in @xcite which do not rigorously include spontaneous emission . generally , spontaneous emission must always arise from random phase fluctuations . these are given by the product of electron and hole densities . while the phase dependent dipole relaxes quickly due to scattering events , the carrier densities are preserved during scattering . therefore @xmath21 is usually a good assumption in semiconductor systems . note that this assumption does fail in the case of low carrier densities and high dipole inducing fields . however , this case only occurs if the light field is injected from an external source . in semiconductor lasers and in light emitting diodes , the major contribution to the dipole - dipole correlations stems from the product of electron and hole densities . to check the statistical independence of photon , electron , and hole densities , it is also convenient to check the corresponding inequality for the three particle coherence represented by the field - dipole correlation , @xmath22 thus , if a calculation does not fulfill this requirement , particle density correlations additionally have to be taken into account . in order to connect the light - carrier interaction to the highly dissipative carrier transport equations , it is practical to transform the carrier and dipole densities using wigner transformations @xcite . replacing the discrete density matrices by continuous ones obtained by polynomial interpolation will allow e.g. for numerical purposes an arbitrary choice of the discretization scales , which generally will be much larger than a lattice constant . analytically , it permits an application of differential operators . physically , the particle densities are smooth functions over distances of several lattice constants . a coherence length shorter than e.g. ten lattice constants would require @xmath23 states with @xmath24 of at least one twentieth of the brilloin zone diameter . in typical laser devices , however , the electrons and holes all accumulate near the fundamental gap at @xmath25 . therefore , it is simply a matter of convenience to define the continuous densities such that @xmath26 these continuous functions may then be transformed into wigner functions by @xmath27 the normalization of these wigner functions has been chosen in such a way that a value of one represents the maximal phase space density possible for fermions , that is one particle per state . since the density of states in the six - dimensional phase space given by @xmath28 and @xmath23 is @xmath29 , a factor of @xmath29 will appear whenever actual carrier densities need to be obtained from the wigner functions . however , the normalization in terms of the maximal possible phase space density is convenient because it represents the probability that a quantum state in a given region of phase space is occupied . therefore , the wigner distribution corresponding to the thermal equilibrium of a given particle density is directly given by the fermi function . to deal with the light field dynamics in the same manner , the field and field - field correlation variables must also be defined on a continuous length scale . in order to obtain photon densities , we define @xmath30 finally , the dipole - field correlation must be defined accordingly , such that @xmath31 with these new definitions , the light - carrier interaction dynamics can now be expressed in a form which considers both the position and the momentum of the electrons and holes . the dynamics of emission and absorption now reads @xmath32 \left.\frac{\partial}{\partial t}{\cal e } ( { \bf r})\right |_{cl } & = & -i g_0\frac{\sqrt{\nu_0}}{8\pi^3}\int d^3{\bf k}\ ; p({\bf r},{\bf k } ) \\ \left.\frac{\partial}{\partial t } p({\bf r},{\bf k})\right |_{cl } & = & ig_0\sqrt{\nu_0}\;\frac{1}{8\pi^3}\!\int\ ! d^3 { \bf x}\!\int\!d^3 { \bf q}\ ; e^{i{\bf qx } } \nonumber \\ & & \times \left(f^e\left({\bf r},{\bf k}+\frac{\bf q}{2}\right)+ f^h\left({\bf r},-{\bf k}+\frac{\bf q}{2}\right)-1\right ) { \cal e}({\bf r}+{\bf x}).\end{aligned}\ ] ] the integrals over @xmath33 and @xmath34 represent seemingly non - local effects introduced by the transformation into wigner functions . this property of the wigner transformation retains the coherent effects in the carrier system . for the interaction of the carriers with the light field , it ensures momentum conservation by introducing a non - local phase correlation in the dipole field corresponding to the total momentum of the electron and hole concentrations involved . effectively , the integral over @xmath34 converts the momentum part of the wigner distributions into a coherence length . this coherence length then reappears in the spatial structure of the dipole field and the electromagnetic field generated by the carrier distribution . however , the coherence length in the carrier system is usually much shorter than the optical wavelength . it can therefore be approximated by a spatial delta function . here , we do this by noting that @xmath35 if the effects of the momentum shift @xmath34 in the wigner functions is neglected , the integrals may then be solved , yielding only local interactions between the carrier system and the light field : @xmath36 \left.\frac{\partial}{\partial t}{\cal e } ( { \bf r})\right |_{cl } & = & -i g_0\frac{\sqrt{\nu_0}}{8\pi^3}\int d^3{\bf k}\ ; p({\bf r},{\bf k } ) \\ \label{eq : endbulk } \left.\frac{\partial}{\partial t } p({\bf r},{\bf k})\right |_{cl } & = & ig_0\sqrt{\nu_0}\ ; \left(f^e({\bf r},{\bf k})+f^h({\bf r},-{\bf k})-1\right){\cal e}({\bf r}).\end{aligned}\ ] ] these equations now provide a compact description of the light - carrier interaction in a three dimensional semiconductor medium , including the incoherent quantum noise term which is the source of spontaneous emission . similar equations may also be formulated for a quantum well structure by replacing the phase space density of @xmath29 with @xmath37 , reducing the spatial coordinates of the carrier system to two dimensions , and introducing a delta function for the coordinate perpendicular to the quantum well at the points where field coordinates correspond to dipole coordinates . of course , the electromagnetic field remains three dimensional , even though the dipole it originates from is confined to two dimensions . in particular , the correlation @xmath38 has both a three dimensional coordinate @xmath28 and a two dimensional coordinate @xmath39 . it is therefore useful to distinguish the two dimensional and the three dimensional coordinates . in the following , the two dimensional carrier coordinates will be marked with the index @xmath40 . note that in some cases , both @xmath28 and @xmath41 appear in the equations . in those cases , the in - plane coordinates @xmath42 and @xmath43 are equal , while the perpendicular coordinate @xmath44 must be equal to the quantum well coordinate @xmath45 . the equations for the interaction of the three dimensional light field with the two dimensional electron - hole system in a single quantum well subband then read @xmath46 \left.\frac{\partial}{\partial t}{\cal e } ( { \bf r})\right |_{cl } & = & -i g_0\frac{\sqrt{\nu_0}}{4\pi^2}\int d^2{\bf k}_\parallel\ ; p({\bf r}_\parallel,{\bf k}_\parallel)\delta(r_z - z_0 ) \\ \label{eq : endqw } \left.\frac{\partial}{\partial t } p({\bf r}_\parallel,{\bf k}_\parallel)\right |_{cl } & = & ig_0\sqrt{\nu_0}\ ; \left(f^e({\bf r}_\parallel,{\bf k}_\parallel ) + f^h({\bf r}_\parallel,-{\bf k}_\parallel)-1\right ) { \cal e}({\bf r})_{r_z = z_0}.\end{aligned}\ ] ] note that the value of @xmath1 will usually be slightly lower than the bulk value because the overlap of the spatial wavefunctions of the electrons and the holes in the lowest subbands is less than one . the equations derived above represent the interaction of a single conduction band and a single valence band with a single scalar light field . neither the spin degeneracy of the carriers nor the polarization of the light field has been considered . since the geometry of light field emission is highly dependent on polarization effects such effects should also be taken into account in the framework of this theory . the basic interaction between a single conduction band , a single valence band and a single light field polarization are accurately represented by equations ( [ eq : bulk]-[eq : endbulk ] ) and ( [ eq : qw]-[eq : endqw ] ) . by adding the contributions of separate transitions , any many band system may be described based on these equations . in semiconductor quantum wells the situation is considerably simplified if only the lowest subbands are considered . then there are only two completely separate transitions involving circular light field polarizations coupled to a single one of the two electron and hole bands . the quantum well structure does not interact with light fields which are linearly polarized in the direction perpendicular to the plane of the quantum well . the equations for quantum wells are therefore completed by adding an index of @xmath47 or @xmath48 to each variable . the situation in the bulk system is much more involved . the transitions occur between the two fold degenerate spin 1/2 system of the electrons and the four fold degenerate spin 3/2 system of the holes . all three polarization directions of the light field are equally possible , connecting each of the electron bands with three of the four hole bands . however , since the effective mass of the two heavy hole bands is much larger than the effective mass of the light holes ( e.g. by a factor of eight in gaas ) , only a small fraction of the holes will be in the light - hole bands ( about 6% in gaas for equilibrium distributions ) . consequently , the carrier subsystem can again be separated into two pairs of bands . however , the light field polarization emitted by the electron - heavy hole transitions in bulk material is circularly polarized with respect to the relative momentum @xmath49 of the electron and the heavy hole . since usually there is no strong directional anisotropy in the @xmath23 space distribution of the carriers , it can be assumed that one third of the @xmath23 space volume contributes to each polarization direction and the equations may be formulated accordingly . in the following , we will assume that the wigner distributions of the two pairs of bands considered are approximately equal at all times . note that this means that hole burning effects in the spin and polarization dynamics which may occur in vertical cavity surface emitting lasers @xcite are ignored . however , such effects have been investigated in several other studies @xcite and are found to be fairly weak in some devices @xcite . the complete set of dynamical equations can now be formulated by adding the carrier dynamics and the linear part of maxwell s equations to the light - carrier interaction . modeling the carrier dynamics of a semiconductor system can be a formidable task all by itself . a number of approximations and models have been developed to deal with the effects of many - particle interactions and correlations and with the dissipation caused by the electron - phonon interactions @xcite . in the following , we choose a simple diffusion model . many particle effects such as the band gap renormalization or the coulomb enhancement are not mentioned explicitly , but can be added in a straightforward manner @xcite . we assume that the electron and hole densities will be kept equal by the coulomb interaction , which will induce a current whenever charges are separated . therefore , it is possible to define the ambipolar carrier density @xmath50 with @xmath51 note that the two fold degeneracy of the electron and heavy hole bands has been included by choosing a density of states of @xmath52 instead of @xmath53 . this includes the assumption that the wigner distribution does not depend on the spin variable of the electrons and holes as mentioned above . the light carrier interaction of this carrier density is @xmath54 where the index @xmath55 denotes the component of the light field or dipole density corresponding to the linear polarization direction of @xmath56 . in the case of @xmath57 the index @xmath55 refers to the field polarization and the second index @xmath58 denotes the vector component of the dipole vector . equation ( [ eq : trans ] ) shows how the field - dipole correlation converts electron - hole pairs into photons . the total carrier density dynamics can now be formulated as @xmath59 where @xmath60 is the ambipolar diffusion constant , @xmath61 is the injection current density , and @xmath62 is the rate of spontaneous recombinations by non - radiative processes and/or spontaneous emission into modes not considered in @xmath63 , e.g. if the paraxial approximation is applied . the @xmath23 dependence of the distribution functions @xmath64 and @xmath65 may be approximated by assuming that the electrons and holes will always be in thermal equilibrium . the distribution functions are then given by fermi functions @xmath66 + 1 \right)^{-1},\ ] ] where @xmath67 are the effective masses of electrons and heavy holes , respectively . the chemical potential @xmath68 is a function of the carrier density @xmath50 . a useful estimate of this relationship is given by the pade approximation @xcite . spectral holeburning may be taken into account by introducing a relaxation time @xmath69 and converting the dynamics of the distribution function due to the light - carrier interaction into a deviation from the equilibrium distribution by adiabatic elimination of the relaxation dynamics : @xmath70 finally , the carrier dynamics of the dipole @xmath71 and the dipole part of the field - dipole correlation @xmath72 needs to be formulated . since both depend on a correlation of the electrons with the holes , they will necessarily relax rather quickly at a rate of @xmath73 which should be of the same order of magnitude as @xmath74 . physically , @xmath73 may be interpreted as the total momentum dependent scattering rate in the carrier system . the remainder of the dynamics can be derived from the single particle dynamics . this unitary contribution to the evolution of the dipole may be expressed by a momentum dependent frequency @xmath75 . for parabolic bands and isotropic effective masses @xmath76 , this frequency is given by @xmath77 many - particle effects due to the coulomb interaction between the carriers may be included by introducing a carrier density dependence in @xmath78 and @xmath79 . such renormalization terms representing the mean field effects of the carrier - carrier interaction have been derived and discussed e.g. in @xcite . in the following , this many particle renormalization will not be mentioned explicitly , although it can be included in a straightforward manner . with the rates @xmath73 and @xmath75 the dipole dynamics reads @xmath80 note that the phase dynamics is formulated relative to the band gap frequency @xmath81 . the real physical phase oscillations of @xmath71 would include an additional phase factor of @xmath82 $ ] . however , the only physical effect of this oscillation is to establish resonance with the corresponding frequency range in the electromagnetic field , the dynamics of which we consider next . the heisenberg equations of motion describing the operator dynamics of the electromagnetic field operators are identical to the classical maxwell s equations . in terms of the electromagnetic field @xmath83 and the dipole densities @xmath84 , the equation reads @xmath85 where @xmath86 and @xmath87 are the dielectric constant and the speed of light in vacuum , respectively , and @xmath88 is the dielectric constant in the background semiconductor medium . maxwell s equation describes the light field dynamics for all frequencies . since we are only interested in frequencies near the band gap frequency @xmath81 , it is useful to separate the phase factor of @xmath82 $ ] , defining @xmath89{\bf e_0}({\bf r})$ ] . now @xmath90 can be considered to vary slowly in time relative to @xmath82 $ ] . therefore , the time derivatives may be approximated by @xmath91 \frac{\partial^2}{\partial t^2}\exp[-i\omega_0 t]{\bf e}_0({\bf r } ) \approx -\omega_0 ^ 2 { \bf e}_0({\bf r } ) - 2i\omega_0 \frac{\partial}{\partial t } { \bf e}_0({\bf r}).\ ] ] similarly , @xmath92 may be defined such that @xmath93{\bf p}_0({\bf r})$ ] . the approximation used here may even be of zero order , since we are primarily interested in the dynamics of the electromagnetic field : @xmath91 \frac{\partial^2}{\partial t^2}\exp[-i\omega_0 t]{\bf p}_0({\bf r } ) \approx -\omega_0 ^ 2 { \bf p}_0({\bf r}).\ ] ] the temporal evolution of the electromagnetic field now reads @xmath94 where @xmath95 is the vacuum wavevector length corresponding to @xmath81 . in ( [ eq : edynparax1 ] ) , the field dynamics is described in terms of electromagnetic units , that is the fields represent forces acting on charges . to switch scales to the photon densities represented by @xmath96 , energy densities have to be considered . since the energy of each photon will be close to the bandgap energy @xmath97 , the energy density of the electromagnetic field is given by @xmath98 therefore , the field may be expressed as photon density amplitude using @xmath99 the dipole density @xmath92 may be expressed in terms of @xmath100 and @xmath101 by noting that the density @xmath100 is the dipole density in units of one - half the atomic dipole given by the interband dipole matrix element @xmath102 at @xmath103 . the factor of one - half is a logical consequence of the property that @xmath104 . thus , a fully polarized lattice would have a dipole density of @xmath105 which must correspond to @xmath106 . note that @xmath102 contains an arbitrary phase factor depending only on the definition of the states used for its determination . for convenience , we assume a definition of phases such that @xmath102 is real . the dipole density @xmath92 may then be written as @xmath107 written in terms of @xmath108 and @xmath101 , the complete field dynamics now reads @xmath109 the coupling frequency @xmath1 introduced in equation([eq : hamilton ] ) may be expressed in terms of the dipole matrix element @xmath110 : @xmath111 with this equation , the operator dynamics of the light field operator @xmath2 corresponds to the field dynamics of the maxwell - bloch equations for classical fields . by applying the linear propagation dynamics of the field to @xmath112 and @xmath57 as well , it is now possible to formulate a complete set of quantum maxwell - bloch equations . on the basis of the discussion in the previous sections , the quantum maxwell - bloch equations for a bulk gain medium in three dimensions read @xmath113 \frac{\partial}{\partial t } p_i({\bf r},{\bf k } ) & = & - \left(\gamma ( { \bf k } ) + i\omega({\bf k})\right ) p_i({\bf r},{\bf k } ) \nonumber \\ & & + i g_0\frac{2\sqrt{\nu_0}}{3}\ ; \left ( f^e_{eq}\left(k;n({\bf r})\right ) + f^h_{eq}\left(k;n({\bf r})\right ) -1 \right ) { \cal e}_i({\bf r } ) \\ \frac{\partial}{\partial t } { \cal e}_i({\bf r } ) & = & i\frac{\omega_0}{2k_0 ^ 2}\left(\sum_k \frac{\partial}{\partial r_k}\epsilon_r^{-1 } \frac{\partial}{\partial r_k } { \cal e}_i({\bf r } ) -\frac{\partial}{\partial r_i}\epsilon_r^{-1 } \frac{\partial}{\partial r_k}{\cal e}_k({\bf r } ) + k_0 ^ 2 { \cal e}_i({\bf r})\right ) \nonumber \\ & & - i g_0\frac{\sqrt{\nu_0}}{8\pi^3 } \int d^3{\bf k } \ ; p_i({\bf r},{\bf k}).\end{aligned}\ ] ] in order to describe a realistic diode one needs to describe not only the volume of the active region but also the propagation of light out of this region . this may either be achieved by defining realistic boundary conditions or by setting all material properties equal to zero outside a finite active volume and calculating the light field propagation into the outside medium by varying @xmath114 in space . next , we will formulate the equations for a quantum well structure . for this case , we will also describe different cavity structures and the appropriate paraxial approximations possible for the various types of laser devices . using the terminology of equations ( [ eq : qw]-[eq : endqw ] ) , the quantum maxwell - bloch equations for a multi quantum well structure with q quantum wells read @xmath115 & & + i g_0 q \sqrt{\nu_0}\;\delta({\bf r}_\parallel-{\bf r^\prime}_\parallel ) \delta(r_z - z_0 ) \delta_{ij_\parallel } f^e_{eq}\left(k_\parallel;\frac{n({\bf r}_\parallel)}{q}\right ) \cdot f^h_{eq}\left(k_\parallel;\frac{n({\bf r}_\parallel)}{q}\right ) \\ \frac{\partial}{\partial t } i_{ij}({\bf r};{\bf r^\prime } ) & = & - i\frac{\omega_0}{2k_0 ^ 2 } \sum_k\left(\frac{\partial}{\partial r_k}\epsilon_r^{-1 } \frac{\partial}{\partial r_k } -\frac{\partial}{\partial r^\prime_k}\epsilon_r^{-1 } \frac{\partial}{\partial r^\prime_k}\right ) i_{ij}({\bf r};{\bf r^\prime } ) \nonumber \\ & & + i\frac{\omega_0}{2k_0 ^ 2}\sum_k\left ( \frac{\partial}{\partial r_i}\epsilon_r^{-1 } \frac{\partial}{\partial r_k } i_{kj}({\bf r};{\bf r^\prime } ) - \frac{\partial}{\partial r^\prime_j}\epsilon_r^{-1 } \frac{\partial}{\partial r^\prime_k } i_{ik}({\bf r};{\bf r^\prime})\right ) \nonumber \\ & & -i g_0\frac{\sqrt{\nu_0}}{4\pi^2 } \int d^2{\bf k}_\parallel \left ( c_{ij_\parallel}({\bf r};{\bf r^\prime}_\parallel,{\bf k}_\parallel ) \delta(r_z^\prime - z_0 ) \right . \nonumber \\ & & \left . \hspace{3 cm } -c_{ji_\parallel}^*({\bf r^\prime};{\bf r}_\parallel,{\bf k}_\parallel ) \delta(r_z - z_0 ) \right ) \\[1 cm ] \frac{\partial}{\partial t } p_{i_\parallel}({\bf r}_\parallel,{\bf k}_\parallel ) & = & - \left(\gamma ( { \bf k}_\parallel ) + i\omega({\bf k}_\parallel)\right ) p_{i_\parallel}({\bf r}_\parallel,{\bf k}_\parallel ) \nonumber \\ & & + i g_0 q \sqrt{\nu_0}\ ; \left ( f^e_{eq}\left(k_\parallel;\frac{n({\bf r}_\parallel)}{q}\right ) + f^h_{eq}\left(k_\parallel;\frac{n({\bf r}_\parallel)}{q}\right ) -1 \right ) { \cal e}_i({\bf r})_{r_z = z_0 } \\ \frac{\partial}{\partial t } { \cal e}_i({\bf r } ) & = & i\frac{\omega_0}{2k_0 ^ 2}\left(\sum_k \frac{\partial}{\partial r_k}\epsilon_r^{-1 } \frac{\partial}{\partial r_k } { \cal e}_i({\bf r } ) -\frac{\partial}{\partial r_i}\epsilon_r^{-1 } \frac{\partial}{\partial r_k}{\cal e}_k({\bf r } ) + k_0 ^ 2 { \cal e}_i({\bf r})\right ) \nonumber \\ & & - i g_0\frac{\sqrt{\nu_0}}{4\pi^2 } \int d^2{\bf k}_\parallel \ ; p_{i_\parallel}({\bf r}_\parallel,{\bf k}_\parallel)\delta(r_z - z_0).\end{aligned}\ ] ] note that @xmath116 is the total carrier density . therefore , the density per quantum well which determines the chemical potential of the carrier distribution functions is @xmath117 . again , the structure of an external cavity may be considered either by boundary conditions or by spatially varying @xmath114 . in particular laser diodes may be described by distinguishing between reflective and non - reflective edges . if the reflective surface is perpendicular to the plane of the quantum wells , the laser is an edge emitter . if the reflectivity is very high on the surface planes parallel to the quantum wells , the laser is a vertical cavity surface emitting laser ( vcsel ) . in an edge emitting laser , the laser light field propagates in the plane of the quantum well . since the @xmath118-direction is already defined as the one perpendicular to the quantum well we will define the @xmath119-direction as the direction along which the laser light propagates . a schematic representation of this type of laser geometry is shown in figure [ edge ] . it is possible to drastically reduce the dimensionality of the equation describing the edge emitting laser geometry by noting that the light field polarization of the amplified fields will be in the plane of the quantum well and by limiting the analysis to a single longitudinal mode . effectively , this corresponds to a light field @xmath108 with the following properties : @xmath120 the envelope function @xmath121 describes both the propagation along the @xmath119-direction and the confinement along the @xmath118-direction . it represents an approximate solution of the wave equation in the @xmath122-plane normalized by @xmath123 the equations are then limited to light field modes with the two dimensional envelope @xmath121 . spontaneous emission into other light field modes must be considered by including the rate of emission in the carrier recombination rate @xmath62 . since the length @xmath124 of the laser in the @xmath119-direction is also an important property of the device , it is included by considering the openness of the optical cavity . with the reflectivities of the laser mirrors given by @xmath125 and @xmath126 , the light field in the cavity is damped by losses through the mirrors at a rate of @xmath127.\ ] ] the new one - dimensional variables are now defined as follows : @xmath128 the carrier density is now given in terms of a one dimensional density . to obtain the two dimensional carrier density per quantum well , this density is to be divided by @xmath129 . note that the intensity is also given in terms of photons per unit length . the dynamics of the edge emitter then reads @xmath130 \frac{\partial}{\partial t } p_0(r_x,{\bf k}_\parallel ) & = & - \left(\gamma ( { \bf k}_\parallel ) + i\omega({\bf k}_\parallel)\right ) p_0(r_x,{\bf k}_\parallel ) \nonumber \\ & & + i g_0 \sigma \sqrt{\nu_0}\ ; \left ( f^e_{eq}\left(k_\parallel;\frac{n_{1d}(r_x)}{ql}\right ) + f^h_{eq}\left(k_\parallel;\frac{n_{1d}(r_x)}{ql}\right ) -1 \right ) { \cal e}_0(r_x ) \\ \label{eq:1ddynamics_b } \frac{\partial}{\partial t } { \cal e}_0(r_x ) & = & -\kappa { \cal e}_0(r_x ) + i\frac{\omega_0}{2k_0 ^ 2\epsilon_r } \frac{\partial^2}{\partial r_x^2 } { \cal e}_0(r_x ) -i g_0\frac{\sqrt{\nu_0}}{4\pi^2 } \int d^2{\bf k}_\parallel \ ; p_0(r_x,{\bf k}_\parallel),\end{aligned}\ ] ] where @xmath131 is the confinement factor which determines the overlap between the quantum wells and the light field mode , @xmath132 in a vcsel the light field is strongly confined by highly reflective mirrors above and below the quantum wells . the light field propagates perpendicular to the quantum well structure as shown in figure [ vcsel ] . therefore both the possible polarization directions and the spatial dynamics remain two dimensional . only the @xmath118-direction may be eliminated by averaging over a single longitudinal mode . coupling terms between the polarization directions should be taken into account , even if they are small . for vcsels , the assumptions read @xmath133 the latter condition takes into account the divergence of @xmath134 caused by the polarization @xmath135 . this is an important contribution to the quantum maxwell - bloch equations , since it coherently couples orthogonal polarizations . the properties of the envelope function @xmath136 are defined as @xmath137 the cavity loss rate @xmath138 is defined as in equation ( [ eq : kappa ] ) . however , the experimentally observed polarization stability is taken into account by using slightly different reflectivities for the x and y polarizations . the cavity loss rate is therefore given by @xmath139 and @xmath140 . experimental results @xcite suggest that @xmath141 . a birefringence of @xmath142 is also included to denote the difference between the band gap frequency and the longitudinal frequencies of the confined light field for the two polarization directions . the birefringence @xmath143 is usually in the ghz range . since all coordinates are two dimensional , the index @xmath144 which marked the two dimensional coordinates in the quantum well equations will be omitted . instead , the two dimensional variables are marked with a tilde . all indices and coordinates of such variables are only defined in two dimensions . the variables are defined as @xmath145 note that the dipole variable @xmath146 is rescaled by the field density of the envelope function @xmath136 at @xmath147 . the dipole given by @xmath148 is therefore the average dipole density within the whole cavity , not just within the quantum wells . the quantum maxwell - bloch equations for the vcsel now read [ eq : qmbevcsel ] @xmath149 & & + i g_0 \sigma \sqrt{\nu_0}\;\delta({\bf r}-{\bf r^\prime})\delta_{ij } f^e_{eq}\left(k;\frac{\tilde{n}({\bf r})}{q}\right ) \cdot f^h_{eq}\left(k;\frac{\tilde{n}({\bf r})}{q}\right ) \\ \frac{\partial}{\partial t } \tilde{i}_{ij}({\bf r};{\bf r^\prime } ) & = & - ( \kappa_i + \kappa_j)\tilde{i}_{ij}({\bf r};{\bf r^\prime } - i\frac{\omega_0}{2k_0 ^ 2\epsilon_r } \left(\delta_{\bf r}-\delta_{\bf r^\prime}\right ) \tilde{i}_{ij}({\bf r};{\bf r^\prime } ) \nonumber \\ & & -i g_0\frac{\sqrt{\nu_0}}{4\pi^2 } \int d^2{\bf k } \left ( \tilde{c}_{ij}({\bf r};{\bf r^\prime},{\bf k } ) -\tilde{c}_{ji}^*({\bf r^\prime};{\bf r},{\bf k})\right . \nonumber \\ & & + \sum_k \frac{1}{k_0 ^ 2\epsilon_r}\left.\left ( \frac{\partial^2}{\partial r^\prime_j r^\prime_k } \tilde{c}_{ik}({\bf r};{\bf r^\prime},{\bf k } ) - \frac{\partial^2}{\partial r_i r_k } \tilde{c}_{jk}^*({\bf r^\prime};{\bf r},{\bf k } ) \right)\right ) \\[1 cm ] \frac{\partial}{\partial t } \tilde{p}_{i}({\bf r},{\bf k } ) & = & - \left(\gamma ( { \bf k } ) + i\omega({\bf k})\right ) \tilde{p}_{i}({\bf r},{\bf k } ) \nonumber \\ & & + i g_0 \sigma \sqrt{\nu_0}\ ; \left ( f^e_{eq}\left(k;\frac{\tilde{n}({\bf r})}{q}\right ) + f^h_{eq}\left(k;\frac{\tilde{n}({\bf r})}{q}\right ) - 1 \right ) \tilde{\cal e}_i({\bf r } ) \\ \frac{\partial}{\partial t } \tilde{\cal e}_i({\bf r } ) & = & -(\kappa_i+i\delta\omega_i ) \tilde{\cal e}_i({\bf r } ) + i\frac{\omega_0}{2k_0 ^ 2\epsilon_r}\delta \tilde{\cal e}_i({\bf r } ) \nonumber \\ & & - i g_0\frac{\sqrt{\nu_0}}{4\pi^2 } \int d^2{\bf k}\;\left(\tilde{p}_{i}({\bf r},{\bf k } ) + \sum_j \frac{1}{k_0 ^ 2\epsilon_r}\frac{\partial^2}{\partial r_i r_j } \tilde{p}_{j}({\bf r},{\bf k})\right),\end{aligned}\ ] ] with @xmath131 being the confinement factor along the @xmath118-direction , @xmath150 equations ( [ eq : qmbevcsel ] ) present a starting point for the study of spatial polarization patterns and fluctuations in vcsels . for more realistic models , it may also be desirable to include a spatial dependence of the birefringence and the dichroism . also , nonlinear effects may be introduced , e.g. by separating the carrier densities for right and left circular polarization @xcite . using the equations presented above , it is now possible to calculate the emergence of a spatially coherent light field in semiconductor laser diodes both above and below threshold . note that the average light field @xmath151 will remain zero at all times due to the random phases of spontaneous emission processes . the ( average ) spatial coherence of the light field , however , does not vanish and is fully described by the non - local field - field correlations @xmath152 which emerge due to the propagation and/or amplification of the originally incoherent local spontaneous emissions . this emergence of coherence as a concequence of incoherent emissions has been discussed in a temporal context using nonequilibrium green s functions in @xcite . in order to understand the physical implications of the well known absence of an average light field in lasers , it should be recalled that all the results of the quantum maxwell - bloch equations represent averages which have to be interpreted in terms of statistical physics . for example , the field - field correlation @xmath153 represents a variance of the probability distribution with respect to the possible spatial electromagnetic field values . the coherent field observed in experimental time - resolved measurements will vary randomly from measurement to measurement according to this probability distribution . indeed , the intensity distribution itself will vary depending on the random phase interference of the eigenmodes given by @xmath153 . the calculated average spatial intensity distribution @xmath154 only describes the average near field pattern , which is likely to be close to but not identical with the one actually observed . the fluctuations of the actual intensity distribution around this average , however , are disregarded as a consequence of the factorization performed in section [ sec : lci ] . moreover , the fluctuations in the carrier density distribution induced by spatial holeburning associated with these fluctuations of the intensity distribution have also been disregarded . while the average spatial coherence of the light field is fully described by the quantum maxwell - bloch equations , the average expected temporal coherence has not yet been explicitly considered . indeed , it is not necessary to consider temporal coherence in the closed sets of quantum maxwell - bloch equations given above because all the information required to obtain the correct emission and absorption rates are incorporated in the field - dipole correlation @xmath155 . as shown in section [ sec : ase ] , the spectra of gain and spontaneous emission are implicitly given by the coherent dipole dynamics which enters into the temporal evolution of this field - dipole correlation . if explicit information about the two time correlations is desired , however , such correlations may be included in the dynamics by noting that the temporal evolution of the two time correlations of the field @xmath156 and the two time correlations of the field - dipole correlation @xmath157 is equivalent to the dynamics of the field and dipole expectation values@xcite . while the quantum maxwell - bloch equations for the correlations at @xmath158 remain unchanged , the evolution of the two time correlations as a function of @xmath159 is then given by an additional pair of equations which depend on the carrier dynamics given by the solution of the original system of quantum maxwell - bloch equations as presented above . in the case of vcsels these additional equations supplementing equations ( [ eq : qmbevcsel ] ) read [ eq : lin ] @xmath160 if the carrier density changes slowly , the equations describe the linear response of the medium caused by the initial intensity distribution @xmath161 and the initial field - dipole correlation @xmath162 determined from equations ( [ eq : qmbevcsel ] ) . in this case it is possible to derive the spectrum and the gain from the eigenmodes and the associated eigenvalues of the quasi - stationary linear optical system . note that the eigenmodes of the linearized dynamics are not necessarily identical with the eigenmodes of the intensity distribution @xmath161 , since fast variations in the carrier density distribution may have induced phase locking between the dynamical eigenmodes . in general , the frequency spectra of the light field are given by fourier transforms of the two - time correlations obtained from equations ( [ eq : lin ] ) . however , as noted above , such spectra do not comprise any effects which arise from carrier density fluctuations . these effects are known to be quite significant . well known examples of carrier fluctuation effects in the frequency spectrum of semiconductor lasers are the linewidth enhancement phenomenologically described by the linewidth enhancement factor @xmath163 and the relaxation oscillation sidebands observed in stable single mode operation . moreover , spatial carrier density fluctuations may also significantly modify the multi mode spectra of semiconductor laser devices , as pointed out recently for the case of semiconductor laser arrays @xcite . for a given carrier distribution @xmath164 , the dynamics of the optical field @xmath108 and the dipole density @xmath165 are linear . in this case , it is possible to integrate the equations of motion to obtain a green s function for the field dynamics . the equation for bulk material reads @xmath166 correspondingly , the equation for a multi quantum well structure of @xmath167 quantum wells has the form @xmath168 an expression for the rate @xmath169 at which a light field mode of frequency @xmath170 is amplified can be derived by solving the integral over @xmath171 using @xmath172 . the real part of the result is the gain spectrum given in terms of amplification per unit time , @xmath169 . for bulk material , this amplification rate is given by @xmath173 and for quantum wells , the corresponding amplification rate reads @xmath174 the gain per unit length can be obtained by dividing the rate @xmath169 by the speed of light in the semiconductor medium , @xmath175 . however , in order to establish the connection between the gain spectrum and the spectral density of spontaneous emission , it is more convenient to use the amplification rate as a starting point . the quantum maxwell - bloch equations for the field - dipole correlation @xmath176 show that the ratio between the spontaneous contributions and the stimulated contributions is @xmath177 in this equation , @xmath178 corresponds to a photon density of one photon per mode . the spectral density of the spontaneous emission may therefore be obtained by replacing @xmath179 with @xmath180 in equations ( [ eq : bulkgain ] ) and ( [ eq : qwgain ] ) , respectively , and multiplying the resulting rates with twice the density of light field modes which couple to the medium ( the factor of two being a result of considering intensities instead of fields ) . at the band edge frequency @xmath81 , the density of light field modes per volume and frequency interval in a continuous medium is @xmath181 the density of the spontaneous emission rate for bulk material @xmath182 thus reads @xmath183 for quantum wells , the density of modes is only @xmath184 of @xmath185 , since the dipole component perpendicular to the quantum well is zero . also , the delta function @xmath186 may be omitted to obtain the emission density per area . the spontaneous emission density @xmath187 is then given by @xmath188 the total rate of spontaneous emission per unit volume or area may be obtained by integrating over all frequencies . this integral removes the dependence on @xmath73 . for bulk material , @xmath189 and for quantum wells , @xmath190 for zero temperature , the spontaneous emission rate may be derived by noting that @xmath191 . the integral over @xmath23 may thus be solved , resulting in @xmath192 for bulk material and @xmath193 for quantum wells . for both bulk and quantum wells the density of spontaneous emission now may be expressed as @xmath194 where @xmath195 is the rate of spontaneous emission given by @xmath196 using equation ( [ eq : gd ] ) , the rate of spontaneous emission may also be expressed in terms of the dipole matrix element @xmath197 , @xmath198 where @xmath199 is the classical power radiated by an oscillating dipole of the amplitude @xmath197 . note that the factor of @xmath200 is consistent with a quantum noise interpretation of spontaneous emission such as the one represented by the semiclassical langevin equations . according to this interpretation , the fluctuations of each dipole are given by @xmath201 because both the real and the imaginary part of the dipole contribute . the spontaneous emission of an excited atom is then composed of one half amplified field noise and one half dipole fluctuations . therefore , each one of the two dipole components contributes one quarter of the total spontaneous emission of an excited state . numerical values of @xmath202 and @xmath1 for gaas may be determined by assuming a spontaneous lifetime of @xmath203 ns , a band gap of @xmath204 ev and @xmath205 . the dipole matrix element is then @xmath206 which corresponds to a distance of @xmath207 m times the electron charge and the coupling frequency is @xmath208 . for quantum wells at zero temperature , the integrals over @xmath209 can be solved analytically using equation ( [ eq : bandshape ] ) and assuming that @xmath210 is independent of @xmath209 . the resulting gain spectrum is then given by @xmath211 where @xmath212 is the transition frequency at the fermi surface of the electrons and holes . it is related to the carrier density @xmath213 by @xmath214 the spectral density of spontaneous emission is thus given by @xmath215 typical spectra of gain and spontaneous emission of an active layer containing @xmath216 quantum wells obtained from the analytic approximations ( [ eq : wqw ] ) and ( [ eq : sqw ] ) , respectively , are presented in fig . [ spectra ] for characteristic values of the carrier density . for the spectra in fig . [ spectra ] we have assumed a total spontaneous emission lifetime of @xmath203 ns and a band gap of @xmath217 ev . other parameters are the effective mass of electrons ( @xmath218 ) and holes ( @xmath219 ) , given in units of the electron mass @xmath220 as well as the dipole damping rate @xmath221 mev , and dielectric constant @xmath205 . gain and spontaneous emission are both displayed relative to the peak values @xmath222 and @xmath223 . the gain value may be interpreted by calculating the gain of a light beam of width @xmath224 cm with an incidence perpendicular with respect to the quantum well and traveling at a speed of @xmath225 in the plane of the quantum well . the maximal gain is then given by @xmath226 . the five spectra displayed in fig . [ spectra ] are the gain at @xmath227 ( @xmath1 ) , the gain at @xmath228 ( @xmath229 ) , the spontaneous emission at @xmath228 ( @xmath230 ) , the gain at @xmath231 ( @xmath232 ) and the spontaneous emission at @xmath233 ( @xmath234 ) . [ spectra ] clearly shows the influence of the carrier density @xmath213 . in the absence of charge carriers ( @xmath1 ) , the laser is purely absorptive . with increasing carrier density @xmath213 , transparency is reached at @xmath228 ( @xmath229 ) . this density is characterized by vanishing gain ( i.e. there is neither gain nor absorption ) at a frequency of @xmath235 . at the same time , however , there is a significant contribution of spontaneous emission ( @xmath230 ) with a maximum at a frequency of @xmath236 mev . finally , at high values of the carrier density ( @xmath233 ) , both gain ( @xmath232 ) and spontaneous emission ( @xmath234 ) have a maximum above the band - gap frequency . in the optical cavity of a laser the equations for gain and spontaneous emission are modified by the mode structure . in particular , the total linear response of an electromagnetic field mode inside the cavity includes the cavity loss rate @xmath138 . for the edge emitting semiconductor laser , the gain function of the cavity modes is given by @xmath237 where the confinement factor @xmath131 is defined according to equation ( [ eq:1dsigma ] ) . spontaneous emission into the cavity modes passes through the gain medium and is thereby absorbed or amplified accordingly . thus for an edge emitting laser , the rate of spontaneous emission into a cavity mode of frequency @xmath170 is given by @xmath238 this rate represents the total rate of spontaneous emission events per mode , regardless of the actual width of the laser . on the other hand , equation ( [ eq : stot ] ) gives the total rate of spontaneous emission per quantum well area . in a laser of length @xmath124 and total width @xmath239 ( c.f . [ edge ] ) , the spontaneous emission rate into free space is @xmath240 . the spontaneous emission factor @xmath0 , which is generally defined as the fraction of spontaneous emission being emitted into the cavity mode @xcite , is on the basis of our theory given by the expression @xmath241 note that the two dimensional carrier density @xmath213 in the quantum well is related to the one dimensional carrier density @xmath242 by @xmath243 . the spontaneous emission factor @xmath0 is a function of both frequency and carrier density . consequently , the common assumption of the spontaneous emission factor @xmath0 being independent of the carrier density @xcite may be regarded as an approximation similar to the assumption of linear gain . an analytical expression for the spontaneous emission factor may be obtained for zero temperature . the k - space integrals may then be solved analytically using equation ( [ eq : bandshape ] ) and assuming that @xmath210 is independent of @xmath209 . for the spontaneous emission factor @xmath0 , the analytical result reads @xmath244 the fermi frequency @xmath212 is defined by equation([eq : fermif ] ) and expresses , in particular , the carrier density dependence of @xmath0 . for @xmath245 we recover the result typically given in the literature ( e.g. @xcite ) which does not depend on @xmath213 . [ beta ] shows the deviation of the spontaneous emission factor from this value as the fermi frequency @xmath212 passes the point of resonance with the cavity mode . [ beta ] illustrates , in particular , the carrier density dependence of @xmath0 for three modes with frequencies above the band gap frequency given by ( a ) @xmath246 , ( b ) @xmath247 , and ( c ) @xmath248 . most notably , @xmath0 is always smaller than the usual estimate given by @xmath249 , which is based on the assumption of ideal resonance between the transition frequency and the cavity mode . with equations ( [ eq:1dgain ] ) and ( [ eq:1dspont ] ) , it is possible to find the steady state intensity @xmath250 of a mode with frequency @xmath170 , @xmath251 note that this result may also be obtained directly from equations ( [ eq:1ddynamics_a]-[eq:1ddynamics_b ] ) . in a wide cavity , the cavity modes are approximately plane wave modes and the relation between @xmath252 and @xmath253 in that case reads @xmath254 the steady state intensity distribution is characterized by the spatial coherence derived from the intensity distribution of the plane wave modes of the cavity . generally , the intensity distribution of plane waves corresponds in the far field to an optical field at angles relative to the axis of emission in the plane of the quantum well . the angular distribution of intensity and coherence in the far field is thus given by @xmath255 therefore , the far field intensity distribution may be determined directly from the frequency dependence of the intensities by @xmath256 the intensity is given in units of @xmath257 per unit angle . for @xmath258 we may in analogy to ( [ eq : beta ] ) solve the integral in ( [ eq:1dspont ] ) by assuming @xmath210 to be independent of @xmath23 . as an analytical expression we then obtain for the far - field intensity distribution of a broad area semiconductor laser @xmath259 the parameter @xmath260 represents the ratio between the cavity loss rate @xmath138 and the maximum amplification rate of the gain medium . laser activity is only possible if @xmath261 . the classical laser threshold is defined by the carrier density for which the denominator of @xmath262 is zero for a single specific frequency @xmath263 . consequently , the carrier density at which this occurs is pinned . figure [ farfield ] shows the far field intensity distribution for different carrier densities below this pinning density . in fig . [ farfield ] ( a ) , the wide intensity distribution of amplified spontaneous emission for carrier densities is much lower than the pinning density . the intensity maximum is clearly located at @xmath264 . figure [ farfield](a ) shows the intensity distribution for carrier densities halfway towards threshold . already , the intensity maxima move to angles of @xmath265 , corresponding to the frequency at which the gain spectrum has its maximum . in the case of fig . [ farfield](c ) , the threshold region is very close to the pinning density . the peaks in the far field pattern narrow as the laser intensity is increased . consequently the far field pattern indeed is a measure of the spatial coherence similar as the linewidth of the laser spectrum is a measure of temporal coherence . it is therefore desirable to consider quantum noise effects in the spatial patterns of optical systems . in the context of squeezing , such patterns have been investigated by gatti and coworkers @xcite based on the general formulation of lugiato and castelli @xcite . the laser patterns presented here are based on the same principles . usually , however , the strong dissipation prevents squeezing in laser systems unless the pump - noise fluctuations are suppressed @xcite . the quantum maxwell - bloch equations ( qmbe ) for spatially inhomogeneous semiconductor lasers derived in this paper take into account the quantum mechanical nature of the light field as well as that of the carrier system . the only approximation used in the derivation of the intensity and correlation dynamics is that of statistical independence between the two carrier systems and the light field . in the qmbe presented here , the effects of coherent spatiotemporal quantum fluctuations which are generally not considered in the semiclassical maxwell - bloch equations for semiconductor laser devices have thus been taken into account . the spontaneous emission term appears side by side with the gain and absorption term in the dynamics of the field - dipole correlation . in this way the spatial coherence of spontaneous emission and amplified spontaneous emission is consistently described by the quantum maxwell - bloch equations . typical features of the model have been illustrated by the spectra of gain and spontaneous emission . an example of the spatial coherence characteristics described by the quantum maxwell - bloch equations has been presented by analytically obtaining the spontaneous emission factor @xmath0 and the far field distribution for the example of a broad area edge emitting laser . in general the quantum maxwell - bloch equations derived for edge emitting and vertical cavity surface emitting lasers provide a starting point for a detailed analysis of spatial coherence patterns in diverse semiconductor laser geometries such as broad area or ultra - low threshold lasers . as explained at the end of section [ sec : qmbe ] , the field average of zero is an expression of our lack of knowledge about the actual physical field present in the laser device . this lack of knowledge could be removed by performing a measurement on the light field . for example , the phase information could be obtained by measuring the interference of the light field from the diode with light from a separate laser . a simple theoretical simulation of such a measurement is obtained by assuming that @xmath153 defines the variance of a gaussian distribution of coherent states ( gaussian p representation ) . the measurement of the actual field @xmath266 may then be simulated by randomly selecting a coherent field from this gaussian distribution . since the field and the dipole density are correlated , the statistical selection of @xmath266 shifts the dipole density average @xmath101 from zero to @xmath267 because the field has been determined with a precision equal to the quantum limit , the correlations factorize and the new values of @xmath153 and @xmath268 are given by the respective products of @xmath266 and @xmath101 . the drastic changes in the intensity distribution and in the dipole dynamics which may result from such a simulated measurement cause spatial holeburning and give rise to relaxation oscillations . by performing several stochastical simulations starting from the same initial probability distribution , the statistics of the carrier density fluctuations may be obtained . it is then possible to derive the correct linewidth enhancement factor @xmath163 as well as the relaxation oscillation sidebands directly from the dynamics of the quantum maxwell - bloch equations . an alternative approach to the problem of quantum noise is given by the langevin equations @xcite . langevin equations simulate quantum noise by adding classical noise sources which reproduce the statistical properties given by the uncertainty relations . in particular , the exponential damping terms given by the cavity loss rate @xmath138 and the dipole relaxation rate @xmath210 must be compensated by a noise input maintaining the quantum fluctuations in the field and dipole densities . for quantum well lasers , these fluctuations are given by note that the variables @xmath270 and @xmath271 are classical quantities in this context . the langevin equations presented in the following do not describe quantum mechanical coherent states . instead , semiclassical light fields and dipole densities are calculated with a precision violating the uncertainty relations . while the statistics thus derived correspond to the quantum statistics , the theory applied is fully classical . @xmath272 \frac{\partial}{\partial t } \tilde{p}_{i}({\bf r},{\bf k } ) & = & - \left(\gamma ( { \bf k } ) + i\omega({\bf k})\right ) \tilde{p}_{i}({\bf r},{\bf k } ) \nonumber \\ & & + i g_0 \sigma \sqrt{\nu_0}\ ; \left ( f^e_{eq}\left(k;\frac{\tilde{n}({\bf r})}{q}\right ) + f^h_{eq}\left(k;\frac{\tilde{n}({\bf r})}{q}\right ) - 1 \right ) \tilde{\cal e}_i({\bf r } ) + q^p_i ( { \bf r},{\bf k},t ) \\ \frac{\partial}{\partial t } \tilde{\cal e}_i({\bf r } ) & = & -(\kappa_i+i\delta\omega_i ) \tilde{\cal e}_i({\bf r } ) + i\frac{\omega_0}{2k_0 ^ 2\epsilon_r}\delta \tilde{\cal e}_i({\bf r } ) \nonumber \\ & & - i g_0\frac{\sqrt{\nu_0}}{4\pi^2 } \int d^2{\bf k}\;\left(\tilde{p}_{i}({\bf r},{\bf k } ) + \sum_j \frac{1}{k_0 ^ 2\epsilon_r}\frac{\partial^2}{\partial r_i r_j } \tilde{p}_{j}({\bf r},{\bf k})\right ) + q^{\cal e}_i({\bf r},t)\end{aligned}\ ] ] these langevin equations allow a stochastical simulation of quantum noise in a semiclassical context . however , it is necessary to perform a large number of runs before statistical results can be obtained . in particular , the unrealistic intensity fluctuations of the light field vacuum need to average out before the results are consistent with quantum theory . 6 o. hess and t. kuhn , prog . electr . * 20 * , 85 ( 1996 ) . ebeling , _ integrated optoelectronics _ ( springer , berlin 1993 ) . t. lee , c.a . burrus , j.a . copeland , a.g . dentai and d. marcuse , ieee j. quantum electron . qe-*18 * , 1101 ( 1982 ) . k. petermann , ieee j. quantum electron . qe-*15 * , 566 ( 1979 ) . a. yariv and s. margalit , ieee j. quantum electron . qe-*18 * , 1831 ( 1982 ) . a.e . siegmann in _ coherence and quantum optics vii _ edited by eberly , mandel and wolf ( plenum , new york 1996 ) . g. bjrk , a. karlsson , and y. yamamoto , phys . a * 50 * , 1675 ( 1994 ) . y. yamamoto and r.e . slusher , physics today , june 1993 , 66 ( 1993 ) . f.jahnke , k.henneberger , w.schfer and s.w . koch , j.opt.soc.am . b * 10 * , 2394 ( 1993 ) . f.jahnke and s.w . koch , phys . a * 52 * , 1712 ( 1995 ) . k. henneberger and s.w . koch , phys . lett . * 76 * , 1820 ( 1996 ) . o. hess and t. kuhn , phys . a * 54 * , 3347 ( 1996 ) . yu and m. cardona , _ fundamentals of semiconductors _ , ( springer , berlin 1996 ) . e. wigner , phys . rev . * 40 * , 749 ( 1932 ) . m. san miguel , q. feng and j.v . moloney , phys . a * 52 * , 1728 ( 1995 ) . j. martin - regalado , m. san miguel , n.b . abraham and f. prati , opt . lett * 21 * , 351 ( 1995 ) . hofmann and o. hess , phys . a * 56 * , 868 ( 1997 ) . h. van der lem and d. lenstra , opt . lett . * 22 * , 1698 ( 1997 ) . hofmann and o. hess , quantum semiclass . opt . * 10 * , 87 ( 1998 ) . van doorn , m.p . van exter , a.m. van der lee and j.p . woerdman , phys . a * 55 * , 1473 ( 1997 ) . w.w . chow , s.w . koch , and m. sargent , _ semiconductor - laser physics _ , ( springer , berlin 1994 ) . t. kuhn and f. rossi , phys . lett . * 69 * , 977 ( 1992 ) , _ ibid . b * 46 * , 7496 ( 1992 ) . c. ell , r. blank , s. benner and h.haug , j.opt.soc.am . b * 6 * , 2006 ( 1989 ) . hofmann and o. hess , opt . lett . * 23 * , 391 ( 1998 ) . d.f . walls and g.j . milburn , _ quantum optics _ ( springer , berlin 1994 ) . a. gatti , h. wiedemann , l.a . lugiato , i.marzoli , g. oppo , and s.m.barnett , phys.rev . a * 56 * , 877 ( 1997 ) . lugiato and f. castelli , phys . lett . * 68 * , 3284 ( 1992 ) .

time travel is a phenomenon which has been attracting interest both in fiction and general discussions for a long time . however , only after a formulation of the theory of relativity such considerations could be investigated on a more scientific and solid basis . already the special relativity shows that different observers experience different times and one of them can `` travel '' to the future of others by means of his relative motion . thanks to the general theory of relativity a possibility opens that an observer could travel even to his own past his worldline could pass through a geometrically or topologically nontrivial area to a region where the worldline originally started @xcite . worldlines which even cross themselves are called _ closed timelike curves ( ctcs ) _ and it is customary to say that spacetimes with ctcs contain _ time machines _ @xcite . spacetimes with time machines are causally nontrivial in such spacetimes you can send a signal to your own past or even try to influence the past which immediately opens a question of consistency of standard physical laws as we know them . on a formal level it is the question of the existence of solutions of physical equations of motion and the question whether the initial value problem is well possessed . on a less formal level these problems can be phrased as the well - known `` grandfather paradox '' , suggested , e.g. , in @xcite : in spacetimes with time machines one has to face a logical riddle what happens if one travels to his own past and kills his grandfather . consequently one would never be born and therefore one could not travel to the past . this is a clearly inconsistent situation which suggests that spacetimes with ctcs are pathological and they should be excluded from a serious scientific consideration . however , a system containing live beings is too complicated by too many unknown physical laws and therefore one can not be sure that the inconsistency of the `` grandfather paradox '' is really inescapable . therefore , people tried to formulate analogous situations for much simpler systems which could be studied exactly @xcite . as an example , in @xcite , and especially in @xcite , the system of the billiard balls in spacetime with wormholes has been studied . this system allows a straightforward reformulation of the grandfather paradox : the ball could be sent through the time machine in such a way that it hits itself and thus prevents inconsistently its entry to the time machine , cf . [ fig : dangerous - px ] . it seems that such `` paradoxical '' initial conditions do not lead to a consistent evolution of the system . * the ball self - colliding inconsistently . * the ball ( in gray ) comes from a distant region and enters the wormhole without any self - collision . it leaves the time machine ( now in white ) in such a way that it inconsistently hits itself . thus , this situation represents an inconsistent evolution which is the direct analogue of the grandfather paradox.,width=316 ] maybe surprisingly , the extensive studies of systems with ctcs during the 1980s and the 1990s showed , that for a simple physical system pathology of spacetimes is not so severe and the equations of motion can be consistently solved . let us formulate this point more precisely . we consider a spacetime containing a time machine and we want to study a system with well known local physical laws ( e.g. , a particle or electromagnetic field ) . we do not change these local laws , i.e. , we require that they hold locally in any small spacetime domain . however , in addition to the local laws , we also require the so - called _ principle of self - consistency _ @xcite . namely , a _ globally consistent _ solution of local laws must exist . it means that we allow the system to propagate to its own past , however , it must be done in a _ consistent _ way with the original evolution in the past . the past can not be changed since it has been already changed . the key question of studies of time machines is whether such globally consistent evolutions exist for given local laws and whether these global evolutions are sufficiently generic . more accurately , we would like to show that there exist consistent solutions for all , or , at least , for almost all standard initial data . otherwise , if the local laws have no globally consistent solution , the spacetime would be clearly pathological and we could rule it out from our consideration . similarly , the pathology would be serious if the local physical laws had only few globally consistent solutions . as we have said , the studies of different systems show that spacetimes with ctcs are not necessarily causally pathological . let us mention the results for a system of interacting particles @xcite or the scalar field theory @xcite where it was shown that standard local laws have generic globally consistent solutions even in the presence of ctcs . another surprising result of such studies is that the existence of time machines does not usually restrict a number of consistent solutions , but on the contrary , it leads to a possibility of more than one globally consistent solutions for given initial values . in spacetimes with time machines we thus usually lose the uniqueness of the evolution @xcite . especially for the apparently paradoxical initial conditions from the `` grandfather paradox''-like situation it was shown @xcite that a consistent evolution exists , although it can be rather nonintuitive . on less formal level of the human version of the paradox it could be rephrased as a conjecture of a hidden law which always prevents the grandson from killing his grandfather @xcite . in the present work we want to study the system of a ball moving in a non - relativistic space with ctcs which is very close to that of @xcite . however , we have chosen a different time - machine configuration which significantly simplifies the analysis . we will be able to solve the equations of motion explicitly and we will confirm the behavior described above : `` paradoxical '' initial conditions are not really paradoxical , and the evolution is not , in general , unique . the contribution of our analysis is that it can be done very explicitly , since our model is sufficiently simple ; at the same time it is sufficiently non - trivial to reproduce the most important features of systems with time machines . our work proceeds as follows . in the next section we describe the system of a ball moving in the space with wormhole - based time machine . we derive the equations characterizing trajectories . in sec . [ sc : numbsol ] we discuss the character and number of solutions for various initial conditions and also the resolution of the `` grandfather paradox '' . [ sc : ppart ] describes geometry of the trajectories and the paper is summarized in the conclusion . * spatial representation of two simple wormholes . * a wormhole obtained by gluing two planar sections cut from otherwise euclidian space . planar wormhole mouths could be obtained , e.g. , by squeezing the mouths of the spherical wormhole into very thin planes ) . the external curvature is vanishing with the exception of the boundary of the planar sections and the geometry through the wormhole is thus flat.,width=316 ] the simplest and most natural way how to construct a spacetime with ctcs is to use a wormhole @xcite . the wormhole can be viewed as a shortcut between two location of the spacetime . it forms ctcs if one of the mouths of the wormhole lies in the past of the other mouth . in the nonrelativistic setting we can consider spatial wormholes connecting two places in space with an additional time - shift . the speed of light is infinite and it determines a unique notion of simultaneity and thus it allows us to define a global time at least , before introducing the time difference . introducing the time - shift means that traveling through the wormhole not only sends the observer to a different place in space but also to a different time . a simple spatial wormhole in otherwise euclidian space can be obtained by the cutting and gluing method . for example , we can cut out two spheres and glue their surfaces together , cf . @xcite , or @xcite . we thus obtain a topologically and geometrically nontrivial space it is not a simply connected space and the geometry on the glued surface is not flat . in this work we consider even a simpler situation of the wormhole with planar mouths instead of spherical ones . namely , we cut out from the space two planar sections which we identify as in fig . [ fig : wh - s ] . since we use flat planar sections , their identification is geometrically trivial . the whole curvature of the mouths is squeezed to the borders of the planar sections which can be understood as a kind of solid frames on which the traversable parts of the wormhole are spanned . to avoid a discussion of the wormhole boundary we assume that the planar sections are much larger than the scales of our experiments . as idealization , we consider the mouths of our wormhole to be two half - planes which form an angle @xmath0 with a common boundary line , called the axis . if we identify these half - planes ( first , say , at the same moment of time ) , the space between them becomes a locally euclidean space with a conical singularity localized on the axis . indeed , if we restrict ourselves to the two dimensional picture and ignore the direction parallel to the axis , our space forms a cone with the angle @xmath0 around the vertex . of course , this is a over - idealized situation . we should keep in mind that the mouths of the wormhole are large but finite , so somewhere very far from the axis the conical part of the space ends and goes over to the full euclidean space . but in most of our discussions we restrict ourselves only to the part of the space between the mouths of the wormhole . we thus effectively work in the conical space with angle @xmath0 around the axis . let us stress that in our construction the mouths of the wormhole are special and privileged . however , after enlarging them to the semi - infinite size and restricting ourselves only to the conical space between mouths , we can no longer localize the position of the mouths by local experiments . the geometry through the mouths is locally euclidean as anywhere else . we thus obtained a space which is axially symmetric with respect to the rotation around the axis ( it has also translation symmetry along the axis and it is static ) . the position of the wormhole can be identified only on scales larger than the wormhole , from the surrounding globally euclidean space . finally , we assume that the angle @xmath1 . only for @xmath0 smaller than @xmath2 , we obtain interesting situations when the straight trajectory of a free particle intersects itself . for @xmath3 the straight trajectory intersects itself only once . the second condition is not a crucial assumption , it just simplifies the discussion . since we want to study a space with a time machine , we have to identify the mouths of the wormhole with a time - shift @xmath4 . of course , it destroys the standard causal structure of the non - relativistic spacetime ( a clear distinction between future and past ) . despite this , we will keep using and referring to the original notion of the simultaneity and to the global time of the surrounding space , which is , certainly , reasonable for small wormholes , but we admit , it can be slightly confusing and less founded in the idealized case considered below . indeed , the constructed spacetime is still locally euclidian ( of course , except the axis ) , but endowed with a strange causal structure . hypersurfaces of simultaneity ( visualized in the standard non - relativistic spacetime diagram as horizontal planes ) propagate through the wormhole and form `` helical '' surfaces winding around the axis . this indicates that the spacetime contains ctcs . the particle moving towards the wormhole reads that the external time @xmath5 increases continuously until it enters the wormhole mouth . by crossing one of the mouths the external time @xmath5 either increases by @xmath6 into @xmath7 , or decreases by @xmath8 into @xmath9 . * a spacetime diagram of the conical time machine . * the vertical direction is temporal , horizontal planes correspond to the hypersurface of simultaneity of the original spacetime ( the third spatial direction is suppressed ) . two half - planes on the boundary of the conical spacetime represent the history of the mouths of the wormhole and they are identified with a time - shift @xmath4.,width=316 ] this can be visualized in the spacetime diagram in fig . [ fig : wh - st ] . here , two spatial directions perpendicular to the axis are shown in horizontal directions , the direction parallel to axis is suppressed . the vertical direction corresponds to time . semi - planar mouths of the wormhole at one moment are thus depicted as horizontal semi - lines , their time evolution as vertical half - planes . the identification of such two half - planes is not on the same vertical level , but with the vertical shift @xmath4 . we assume that going through the wormhole in anticlockwise direction takes us time @xmath10 to the past , in clockwise direction to the future . now , we want to derive equations of motion for a ball moving in the non - relativistic spacetime which has just been described . we assume that the motion satisfies classical local laws of motion . additionally , we restrict ourselves to motions perpendicular to the axis of the wormhole because the ball can self - interact only for such a motion . clearly , a free particle moves uniformly along a straight line . however , for @xmath11 , a straight line in a conical space must intersect itself . if the trajectory of the particle crosses itself in different times , we speak about _ self - intersection_. if the particle intersects its trajectory exactly at the same time which is allowed thanks to time - shift @xmath8 gained in the wormhole it hits itself and we speak about _ self - collision_. our goal is to describe trajectories with exactly one consistent self - collision . * character of self - collision of the ball with itself . * the incoming velocity @xmath12 of the younger version of the ball ( in grey ) is complementary to mirror reflection of the outgoing velocity @xmath13 of the older version of the ball ( white ) and similarly for velocities @xmath14 and @xmath15.,width=316 ] * physical self - collisions of types i and ii . * ( a ) the configuration of type i represents situation when the younger ball ( in grey ) is on the left side while the older ball ( in white ) is on the right side with respect to the radial line . the self - collision is physical when the momentum transfer is positive and the trajectory of the particle is deflected leftward . ( b ) the configuration of type ii represents situation when the older ball is on the left side while the younger ball is on the right side with respect to the radial line . the self - collision is physical if the trajectory of the particle is deflected rightward.,title="fig:",width=316 ] + ( a ) + * physical self - collisions of types i and ii . * ( a ) the configuration of type i represents situation when the younger ball ( in grey ) is on the left side while the older ball ( in white ) is on the right side with respect to the radial line . the self - collision is physical when the momentum transfer is positive and the trajectory of the particle is deflected leftward . ( b ) the configuration of type ii represents situation when the older ball is on the left side while the younger ball is on the right side with respect to the radial line . the self - collision is physical if the trajectory of the particle is deflected rightward.,title="fig:",width=316 ] + ( b ) + we assume that an elastic collision occurs when the ball collides with itself . the classical elastic impact of two balls is determined by the momentum and energy conservation and the assumption that it occurs in one plane . in our case two versions of the same ball collide : the _ younger _ version of the ball coming from infinity hits the _ older _ version coming from the wormhole ; after the collision the younger ball flies to the wormhole and the older one away to infinity . since the traverse of the wormhole does not change the velocity of the ball , magnitude @xmath16 of the outgoing velocity @xmath15 of the younger ball is the same as the magnitude of the incoming velocity @xmath13 of the older ball . all these conditions determine @xcite that the impact of the ball with itself must have the form depicted in fig . [ fig : impact ] . namely , in the rest frame of the wormhole the incoming velocity @xmath12 of the younger ball is complementary to the mirror reflection of the outgoing velocity @xmath14 of the older ball with respect to the plane tangent to balls at the point of impact . we call this plane the _ impact plane_. similar property holds for velocities @xmath15 and @xmath13 . moreover , projections of the incoming and outgoing velocities @xmath12 and @xmath15 ( respectively @xmath13 and @xmath14 ) to the impact plane are the same . * geometry of self - colliding point - particle trajectory . * ( a ) a point particle is approaching the wormhole from infinity with a velocity @xmath17 and impact parameter @xmath18 , it collides with the version of itself which already passed through the time machine , and with a velocity @xmath19 moves toward the wormhole . after passing it and self - colliding with itself , it moves with the velocity @xmath20 back to infinity . the self - collision occurs at the distance @xmath21 from the axis and the outgoing trajectory forms with the radial direction the orientated angle @xmath22 . thanks to symmetry of the self - collision with respect to radial direction , the angle @xmath22 is a half of the angle between the incoming and outgoing trajectory . ( b ) the same situation depicted in a map which cuts the conical space not along the wormhole but along the radial direction through the self - collision . the length @xmath23 of the straight trajectory between its self - intersection is given by @xmath24.,title="fig:",width=316 ] + ( a ) + * geometry of self - colliding point - particle trajectory . * ( a ) a point particle is approaching the wormhole from infinity with a velocity @xmath17 and impact parameter @xmath18 , it collides with the version of itself which already passed through the time machine , and with a velocity @xmath19 moves toward the wormhole . after passing it and self - colliding with itself , it moves with the velocity @xmath20 back to infinity . the self - collision occurs at the distance @xmath21 from the axis and the outgoing trajectory forms with the radial direction the orientated angle @xmath22 . thanks to symmetry of the self - collision with respect to radial direction , the angle @xmath22 is a half of the angle between the incoming and outgoing trajectory . ( b ) the same situation depicted in a map which cuts the conical space not along the wormhole but along the radial direction through the self - collision . the length @xmath23 of the straight trajectory between its self - intersection is given by @xmath24.,title="fig:",width=316 ] + ( b ) + now we want to find a location of the self - collision in such a way that the outgoing younger ball consistently passes to the incoming older ball . it can be done @xcite in two ways depicted in fig . [ fig : bcollisions ] . the key property is that the impact plane must be radial , i.e. , it contains the axis ( dashed line in the figure ) . type i represents the situation when the older ball touches the younger one by its rear part , i.e. , the younger version must collide from the left side . for type ii the younger ball touches its older version by its front part and is incoming from the right side . for both types we can distinguish the physical collisions with a positive momentum transfer from the younger ball to the older ball from `` spurious '' collisions which would need a negative momentum transfer . the latter are unphysical not only for the sign of the exchanged momentum but also because the ball would not fit geometrically into space for such collisions it would have to fly through itself . for the physical self - collision of type i the younger ball must deflect leftward , i.e. , closer to the vertex than if it followed the collision - free trajectory . for the physical self - collisions of type ii the trajectory of the younger ball is deflected rightward , cf . [ fig : bcollisions ] . to determine parameters of the self - collision , we start with a simpler case of a point particle , i.e. , taking radius of the ball @xmath25 . in this case the distinction between the two types described above disappears and the geometry of the self - collision is depicted in fig . [ fig : model - parameters ] . the trajectory of the incoming particle can be determined by two initial parameters : the _ impact parameter _ @xmath26 which gives the distance of the incoming trajectory from the vertex of the cone , and the _ magnitude of incoming velocity _ @xmath27 . we adopt the convention that the impact parameter @xmath26 is positive if the particle circles the cone in the counterclockwise direction and it is negative if it circles the cone in the clockwise direction . of course , the parameters @xmath20 and @xmath26 do not determine the incoming trajectory uniquely since they do not specify its angular location around the axis and its temporal location we would need additional two parameters for that . however , the conical space has the rotational symmetry and it is static , so the exact angular and temporal location is irrelevant for the character of solutions . therefore we can ignore the additional initial parameters when we investigate the geometry of the self - collision . the self - collision can be parametrized by its _ radial distance _ @xmath28 from the axis and by the oriented _ scattering angle _ @xmath29 between the outgoing trajectory and the radial direction , see fig . [ fig : model - parameters]a . thanks to the symmetry of the self - collision with respect to the radial direction , the angle @xmath22 also determines the direction of incoming trajectory . for the same reason , the outgoing particle also has velocity @xmath17 and impact parameter @xmath18 . the angle between the inner trajectory and the radial direction is determined by the conical geometry and it is equal to @xmath30 , cf . fig . [ fig : model - parameters]b . the length @xmath23 of the inner trajectory is then @xmath31 . the collision parameters @xmath21 and @xmath22 encode the same information as the initial parameters @xmath17 and @xmath18 . indeed , @xmath18 in terms of @xmath21 and @xmath22 is given by the simple geometry @xmath32 since the radial projection of the particle velocity before and after the collision is the same ( as a consequence of the laws of the elastic impact as we discussed for finite balls ) , the incoming velocity @xmath17 is related to the inner velocity @xmath16 along the trajectory between the self - collision as @xmath33 however , the inner velocity must be such that the particle passes the inner trajectory exactly in the time @xmath4 gained in the wormhole , @xmath34 we thus obtain relation for @xmath17 : @xmath35 it will be also useful to write down the relation for @xmath18 as function of @xmath22 and @xmath17 @xmath36 equations and thus relate the parameters of the incoming trajectory @xmath17 , @xmath18 and the parameters of the self - collision @xmath21 , @xmath22 . their consequences will be discussed in more detail in sec . [ sc : ppart ] . before we return to the case of finite balls , let us note that we can introduce similar parameters also for the free trajectory . the only difference is that @xmath21 and @xmath22 refer , in general , to self - intersection ( i.e. , not necessarily to self - collision ) of the free trajectory . clearly , the angle @xmath22 is now given just by the conical geometry @xmath37 the solutions , for which the younger version of the particle passes through the point of self - intersection later than the older version ( coming from the time machine ) , are the solutions of type i. when the younger particle passes the point of self - intersection earlier , we speak about the solutions of type ii . the self - intersection threatens to become a self - collision if time spent on the inner part of the free trajectory is equal to the time - shift of the time machine . since for a free trajectory @xmath38 , conditions and give @xmath39 such values @xmath40 , @xmath41 correspond to the `` paradoxical '' free trajectory for which the particle occurs at the point of self - intersection twice at the same time . for a point particle , it is not possible to decide plausibly what happens in such a situation . we can only observe , that the same parameters @xmath40 , @xmath41 also describe the trajectory with one self - collision , which is given by the same parameters as `` paradoxical '' self - intersection of the free trajectory , namely @xmath42 and @xmath43 . a detailed analysis of this situation will be done for finite balls in sec . [ sc : numbsol ] . the direction @xmath44 also plays a role of the limiting value between physical collisions of type i and ii . physical self - collisions of type i , for which the trajectory deflects leftward from the free trajectory , are bounded within the interval @xmath45 ; while physical self - collisions of type ii ( with rightward deflection ) sweep out the interval @xmath46 . we have to be more precise to define the parameters of the self - collision for a finite ball . let @xmath21 be the radial distance from the axis of the intersection of the incoming and outgoing trajectories ( extended beyond the actual self - collision ) , and @xmath47 be the radial distance from the axis of the self - intersection of the inner trajectory ( for type ii extended beyond the self - intersection ) , see fig . [ fig : bcollisions ] . clearly , eq . still holds , and @xmath48 , with upper sign for type i and lower sign for type ii . the geometry of the inner trajectory of the finite balls is identical with that of a point particle with modified impact parameter @xmath49 . the path @xmath23 traveled during time @xmath4 gained in the time machine must be corrected due to finite radius of balls , @xmath50 , where , analogously to the point - particle case , @xmath51 is the length of the inner trajectory between its self - intersection . the corrected relation together with finally leads to equation between @xmath17 , @xmath18 , and @xmath22 : @xmath52 again , with upper / lower sign for type i / ii , respectively . physical self - collisions of type i or ii are restricted by the conditions @xmath53 otherwise the ball would deflect to a wrong side of the free trajectory and the momentum transfer from the younger to older version of the ball would be negative . . supersedes relation for the point particle , as can be also seen by taking limit @xmath54 . this can be regarded as the key equation of motion which determines the scattering angle @xmath22 in terms of the initial conditions @xmath17 and @xmath18 . the position of the self - collision is then determined by the parameter @xmath21 which is given by eq . . * the @xmath55 diagram of consistent solutions for a point particle . * each point of the graph in the @xmath18-@xmath22 plane represents a consistent solution for a point particle with at most one self - collision . the velocity @xmath17 is fixed ( it changes only a scale of the solutions ) and the parameter @xmath21 is given by eq . . collision - free solutions are represented by the straight vertical lines . the line @xmath56 ( in green ) represents collision - free trajectories to the future , the line @xmath43 represents free trajectories to the past . the points below the value @xmath57 ( in red ) corresponds to the solutions of type i , for which the younger particle passes the point of self - intersection later than that coming from the wormhole . the solutions with @xmath58 ( in blue ) are of type ii . the sinusoidal part of the graph represents solutions with one self - collision . the part with @xmath59 ( in red ) represents self - collisions of the type i , the part with @xmath60 ( in blue ) represent self - collisions of type ii . self - colliding solutions are thus possible only for @xmath61 . the point @xmath62 represents both the paradoxical collision - free solutions for which younger and older versions of the particle meet at the point of intersection at the same moment , and the consistent self - colliding solution ( which is geometrically identical to the paradoxical solutions ) . exact relation of these solution is clarified in the case of balls with a finite radius.,title="fig:",width=316 ] + in the case of the point particle there is only one length scale @xmath63 given by the initial velocity @xmath17 and the time - shift @xmath4 . the velocity @xmath17 thus changes only the scale of the whole experiment . therefore , we can fix @xmath17 and study only the relations among @xmath18 , @xmath21 , and @xmath22 . the relation between @xmath22 and @xmath18 is given by eq . and depicted in fig . [ fig : p - rho(omega ) ] , the relation to the parameter @xmath21 is given by ( or , alternatively , implicitly by eq . ) . we immediately see that the self - collision can happen only for @xmath64 for larger @xmath65 the particle would not be fast enough to travel through the wormhole and hit itself consistently it would be too far from the axis of the wormhole and the path through the wormhole would be too long . * three possible solutions for given initial conditions of a point - like particle . * for @xmath61 , @xmath66 there exist three possible solutions : the collision - free trajectory ( black ) and two self - colliding trajectories ( red / gray and blue / light - gray ) . one of the self - colliding trajectories is close to the free trajectory ( it coincides in the limit @xmath67 ) , another is rather different.,width=316 ] for @xmath61 , the map @xmath68 is not a unique relation : for a given @xmath18 we have , in general , two @xmath22 satisfying . for the given initial conditions , we thus obtain , in addition to the collision - free trajectory , two self - colliding solutions . these two solutions have the scattering angles complementary to @xmath69 and represent completely different evolutions of the system . note , that the self - colliding solutions exist even if @xmath70 , i.e. , if the free trajectory passes the wormhole to the future . all three possible solutions for the given initial conditions ( with @xmath71 ) are shown in fig . [ fig : p.two.solutions ] . for the paradoxical value @xmath72 , eq . , the `` pardoxical '' collision - free solution geometrically coincides with one of the self - colliding solutions . it is called `` paradoxical '' , since for this initial values the free moving particle meets itself at the point of self - intersection exactly at the same moment , so it can not be the collision - free solution . however , since for a point - like particle we can not distinguish the non - colliding solution from the colliding one , it is dubious to study the nature of this `` paradoxical '' situation . therefore , we will discuss the details of the `` paradoxical '' initial conditions in the case of finite balls . after that we will return to the simpler case of a point - like particle to study the geometry of the trajectories in more detail . in the case of finite balls the system has an additional length scale given by the radius @xmath73 of the ball . therefore , the dependence on the velocity @xmath17 is not trivial any more . nevertheless , we still fix the value of the initial velocity and discuss the structure of the corresponding solutions . the values of various quantities can , however , depend on @xmath17 in a more complicated way than just a rescaling . collision - free trajectories of a finite ball are again characterized by the angle @xmath74 and they can be also divided into types i and ii , see fig . [ fig : collision - free ] . however , due to the finite radius of the balls , there is a whole interval of the impact parameter @xmath18 , for which the collision - free trajectory becomes `` paradoxical''if the ball hits itself inconsistently . checking the geometry of the free trajectory of a ball , it turns out @xcite , that the boundary of this interval is given by values of @xmath18 determined by eqs . for types i and ii with @xmath43 , @xmath75 explicitly , the `` paradoxical '' values of the impact parameter are given by @xmath76 for @xmath77 the collision - free trajectory is of type i , i.e. , the older version of the ball overtakes the younger one at the point of intersection . if @xmath78 , the collision - free trajectory is of type ii , cf [ fig : collision - free ] . * two types of collision - free trajectories passing the time machine to the past . * ( a ) trajectory of type i. for @xmath77 , time @xmath79 spent by the ball between the self - collision is smaller than the time - shift @xmath4 , i.e. , the younger version of the ball goes through the point of self - intersection later than the older version . ( b ) trajectory of type ii . for @xmath78 the younger version of the ball goes through the point of intersection earlier than the older version.,title="fig:",width=316 ] + ( a)(b ) the parameters of the self - colliding trajectories are characterized by relations and . we can depict the physical solutions ( cf . conditions ) with at most one self - collision in a diagram analogous to the one we used in the point - particle case , see fig . [ fig : rho&r(omega)-a ] . we see here that the solutions split into two branches : the physical self - colliding solutions of type i together with collision - free solutions of type i , and self - colliding and collision - free solutions of type ii . the collision - free solution of type i becomes self - colliding solution of type i for @xmath80 , @xmath81 . it is limiting case when the ball on the free trajectory just touches itself but it does not exchange any momentum . it is thus identical to the limiting case of physical self - colliding solutions of type i. similarly , for @xmath80 , @xmath82 the ball on the free trajectory just touches itself , now from the other side , and it becomes the self - colliding solution of type ii . * physical solutions in the @xmath55 plane . * the points on the curves represent solutions characterized by the parameters @xmath18 and @xmath22 . the initial velocity @xmath17 is fixed . the physical solutions correspond to the solid curves , the spurious one to the dashed curves . vertical lines are collision - free solutions . sinusoidal curves , determined by eqs . , represent the solution with one self - collision . the type of the solution is indicated in a similar way as in fig . [ fig : p - rho(omega)].,width=316 ] inspecting fig . [ fig : rho&r(omega)-a ] , we can also conclude , that for the initial conditions with @xmath83 ( where @xmath84 are given by the maximal values of the expression for type i and ii , respectively ) , there exist just one consistent solution . it is typically a collision - free trajectory , however , for certain values of parameters it can also be a self - colliding solution , cf . [ fig : ball - structure - dangerous]b . for initial conditions with @xmath85 , there are three possible solutions ; typically , one collision - free and two self - colliding . however , for the paradoxical values of the impact parameter @xmath86 , the collision - free solution is superseded by a self - colliding solution , cf . [ fig : ball - structure - dangerous ] . in this case , we obtain two self - colliding solutions of type i and one solution of type ii , see fig . [ fig : bdangerousregion ] . we can conclude that the `` paradoxical '' initial conditions are not paradoxical in any dangerous way . these initial conditions lead to the same number of solutions as other sufficiently close initial conditions . only , the character of solutions is different : the collision - free solution changes to the self - colliding solution . the number of solutions depends on other characteristics , namely if @xmath85 : for a large @xmath65 only the collision - free solution is admissible . however , it should be said that the situation changes if the angle between the wormhole mouths is smaller . for @xmath87 , the equations for the trajectory remain the same but the value of the angle @xmath74 is larger than @xmath88 . the structure of the solutions thus changes as depicted in fig . [ fig : paradox ] . clearly , for @xmath86 the collision - free trajectory is not possible and it is not superseded by another solution . moreover , for such @xmath18 , there exists only one self - colliding solution . in this case we can indeed speak about paradoxical initial conditions since the solutions evolved from these conditions are really restricted . however , in the case @xmath87 the whole discussion is more complicated since the particle can self - interact in a more complicated way ( there is possibility of self - intersection between several self - collisions , etc . ) . we do not have any indications that these more complicated processes could improve the discussed behavior , however we have not excluded it completely @xcite . in the following we restrict again to the larger angles , @xmath89 . * number of solutions for given initial conditions . * the number of solutions for chosen initial parameter @xmath18 can be determined from fig . [ fig : rho&r(omega)-a ] by intersecting the graph with the horizontal line corresponding to @xmath18 . the choice of the initial velocity @xmath17 can slightly modify a shape of the diagrams . here , two representative cases are shown . only the part with @xmath90 is depicted it approximately corresponds to the initial conditions @xmath91 with the ball directed to the time machine toward the past . clearly , the number of solutions steps from one to three with @xmath18 becoming smaller than @xmath92 . in the `` paradoxical '' interval @xmath93 the number of solutions remains the same . only the character of the solutions is different : the collision - free solution is superseded by the self - colliding solution . in the case ( b ) the limiting value @xmath92 belongs to the `` paradoxical '' interval . for @xmath94 there exists just one self - colliding solution.,title="fig:",width=316 ] ( a)(b ) * three self - colliding solutions for @xmath95 from the `` paradoxical '' interval . * unlike the point - like case , there is a whole interval @xmath93 , where collision - free trajectories do not exist . as we can see from fig . [ fig : ball - structure - dangerous]a , the collision - free trajectory is replaced by a self - colliding one . the diagram depicts all three self - colliding solutions , two of type i ( in light and dark red ) and one of type ii ( in blue).,width=316 ] * solutions from `` paradoxical '' interval for @xmath96 . * in this case the collision - free angle @xmath97 . for the impact parameter @xmath18 from the corresponding paradoxical interval @xmath86 we find a gap in the solution curves . for a paradoxical @xmath18 we obtain only one self - colliding solution . moreover , this solution corresponds to a value of @xmath22 which is substantially different from @xmath74.,title="fig:",width=316 ] + clarifying the character of solutions with `` paradoxical '' initial conditions , in this section we describe the motion of a point - like particle in more detail . the geometry of the inner trajectory is given only by the parameter @xmath21 and by the angular position of the self - collision with respect to the wormhole i.e . , it is independent of the scattering angle @xmath22 . the geometry is such that the point of self - collision , the point of the entry to wormhole , the point of the departure from the wormhole , and the axis itself , belong to a common circle , cf . [ fig : model - parameters]a . the inner particle velocity @xmath16 is given by eq . . for the fixed point of self - collision one can then choose any @xmath29 , which determines @xmath18 through eq . , and calculate the initial velocity using eq . . taking into account the rotational symmetry of the conical space between the mouths of the wormhole , we can ignore the angular position with respect to the wormhole in the discussion about many features of the motion . but , when discussing the asymptotic behavior of the incoming and outgoing trajectories , the angular information must be taken into account . namely , one has to track if the outer trajectories pass through the wormhole . clearly , it must occur for the scattering angle @xmath98 , but , depending on the angle between the self - collision and wormhole , it can happen also for smaller angles @xmath22 . however , we first describe the motion without a reference to the wormhole mouths . it can be done in the simplest way employing the totally covering space for our conical space . namely , instead of the conical space with angular coordinate @xmath99 , we use the space without any restriction on @xmath100 , i.e. , a helical surface winding around the axis infinitely . the original conical space is then obtained by the identification of points which differ in coordinate @xmath100 by an integer multiple of @xmath0 . in this covering space the trajectory can be described as follows . let us assume a particle incoming along the direction @xmath101 with the impact parameter @xmath18 and the initial velocity @xmath17 . the self - collision @xmath102 for such a trajectory always happens on the circle ( let us call it the _ collision circle _ ) which has the center @xmath103 on the radial line @xmath104 , it passes through the axis , and its radius is @xmath105 , cf . [ fig : trajgeometry ] . for @xmath106 , the incoming trajectory intersects this circle twice , which corresponds to two possible self - colliding solutions . the angular coordinate @xmath100 of self - collision is given exactly by the scattering angle @xmath22 . at the point of self - collision @xmath102 the trajectory is deflected , and it continues toward the point @xmath107 , which can be obtained by a contra - clockwise rotation of the point @xmath102 around the axis by angle @xmath0 , cf . [ fig : trajgeometry ] . in the covering space @xmath102 and @xmath107 are different points , however , in the original conical space these points are identified as the point of self - collision of course , the particle must pass the point of self - collision twice . finally , from @xmath107 the particle continues through the covering space in the direction which aims from the center @xmath108 of the rotated collision circle , cf . [ fig : trajgeometry ] . * self - colliding trajectory in the totally covering space . * the particle incoming along the direction @xmath101 with the impact parameter @xmath18 is deflected at the point of self - collision @xmath102 . the point @xmath102 must lie on the collision circle with the center @xmath103 on the radial line @xmath104 , it passes through the axis @xmath109 , and its radius is @xmath105 . the trajectory continues toward the point @xmath107 , which is obtained by a contra - clockwise rotation of the point @xmath102 around @xmath109 by angle @xmath0 . in the original conical space points @xmath102 and @xmath107 are identified and correspond to the point of self - collision . from @xmath107 the trajectory continues in the direction which aims from the center @xmath108 of the rotated collision circle . the direction of the inner trajectory goes through the focusing point @xmath110 . the diagram shows only a part of the totally covering space.,width=316 ] * scattering of the plane - wave congruence of particles in the totally covering space . * congruence of particles coming along @xmath101 direction in `` plane - wave '' configuration ( i.e. , aligned at an initial moment on a planar `` wavefront '' perpendicular to the direction of motion ) approaches the time machine . the particles scatters on the collision circle and moves toward the corresponding points on the rotated collision circle . here , they are deflected in the directions coming from the center @xmath111 of the rotated collision circle . the dotted curve depicts the wavefront after the scattering in the totally covering space . the particles do not scatter on the collision circle at the same time ; however , for large final times the wavefront after scattering approaches the circle . the real wavefront projected back to the conical space is drawn as dashed curves . segments projected from various sheets of the totally covering space gain additional time - shift thanks to the passage through the time machine . therefore , these parts of the wavefront are larger since the particle had more time for their motion . inner trajectories focus at one focusing point , for @xmath112 they pass through this point . the diagram shows only part of the totally covering space , given approximately by @xmath113.,width=316 ] + + + + + + + ( d)(e)(f)(g ) + it means , that particles which approach the time machine along the same direction @xmath104 ( with various @xmath106 , forming thus congruence with `` plane - wave '' wavefront ) self - collide on the common colliding circle and , in the end , leave in the directions which point out from the common point @xmath108 , cf . [ fig : congruence ] . in the covering space , the original plane - wave congruence of particles scatters to the circular - wave congruence , but the particles are phase shifted . they fly in radial directions , but the wavefront of the congruence at one moment does not form a circle since the particles start to move in the radial directions in various times . indeed , the incoming particles do not self - collide at the same time : they reach the collision circle in various times . let us mention an interesting feature of the trajectories of the discussed congruence : all inner trajectories ( between the self - collisions ) have the direction going through one focusing point @xmath110 , cf . [ fig : congruence ] . for @xmath112 they even pass through this point . the congruence thus focuses in this point , however , the particles do not pass the focusing point at the same moment . we have obtained a simple picture of the motion in the covering space . it complicates slightly if we return to the conical space by choosing the orientation of the wormhole ( i.e. , setting mouths at the angles @xmath114 and @xmath115 with @xmath116 ) and making the identification of points with @xmath100 differing by an integer multiple of @xmath0 . the resulting picture depends on the values of @xmath0 , @xmath117 and of the impact parameter @xmath18 ; some of the representative trajectories are depicted in fig . [ fig : trajexamples ] . the typical quantity , which depends on the angular position of the self - collision , is the total time @xmath118 gained in the wormhole during the whole scattering process . it can be read out in the covering space from the angular coordinate @xmath119 of the outgoing trajectory : and its angular position with respect to the wormhole is given by the parameter @xmath117 . alternatively , we could set @xmath120 and admit @xmath121 . ] if it belongs to @xmath122-th copy of the wormhole in the covering space , @xmath123 , the particle gains the time - shift @xmath124 : @xmath125 taking into account the restrictions on @xmath0 , @xmath117 and @xmath22 , the total time - shift can be @xmath126 . see fig . [ fig : trajexamples ] for examples . the total time - shift can be similarly calculated for the collision - free trajectories . in this case the direction of the outgoing trajectory in the covering space is @xmath127 for the trajectories passing the wormhole to the past , and @xmath128 for the trajectories passing the wormhole to the future . the condition gives that the trajectory traveling to the past , @xmath91 , gains the time - shift @xmath129 for @xmath130 , or it can pass the wormhole twice , @xmath131 , if @xmath132 . similarly , for @xmath70 , the particle gains the time - shift @xmath133 for @xmath134 and @xmath135 for @xmath136 . we have analyzed a simple interacting system in the space with a nontrivial causal structure . we have assumed the principle of self - consistency , i.e. , we have required the validity of standard local physical laws and we searched for consistent global solutions . in such setting one can formulate the analogue of the so called `` grandfather paradox '' as a question whether all natural initial conditions lead to the global consistent solutions of the local physical laws . initial conditions which do not lead to such solutions would be paradoxical and the system with nontrivial set of paradoxical initial conditions would be logically inconsistent . we have shown that the investigated system of finite billiard balls in the non - relativistic space with a time machine realized by a wormhole with a time - shift is not logically inconsistent in this sense . a nave guess can identify potentially paradoxical initial conditions for which the ball sent through the time machine hits itself and changes inconsistently its own motion . the detailed analysis has shown that even for such initial conditions there exists a global solution satisfying locally all physical laws . this result was anticipated since it agrees with the previous results obtained for various other systems , in particular , for the very closely related system studied in @xcite . our results thus endorse one of the main messages of the study of the system with a nontrivial causal structure : that the presence of time machines does not necessarily imply a drastic reduction of space of the classical solutions of the equations of motion . however , the previous results have been obtained mainly for linear ( non - interacting ) systems or through a rather complicated and cumbersome analysis of interacting systems . therefore , the confirmation for the interacting system allowing a detailed explicit analysis is valuable . we have confirmed also another similarly interesting result discussed in @xcite . the presence of the nontrivial causal structure in our system has enlarged the space of solution . we have found that for a wide class of initial conditions ( namely for @xmath137 with given @xmath17 ) more than one classical evolutions exist . in addition to the `` standard '' collision - free trajectory the particle can also move along two different self - colliding trajectories , see fig . [ fig : p.two.solutions ] . the evolution of the system thus fails to be deterministic and the classical theory does not have any means to determine which of the different evolutions would be realized . one can speculate that the quantum theory would be a more complete description . it does not determine exact trajectory of the particle it just estimates a probability for various possible evolutions ( specified up to a quantum uncertainty ) . in the classical limit only evolutions close to the solutions of the classical laws would have a non - trivial probability . in systems with standard causal structure , there is usually only one such a solution . however , in our system , one can expect that all three classical solutions would have non - trivial probabilities and a semi - classical approximation could give an estimate for these probabilities . nevertheless , one has to emphasize that it s not trivial at all to complete this program . the status of quantum theory in the presence of time machines is unclear . the common formulations of the quantum theory are heavily based on the standard notion of time , so a nontrivial causal structure would change the theory substantially . it would be extremely interesting to find a modification of the quantum mechanics for a space with time machines , but it is a hard challenge for further work . the simplicity of our model could give hope that such a challenge is treatable . however , one could ask what a key ingredient is for the existence of more solutions of the classical equations of motion . is it really the presence of the time machine ? one could also suspect a peculiar geometrical structure of our conical space . the conical space breaks the uniqueness of the classical solutions even without the time machine . indeed , for given initial and final points , in the conical space there exist more trajectories which join them . this effect is purely geometrical and does not need a nontrivial causal structure . however , we have identified a different kind of uniqueness : we have found that the specification of both the initial position and the momentum ( the complete initial conditions ) admits more different evolutions , which is certainly surprising when the local evolution is given by standard differential equations which are generally accepted as deterministic . non - uniqueness appears here as a strange result of the interplay between a local deterministic evolution and a nontrivial causal structure on a global scale for which the time machine is the key ingredient . let us finally mention , that further study @xcite of our system revealed , that if one takes into account a possibility of multiple self - collisions , the situation gets even more interesting . in this work we have studied only the collision - free trajectories , and the trajectories with one self - collision . however , the incoming and outgoing trajectories from the self - collision can easily self - intersect and for a special choice of parameters even self - collide again . it is possible to show that when the impact parameter @xmath18 gets smaller ( with fixed @xmath17 ) , there exist more and more multi - self - colliding solutions . for a point - like particle , there exist a finite value @xmath138 under which there exist infinite number of possible evolutions for single initial parameters @xmath18 and @xmath17 , cf . we leave further details to the subsequent publication . summarizing , the studied system with a wormhole - based time machine does not suffer from the paradoxical initial conditions . on the contrary , it breaks the deterministic character of the theory and offers more solutions for single initial conditions .

short clean one - dimensional ( 1d ) mesoscopic wires , often referred to as quantum point contacts , show conductance quantization@xcite as a function of the channel width . the quantization is well described by the theory of adiabatic propagation of free electrons.@xcite for non - interacting particles , conductance quantization should occur in longer channels too , as long as there is no backscattering off inhomogeneities within the channel . a lot is known about the role of electron - electron interaction of 1d channels . electron - electron repulsion in a wire enhances dramatically the reflection coefficient , making it energy - dependent.@xcite however , interaction between electrons does not alter the quantization ( in units of @xmath1 ) of an ideal channel conductance in the limit of zero temperature.@xcite what is still an open question is whether there are other manifestations of interactions due to inelastic processes , which influence the transport properties . in the absence of interactions , left- and right - moving particles in a wire are at equilibrium with the reservoirs they originate from . if a bias is applied between the reservoirs , then these equilibria differ from each other , giving rise to a particular form of the non - equilibrium distribution inside the channel . on the other hand , in a long ideal channel and in the presence of interactions one may expect equilibration to occur between the left- and right - movers into a single distribution characterized by an equilibrium with respect to a reference frame moving with some drift velocity . interestingly , in a model with momentum - independent electron velocity for left- and right - movers ( as it is the case in tomonaga - luttinger model ) there is no difference between the two distributions . effects originating from the particle - hole asymmetry , however , may discriminate between the two . thermopower and coulomb drag @xcite are examples of such effects . . this fabrication method has the advantage of producing reflectionless contacts to the leads,@xcite so that the boundary conditions of the distribution function is given by the fermi function of the reservoirs . we define the thermopower as @xmath2 , i.e. the voltage @xmath3 required to counteract a current due to the temperature difference @xmath4 . [ fig : qpc - picture],scaledwidth=40.0% ] at present , little is known about equilibration in a 1d electron system . in higher dimensions the electron - electron interaction provides the most effective relaxation mechanism at low temperatures and therefore we include this relaxation mechanism as the first approach . however , in 1d pair collisions can not change the distribution function for quadratic dispersion , since the momentum and energy conservation@xcite laws result in either zero momentum exchange or an interchange of the two momenta . @xcite in either case the distribution function remains the same . thus , the leading equilibration mechanism is due to three - particle collisions , which we study in this paper.@xcite we investigate here the effects of three - particle collisions in reasonably short wires ( see fig . [ fig : qpc - picture ] ) , where electron - electron scattering can be considered perturbatively . as measurable quantities , we evaluate the temperature dependence of the thermopower and conductance . note that for more than one mode pair collisions become important for certain fillings.@xcite the paper is organized as follows : first we review the non - interacting limit of thermopower and give a qualitative explanation of the effects due to three - particle collisions . then we describe how to include the electron interactions using the boltzmann equation . next we calculate the main ingredient for our perturbation theory , namely the three - particle matrix element and scattering rate using a @xmath5-matrix expansion . we note several interesting properties of this scattering rate . finally , we derive the conductance and thermopower corrections and discuss the deviation from the so - called mott formula . furthermore , some technical details are put in two appendices , and in appendix [ appendix - a - meso - princip ] we show that the number of left and right movers have to change in a scattering event for the current to change . for a wire without interactions the distribution function @xmath6 is determined solely by the electron reservoirs @xmath7 where @xmath8 is the dispersion relation for momentum @xmath9 and spin @xmath10 ( suppressed in the notation ) , and @xmath11^{-1}$ ] is the fermi function with @xmath12 and @xmath13 denoting the chemical potential and temperature of the left / right contact , respectively ( see fig . [ fig : qpc - picture ] ) . the electric current for low temperature @xmath14 and in linear response to the applied bias @xmath3 and temperature difference @xmath15 then follows as ( @xmath16)@xcite @xmath17 from this the well - known leading - order results for conductance @xmath18 and for thermopower @xmath19 for a fully open channel are obtained . here @xmath20 is the fermi temperature . ) and cold right moving electrons ( @xmath21 ) . due to the temperature difference of the initial distributions the scattering process creating left movers dominate compared to the opposite scattering and therefore it gives a positive correction to the thermopower . [ fig : scattering - picture],scaledwidth=45.0% ] one of the main results of this paper is that the three - particle collisions give a _ positive _ contribution to the thermopower , i.e. the current due to a temperature difference is _ increased _ by the three - particle scattering . this can be explained in simple terms . firstly , to change the current the number of left and right moving electrons need to change , since it is the number of electrons going through a mesoscopic structure that determines the current and not their velocity ( see the appendix [ appendix - a - meso - princip ] ) . secondly , we find the dominant scattering process at low temperature to only involve a single electron changing direction . this occurs near the bottom of the band as pictured on fig . [ fig : scattering - picture](a ) . for the initial electronic distribution the left moving electrons have a higher temperature then the right moving ones , which favors scattering into the warmer distribution as seen on fig . [ fig : scattering - picture](b ) . this thus creates more left moving electrons and thereby increases the particle current towards the colder reservoir , i.e. increasing the thermopower . another important point is that the thermopower and conductance corrections are exponential in temperature , i.e. proportional to @xmath22 . this is a direct consequence of the dominant three - particle scattering process requiring an empty state near the bottom of the band . we find the form of the thermopower correction at low temperatures due to the three particle scattering to be given by @xmath23 where @xmath3 is the electron - electron interaction strength and @xmath24 the fermi temperature . this is found perturbatively in the short wire limit . the long - wire limit remains an open question , and we expect that the length dependence of thermopower saturates once @xmath0 exceeds some relaxation length ( which increases for decreasing temperature ) . in contrast , the conductance correction is negative . to understand this , note that the chemical potential of the initial distribution is higher for the right moving electrons then the left moving ones . this favors scattering into the left moving branch ( still with the process shown in fig . [ fig : scattering - picture](a ) ) for non - zero temperature and thereby decreasing the current . the form of the conductance correction is similar to the thermopower correction : @xmath25 to model the current through a short 1d quantum wire including perturbatively the three particle interactions , we use the boltzmann equation @xmath26 , \label{eq : boltzmann - eq}\end{aligned}\ ] ] where @xmath27 is the distribution function at a space point @xmath28 between zero and @xmath0 ( see fig . [ fig : qpc - picture ] ) , @xmath29 is the velocity and @xmath30 $ ] is the three - body electronic collision integral , i.e. no impurity or interface roughness effects are included here . we include the voltage and temperature difference in the boundary conditions of the reflectionless contacts,@xcite i.e. [ eq : boundary - conditions ] @xmath31 and therefore omit the term @xmath32 in the boltzmann equation , allowed in the linear response regime.@xcite a similar method has been used to investigate electron - phonon interactions in short quantum wires,@xcite quantum hall effect in quantum wires @xcite and ballistic coulomb drag.@xcite the three - particle collision integral is assumed to be local in space and is given by @xmath33= & -\!\!\!\!\!\!\sum_{\substack{\sigma_2\sigma_3\\ \sigma_{1^{\prime}}\sigma_{2^{\prime}}\sigma_{3^{\prime } } } } \sum_{\substack{k_2k_3\\k_{1^{\prime}}k_{2^{\prime}}k_{3^{\prime}}}}\!\!\!\ ! w_{123;1^{\prime}2^{\prime}3^{\prime } } \left[f_1f_2f_3(1-f_{1^{\prime}})(1-f_{2^{\prime}})(1-f_{3^{\prime } } ) -f_{1^{\prime}}f_{2^{\prime}}f_{3^{\prime}}(1-f_1)(1-f_2)(1-f_3)\right ] , \label{eq : collision - integral - full}\end{aligned}\ ] ] where the quantum numbers are primed / unprimed after / before the scattering event , @xmath34 and the scattering rate @xmath35 is found in the next section . without interactions ( @xmath36 ) the solution of the boltzmann equation is simply given by @xmath6 in eq . ( [ eq : f-0 ] ) . when interactions are included it becomes a very difficult task to solve the boltzmann equation to all orders in the interaction . however , for a short wire the interactions only have short time to change the distribution function away from the initial distribution @xmath6 and therefore we expand the distribution function in orders of @xmath35 as @xmath37 to find @xmath38 to the first order in @xmath39 , we insert the expansion of @xmath40 in the boltzmann equation and realize that only @xmath6 is necessary in the collision integral . since @xmath41=\mathcal{i}_{k}[f^{(0)}]$ ] is independent of @xmath28 , we find that [ eq : f-1kk ] @xmath42 \quad & \textrm{for}\quad & k>0,\\ f_{k}^{(1)}(x)&=\frac{x - l}{v_{k}}\mathcal{i}_{k}[f^{(0 ) } ] \quad & \textrm{for}\quad & k<0 , \label{eq : f-1}\end{aligned}\ ] ] using the boundary conditions eq . ( [ eq : boundary - conditions ] ) . therefore the current to the first order in @xmath39 is @xmath43\equiv i^{(0)}+i^{\textrm{int } } , \label{eq : i - int - def}\end{aligned}\ ] ] where @xmath44 is the non - interacting ( landauer ) part of the current from eq . ( [ eq : i-0 ] ) and @xmath45 is the part due to interactions . the form of the interacting part of the current is now known and the next step is therefore to evaluate it to linear response to @xmath3 and @xmath4 to obtain the thermopower and conductance corrections . to this end , we define @xmath46 via @xmath47 where @xmath48 is the fermi function with temperature @xmath5 and fermi level @xmath49 . it turns out that @xmath46 is proportional to either @xmath3 or @xmath4 . this is seen by using the identity @xmath50 so we can identify @xmath46 by expanding the non - interacting distribution function @xmath51 ( see eq . ( [ eq : f-0 ] ) and fig . [ fig : qpc - picture ] ) @xmath52ev,\\ f^0_{\textrm{r}}({\varepsilon}_k)&\simeq f^0({\varepsilon}_k)+\left[-{\partial}_{{\varepsilon}}f^0({\varepsilon}_k)\right]({\varepsilon}-{\varepsilon_{\textrm{f}}^{}})\frac{\delta t}{t } , \ ] ] i.e. @xmath53 therefore to get @xmath45 in linear response to @xmath3 and @xmath4 , we linearized the collision integral @xmath54 $ ] eq . ( [ eq : collision - integral - full ] ) with respect to @xmath55 and insert it into @xmath45 eq.([eq : i - int - def ] ) to obtain @xmath56 , \label{eq : i-1-linear - response}\end{aligned}\ ] ] where we defined @xmath57 using the shorthand notation @xmath58 and @xmath59 . to linearize the collision integral and thereby the correction to the current due to interactions @xmath45 , we have used the relation @xmath60 valid at @xmath61 . since @xmath62 is different for positive and negative @xmath63 , we need to divide the summation in @xmath45 eq . ( [ eq : i-1-linear - response ] ) into positive and negative @xmath9 sums , which gives @xmath64 terms . for this purpose , we introduce the notation @xmath65 and similarly for other combinations of the summation intervals . the 32 terms can be simplified to only three terms using energy conservation and symmetry properties of @xmath66 in eq . ( [ eq : delta - def ] ) under interchange of indices . there are pairwise exchanges of indices @xmath67 etc . and interchanges between primed and unprimed indices , @xmath68 , using eq . ( [ eq : h - teorem ] ) and the fact that @xmath35 contains a matrix element squared . this leads to six terms . furthermore , @xmath66 is invariant under @xmath69 for all @xmath70 simultaneously due to time reversal symmetry , also seen explicitly from the form of @xmath35 ( derived below ) . an example of how the simplifications occurs can be seen in eq . ( [ eq : simplification - of - sums ] ) . thus we obtain the more compact result @xmath71\nonumber\\ & + 4(-e)\sum_{\textrm{spin}}\sum_{\substack{--+\\+++}}\delta_{123;1^{\prime}2^{\prime}3^{\prime } } \left[\frac{\delta t}{{k_{\textrm{b}}^{}}t^2 _ { } } \left\{({\varepsilon}_1-{\varepsilon_{\textrm{f}}^{}})+({\varepsilon}_2-{\varepsilon_{\textrm{f}}^{}})\right\}-\frac{2ev}{{k_{\textrm{b}}^{}}t}\right]\nonumber\\ & + 3(-e)\sum_{\textrm{spin}}\sum_{\substack{++-\\+-- } } \delta_{123;1^{\prime}2^{\prime}3^{\prime } } \left[\frac{\delta t}{{k_{\textrm{b}}^{}}t^2 _ { } } \left\{-({\varepsilon}_3-{\varepsilon_{\textrm{f}}^{}})+({\varepsilon}_{2^{\prime}}-{\varepsilon_{\textrm{f}}^{}})+({\varepsilon}_{3^{\prime}}-{\varepsilon_{\textrm{f}}^{}})\right\}-\frac{ev}{{k_{\textrm{b}}^{}}t}\right ] , \label{eq : i-1-linear - simplificeret - exact}\end{aligned}\ ] ] where the definition of @xmath62 in eq . ( [ eq : psi-0-value ] ) was inserted . an important point is that the number of positive / negative wave - vector intervals is not the same before and after the scattering . therefore , we note that only scattering events that change the number of left and right moving electrons contributes to the interaction correction to the current . the origin of this is the cancellation of the velocity in the definition of the current and in the distribution functions . this cancellation thus leads to an expression for the interaction correction to the current in eq . where all the in - going and out - going momenta enter on equal footing . in the appendix [ appendix - a - meso - princip ] , we show that this is valid to all orders in perturbation theory . due to this property and momentum conservation , there are no processes that alter the current possible near the fermi level . consequently , states far away from the fermi level have to be involved in the scattering , which , as we will see , leads to a suppression of @xmath45 by a factor @xmath22 . the distribution function on the other hand , _ can _ be changed by scattering processes near the fermi level . to identify the important processes we find in the next section the scattering rate @xmath35 . the three particle scattering rate @xmath35 is calculated using the generalized fermi golden rule inserting the @xmath5-matrix , @xmath72 , iterated to second order in the interaction @xmath3 to get the three particle interaction amplitude , i.e. , @xmath73 where @xmath74 is the initial energy , @xmath75 the final energy , @xmath76 is the resolvent operator ( or free green function ) , @xmath77 is short hand for @xmath78 , and the subscript @xmath79 means connected in the sense that the scattering process can not be effectively a two particle process , where one of the incoming particles does not participate in the scattering . explicitly @xmath76 and @xmath3 are given by @xmath80 here @xmath81 is the unperturbed hamiltonian ( i.e. kinetic energy with some dispersion ) , @xmath82 the fourier - transformed interaction potential , and @xmath83 ( @xmath84 ) is the annihilation ( creation ) operator . to calculate the matrix element @xmath85 , we write the initial and final states as @xmath86 where @xmath87 is the empty state . using the anti - commutator algebra @xmath88 , we obtain @xmath89 where we introduced @xmath90 ( the last equality is only valid for a quadratic dispersion ) , and where the set of permutations is given by @xmath91 . here the signs of the permutation , @xmath92 , are shown as superscripts . in order to exclude the effectively two - particle processes when multiplying eq . ( [ eq : vg|i > ] ) by @xmath93 from the left , @xmath94 ( @xmath95 ) needs to be different from @xmath96 ( @xmath97 ) . the result is @xmath98 where @xmath99 is the symmetrized interaction . the matrix element consist of 36 terms and the scattering rate thus has @xmath100 terms . to obtain this result we did not use energy conservation . for a quadratic dispersion , the denominator is only zero if we have an effective pair collision or if the momentum transfer is zero , as seen from the expression @xmath101 . a picture of the matrix element is found in fig . [ fig : diagram](b ) , where the exchange processes ( including the sign ) are visualized as different ways to connect two interaction lines and an intermediate propagation ( @xmath76 ) seen on fig . [ fig : diagram](a ) . the inclusion of the fermi statistics makes a substantial difference for the properties of the scattering rate as compared to the case described in ref . [ ] , which is obtained by setting all @xmath102 . we can rewrite the matrix element eq . ( [ eq : matrix - element - vgv - genneral ] ) in a more transparent way in terms of quantum mechanical exchange symmetry . first , we introduce the following combination of three - particle scattering amplitudes @xmath103 , \label{eq : v - def}\end{aligned}\ ] ] and after some rewriting , we then obtain @xmath104.\nonumber\end{aligned}\ ] ] we interpret this result in a way similar to a two - particle matrix element , @xmath105 , \label{eq:2-particle - matrix - element}\end{aligned}\ ] ] which contains a direct ( first term ) and a exchange term ( where @xmath106 ) . in the three - particle case @xmath107 is the direct term and one can make five exchange processes ( instead of one ) by exchanging the three final states @xmath108 , @xmath109 and @xmath110 . this gives eq . ( [ eq : matrix - element - vha - v ] ) . the sign in front of each @xmath111 is determined by the number of exchanges made , e.g. in @xmath112 a single exchange , @xmath113 , gives a minus @xmath114 whereas for @xmath115 two exchanges ( @xmath116 followed by @xmath117 ) gives a positive sign @xmath118 . furthermore , the arguments in @xmath107 are ordered in three pairs such that the differences between the elements in each pair are the only arguments of the interaction potential , see eq . ( [ eq : v - def ] ) . this is useful when constructing approximations having a specific scattering process in mind . how the matrix element was rewritten into the form of eq . ( [ eq : matrix - element - vha - v ] ) can also be described in terms of the drawings of fig . [ fig : diagram ] . the direct term @xmath107 is the sum of the six terms having mirror - symmetric exchanges before and after the scattering . the other terms in eq.([eq : matrix - element - vha - v ] ) then can be obtained by suitable changes of outgoing lines . the expressions we obtain for the three - particle scattering rates eq . ( [ eq : w - fermi - rule ] ) are quite cumbersome . nevertheless , the obtained results allow for some consistency checks . remarkably , for some two - body potentials , scattering of the particles of an @xmath119-body system is exactly equivalent to a sequence of two - body collisions . such `` special '' potentials were studied in the context of integrable quantum many - body problems.@xcite we recall now that for a quadratic band a pair collision does not change the momenta of the incoming particles or simply permutes the two momenta . therefore , three - particle scattering for a the integrable potentials may result _ only _ in permutations within the group of three momenta of the colliding particles ; all other three - particle scattering amplitudes must be zero for such potentials . in the context of this work it means that even three - particle ( or higher - order ) collisions would not bring electron equilibration for such types of electron - electron interaction . in this section we check that the three - particle scattering amplitudes are indeed zero for two special potentials . in the case of contact interaction , @xmath120 , and for any kind of electron dispersion relation ( i.e. , not necessarily quadratic ) , we find by using the energy conservation law that @xmath121 , \label{eq : v - const - e - general}\end{aligned}\ ] ] where @xmath122 . this is a major simplification from @xmath100 to @xmath123 terms by performing the spin summation . if furthermore , the dispersion is quadratic , @xmath124 , then we find the ( at first sight ) surprising cancellation @xmath125 this can be seen directly from eq.([eq : v - const - e - general ] ) or by noting that @xmath126 for a quadratic dispersion and constant interaction for @xmath127 , i.e. each term of eq . ( [ eq : matrix - element - vha - v ] ) is zero . @xmath107 cancels in such a way , that the three first terms of eq . ( [ eq : v - def ] ) cancel each other ( the even permutations of @xmath128 combined with the same primed permutation ) and the three last terms cancel each other ( the odd permutations of @xmath128 combined with the same primed permutation ) . in fact , the described above cancellation is in agreement with the general factorization results for the @xmath129matrix of one - dimensional @xmath119body problem with @xmath130function interaction in real space.@xcite in this context , it is crucial that the particles have quadratic dispersion relation ; if we use e.g. @xmath131 , then the cancellation does not occur . notice also that the cancellation we demonstrate is not a trivial zero . indeed , the underlying two - particle amplitudes eq . ( [ eq:2-particle - matrix - element ] ) are finite for a @xmath132independent potential , if one includes spins . ( for spinless fermions and contact interaction the matrix element would be zero because the direct and the exchange terms cancel in accordance with the pauli principle . ) we checked also that the energy conserving part of the matrix element @xmath85 in the case of _ spinless _ fermions , quadratic dispersion , and the fourier transformed interaction potential of the form @xmath134 becomes equal to zero . this is also possible to expect because of the relation of the potential eq . ( [ liebliniger ] ) to the integrable 1d bosonic lieb - liniger model @xcite . indeed , the bosonic model with contact interaction potential @xmath135 may be exactly mapped @xcite onto the spinless fermionic model with interaction @xmath136 . the integrability of the bosonic model guarantees the integrability of the corresponding fermionic one . adding a contact interaction to @xmath137 does no harm , as we are considering spinless fermions . finally , fourier transformation takes us to eq . ( [ liebliniger ] ) . we checked that including the spin degree of freedom , spoils the remarkable cancellation for a three - particle amplitude . in the following sections , we assume a general case interaction potential for which the three - particle scattering amplitudes lead to a non - trivial re - distribution of the momenta between the particles . in this section , we go though the main ideas and approximations in evaluating the current correction due to interactions @xmath45 eq . ( [ eq : i-1-linear - simplificeret - exact ] ) to lowest order in the temperature , @xmath14 . we give a more detailed calculation in the appendix [ appendix - b - detailed - calcualtion ] . as noted previously , all three terms in @xmath45 eq . ( [ eq : i-1-linear - simplificeret - exact ] ) are exponentially suppressed , since momentum conservation @xmath138 forbids scattering processes near the fermi level for the given combinations of positive and negative @xmath9 intervals . to be more specific , it is the phase space restrictions of the fermi functions that give the exponential suppression , i.e. @xmath139 we begin by identifying the most important three particle scattering process . the three terms in @xmath45 eq . ( [ eq : i-1-linear - simplificeret - exact ] ) are : ( i ) two right movers backscattering a left mover while remaining right movers ; ( ii ) one right mover keeping its direction , while backscattering two left movers , and ( iii ) a left and a right mover keeping their directions , while backscattering the third particle . from now on , we will concentrate on the case of coulomb interaction @xmath140 , which is the largest for small @xmath132 , therefore we want to identify processes where the initial and final state are close in momentum space @xcite . further , the process(es ) should not require more than one electron in states suppressed exponentially by the fermi functions . one can see that due to the constraints stemming from momentum and energy conservation , in fact only process ( iii ) allows both initial and final states to be close to each other in momentum space and at the same time having only a single exponentially suppressed factor . the corresponding scattering process is of the type shown in fig . [ fig : scattering - picture](a ) . therefore to the first order in @xmath22 , we include only the third one in eq . ( [ eq : i-1-linear - simplificeret - exact ] ) . this leads to @xmath141 . \label{eq : i - int - vigtigt}\end{aligned}\ ] ] here @xmath66 expresses the available phase space in form of the fermi functions and the three - particle scattering rate , see eq . ( [ eq : delta - def ] ) . one essential approximation is that for the scattering process depicted in fig . [ fig : scattering - picture](a ) , we may replace the full fermi distribution functions by the exponential tales or the low - temperature limit expressions , i.e. [ f0s ] @xmath142 note that @xmath143 , @xmath144 and @xmath145 are all positive . we see that the product of the fermi functions is indeed exponentially suppressed , i.e. @xmath146 . the second essential approximation is that for the scattering process seen in fig . [ fig : scattering - picture](a ) the initial and final states differ by a small momentum . therefore the matrix element in the transition rate @xmath35 is dominated by the direct term @xmath107 in eq . ( [ eq : matrix - element - vha - v ] ) , since the five exchange terms are suppressed by the coulomb interaction @xmath147 , i.e. @xmath148 the direct term eq . ( [ eq : matrix - element - direct - only ] ) would be zero for @xmath149 . in the case of quadratic dispersion relation and general @xmath150 , it vanishes in the limit @xmath151 and @xmath152 for @xmath153 due to the pauli principle . for a quadratic dispersion and for a general symmetrized interaction @xmath154 , the direct term @xmath107 simplifies to the following expression @xmath155}{(k_1-k_3+q_1)q_1q_3(k_1-k_3-q_3 ) } \label{eq : v - direct - quadratic - sec4}\end{aligned}\ ] ] where we used energy conservation and introduced @xmath156 and @xmath157 . next , we give a qualitative explanation for the power law in @xmath5 for the interacting current correction eq . ( [ eq : i - int - vigtigt ] ) using the quadratic dispersion . first , we consider the phase space constraint . to do the sum over all @xmath9 in eq . ( [ eq : i - int - vigtigt ] ) we use the momentum and energy conservation and introduce new variables @xmath156 and @xmath157 , i.e. , change the summation variables , @xmath158 the energy conservation for a quadratic dispersion gives a factor of @xmath159 ( see e.g. eq . ( [ eq : energy - conservation ] ) ) . for the process at hand , the @xmath143 and @xmath160 are close to the fermi level and each of their sums contribute with a factor of @xmath161 and @xmath162 , respectively . the fermi functions give the exponential suppression and a contribution to the phase space in form of an exponential tail , i.e. @xmath163 see eq . ( [ eq : fermi - function-1 - 1m]-[eq : fermi - function-3 - 3 m ] ) . to get the low temperature result for @xmath45 , eq . ( [ eq : i - int - vigtigt ] ) , we use the method of steepest decent to calculate the integral . to this end , we note that the exponent @xmath164 is a function of @xmath161 and @xmath162 and in the limit @xmath165 the most important part is around the origin @xmath166 . here @xmath164 vanishes as @xmath167 ( see appendix [ appendix - b - detailed - calcualtion ] for details ) . therefore collecting the phase space factors the current correction due to three particle interactions eq . ( [ eq : i - int - vigtigt ] ) becomes @xmath168.\end{aligned}\ ] ] in the limit @xmath14 . furthermore , it turns out that the constraints @xmath169 and @xmath170 in the sum eq . ( [ eq : i - int - vigtigt ] ) only leaves phase space close to @xmath171 for @xmath165 , so we can set @xmath172 in the integrand and do the integral over @xmath162 , which is @xmath173 due to the phase space limits . to lowest order in temperature this yields @xmath174 . \label{eq : i - int - uden - fase - rum}\end{aligned}\ ] ] from this we conclude that phase space alone ( i.e. assuming @xmath175 to be a constant ) gives a temperature dependence of the form @xmath176 \quad \textrm{(phase space only)}.\end{aligned}\ ] ] however , as we have seen the three - particle interaction rate has delicate momentum dependence that needs to be taken into account . therefore , to calculate the direct interaction term @xmath107 we expand the symmetrized potential @xmath140 for small @xmath132 , as @xmath177 , \label{eq : vq - taylor - series}\end{aligned}\ ] ] where the parameter @xmath178 describes the screening due to the metallic gates near the quantum wire and @xmath179 is ( twice ) the @xmath180 fourier transform of the coulomb potential cut - off by the screening . setting @xmath181 into the three - particle scattering rate eq . ( [ eq : v - direct - quadratic - sec4 ] ) , we obtain @xmath182 to lowest order in @xmath132 . inserting this into the eq . ( [ eq : i - int - uden - fase - rum ] ) the final result for the current correction , including both phase space factors and the momentum dependent scattering rate , becomes @xmath183.\end{aligned}\ ] ] ( here we noticed that the non - constant three particle scattering rate gave rise to four extra powers in temperature . ) the detailed calculation given in appendix [ appendix - b - detailed - calcualtion ] yields a prefactor , and the end result is @xmath184 + \mathcal{o}\left[\bigg(\frac{t}{{t_{\textrm{f}}}}\bigg)^8\right ] . \label{eq : i - int - resultat}\end{aligned}\ ] ] combining this result with the zero - order in the interaction terms , see eqs . ( [ g0 ] ) and ( [ s0 ] ) , we find for the thermopower and conductance in the low temperature limit , @xmath185 , \label{eq : s - result}\\ g= & \frac{2e^2}{h } -\frac{2e^2}{h}e^{-{t_{\textrm{f}}}/t}\left[1+\frac{l}{\ell_{\textrm{eee}}}\right ] . \label{eq : g - result}\end{aligned}\ ] ] here we introduced the effective length @xmath186 by the relation@xcite @xmath187 which may be viewed as a mean free path with respect to backscattering for a hole near the bottom of the band . to recapitulate , the temperature dependence @xmath188 in eq . can be understood in the following way : the three - particle scattering of a single particle leaves five free momenta , and since two are taken by energy and momentum conservation this gives @xmath189 . in addition , the interaction , @xmath82 , is proportional to @xmath190 , and when squared it gives rise to four more powers , which results in the @xmath191 dependence . in the limit of a point - like interaction , @xmath192 , the corrections are zero in agreement with the result of section [ subsec : point - like - int ] . it is known from the luttinger liquid theory that in the limit of linear spectrum , which corresponds to @xmath193 , the conductance remains finite even if the wire is infinitely long ( @xmath194 ) . therefore it is tempting to speculate that the two terms in the square brackets of eq . ( [ eq : g - result ] ) are the first terms of an expansion in @xmath195 of some function @xmath196 which saturates at a constant value in the limit @xmath197 . one may also have a similar speculation generalizing eq . ( [ eq : s - result ] ) for the thermopower , @xmath198\to f_s(l/\ell_{\textrm{eee}})$ ] . as a final remark , we note that the so - called mott formula @xcite relating the thermopower to the low - temperature conductance , @xmath199 is clearly violated by eqs . ( [ eq : g - result ] ) and ( [ eq : s - result ] ) . this violation could be expected , because the conventional derivation of the mott formula ( for the non - interacting case ) assumes that the main contribution to the conductance _ and _ thermopower comes from the states around the fermi level , in an energy interval of the order of temperature.@xcite however , in the considered case the main contribution to @xmath129 comes from the `` deep '' states , even in the zeroth order with respect to the interaction potential . correspondingly there is no surprise that eqs . ( [ eq : s - result ] ) and ( [ eq : g - result ] ) being substituted , respectively , in the left- and right - hand sides of eq . ( [ mott ] ) produce a parametrically large mismatch @xmath200 . we have calculated the leading interaction correction to the transport properties of a clean mesoscopic wire adiabatically connected to the leads , using perturbation theory in the length of the wire . for a single - mode wire , the leading interaction corrections turns out to be given by three - particle scattering processes . this is because two - particle processes can not change the current due to momentum and energy conservation . to calculate the effect of the three - body processes , we have utilized the boltzmann equation formalism , with three - particle scattering events defining the collision integral . we have identified the leading - order scattering processes and found that they involve at least one state near the bottom of the band , i.e. far from the fermi level . the involvement of such `` deep '' states results in an exponentially small , @xmath201 , interaction - induced correction to thermopower and conductance at low temperatures . the account for interaction in this paper is performed for relatively short wires , where perturbation theory in the interaction or equivalently in the wire length is valid . for longer wires one needs to find the distribution function by treating the collision integral in the boltzmann equation non - perturbatively . it is not clear whether the relaxation of the distribution function would instead yield non - exponential corrections to the transport coefficients for longer wires . however , since the scattering processes that contribute to the current must involve a particle that changes direction ( which is proven in appendix [ appendix - a - meso - princip ] ) , one might speculate that the exponential suppression is valid for all lengths , as long as electron - electron scattering is the only active relaxation mechanism . the question of what the relaxed distribution function looks like for a mesoscopic wire is an interesting and unsolved problem . here we have only given a partial answer for the leading contributions for a short wire , i.e. to lowest order in the interaction . further studies should involve a self - consistent determination of the distribution function . since thermopower is sensitive to the electron distribution function , it might be a good experimental tool for answering the fundamental questions regarding the effect of electron - electron collisions . indeed refined measurements of thermopower of short 1d quantum wires have been performed yielding a reasonably good agreement with the free - electron theory.@xcite it remains an open question whether the accuracy of thermopower measurements is high enough to see the interaction effects in longer wires . we acknowledge illuminating discussions with m. garst , a. kamenev , m. khodas , t. l. larsen and m. pustilnik . a.m.l . appreciates and enjoyed the hospitality of the william i. fine theoretical physics institute , university of minnesota . this work is supported by nsf grants dmr 02 - 37296 and dmr 04 - 39026 . in this appendix , we show that _ the particle current change due to electronic scattering if and only if the scattering changes the number of left and right moving electrons_. in the main text ( see eq.([eq : i-1-linear - simplificeret - exact ] ) ) , this was shown to first order in the transition rate , but here we show it to _ all orders in the interaction_. we show it explicitly in the boltzmann equation framework , however , suspect it to be a general feature of mesoscopic systems . intuitively , the statement means that it is the _ number _ of particles that passes though the mesoscopic system that matters and not their velocity . in contrast to this is e.g. a long 1d wire or a bulk metal , where a velocity change of the particles is enough to change the current . to show the above statement explicitly , we formally rewrite the boltzmann equation ( [ eq : boltzmann - eq ] ) including the boundary conditions eq.([eq : boundary - conditions ] ) as @xmath202}{v_{k } } \ \ \ \textrm{for}\ k>0 , \label{eq : solution - to - be - genneral - positive - k}\\ f_{k}(x)&=f^0_{\textrm{r}}({\varepsilon}_{k})+\int_l^x\!\!\ ! { \textrm{d}}x^{\prime}\frac{\mathcal{i}_{kx^{\prime}}[f]}{v_{k } } \ \ \ \textrm{for}\ k<0 . \label{eq : solution - to - be - genneral - negative - k}\end{aligned}\ ] ] note that this is not a closed solution of the boltzmann equation , since the distribution function is still contained inside the collision integral . however , this rewriting enables us to find the current without finding the distribution function first , i.e. by inserting eq . ( [ eq : solution - to - be - genneral - positive - k],[eq : solution - to - be - genneral - negative - k ] ) into the current definition @xmath203 and obtain ( after a few manipulations ) : @xmath204\nonumber\\ & -\frac{(-e)}{l } \int_0^l\!\!\ ! { \textrm{d}}x \sum_{\sigma k<0}\mathcal{i}_{kx}[f ] + \frac{(-e)}{l } \int_0^x\!\!\ ! { \textrm{d}}x^{\prime } \overbrace{\sum_{\sigma k}\mathcal{i}_{kx^{\prime}}[f]}^{=0}\nonumber\\ \equiv&\ i^{(0)}+i^{(\textrm{int})},\end{aligned}\ ] ] where the @xmath28-dependent part can be seen to be zero by changing variables . we note the cancellation of the velocity in the distribution function eqs . ( [ eq : solution - to - be - genneral - positive - k ] ) , ( [ eq : solution - to - be - genneral - negative - k ] ) and the current definition eq . ( [ eq : current - def ] ) , which is the origin of the statement we are showing ( as in the first order calculation ) . a similar cancellation occurs in the landauer formula thus relating the transmission to the conductance . by using the explicit form of the collision integral eq . ( [ eq : collision - integral - full ] ) the current from the interactions is @xmath205.\end{aligned}\ ] ] we can divide the summation over @xmath9 quantum number into positive and negative intervals as in the main text ( see section [ sec : linear - response - limit ] ) . the essential point is now , that all terms that have the same number of positive ( and negative ) intervals for the primed and unprimed wave numbers @xmath9 are zero . in other words , if the number of left and right moving electrons does not change then the contribution is zero by symmetry of the transition rate . we show this cancellation in practice by an example ( using the notation of eq.([eq : notation - def-++ ] ) ) : @xmath206\label{eq : simplification - of - sums}\\ = & \sum_{\textrm{spin } } \sum_{\substack{-+-\\-+- } } w_{123;1^{\prime}2^{\prime}3^{\prime } } \left[f_1f_2f_3(1-f_{1^{\prime}})(1-f_{2^{\prime}})(1-f_{3^{\prime } } ) -f_{1^{\prime}}f_{2^{\prime}}f_{3^{\prime}}(1-f_1)(1-f_2)(1-f_3)\right]\nonumber\\ = & \sum_{\textrm{spin } } \sum_{\substack{-+-\\-+- } } w_{123;1^{\prime}2^{\prime}3^{\prime } } f_1f_2f_3(1-f_{1^{\prime}})(1-f_{2^{\prime}})(1-f_{3^{\prime } } ) - \sum_{\textrm{spin } } \sum_{\substack{-+-\\-+- } } \underbrace{w_{123;1^{\prime}2^{\prime}3^{\prime } } f_{1^{\prime}}f_{2^{\prime}}f_{3^{\prime}}(1-f_1)(1-f_2)(1-f_3)}_{\textrm{interchange}\ ( 123)\leftrightarrows(1^{\prime}2^{\prime}3^{\prime})}=0,\nonumber \ ] ] interchanging @xmath108 and @xmath109 at the first equality using @xmath207 and interchanging @xmath208 in the second term as indicated . thereby we have shown to all orders that to change the current by electronic interactions the number of left and right movers have to change . the statement is _ not limited _ to only three particle scattering and can be show equivalently for pair interaction including several bands , electron phonon coupling or any other interaction with the same kind of symmetry under particle interchange . furthermore , the statement is still true if the collision is non - local in space , since that only introduce some spatial integrals in the collision integral that can be handled similarly . note however that the distribution function can be changed by processes that does not change the number of left and right movers . the purpose of this appendix is to calculate @xmath45 in eq . ( [ eq : i - int - vigtigt ] ) @xmath141 . \label{eq : i - int - vigtigt - appendix - b}\end{aligned}\ ] ] in the low - temperature limit , @xmath14 , step by step to find the prefactor given in eq . ( [ eq : i - int - resultat ] ) . as already mentioned , we preform the calculation with the scattering process seen in fig . [ fig : scattering - picture](a ) in mind . therefore we use the fermi functions as given in eq . and the matrix element entering in the scattering rate from eq . ( [ eq : matrix - element - direct - only ] ) and eq . ( [ eq : v - direct - quadratic - sec4 ] ) , i.e. using a quadratic dispersion . we preform the summation over all the @xmath9 in eq . ( [ eq : i - int - vigtigt - appendix - b ] ) in the following way : first of all , we note that due to the momentum and energy conservation in the interaction process described , the scattering of @xmath160 to @xmath209 has to be from above to below the fermi level , i.e. @xmath210 this is due to the signs of @xmath145 and @xmath211 and can be understood as a sign of the difference between the curvature of the dispersion near the bottom of the band and near the fermi level . next , we introduce the momentum transfer around the fermi level @xmath212 for @xmath213 and using the momentum conservation to do the @xmath211 summation , we obtain @xmath214 remembering the constraint @xmath215 , @xmath216 and @xmath217 . the fermi factors @xmath218 and @xmath219 restrict the momentum transfer @xmath161 and @xmath162 to be much smaller then @xmath220 and the @xmath143 and @xmath160 to be near the fermi level for the process in mind . therefore we can use the fermi functions @xmath218 to do the summation over @xmath143 . assuming slow variation of the scattering rate over a range of @xmath221 at the fermi level , the @xmath143 summation becomes @xmath222 similarly the @xmath160 summation is done using the phase space constraint in eq . ( [ eq : fase - rum - restriktion - k3 ] ) @xmath223 we see that since @xmath143 and @xmath160 are restricted to the fermi level , we can insert @xmath224 and @xmath225 in the rest of the integrand . to do the @xmath169 summation , we use the energy conservation contained in the scattering rate . it is rewritten as ( inserting @xmath226 and @xmath227 ) : @xmath228 . \label{eq : energy - conservation}\end{aligned}\ ] ] we have now done the summation over @xmath143 , @xmath145 and @xmath160 and are left with the summation over @xmath161 and @xmath162 of the scattering rate , some fermi functions and the phase factors described above . to this end , we introduce @xmath229 by inserting @xmath226 and @xmath230 in eq . ( [ eq : v - direct - quadratic - sec4 ] ) @xmath231 for a general symmetrized interaction @xmath154 . furthermore , we collect the exponential tales of the fermi functions eq . ( [ eq : fermi - function-2 - 2m]-[eq : fermi - function-3 - 3 m ] ) , @xmath232 defining @xmath233 inserting @xmath234 , @xmath145 from the energy conservation eq . ( [ eq : energy - conservation ] ) and @xmath235 and @xmath230 . therefore we finally get the interacting contribution to the current in eq . ( [ eq : i - int - vigtigt - appendix - b ] ) as : @xmath236.\nonumber\end{aligned}\ ] ] here only the step functions , that restricts the integral are included . next we introduce the dimensionless integration variables @xmath237 for @xmath213 and the dimensionless functions @xmath238 in the integral : @xmath239 , \label{eq : three - particle - contribution - to - current - midway-2-dimless}\end{aligned}\ ] ] where @xmath240 is the integration area shown in fig . [ fig : int - area](a ) . note that this expression is valid for a general interaction @xmath140 and that it is not possible to extract a power law in temperature times some integral by defining new integration variables . for the integral eq . ( [ eq : three - particle - contribution - to - current - midway-2-dimless ] ) to calculate the current due to interactions . the two boundaries for the integration area _ close to the origin _ stemming from the signs of @xmath145 and @xmath211 are indicated . ( b ) the @xmath241 function , a dimensionless version of @xmath242 eq . ( [ eq : xi - def ] ) , important in the calculation using the method of steepest decent . [ fig : int - area],scaledwidth=35.0% ] to proceed , we consider the low temperature limit @xmath243 by using the method of steepest decent : due to the exponential function @xmath244 the maximum of @xmath245 will dominate the integral for @xmath246 , since @xmath247 . the maxima are @xmath248 and @xmath249 and @xmath245 is shown in fig . [ fig : int - area](b ) . for a decreasing interaction the area of @xmath250 and @xmath251 dominates even though the integrand is zero for @xmath252 . therefore we expand around the maximum @xmath253 to get the lowest order result in @xmath254 . in view of the integration region fig . [ fig : int - area](a ) , we use @xmath255 in the integral eq.([eq : three - particle - contribution - to - current - midway-2-dimless ] ) and thereby do the @xmath256 integral using the approximate limits seen in fig . [ fig : int - area](a ) , i.e. @xmath257 to model the symmetrized potential @xmath140 for small @xmath132 , we include the deviation from a constant as described in eq . ( [ eq : vq - taylor - series ] ) . this gives @xmath258 to lowest order in @xmath259 . in the exponential we keep @xmath260 to lowest order in @xmath259 , i.e. @xmath261 so using the lowest order in @xmath259 in the integrand ( leading to lowest order in @xmath5 ) the interacting contribution to the current is : @xmath262\\ \simeq & \frac{8505}{2048\pi^4}e^{-\frac{{t_{\textrm{f}}}}{t } } \frac{e}{{\hslash}}\frac{(v_0{k_{\textrm{f}}^{}})^4}{{\varepsilon}_{\textrm{f}}^3}\nonumber \\ & \times ( l{k_{\textrm{f}}^ { } } ) \bigg(\frac{{k_{\textrm{f}}^{}}}{q_0}\bigg)^4 \bigg(\frac{t}{{t_{\textrm{f}}}}\bigg)^7 \bigg[\frac{\delta t}{t}+\frac{ev}{{\varepsilon_{\textrm{f}}^ { } } } \bigg]\end{aligned}\ ] ] to lowest order in temperature . this is the result stated in the text in eq . ( [ eq : i - int - resultat ] ) . this statement is still true for energy bands like @xmath263 or @xmath264 for @xmath265 . a special case is linear dispersion where two - particle collisions _ can _ change the distribution . however , even for this case the two - particle collisions _ can not _ change the number of left- and right movers and hence not the current , as shown in appendix a. see _ e.g. _ , m.a . cazalilla , physical review a * 67 * , 053606 ( 2003 ) . the interaction potential between fermions is written there in a slightly different form , which however can be reduced ( m. khodas , a. kamenev , l.i . glazman , unpublished ) to the one we present in the text . we assume that electrons in the channel produce some image charges in a distant gate . as the result , the coulomb interaction is cut off at some distance @xmath266 large compared to the fermi wavelength . note here that the sign of the conductance correction is negative even though there is a positive sign in eq.([eq : i - int - resultat ] ) . this is due to the way the bias is defined the left contact having the higher chemical potential . surprisingly , mott formula still yields a reasonably good approximation for thermopower even if the temperature is comparable to the energy scale over which the transmission amplitude changes value , see a. m. lunde and k. flensberg , j. phys . : condens . matter * 17 * , 3879 ( 2005 ) .

woven graph codes can be considered as a generalization of low - density parity - check ( ldpc ) block codes @xcite . their structure as graph codes makes them suitable for iterative decoding . moreover , the ldpc block codes are known as codes with low - complexity decoding and they can be considered as competitors to the turbo codes @xcite which are sometimes called parallel concatenated codes . as mentioned in @xcite , the underlying graph defines a permutation of the information symbols which resembles the interleaving in turbo coding schemes . on the other hand , similarly to the ldpc codes , graph codes usually have minimum distances essentially smaller than those of the best known linear codes of the same parameters . at a first glance , the minimum distance of a graph code does not play an important role in iterative decoding since the error - correcting capability of this suboptimal procedure is often less than that guaranteed by the minimum distance . however , in general , the belief - propagation decoding algorithms work better if the _ girth _ of the underlying graph is large , that is , if the minimum distance of the graph code is large @xcite . in the sequel we distinguish between graph , graph - based , and woven graph codes . we say that a _ graph code _ is a block code whose parity - check matrix coincides with the incidence matrix of the corresponding graph . _ graph - based codes _ constitute a class of concatenated codes with constituent block codes concatenated with a graph code ( see , for example , @xcite ) . each vertex in the underlying graph corresponds to a constituent block code . the main feature of these codes is that the block length of their constituent block codes coincides with the degree of the underlying graph . we introduce _ woven graph codes _ which are , in fact , graph - based codes with constituent block codes whose block length is a multiple of the graph degree @xmath7 , that is , their block length is @xmath8 , where @xmath9 is an integer . in particular , when @xmath9 tends to infinity we obtain convolutional constituent codes . distance properties of bipartite graph - based codes with constituent block codes were studied in @xcite . it was shown that if the minimum distance of the constituent block codes is larger than or equal to @xmath10 , then there exist asymptotically good codes with fixed constituent codes among these graph - based codes . also it was shown in @xcite that for some range of rates , random graph - based codes with block constituent codes satisfy the vg bound when the block length of the constituent codes tends to infinity . one disadvantage of graph - based codes that becomes apparent in the asymptotic analysis is that good performances can only be achieved when the block length of the constituent block codes ( which in this case coincides with the graph degree @xmath7 ) tends to infinity . in practice this leads to rather long graph - based codes with not only rather high decoding complexity of the iterative decoding procedures but also high encoding complexity . in this paper , we consider a class of the generalized graph - based codes which we call woven graph codes with constituent block and convolutional codes . they are based on _ @xmath0-partite , @xmath0-uniform hypergraphs_. notice that graph - based codes with constituent block codes based on hypergraphs were considered in @xcite , @xcite . it is mentioned in @xcite that gallager s ldpc codes are graph codes over hypergraphs . we consider first woven graph codes with constituent @xmath11 block codes . a product - type lower bound on the minimum distance of such codes is derived . in order to analyze their asymptotic performances we modify the approach used in @xcite to @xmath0-partite , @xmath0-uniform hypergraphs and constituent @xmath11 block codes . it is shown that when @xmath9 grows to infinity in the random ensemble of woven graph codes with binary constituent block codes we can find @xmath12 such that there exist codes satisfying the vg lower bound on the minimum distance for any rate . in order to generalize the asymptotic analysis to woven graph codes with constituent convolutional codes we assume that the binary constituent block code is chosen as a zero - tail ( zt ) terminated convolutional code and consider a sequence of zt convolutional codes of increasing block length @xmath9 . it is shown that when the overall constraint length of the woven graph code tends to infinity in the random ensemble of such convolutional codes we can find @xmath13 such that there exist codes satisfying the costello lower bound on the free distance for any rate . we also describe the constituent convolutional codes as block codes over the field of binary laurent series @xcite . this description as well as the notion of _ block _ hamming distance @xcite of convolutional codes is used to derive a product - type lower bound on the free distance of woven graph codes with constituent convolutional codes and to construct examples of such woven codes with rate @xmath1 . for a given hypergraph the free distance of the woven graph code depends on the numbering of code symbols associating to the hypergraph vertices . by a search over all possible permutations of the constituent code we found an example of a rate @xmath1 woven graph code with overall constraint length @xmath14 and free distance @xmath2 . the rate @xmath1 woven graph code is based on heawood s bipartite graph @xcite , @xcite and contains constituent convolutional codes with overall constraint length @xmath5 and free distance @xmath15 . we consider also the encoding problem for graph and woven graph codes . the traditional encoding technique for graph codes has complexity @xmath16 , where @xmath17 is the blocklength . we show by examples that some regular block graph codes are quasi - cyclic and thereby can be interpreted as tailbiting ( tb ) codes ( see , for example , @xcite , @xcite ) . it is known that the encoding complexity of such codes is proportional to the overall constraint length of the parent convolutional code . by using a tb representation for the graph code we can construct an example of an encoder for a woven graph code that is also represented in the form of a tb code but with overall constraint length less than or equal to @xmath18 , where @xmath19 is the number of constituent convolutional codes with overall constraint length @xmath6 each . in section ii , we consider some properties of @xmath0-partite , @xmath0-uniform , @xmath7-regular hypergraphs . we define woven graph codes with constituent block codes as well as with constituent convolutional codes and obtain product - type lower bounds on their minimum and free distances . then , in section iii , we derive a lower bound on the free distance of the random ensemble of woven graph codes . in section iv , examples of woven graph codes are given . we conclude the paper by considering encoding techniques for graph codes and woven graph codes in section v. a _ hypergraph _ is a generalization of a graph in which the edges are subsets of vertices and may connect ( contain ) any number of vertices . these edges are called hyperedges . a hypergraph is called @xmath0-uniform if every hyperedge has cardinality @xmath0 or , in other words , connects @xmath0 vertices . if @xmath20 the hypergraph is simply a graph . the _ degree of a vertex _ in a hypergraph is the number of hyperedges that are connected to ( contain ) it . if all vertices have the same degree we say that this is the _ degree of the hypergraph_. the hypergraph is _ @xmath7-regular _ if every vertex has the same degree @xmath7 . let the set @xmath21 of vertices of an @xmath0-uniform hypergraph be partitioned into @xmath22 disjoint subsets @xmath23 , @xmath24 . a hypergraph is said to be _ @xmath22-partite _ if no edge contains two vertices from the same set @xmath23 , @xmath24 . in the sequel we consider @xmath0-partite , @xmath0-uniform , @xmath7-regular hypergraphs . such a hypergraph is a union of @xmath0 disjoint subsets of vertices . each vertex has no connections in its own set and is connected with @xmath25 vertices in the other subsets . in fig . [ figure : graph4 ] a @xmath10-partite , @xmath10-uniform , @xmath26-regular hypergraph is shown . it contains three sets of vertices . they are shown by triangles , rectangles , and ovals , respectively . there are no edges connecting vertices inside any of these three sets . the vertices are connected by hyperedges each of which connects three vertices . a _ cycle _ of length @xmath27 in the hypergraph is an alternating sequence of @xmath28 vertices and @xmath27 hyperedges where all vertices are distinct except the initial and the final vertex , which coincide , and all edges are distinct . the _ girth _ of a hypergraph is the length of its shortest cycle . in fig . [ figure : hypergirth ] we show a subgraph that contains the shortest cycle of the @xmath10-partite , @xmath10-uniform , @xmath26-regular hypergraph in fig . [ figure : graph4 ] . it consists of the vertices @xmath29 , @xmath30 , and @xmath29 and has girth equal to 2 . we introduce the notion of a _ compact @xmath31-connected subgraph _ in the hypergraph . it is a connected subgraph in which each vertex is incident with at least @xmath32 hyperedges . we call the length ( number of hyperedges ) of the shortest compact subgraph its _ ( @xmath0,@xmath32)-girth_. in fig . [ figure : hypergirth ] the hyperedges belonging to the shortest @xmath33-compact subgraph are marked by circles . it is easy to see that ( @xmath10,@xmath34)-girth is @xmath35 . a @xmath34-partite , @xmath34-uniform hypergraph is a bipartite graph . for such a hypergraph the ( @xmath34,@xmath34)-girth is equal to the girth and a compact subgraph is a cycle . heawood s bipartite graph @xcite , @xcite with @xmath36 vertices and @xmath37 edges is shown in fig . [ figure : graphh ] . this graph contains a set of @xmath3 black and a set of @xmath3 white vertices . each vertex has no connections within its own set and is connected with @xmath38 vertices from the other set . the girth of the heawood graph is @xmath35 . in order to illustrate the structure of a binary graph - based block code with constituent block codes we represent the heawood bipartite graph using a so - called tanner graph @xcite as shown in fig . [ figure : graph ] . we introduce a set of @xmath39 ( variable ) vertices which correspond to the code symbols . each of the @xmath40 ( constraint ) vertices on the right- and left - hand sides corresponds to one of 14 parity checks . the @xmath38 edges leaving one constraint vertex correspond to a codeword of the constituent @xmath41 block code of rate @xmath42 . the parity - check matrix of the corresponding graph - based code with binary constituent block codes is @xmath43 where the parity - check matrix @xmath44 of size @xmath45 has the form @xmath46 where @xmath47 is a size @xmath48 parity - check matrix of the constituent block code , and @xmath49 is a size @xmath45 parity - check matrix which is the permutation of the columns of @xmath44 determined by the graph . notice that in general by choosing @xmath50 and assigning constituent block codes of different rates @xmath42 to the same graph we can obtain graph - based codes of different rates . in general , since in an @xmath0-partite , @xmath0-uniform , @xmath7-regular hypergraph the total number of parity checks is equal to @xmath51 , the code rate @xmath52 of the graph - based code is @xmath53 with equality if and only if all parity - checks are linearly independent . if @xmath20 , then we get @xmath54 . the simplest example of a heawood graph - based code can be obtained by choosing as constituent block codes a single - parity - check code of rate @xmath55 . then the parity - check matrix @xmath47 has the form @xmath56 and the parity - check matrix of the graph - based code is @xmath57 in this case the graph - based code coincides with the graph code since ( [ incdmatr ] ) is the incidence matrix of the heawood graph . in @xcite it is proved that the minimum distance of the bipartite graph - based code with single - parity - check constituent codes is @xmath58 , where @xmath59 is the girth of the corresponding graph . notice that for the tanner graph we have @xmath60 . the parity - check matrix ( [ incdmatr ] ) is a @xmath61 parity - check matrix . taking into account that one check is linearly dependent on the other , we obtain a @xmath62 binary block code . its minimum distance is @xmath63 . consider the hypergraph shown in fig . [ figure : graph4 ] . its incidence matrix has the form @xmath64 and is a @xmath65 parity - check matrix of a hypergraph - based code which coincides with the parity - check matrix of the hypergraph code . each column represents a hyperedge and each row represents a vertex of this hypergraph . for example , the first four rows represent the vertices 1 , 2 , 3 , and 4 ( triangles ) , the next four rows he vertices 5 , 6 , 7 , and 8 ( rectangles ) , and the last four rows the vertices 9 , 10 , 11 , and 12 ( ovals ) . the first column represents the hyperedge which connects the vertices 1 , 5 , and 9 , the second column the hyperedge connecting vertices 1 , 6 , and 10 _ etc_. the rows of ( [ hyper ] ) are linearly dependent . by removing two parity checks we obtain a @xmath66 linear block code with the minimum distance @xmath67 , where @xmath68 is the ( @xmath10,@xmath34)-girth of the hypergraph . the rate of this hypergraph code is @xmath69 , which satisfies inequality ( [ rate ] ) , @xmath70 the tanner version of this hypergraph is shown in fig . [ figure : tannerhypergraph ] . for an @xmath0-partite , @xmath0-uniform , @xmath7-regular hypergraph - based code with constituent block codes we have the following theorem . [ graph - based ] the minimum distance of a hypergraph - based code based on an @xmath0-partite , @xmath0-uniform , @xmath7-regular hypergraph with ( @xmath0,@xmath71)-girth @xmath72 and containing constituent block codes with minimum distance @xmath73 is @xmath74 . any nonzero codeword in an @xmath0-partite , @xmath0-uniform , @xmath7-regular hypergraph - based code always corresponds to a connected @xmath75-subgraph or a set of disjoint connected subgraphs . these subgraphs are called _ active _ @xcite , @xcite . all hyperedges and vertices in an active subgraph are also called _ the number of hyperedges in the shortest connected subgraph is equal to @xmath72 . any nonzero symbol in a codeword corresponds to an active hyperedge in the graph . by using the arguments given above , we conclude that for any codeword @xmath76 , @xmath77 where @xmath78 is the hamming weight of @xmath76 . minimizing over @xmath76 completes the proof . now assume that the constituent code assigned to the hypergraph vertices is a binary @xmath11 linear block code determined by a parity - check matrix @xmath79 where @xmath80 is a size @xmath81 matrix , @xmath82 is the set of all possible binary matrices of size @xmath81 . let @xmath83 denote such a binary @xmath11 constituent block code determined by the matrix ( [ nonbin ] ) . we call the corresponding hypergraph - based code with @xmath83 as constituent codes a _ woven graph _ code with constituent block codes . consider an example of a woven graph code based on the bipartite graph with girth @xmath84 shown in fig . [ figure : bgraph ] . the tanner version of this the so - called `` utility '' bipartite graph is shown in fig . [ figure : tgraph ] . the incidence matrix of this graph is @xmath85 we use a constituent ( @xmath86 ) linear block code with @xmath87 determined by the parity - check matrix @xmath88 by searching over all possible permutations of the matrices @xmath89 , @xmath90 , and @xmath91 we found the following parity - check matrix of the woven graph code with the best minimum distance @xmath92 the matrix ( [ wovbl ] ) describes a @xmath93 linear block code with @xmath94 . any codeword @xmath95 of the @xmath11 constituent block code can be represented as a sequence of @xmath7 blocks of length @xmath9 , that is , @xmath96 , where @xmath97 , @xmath98 . we define the _ minimum hamming block distance _ between the codewords @xmath95 and @xmath99 of the constituent block code as @xmath100 where @xmath101 , @xmath98 . next we will prove the following theorem . [ block ] the minimum distance of woven graph codes based on @xmath0-partite , @xmath0-uniform , @xmath7-regular hypergraphs with ( @xmath0,@xmath102)-girth @xmath103 and containing constituent block codes with minimum distance @xmath71 and minimum block distance @xmath104 can be lower - bounded by @xmath105 _ proof_. any nonzero codeword corresponds to an active connected subgraph or a set of disjoint connected subgraphs and the number of hyperedges in the shortest subgraph is @xmath103 . any nonzero symbol in a codeword activates a hyperedge in the graph , that is , not less than @xmath0 constituent subcodes correspond to a codeword . since at most @xmath7 hyperedges are connected with any hypergraph vertex then the number of active constituent subcodes can be lower - bounded by @xmath106 taking into account that any codeword of block weight greater than or equal to @xmath102 in the constituent block code has a weight at least equal to @xmath71 we obtain the following inequality @xmath107 for any codeword @xmath76 and the proof is complete . from theorem [ block ] for the woven graph code determined by ( [ wovbl ] ) we obtain that @xmath108 woven graph codes with constituent convolutional codes can be considered as a straightforward generalization of the woven graph code with constituent block codes . assume that the @xmath83 code is chosen as a zero - tail terminated ( zt ) convolutional code and consider a sequence of zt convolutional codes with increasing @xmath9 . it is evident that when @xmath9 tends to infinity the @xmath11 constituent code @xmath83 can be chosen as a rate @xmath42 binary convolutional code with constraint length @xmath6 . then the corresponding woven graph code has rate @xmath109 and its constraint length is at most @xmath110 . another description of woven graph codes with constituent convolutional codes follows from the representation of the constituent convolutional code in polynomial form . let @xmath111 be a minimal encoding matrix @xcite of a rate @xmath42 , memory @xmath112 convolutional code , given in polynomial form , that is , @xmath113 where @xmath114 , @xmath115 , @xmath116 , are binary polynomials such that @xmath117 . the overall constraint length is @xmath118 . the binary information sequence @xmath119 is encoded as @xmath120 where @xmath121 is a binary code sequence . let @xmath122 denote a parity - check matrix for the same code , @xmath123 where @xmath124 is the redundancy of the constituent code . we denote by @xmath125 the field of binary laurent series and regard a rate @xmath42 constituent convolutional code as a rate @xmath42 block code @xmath126 over the field of binary laurent series encoded by @xmath111 . then its codewords @xmath127 are elements of @xmath128 , which is the @xmath7-dimensional vector space over the field of binary laurent series @xcite . the minimum hamming _ block _ distance between the codewords @xmath129 and @xmath130 is defined @xcite as @xmath131 where @xmath132 is the hamming ( block ) weight of @xmath133 . representing a convolutional code as a block code over the field of binary laurent series we can obtain a woven graph code with constituent convolutional codes as a generalization of a graph - based code with binary constituent block codes . for example , a parity - check matrix @xmath134 of the rate @xmath135 heawood s graph - based code with @xmath4 constituent convolutional codes has the form @xmath136 where @xmath137 and @xmath138 are short - hand for @xmath139 and @xmath140 , respectively , and @xmath141 is a parity - check matrix of the rate @xmath4 constituent convolutional code and @xmath142 is one of six possible permutations of @xmath143 . exploiting the above definitions we can interpret this bipartite woven graph - based code with constituent convolutional codes as follows . the left column of vertices in fig . [ figure : graph ] represents @xmath19 parity checks each of which determines one of @xmath19 constituent fixed and identical convolutional codes and their @xmath144 branches represent the elements @xmath145 , @xmath146 even , @xmath147 , @xmath148 . similarly , the right column of vertices represents same convolutional codes and their @xmath144 branches represent the elements @xmath149 , @xmath146 odd , @xmath150 , @xmath148 , where the set @xmath151 is a random permutation of the set @xmath152 determined by the graph . we can also regard the @xmath19 left constituent convolutional codes as a warp with @xmath144 threads . each of the @xmath19 right constituent convolutional codes are tacked on @xmath7 of the threads in the warp such that each thread of the warp is tacked on exactly once . thus , our construction is a special case of a woven code @xcite and we call this graph - based code a _ woven graph _ code . [ wovbound ] the free distance of a woven graph code based on an @xmath0-partite , @xmath0-uniform , @xmath7-regular hypergraph with the ( @xmath0,@xmath102)-girth @xmath103 and containing constituent convolutional codes with free distance @xmath153 and minimum block distance @xmath154 can be lower - bounded by @xmath155 . since woven graph codes with constituent convolutional codes can be considered as a generalization of woven graph codes with constituent block codes , the theorem follows from theorem [ block ] when @xmath9 tends to infinity . for a woven graph code based on a bipartite graph with girth @xmath59 and containing constituent convolutional codes with minimum block distance @xmath156 and free distance @xmath153 by a straightforward generalization of the approach of @xcite we obtain the following tighter bound on the free distance @xmath157 we will show that the ensemble of random woven graph codes based on random @xmath0-partite , @xmath0-uniform , @xmath7-regular hypergraphs with a fixed degree @xmath7 and with a fixed number of vertices @xmath19 in each subgraph contains asymptotically good codes . in order to prove this we will modify the approach in @xcite . first we consider the ensemble of random woven graph codes with rate @xmath42 constituent block codes determined by the edges of a random @xmath0-partite , @xmath0-uniform , @xmath7-regular hypergraph corresponding to the time - varying random parity - check matrix @xmath158 where @xmath159 , @xmath160 , is a block matrix of size @xmath161 ( or a binary matrix of size @xmath162 ) and @xmath163 denotes a random permutation of the columns of @xmath164 , @xmath165 where @xmath166 , @xmath167 , denotes the random parity - check matrix ( [ nonbin ] ) which determines the @xmath11 constituent block code and @xmath19 is the number of constituent codes in each subgraph . _ remark _ : in @xcite a more restricted ensemble of random codes is studied in which all matrices are identical random matrices . in the proof of theorem 1 we need that the syndrome components are independent random variables in the product probability space of random matrices and random permutations . the following simple example shows that this is not always the case if all matrices are identical . consider @xmath168 constituent block codes of block length @xmath169 with @xmath170 information symbols . this example is rather artificial since the rate of the constituent block code @xmath171 and therefore the rate of the graph - based code with @xmath20 is @xmath172 . in this case the parity - check matrix of the code has the form @xmath173 where @xmath174 is a random permutation of @xmath7 elements . first assume that all matrices are identical , that is , @xmath175 . there are only 8 equiprobable elements in the product space , namely , @xmath176 for any vector @xmath177 of weight 1 we have the following set of random equiprobable syndromes : @xmath178 therefore , @xmath179 if @xmath180 and @xmath181 are both random and independent this probability is equal to 1/4 . although this remark contradicts the proof of theorem 3 in @xcite , there exists another ( combinatorial ) way to prove the same statement for identical @xmath182 @xcite . next we prove the following theorem . [ vgtheorem ] ( varshamov - gilbert lower bound ) for any @xmath183 , some @xmath184 , some integer @xmath185 and for all @xmath186 in the random ensemble of length @xmath187 woven graph codes with @xmath11 binary block constituent codes of rate @xmath42 there exist codes of rate @xmath188 such that their relative minimum distance @xmath189 satisfies the inequalities @xmath190 where @xmath191 is a root of the equation @xmath192 and @xmath193 is the solution of @xmath194 , and @xmath195 denotes the binary entropy function . _ let @xmath196 be the hamming weight of the codeword @xmath76 of the random binary woven graph code @xmath197 . we are going to find a parameter @xmath32 such that the probability @xmath198 tends to 0 for all @xmath199 . we can rewrite @xmath200 as @xmath201 where @xmath202 and @xmath203 denotes the number of nonzero constituent codewords in the @xmath146th subgraph corresponding to the codeword of weight @xmath196 . in the ensemble of random parity - check matrices @xmath204 , @xmath205 , of size @xmath206 the probability that a nonzero vector @xmath207 is a codeword of the corresponding constituent random binary code @xmath83 is equal to @xmath208 since the syndromes of the constituent codes are equiprobable sequences of length @xmath209 . taking into account that in the @xmath146th subgraph we have @xmath203 nonzero constituent codewords the probability @xmath210 can be upper - bounded by @xmath211 in order to estimate the probability @xmath212 we prove the following lemma . for the ensemble of binary woven graph codes with constituent block codes described in theorem [ vgbound ] , the probability @xmath212 that a codeword of weight @xmath196 contains @xmath202 nonzero constituent codewords in the @xmath0 subgraphs can be upper - bounded by @xmath213 [ lemma ] _ proof . _ taking into account that in the @xmath146th subgraph the number of nonzero component codewords is equal to @xmath203 and that the subgraphs are random and independent we can rewrite the probability @xmath214 as @xmath215 the probability @xmath216 can be upper - bounded as @xmath217 where @xmath218 . the cardinality of @xmath219 can be upper - bounded as @xmath220 @xmath221 where the sum is upper - bounded by the maximal term times the number of terms @xmath222 . notice that in the above derivations we ignored the fact that @xmath223 can be noninteger since we consider the asymptotic behaviour of ( [ main ] ) . it follows from lemma 1 that @xmath224 @xmath225 @xmath226 @xmath227 consider the asymptotic behaviour of ( [ main ] ) when @xmath228 tends to infinity . introduce the notations @xmath229 and @xmath230 and the function @xmath231 after simple derivations we obtain @xmath232 } ( 1-s)h(\delta)-(1-r_{\rm wg})\gamma+s\gamma h\left(\frac{\delta}{\gamma}\right ) \right\ } \label{asympt}\ ] ] where @xmath188 is the rate of binary woven graph code . maximizing ( [ asympt ] ) over @xmath233 gives @xmath234 inserting @xmath235 and @xmath236 into ( [ asympt ] ) we obtain @xmath237 which coincides with ( 9 ) and ( 10 ) in @xcite for @xmath20 , that is , if the graph is bipartite . for any @xmath238 and @xmath239 from @xmath240 , it follows that there exist codes of rate @xmath238 with relative minimum distance @xmath241 . let @xmath191 denote the solution of the equation @xmath242 for @xmath243 and let @xmath244 be the solution of @xmath194 . solving ( [ fun ] ) for @xmath235 and @xmath236 we obtain that there exist woven graph codes of rate @xmath238 with the relative minimum distance @xmath245 satisfying the inequalities : @xmath246 in fig . [ figure : rd ] the lower bound ( [ vgbound ] ) on the relative minimum distance for the ensemble of binary woven graph codes with block constituent codes as a function of the code rate is shown . it is easy to see that when @xmath0 grows the ensemble of binary woven graph codes contains codes meeting the vg bound for almost all rates @xmath247 . [ figure : rd2 ] demonstrates the gap @xmath248 between the vg bound and the code rate as a function of the relative minimum distance @xmath245 for different values of @xmath0 . it follows from fig . [ figure : rd2 ] that for @xmath249 the difference in code rate compared to the vg bound is negligible . consider a zt convolutional woven graph code with constituent zt convolutional codes of rate @xmath42 . the length of a zt woven graph codeword in @xmath144-tuples is equal to @xmath250 where @xmath9 is the number of @xmath144-tuples influenced by information symbols and @xmath251 is the memory of the woven graph code of rate @xmath188 . denote by @xmath252 the free distance of the corresponding woven graph code . now we can prove the following ( costello lower bound ) for any @xmath183 , some @xmath253 , some integer @xmath13 , and for all @xmath254 in the random ensemble of rate @xmath188 woven graph codes over @xmath0-partite , @xmath0-uniform , @xmath7-regular hypergraphs with constituent convolutional codes of rate @xmath42 there exists a code with memory @xmath251 such that its relative free distance @xmath255 satisfies the costello lower bound @xcite , @xmath256 _ proof _ : analogously to the derivations in the proof of theorem [ vgtheorem ] let @xmath202 where @xmath203 denotes the number of nonzero constituent codewords in the @xmath146th subgraph corresponding to the codeword of weight @xmath196 , @xmath257 . in order to evaluate the number of nonzero constituent codewords among the @xmath19 constituent codewords , notice that the set of such codewords is a union of sets of nonzero constituent codewords belonging to each of the @xmath0 subgraphs . the cardinality of the union is at least @xmath258 . therefore the all - zero `` tail '' required to force the encoder into the zero state has length at least @xmath259 . the total number of redundant symbols consists of two parts : the number @xmath260 of parity - check symbols for the nonzero constituent codewords in the @xmath0 subgraphs and at least @xmath259 redundant symbols required for zero - tail terminating of the woven graph code . thus , formula ( [ redunds ] ) can be rewritten as @xmath261 the statement of the lemma [ lemma ] is changed in a following way @xmath262 instead of ( [ mainf ] ) we now have @xmath263 by introducing the notations @xmath264 we obtain from ( [ redunds_conv])([mainf2 ] ) that @xmath265 } \left\ { ( 1-s)(1+\mu)h\left(\frac{\delta}{1+\mu}\right)-\gamma(1+\mu-\mu r_{\rm wg})+\gamma ( 1+\mu ) sh\left(\frac{\delta}{\gamma(1+\mu)}\right ) \right\}. \label{mainfconv}\end{aligned}\ ] ] maximizing ( [ mainfconv ] ) over @xmath233 , we obtain @xmath266 where @xmath267 if @xmath0 is large enough , then @xmath236 . it follows from ( [ mainfconv ] ) that @xmath268 maximization of @xmath269 over @xmath270 gives @xmath271 where @xmath272 we can find a bound on @xmath273 by solving @xmath274 . thus , we can conclude that for any @xmath275 we can find a woven graph code such that ( [ costellobound ] ) holds we start with considering a graph code determined by the parity - check matrix ( [ incdmatr ] ) . as mentioned before , the matrix ( [ incdmatr ] ) can be considered as a @xmath276 parity - check matrix . since the parity checks defined by the graph are linearly dependent ( the sum of the rows of ( [ incdmatr ] ) is equal to zero ) it turned out that by ignoring one parity check we obtain a parity - check matrix of a @xmath62 linear block code . for simplicity we consider the rate @xmath277 code that is obtained by ignoring the eighth information symbol which yields a @xmath278 subcode of this code . it is easy to see that renumbering the graph vertices by adding to each vertex number some fixed number modulo the total number of vertices preserves both the incidence and adjacency matrices of the graph . for example , in fig . [ figure : graphh ] , by adding 2 modulo 14 we will get exactly the same graph . when we have a similar property for linear codes we call such codes quasi - cyclic codes and these block codes can be described as tailbiting ( tb ) convolutional codes . renumbering the vertices corresponds to permuting the rows of ( [ incdmatr ] ) . by row permutations , ( [ incdmatr ] ) can be reduced to the form @xmath279 it follows from ( [ modifmatr ] ) that the graph shown in fig . [ figure : graphh ] corresponds to a @xmath278 tb code with a parent convolutional code determined by the parity - check matrix @xmath280 it means that this code is `` tail - bitten '' at the 7th level of the trellis diagram . a corresponding polynomial generator matrix of the parent convolutional code has the form @xmath281 the minimum distance of the @xmath278 tb code is equal to the graph girth , that is , @xmath63 . notice that many regular bipartite graphs look very similar to the heawood graph in the sense that by manipulating the incidence ( parity - check ) matrices and truncating lengths we can obtain infinite families of graphs . some properties of these graphs can be easily predicted from the properties of the corresponding parent convolutional codes . consider the parity - check matrix ( [ woven ] ) of the woven graph code based on the heawood bipartite graph with constituent convolutional codes of rate @xmath4 . this woven graph code has the rate @xmath282 . let the rate @xmath4 constituent convolutional code of memory @xmath283 and overall constraint length @xmath5 with @xmath15 be given by the generator matrix @xmath284 a corresponding parity - check matrix @xmath122 is @xmath285 notice that the constituent code @xmath286 considered as a block code over @xmath287 represents a @xmath288 block code with the minimum distance @xmath289 . by using the product - type lower bound ( [ bipartbound ] ) we obtain @xmath290 on the other hand , it was verified by computer search that any codeword of the woven graph code determined by ( [ woven ] ) consists of at least three nonzero codewords of the component code @xmath126 described by ( [ 90 ] ) . moreover , it was found by computer search that each of these nonzero codewords of @xmath126 has the minimum block weight @xmath291 . note that the codewords of the block code over @xmath287 with block weight @xmath291 corresponds to the codewords of the convolutional code belonging to its subcodes of rate @xmath292 . these three subcodes have generator matrices @xmath293 @xmath294 @xmath295 where @xmath296 , @xmath297 , and @xmath298 . the minimum free distance over all these subcodes of rate @xmath292 is equal to 8 . taking into account that all other codewords of the woven graph code contain at least four nonzero codewords of @xmath299 of block weight @xmath300 we obtain an improved lower bound on the free distance of the woven graph code as @xmath301 . in order to obtain an upper bound on the free distance of the woven graph code we consider the parity - check matrix ( [ woven ] ) in more detail . it also describes a quasi - cyclic code and can by row permutations be reduced to a parity - check matrix of a two - dimensional code , a tb ( block ) code in one dimension and a convolutional code in the other , @xmath302 a parity - check matrix of the parent convolutional code for the tb code ( [ modwoven ] ) given in symbolic form is @xmath303 where @xmath304 and @xmath305 are formal variables . the matrix ( [ 97 ] ) can be considered as a parity - check matrix of a two - dimensional convolutional code . the variable @xmath304 corresponds to the parent convolutional code of the heawood graph code ( [ convmatr ] ) , the variable @xmath305 is used for the constituent convolutional code ( [ 90 ] ) . a generator matrix of the two - dimensional convolutional code with the parity - check matrix ( [ 97 ] ) has the form @xmath306 where @xmath307 , @xmath308 , @xmath309 , @xmath310 , @xmath311 , and @xmath312 . the generator matrix ( [ 97gen ] ) tail - bitten over variable @xmath304 at length @xmath37 yields the generator matrix @xmath313 of the code ( [ modwoven ] ) , @xmath314 where @xmath315 is short - hand for @xmath316 . notice that any of the six permutations of the columns @xmath139 , @xmath317 generates a woven graph code . the permutation @xmath318 , @xmath319 , and @xmath320 describes the woven graph code with the largest free distance . the overall constraint length of this generator matrix is equal to 70 but the matrix is not in minimal form . a minimal - basic generator matrix @xcite has the overall constraint length equal to 64 and differs from ( [ tb13 ] ) by one row which can replace any of the rows of @xmath313 and has the form @xmath321 where @xmath322 where @xmath323 and @xmath324 . the matrix ( [ tb13 ] ) is a generator matrix of a convolutional code of rate @xmath325 . by applying the beast algorithm @xcite to the minimal - basic generator matrices corresponding to the different permutations of the columns @xmath139 , @xmath317 we obtained the free distance and a few spectrum coefficients of the corresponding woven graph codes . the parameters of the best obtained woven graph codes are presented in table 1 . generally speaking , encoding of graph - based block codes has complexity @xmath16 , where @xmath17 is the blocklength . this technique implies that we find a generator matrix corresponding to the given parity - check matrix and then multiply the information sequence by the obtained generator matrix . however , we showed by examples that some regular graph codes as well as woven graph codes are quasi - cyclic codes and thereby they can be interpreted as tb codes . for this class of codes the complexity of the encoding is proportional to the constraint length of the parent convolutional code . in this section we are going to illustrate by an example an encoder of a woven graph code with constituent convolutional codes having encoding complexity proportional to the overall constraint length of the corresponding woven graph code @xmath326 . consider again the woven graph code in our example . it is based on the heawood graph and uses constituent convolutional codes of rate @xmath4 and overall constraint length @xmath5 . taking into account the representation ( [ tb13 ] ) of the woven graph code as a rate @xmath325 two - dimensional code , a tb ( block ) code in one dimension and a convolutional code in the other , we can draw its encoder as shown in fig . [ figure : encoder ] . the input symbols @xmath327 enter the encoder once per each cycle of duration seven time instants . at each time moment the contents of each register is rewritten into the next ( modulo seven ) register and the three output symbols @xmath328 are generated . in other words , each of the registers corresponding to the constituent code can be considered as an enlarged delay element of the encoder of the `` tb - dimension '' code determined by the graph . the sequence @xmath327 determines a transition between the states of this encoder . after a cycle of seven time instants we return to the starting state of the enlarged encoder and a tb - codeword ( or a word from one of its cosets ) of length 21 has been generated . then the following seven input symbols @xmath329 enter and after seven time instants another word of length 21 has been generated , etc . the asymptotic behavior of the woven graph codes with block as well as with convolutional constituent codes has been studied . it was shown that in the random ensemble of such codes based on @xmath0-partite , @xmath0-uniform , @xmath7-regular hypergraphs we can find a value @xmath13 such that for any code rate there exist codes meeting the vg and the costello lower bound on the minimum distance and free distance , respectively . product - type lower bounds on the minimum distance of graph - based and woven graph codes have been derived . example of a rate @xmath1 woven graph code with free distance above the product bound is presented . it is shown , by an example , that woven graph codes can be encoded with a complexity proportional to the constraint length of the constituent convolutional code . this work was supported in part by the royal swedish academy of sciences in cooperation with the russian academy of sciences and in part by the swedish research council under grant 621 - 2007 - 6281 . r. g. gallager , _ low - density parity - check codes _ , cambridge , ma : mit press , 1963 . c. berrou , a. glavieux , and p. thitimajshima , `` near shannon limit error - correcting coding and decoding : turbo codes , '' _ in proc . ( icc ) _ , geneva , switzerland , pp . 10641070 , 1993 . a. barg and g. zemor , `` distance properties of expander codes , '' _ ieee trans . inf . theory _ 52 , no.1 , pp . 7890 , jan . 2006 . hu and m. f. fossorier `` on the computation of the minimum distance of low - density parity - check codes , '' _ int . conference on communications ( icc04 ) _ , paris , june 2004 . g. schmidt , v. v. zyablov , and m. bossert , `` on expander codes based on hypergraphs , '' _ isit 2003 _ , yokohama , japan , june 29july 4 , 2003 . s. hoory and y. bilu , `` the codes from hypergraphs , '' _ european journal of combinatorics _ vol . 25 , issue 3 , pp . 339354 , april , 2004 . a. barg , private communication , may 2007 . r. johannesson and k. sh . zigangirov , _ fundamentals of convolutional coding _ , ieee press , piscataway , nj , 1999 . m. bossert , c. medina , and v. sidorenko , `` encoding and distance estimation of product convolutional codes , '' _ in proc . information theory ( isit05 ) _ , adelaide , australia , sept . 4 - 9 , pp . 10631067 , 2005 . j. h. ma and j. k. wolf , `` on tail - biting convolutional codes , '' _ ieee trans . com-34 , pp . 104111 , 1986 . r. m. tanner , `` minimum - distance bounds by graph analysis , '' _ ieee trans . inf . theory _ , 2 , pp . 808820 , feb . r. m. tanner , `` a recursive approach to low complexity codes , '' _ ieee trans . inf . theory _ , 5 , pp . 533547 , sep . 1981 . l. bassalygo , private communication , may 2007 . s. h@xmath330st , r. johannesson , and v. v. zyablov , `` woven convolutional codes i : encoder properties , '' _ ieee trans . inf . theory _ 1 , pp . 149161 , jan . r. m. tanner , d. sridhara , a. sridharan , t. e. fuja , and d. j. costello , jr . , `` ldpc block and convolutional codes based on circulant matrices , '' _ ieee trans . inf . theory _ , 29662984 , dec . 2004 . i. e. bocharova , m. handlery , r. johannesson , and b.d . kudryashov , `` a beast for prowling in trees , '' _ ieee trans . inf . theory _ , 50 , no . 6 , pp . 12951302 , june 2004 .

molecular junctions are electronic devices that consist of an organic quantum dot that is contacted by two macroscopic leads . modern nanotechnology allows for the reliable fabrication of systems where the dots are single aromatic rings , molecular wires@xcite , c@xmath0 fullerenes@xcite or carbon nanotubes.@xcite they are a promising candidates in the search for further miniaturization of electronic and thermoelectric devices.@xcite transport through such systems is determined by the discrete levels of the dot , whose position relative to the fermi energy can be tuned , e.g. , with the help of a third ( gate ) electrode . @xcite the level broadening depends on the dot - lead coupling strength , which can be manipulated via the lead distance or through the choice of different metal - molecule anchoring groups . @xcite in addition to these two external parameters , electron - phonon ( ep ) interaction influences transport through molecular junctions : when it is occupied by charge carriers , the molecule may undergo structural changes or vibrations that correspond to the excitation of local optical phonons of considerable energy . they show up as vibrational signatures in the current - voltage characteristics of the device @xcite . a temperature difference between the leads induces a current of charge carriers across the junction . this thermoelectric effect is measured by recording , for constant temperature difference , the voltage bias necessary to cancel this current . @xcite the quotient of the temperature difference and the thermovoltage , the so - called thermopower , can be used to probe the systems fermi energy@xcite and the vibronic structure of the molecule s state.@xcite most of the experimental findings are well understood within a linear response formulation of the thermopower.@xcite however , the applied temperature differences can be tens of degrees kelvin , i.e. larger than the dot - lead coupling energy . some features of the measured voltage histograms , such as side peaks and temperature dependent widths,@xcite are not accounted for in the linear theory . that is why recently , the discussion of the thermopower has been extended to the nonlinear regime.@xcite then the question arises , how the thermoelectric coefficients can be generalized . our approach in the present work is motivated by the experimental situation : for a given temperature bias we determine the thermovoltage numerically by minimizing the thermally induced charge current . we base our calculations on the anderson - holstein model . here , the organic molecule is represented by a single energy level and a local optical mode , which is linearly coupled to the electron on this level . the quantum dot is connected to two macroscopic leads , while the local mode is coupled to a phonon bath . the current between the leads is given by the interacting dot density of states.@xcite based on such models , different methods have been applied , such as the numerical@xcite and functional renormalization group@xcite , rate equations,@xcite master equations@xcite and nonequilibrium keldysh green functions@xcite to describe transport through the dot for small - to - large dot - lead coupling and weak - to - strong ep interaction . in the antiadiabatic , strong ep coupling regime , typically a lang - firsov transformation@xcite is applied , based on the exact solution of the isolated dot . it predicts the formation of a local polaron , which reduces the effective dot - lead coupling exponentially . this could be beneficial for the thermoelectric response of the system . @xcite for practical applications , however , a moderate level broadening is needed to ensure usable power output . moreover , long electron residence times and strong ep interaction may lead to the accumulation of energy at the dot and , consequently , to its degradation.@xcite because of this , the regime of comparable electronic and phononic time scales and intermediate ep interaction becomes interesting . to account for the polaronic character of the dot - state away from the strong ep coupling , antiadiabatic limit , we use an approach that is based on a variational form of the lang - firsov transformation.@xcite for the polaron problem , the variational lang - firsov approach has been proven to give reliable results in the whole electron - phonon coupling and phonon frequency regime , even in the most physically difficult polaron crossover region.@xcite here , the polaron variational parameter is determined by minimizing the relevant thermodynamical potential . in previous work , @xcite we considered the steady - state current response of the quantum dot to a finite voltage difference between the leads . in doing so , we assumed that the temperatures of the leads and of the phonon bath are all equal . the calculations in these papers were based on the kadanoff - baym formalism , @xcite which relies on the relation between the real - time response functions and the non - equilibrium green functions of the complex time variables . the present work will focus on the thermoelectric effects induced by a finite temperature difference between the leads . section [ secmodel ] introduces the model and the variational ansatz we employ to describe a vibrating quantum dot that is coupled to a phonon heat bath and two macroscopic leads at different temperatures . because for such a setup the temperature is not constant throughout the system , the reasonings of our earlier approach @xcite have to be modified . that is why in sec . [ secresponse ] we generalize the kadanoff - baym method such that the steady - state equation of the response functions , as well as their formal solution deduced in ref . , are applicable to the present case . in sec . [ secself ] we then derive an approximation for the polaronic self - energy , and sec . [ seccurrent ] provides the relation between the polaronic and electronic spectral functions , the latter of which enters the current formula . in sec . [ secvar ] we derive the thermodynamic potential which will be used to determine the variational parameter numerically . section [ secnumerics ] presents our numerical results . the main conclusions and future prospects can be found in sec . [ secsummary ] . is coupled to two macroscopic leads , each in its own thermal equilibrium with different temperatures @xmath1 and chemical potentials @xmath2 . the dot electrons interact with an optical phonon mode of energy @xmath3 that is coupled to a phonon bath at temperature @xmath4 . ] we consider a molecular quantum dot in the three - terminal configuration depicted in fig . [ fig1 ] , which will be modelled by the following hamiltonian : @xmath5 here , the quantum dot is represented by a single energy level @xmath6 with the fermionic operators @xmath7 . it interacts with a local optical phonon mode @xmath8 of energy @xmath3 , where the ep coupling strength is given by the so - called polaron binding energy , @xmath9 by the last term in eq . ( [ equhamiltonianstart ] ) , the dot is coupled to left ( @xmath10 ) and right ( @xmath11 ) leads , each of which contains @xmath12 free electrons with the energies @xmath13 and the corresponding fermionic operators @xmath14 , respectively . in equilibrium , the dot - lead system is characterized by a common chemical potential @xmath15 and a global temperature @xmath16 . in the nonequilibrium situation , a voltage difference between the leads is described by adding to ( [ equhamiltonian ] ) the term @xmath17 with the voltage bias @xmath18 where @xmath19 is the electron charge . in addition , we consider a temperature difference between the macroscopic leads , whereby each lead is supposed to stay in its own thermal equilibrium with the temperatures @xmath20 and @xmath21 , respectively . we also suppose that the local oscillator is coupled to its own heat bath , which has the temperature @xmath22 . in accordance with entin - wohlman @xcite we assume that the coupling to this heat bath ( indicated as @xmath23 in fig . [ fig1 ] ) far exceeds the ep - coupling strength , so that the phonon population at the dot is given by the bose - einstein distribution @xmath24 in the following calculations , we suppose that @xmath25 and @xmath26 in general . moreover , all energies will be taken with respect to the equilibrium chemical potential , i.e. , @xmath27 . we raise the voltage bias symmetrically around @xmath15 , i.e. , @xmath28 and @xmath29 . to account for the polaron localization at the dot , we apply to the hamiltonian ( [ equhamiltonianstart ] ) an incomplete lang - firsov transformation @xmath30 , @xcite introducing the variational parameter @xmath31 $ ] : @xmath32 the transformed hamiltonian then reads @xmath33 where the renormalization of the dot level and the interaction coefficients depends on the parameter @xmath34 : @xmath35 now @xmath36 and @xmath37 are the operators of dressed electrons ( in analogy to polarons ) and the shifted local oscillator . the original electron and oscillator operators read @xmath38 and @xmath39 , respectively . in our previous works , @xcite we considered the electron current response of a molecular quantum dot that was initially in equilibrium with the leads as well as with the phononic bath at a common , fixed temperature . our calculations were based on the kadanoff - baym theory . @xcite at first , we summarize the aspects of this non - equilibrium response theory that are essential for the present setting . the relevant response functions are represented by the real - time green functions of the dot - operators @xmath7 and the lead operators @xmath14 , e.g. @xmath40 the functions @xmath41 of the `` mixed '' operators @xmath14 and @xmath7 are defined in an analogous way . in eqs.([equdefresponse])([equdefresponsegtr ] ) the time dependence of @xmath42 is determined by @xmath43 . the symbol @xmath44 means the standard time - ordering operator so that the function @xmath45 is equal to the functions @xmath46 and @xmath47 for @xmath48 and @xmath49 , respectively . the statistical average @xmath50 corresponds to the equilibrium state at the temperature @xmath16 before the disturbance was turned on . going to the interaction representation , the heisenberg operators @xmath51 are expressed as @xmath52 the real - time response functions may be deduced using the equations of motion for the nonequilibrium green functions of the complex time variables @xmath53 , @xmath54 $ ] , defined as @xmath55 the time dependence of the operators in eq . ( [ equdefg ] ) is determined by @xmath56 only , while the external disturbance is explicit in the time - ordered exponential operator @xmath57 . the operation @xmath58 orders the operators by the imaginary parts of the times @xmath59 and @xmath60 , so that @xmath61 for @xmath62 and @xmath63 for @xmath64 . to find the relation between the functions @xmath65 and @xmath66 , the function @xmath67 is considered for @xmath64 : @xmath68 where @xmath69 the continuation of @xmath70 and @xmath71 to analytic functions of the time variables leads to the identification of @xmath70 with @xmath71 in the limit @xmath72 consequently , the connection of the analytic functions @xmath73 and @xmath74 is given by @xmath75 with similar relations for the functions @xmath76 and @xmath77 . it is evident that the derivation of eq . ( [ equlimitlessgtr ] ) outlined above does not refer to some special properties of the statistical ensemble , nor to the physical meaning of @xmath78 . in this way , it is possible to generalize the definition ( [ equdefg ] ) for @xmath67 assuming the mean value @xmath79 to be unspecified and the complex - time variable to be defined in the interval @xmath80 $ ] , where the time - ordering parameter @xmath81 has no specific physical meaning . we assume that before the disturbance ( [ equhint ] ) was turned on , the system was in a steady state with the temperatures @xmath20 , @xmath21 of the left and right leads , and @xmath4 of the phonon bath . the function @xmath67 defined in this way does not have the properties of the temperature ( @xmath78 ) dependent green function in eq . ( [ equdefg ] ) , it rather represents a functional of the ordered operators which we use to determine the real - time response funcitons @xmath45 . to do this , we define , in analogy to the self energy , the function @xmath82 by the equation @xmath83\bullet & g_{dd } ( \bar t , t_2;u , t_0)\nonumber \\ & = \delta(t_1-t_2)\;,\label{equdyson}\end{aligned}\ ] ] with the inverse zeroth - order function @xmath84 in eq . ( [ equdyson ] ) the matrix multiplication `` @xmath85 '' is defined by the integration @xmath86 containing the time - ordering parameter @xmath81 . the @xmath87-function of complex arguments is understood with respect to this integration . the inverse function to @xmath67 is given as @xmath88 . the deduction of the steady - state equations for the real - time functions @xmath89 and @xmath90 by the limiting procedure @xmath72 is analogous to that given in ref . . defining the fourier transformations according to kadanoff - baym,@xcite @xmath91 and similarly for @xmath92 ) , the solution of the steady - state equations may be written as follows : @xmath93 with the non - equilibrium polaronic spectral function @xmath94 according to the preceeding section , the concrete form of the steady - state solution for the special choice of interactions is determined by the functions @xmath95 . to find an explicit expression for @xmath95 , we start with the equations of motion for @xmath67 and @xmath96 , which are given by the commutators of the operators @xmath7 and @xmath14 with @xmath56 . as a purely formal device , we add to @xmath97 in eq . ( [ equhint ] ) the interaction with fictitious external fields @xmath98 . the equations of motion of @xmath67 and @xmath96 are then expressed by means of the functional derivatives of @xmath99 with respect to these fields . the resulting equations for @xmath95 are solved iteratively . we then let @xmath100 and perform the limit @xmath101 . in the following calculations we will use the self - energy function after the first iteration step:@xcite @xmath102 \big \}\ ; , \label{equsigma1lessgtr}\end{aligned}\ ] ] where we have defined @xmath103 the bose - function @xmath104 in ( [ equsigma1lessgtr ] ) contains the phonon - bath temperature @xmath4 according to eq . ( [ equbosebath ] ) , while the zeroth - order functions of the leads depend on the different temperatures @xmath21 and @xmath20 : @xmath105\;,\end{aligned}\ ] ] with the lead fermi - functions @xmath106 the fourier - transformation of eq . ( [ equsigma1lessgtr ] ) leads to @xmath107 \big \ } \ ; , \nonumber\end{aligned}\ ] ] with the renormalized dot - lead coupling function @xmath108 which depends on the lead density of states : @xmath109 from eq . ( [ equsigmalessfourier1 ] ) , @xmath110 results by interchanging @xmath111 and @xmath112 . note that eq . ( [ equsigmalessfourier1 ] ) is similar to our result for the self - energy in ref . , but now @xmath113 is replaced by the individual lead fermi - functions @xmath114 that were defined in ( [ equfermileads ] ) . if we insert the approximation ( [ equsigmalessfourier1 ] ) into eq . ( [ equdefgamma ] ) , we find @xmath115 \gamma_{a}^{(1)}&(\omega;u)= i_{0}(\kappa ) \widetilde\gamma^{(0)}_a(\omega)\label{equgamma1a}\\ & + \sum_{s\ge 1 } i_{s}(\kappa)2\sinh(s\theta)\nonumber\\ & \times \big [ \widetilde\gamma^{(0)}_a(\omega - s\omega_0)\big(n_b(s\omega_0)+1-f_a(\omega+u_a - s\omega_0)\big ) \nonumber\\ & + \widetilde\gamma^{(0)}_a(\omega+s\omega_0)\big ( n_b(s\omega_0)+f_a(\omega+u_a+s\omega_0 ) \big)\big ] \;.\nonumber\end{aligned}\ ] ] from the functions @xmath116 and @xmath117 , the first order spectral function @xmath118 , the distribution function @xmath119 and the response functions @xmath120 follow according to eqs . ( [ equafrac ] ) , ( [ equbarf ] ) and ( [ equgless ] ) , respectively . the operator of the particle current from lead @xmath121 to the dot reads @xmath122\;.\label{equcurrentop}\end{aligned}\ ] ] its mean value @xmath123 is given by the real - time response functions @xmath124 of the electron operators , which are defined in analogy to eqs . ( [ equdefresponseless ] ) and ( [ equdefresponsegtr ] ) , e.g. @xmath125 we determine @xmath124 based on the equation of motion of the corresponding function @xmath126 of complex - time variables : @xmath127 the current @xmath128 results as @xmath129 where @xmath130 and @xmath131 are the electronic dot spectral function and response function . in the steady state regime our approximation conserves the particle current : @xmath132 . in the following , identical leads are assumed and we work in the wide band limit where @xmath133 . then the total particle current through the dot , @xmath134 , is given by @xmath135\;.\label{equcurrent } \end{aligned}\ ] ] based on eq . ( [ equcurrent ] ) , we determine the electron current through the quantum dot numerically . we note that in the present work we are only interested in the current response to finite voltage biases and temperature differences between the leads . the generation of an electronic response merely by the interaction with a hot phonon bath would require different , energy - dependent densities of states in the leads , as was shown in ref . . in our calculations however , we consider identical lead densities of states . that is why @xmath136 for all @xmath4 , as long as @xmath137 and @xmath138 . to find a relation between the polaronic and electronic functions , we decouple the fermionic and bosonic degrees of freedom in the electronic response function as is customary,@xcite @xmath139 as stated in sec . [ secmodel ] the local oscillator is supposed to be strongly coupled to the heat bath . accordingly we neglect the influence of the ep interaction on the dynamics of the phonon subsystem and evaluate the bosonic correlation function in ( [ equdefgelec ] ) using eq . ( [ equbosebath ] ) . the first order electronic response functions then read @xmath140g_{dd}^{(1)\lessgtr}(\omega\pm s\omega_0;u ) \nonumber\\ & + n_b(s\omega_0)g_{dd}^{(1)\lessgtr}(\omega\mp s\omega_0;u ) \big ) \big\}\;.\nonumber\end{aligned}\ ] ] the corresponding electronic spectral function @xmath141 follows according to the steady - state equation ( [ equdefa ] ) as @xmath142 & \quad\mathrm{e}^{-\widetilde g^2\coth\theta}\big\ { i_{0}(\kappa)a^{(1)}(\omega;u ) + \sum_{s\ge 1 } i_{s}(\kappa)2\sinh(s\theta ) \nonumber \\ & \quad\times\big ( [ n_b(s\omega_0)+\bar f^{(1)}(\omega+s\omega_0;u)]a^{(1)}(\omega+ s\omega_0;u ) \nonumber \\ & \quad+ [ n_b(s\omega_0)+1-\bar f^{(1)}(\omega - s\omega_0;u)]a^{(1)}(\omega- s\omega_0;u)\big ) \big\}\;.\nonumber\end{aligned}\ ] ] to determine the optimal variational parameter @xmath143 , we have to minimize the relevant thermodynamic potential . this poses a problem , since for finite @xmath144 , the effective temperature determining the statistics of the dot electron is not known . it will be given by the constitution of the steady state due to the coupling of the dot with the surroudings . in the present paper , we suppose @xmath34 to be mainly determined by the ep interaction terms contained in eq . ( [ equhamiltonianstart ] ) , since the variational parameter was introduced to characterize the polaron - like quasilocalization of the dot electron . therefore we assume the equilibrium thermodynamic potential of the system , before the temperature- and voltage differences were turned on , to be a reasonable first approximation for the variational function . the use of the well - known `` @xmath145 trick '' to take the interaction terms from @xmath56 into account results in@xcite @xmath146 the function @xmath147 represents the first order equilibrium spectral function if the interaction coefficients in eq . ( [ equhamiltonian ] ) are multiplied by the factor @xmath145 . if we write @xmath147 in terms of @xmath148 , we can carry out the @xmath145-integration and find @xmath149 via @xmath150 and @xmath151 , the thermodynamic potential @xmath152 is a function of @xmath34 . for given parameters @xmath153 , @xmath154 , @xmath3 , @xmath6 and @xmath16 , the optimal variational parameter @xmath143 will be determined in equilibrium , i.e. , for @xmath155 and @xmath138 , by minimizing @xmath152 . we then keep @xmath143 fixed and calculate the self energy , the spectral function and the resulting particle current for finite voltages or temperature differences between the leads . depending on the type of molecular junction , the energies of the characteristic vibrational modes range from the order of @xmath156mev in small molecules@xcite to several @xmath157mev in @xmath158 molecules@xcite . in the following calculations , the corresponding model parameter @xmath3 will be used as the unit of energy , i.e. , we keep @xmath159 fixed and set @xmath160 , @xmath161 and @xmath162 . we assume identical leads and work in the wide band approximation , so that the dot - lead coupling is determined by a single parameter : @xmath133 . furthermore , we consider low equilibrium temperatures , @xmath163 . then , according to eq . ( [ equtildegamma ] ) , the renormalized dot - lead coupling parameter is approximately given as @xmath164 usually , the ratio of the bare dot - lead coupling parameter @xmath154 to the phonon energy @xmath3 is used to determine whether the system is in the adiabatic ( @xmath165 ) or anitadiabatic ( @xmath166 ) regime . recently , eidelstein _ et al._@xcite argued that for strong ep interaction , i.e. , @xmath167 , the antiadiabatic regime can be extended to @xmath168 as long the exponential renormalization of the dot - lead coupling parameter in ( [ equtildegamma2 ] ) is so strong that @xmath169 . in this case , the `` mobility '' of passing charge carriers is reduced so far that the local oscillator is still fast enough to adjust to them individually . because then the physics is essentially those of the antiadiabatic regime , it can be called the `` extended antiadiabatic regime''.@xcite only when @xmath170 approaches @xmath3 the system crosses over to the adiabatic regime . to investigate the latter situation , in the present work we consider comparable electronic and phononic timescales @xmath171 and moderate ep coupling @xmath172 . since in our approach , the polaronic renormalization in eq . ( [ equtildegamma2 ] ) also depends on the value of the variational parameter @xmath31 $ ] , we can interpolate between the extended adiabatic regime and the aforementioned crossover regime . , @xmath173 , and @xmath174 . ( a ) electronic equilibrium spectral function for @xmath175 , @xmath176 , @xmath177 , @xmath178 and @xmath179 . the variational calculation ( black lines ) is compared to the small polaron limit @xmath180 ( red dashed lines ) and to the result for @xmath181 ( blue dot - dashed lines ) . in the latter two cases we set @xmath6 in such a way that @xmath150 is the same as in the variational calculation . ( b ) same as ( a ) , but for @xmath182 , @xmath183 , @xmath184 , @xmath185 and @xmath186 . ( c ) and ( d ) for the same parameters as in ( a ) and ( b ) , thermally induced current @xmath187 as a function of the temperature difference @xmath144 between the leads . , title="fig : " ] + , @xmath173 , and @xmath174 . ( a ) electronic equilibrium spectral function for @xmath175 , @xmath176 , @xmath177 , @xmath178 and @xmath179 . the variational calculation ( black lines ) is compared to the small polaron limit @xmath180 ( red dashed lines ) and to the result for @xmath181 ( blue dot - dashed lines ) . in the latter two cases we set @xmath6 in such a way that @xmath150 is the same as in the variational calculation . ( b ) same as ( a ) , but for @xmath182 , @xmath183 , @xmath184 , @xmath185 and @xmath186 . ( c ) and ( d ) for the same parameters as in ( a ) and ( b ) , thermally induced current @xmath187 as a function of the temperature difference @xmath144 between the leads . , title="fig : " ] to understand the thermoelectric response of the dot we first investigate the dots electronic spectral function in equilibrium . thereby we keep the parameters @xmath159 , @xmath173 , @xmath138 , @xmath155 , and @xmath174 fixed and consider two parameter sets for @xmath154 and @xmath6 . the first set is @xmath175 and @xmath176 , which means that the electronic and phononic subsystems react on a similar timescale and the dot level acts as a tunneling barrier between the leads . for these parameters our variational calculation yields an optimal @xmath177 . as a consequence , the effective dot level given in eq . ( [ equdefeta ] ) is lowered to near resonance , @xmath178 . moreover , the effective dot - lead coupling is reduced by about half , @xmath188 , and we are in the aforementioned crossover regime.@xcite for the second parameter set we , reduced the dot - lead coupling to @xmath182 ( meaning the phononic subsystem is the faster one ) and raised the bare dot level to @xmath183 . this results in @xmath184 , @xmath186 , and @xmath185 , so that the renormalized dot level is still far from resonance . the electronic spectral functions for the first and second parameter set are presented as the black curves in figs . [ fig2](a ) and [ fig2](b ) , respectively . we compare them to two limiting cases : the red dashed curves represent the small polaron limit , where we set @xmath180 instead of the optimal @xmath143 . then @xmath30 in eq . ( [ equlft ] ) corresponds to the complete lang - firsov transformation . in this case , we have @xmath189 and @xmath190 , respectively , and the system is described as being in the extended antiadiabatic regime . the blue dot - dashed curves follow from setting @xmath181 , which represents a rigid quantum dot without ep interaction . in both limiting cases we set @xmath6 in such a way , that @xmath150 is the same as in the respective variational calculation . in general , for @xmath181 , the electronic spectral function features a single broad band centered at @xmath191 . from setting @xmath192 in eqs . ( [ equgamma1 ] ) and ( [ equafrac ] ) , we see that it has a lorentzian shape and its width is given by the bare dot - lead coupling parameter @xmath154 . for finite ep interaction and @xmath180 , the polaronic character of the dot state is signalled by the appearance of several narrow side bands , which are given by the terms with @xmath193 in eq . ( [ equelectronic ] ) . these side bands represent the emission and absorption of optical phonons by the incident electrons . their width is determined mainly by the small parameter @xmath170 and their maxima are located multiple integers of @xmath3 away from the central ( zero - phonon ) peak . note however , that the phonon peak corresponding to @xmath194 is suppressed in figs . [ fig2](a ) and [ fig2](b ) , and only a small shoulder at @xmath195 remains . this is a consequence of pauli blocking and can be understood from eqs . ( [ equsigmalessfourier1 ] ) and ( [ equgamma1a ] ) ; for small @xmath144 , we have @xmath196 and the fermi functions of both leads are nearly step - like . then the third and fourth lines in eqs . ( [ equsigmalessfourier1 ] ) and ( [ equgamma1a ] ) do not contribute in the region @xmath197 $ ] . from eq . ( [ equbarf ] ) we see that @xmath198 in this region and , consequently , the third and fourth line in eq . ( [ equelectronic ] ) do not contribute to the spectral function . this `` floating '' of the phonon bands has been discussed for linear electric transport through molecular quantum dots.@xcite as we will see , it is also the reason why the low temperature thermoelectric response is determined by the shape and position of the zero - phonon peak alone . the variational dot state , due to the moderate renormalization , features broad , overlapping side bands . still , we find a considerable shift of spectral weight to higher energies and a suppression of the zero - phonon peak . the resulting reduction of the low - energy tunneling rate is a main consequence of the polaron formation and is known as the franck - condon blockade.@xcite in the following we calculate the particle current through the junction that results from a finite temperature difference @xmath144 between the leads . let us again consider the two sets of parameters used in sec . [ secstate ] . keeping the respective variational parameters @xmath143 fixed we increase the temperature difference @xmath144 , whereby @xmath199 and @xmath25 . this induces a net particle current through the junction , which is depicted as a function of @xmath144 in figs . [ fig2](c ) and [ fig2](d ) . for the first parameter set , we find that the ep interaction enhances the thermally induced current for small temperature differences . however , it reduces the current throughout the temperature range for the second parameter set . this can be understood from eq . ( [ equcurrent ] ) : the net current through the dot depends on the electronic nonequilibrium spectral function and the difference of the fermi distribution functions of the left and right lead . for zero voltage bias and small temperature differences , the term @xmath200 in eq . ( [ equcurrent ] ) differs from zero only in the narrow region @xmath201 $ ] and changes sign at @xmath202 . physically , this means that the temperature difference between the leads induces a flow of hot charge carriers from the left to the right lead , since @xmath200 is positive for @xmath203 . this current , however , is compensated by a counterflow of cold carriers at @xmath204 , where @xmath200 is negative . since @xmath205 in our calculations , dot states above the fermi energy have more spectral weight than the states below the fermi energy . the net particle current is positive , i.e. , it goes from the left to the right lead . the magnitude of the current is determined by the relative weight of the two flows , i.e. , by the slope of the dot spectral function around the fermi energy . that is why the effect of the polaron formation on the thermally induced current crucially depends on the specific values of @xmath150 and @xmath170 . for the first parameter set , the dot level @xmath178 lies near the fermi surface . in this situation , the polaronic renormalization of the effective dot - lead coupling increases the slope of @xmath206 at @xmath202 , as can be seen in fig . [ fig2](a ) . consequently , in fig . [ fig2](c ) the low temperature thermoelectric current grows with respect to the noninteracting case . for the second parameter set ( corresponding to the non - resonant situation ) the effect is reversed : the slope of @xmath206 decreases and the thermoelectric response of the vibrating molecule is smaller than that of the rigid quantum dot . for both parameter sets we find that , as @xmath207 , the maximum thermocurrent is largest in the case of zero ep interaction . this can also be understood from eq . ( [ equcurrent ] ) . for large @xmath144 , the region @xmath201 $ ] grows and @xmath200 is finite far from the fermi surface . now the thermoelectric current is given not by the slope of @xmath206 at @xmath202 but by the ratio of the integrated spectral weight below and above the fermi surface . because @xmath4 is low , the phonon bands in @xmath208 are weighted according to a poisson distribution with the parameter @xmath209 . as @xmath34 grows , a considerable portion of the total spectral weight is shifted to the phonon bands at @xmath210 . even for large temperature differences , their contribution to the thermocurrent will be exponentially small , and the maximum current decreases with respect to the rigid quantum dot . ( b ) and ( d ) , i.e. , @xmath159 , @xmath173 , @xmath174 , @xmath182 , @xmath183 , @xmath184 , @xmath185 and @xmath186 . ( a ) electronic spectral functions for fixed @xmath211 and @xmath138 . we compare the variational calculation ( black line ) to the small polaron limit @xmath180 ( red dashed line ) and to the result for @xmath181 ( blue dot - dashed line ) . ( b ) current @xmath187 as a function of the voltage difference @xmath212 between the leads for fixed @xmath211 . ( c ) thermovoltage @xmath213 as a function of @xmath144.,title="fig : " ] + ( b ) and ( d ) , i.e. , @xmath159 , @xmath173 , @xmath174 , @xmath182 , @xmath183 , @xmath184 , @xmath185 and @xmath186 . ( a ) electronic spectral functions for fixed @xmath211 and @xmath138 . we compare the variational calculation ( black line ) to the small polaron limit @xmath180 ( red dashed line ) and to the result for @xmath181 ( blue dot - dashed line ) . ( b ) current @xmath187 as a function of the voltage difference @xmath212 between the leads for fixed @xmath211 . ( c ) thermovoltage @xmath213 as a function of @xmath144.,title="fig : " ] + ( b ) and ( d ) , i.e. , @xmath159 , @xmath173 , @xmath174 , @xmath182 , @xmath183 , @xmath184 , @xmath185 and @xmath186 . ( a ) electronic spectral functions for fixed @xmath211 and @xmath138 . we compare the variational calculation ( black line ) to the small polaron limit @xmath180 ( red dashed line ) and to the result for @xmath181 ( blue dot - dashed line ) . ( b ) current @xmath187 as a function of the voltage difference @xmath212 between the leads for fixed @xmath211 . ( c ) thermovoltage @xmath213 as a function of @xmath144.,title="fig : " ] we have seen how , in general , a temperature difference @xmath144 between the leads will induce a particle current through the dot . in a typical experiment , this thermoelectric effect is measured by applying a voltage difference @xmath212 in such a way that the thermally induced current is compensated : @xmath214 for finite @xmath144 and @xmath212 . the determination of the so - called thermovoltage , @xmath213 , and its dependence on the ep coupling will be the subject of the following numerical calculations . thereby , we use the second parameter set from sec . [ secstate ] , i.e. @xmath159 , @xmath173 , @xmath174 , @xmath183 and @xmath182 , resulting in @xmath184 , @xmath186 and @xmath185 . to understand the mechanism , we first consider a finite but fixed temperature difference @xmath211 . for @xmath138 , the corresponding electronic spectral function is shown as the black curve in fig . [ fig3](a ) , where it is again compared to the noninteracting case and the small polaron limit . we calculate the total current @xmath187 as a function of the voltage bias @xmath212 , whereby @xmath215 and @xmath216 . the results are depicted in fig . [ fig3](b ) . for negative voltages , the current signal is in qualitative agreement with the results of previous works.@xcite in a nutshell , the formation of a polaron like state at the dot reduces the zero - bias conductance due to the franck - condon blockade , but for growing voltage we find steps in the current signal whenever @xmath217 , with @xmath218 $ ] . here the chemical potential of the right lead crosses the phonon side bands in @xmath208 and resonant transport of electrons takes place via the emission and subsequent absorption of an equal number of phonons . the right lead is the cold one and has a step - like fermi function . because of this , the width of the current steps is mainly determined by the width of the phonon bands in the spectral function , i.e. , by @xmath170 . in the variational calculation the bands overlap considerably ( see fig . [ fig3](a ) ) and the resonant steps in the current - voltage signal are smeared out . for positive voltage bias we find no current steps even in the small polaron limit . now it is the chemical potential of the hot left lead that crosses the renormalized dot level and the phonon bands . the width of the resonant tunneling steps in @xmath219 does not depend on @xmath170 alone but also on the width of the soft fermi - surface . since @xmath211 is of the order of @xmath3 , the current steps are smeared out and therefore no longer discernible . based on the current - voltage signal we now determine the thermovoltage @xmath213 as a function of @xmath144 numerically and show the result in fig . [ fig3](c ) . we find that for small temperature differences , the absolute value of @xmath213 grows strongest for @xmath180 . this is a consequence of the reduced electrical conductance of the quantum dot in the small polaron picture ; for a given @xmath144 , relatively large voltages are necessary to compensate the corresponding thermocurrent . for large temperature differences , we find a maximum value @xmath220 for all three data sets given in fig . [ fig3](c ) . this voltage corresponds to the postition of the zero - phonon step in @xmath219 in fig . [ fig3](b ) . here , the systems conductance grows considerably and any thermally induced current can easily be compensated by a slight growth in @xmath212 . , @xmath173 , @xmath174 , @xmath182 , as in fig . ( a ) thermovoltage @xmath213 as a function of the bare dot level @xmath6 for @xmath221 . we compare the variational calculation ( black line ) to the small polaron limit @xmath180 ( red dashed line ) and the case with zero ep coupling ( blue dash - dot line ) . ( b ) optimal variational parameter @xmath143 as a function of the bare dot level . ( c ) same as ( a ) but with @xmath211 . ( d ) electronic spectral functions for @xmath222 , @xmath223 and several temperature differences.,title="fig : " ] + , @xmath173 , @xmath174 , @xmath182 , as in fig . [ fig3 ] . ( a ) thermovoltage @xmath213 as a function of the bare dot level @xmath6 for @xmath221 . we compare the variational calculation ( black line ) to the small polaron limit @xmath180 ( red dashed line ) and the case with zero ep coupling ( blue dash - dot line ) . ( b ) optimal variational parameter @xmath143 as a function of the bare dot level . ( c ) same as ( a ) but with @xmath211 . ( d ) electronic spectral functions for @xmath222 , @xmath223 and several temperature differences.,title="fig : " ] + , @xmath173 , @xmath174 , @xmath182 , as in fig . [ fig3 ] . ( a ) thermovoltage @xmath213 as a function of the bare dot level @xmath6 for @xmath221 . we compare the variational calculation ( black line ) to the small polaron limit @xmath180 ( red dashed line ) and the case with zero ep coupling ( blue dash - dot line ) . ( b ) optimal variational parameter @xmath143 as a function of the bare dot level . ( c ) same as ( a ) but with @xmath211 . ( d ) electronic spectral functions for @xmath222 , @xmath223 and several temperature differences.,title="fig : " ] + , @xmath173 , @xmath174 , @xmath182 , as in fig . [ fig3 ] . ( a ) thermovoltage @xmath213 as a function of the bare dot level @xmath6 for @xmath221 . we compare the variational calculation ( black line ) to the small polaron limit @xmath180 ( red dashed line ) and the case with zero ep coupling ( blue dash - dot line ) . ( b ) optimal variational parameter @xmath143 as a function of the bare dot level . ( c ) same as ( a ) but with @xmath211 . ( d ) electronic spectral functions for @xmath222 , @xmath223 and several temperature differences.,title="fig : " ] in the previous sections , we chose the bare dot level @xmath6 in such a way , that the renormalized level @xmath150 was equal in the scenarios with and without ep interaction . in this way , we concentrated on the influence of the renormalized dot - lead coupling on the thermovoltage . however , in an experimental situation , the dot level may be manipulated , e.g. , by applying a gate voltage through a third electrode , in order to optimize the thermoelectric effect . to investigate this situation , we now consider @xmath6 as our free parameter . for each @xmath6 we temporarily set @xmath138 and @xmath155 to determine the optimal parameter @xmath143 . we then keep @xmath143 fixed , set @xmath144 to a finite value and calculate the thermovoltage @xmath213 . the remaining parameters are the same as in sec . [ seccompvol ] , i.e. , @xmath159 , @xmath173 , @xmath174 , and @xmath182 . first , we consider a small temperature difference @xmath221 . our result for the thermovoltage as a function of the bare dot level is presented in fig . [ fig4](a ) . in general , @xmath224 features two resonances of opposite sign and goes to zero when the renormalized dot level crosses the fermi surface , i.e. , when @xmath225 . at this point the spectral function is symmetrical around @xmath202 , and the thermoelectric flow and counterflow between the leads cancel exactly . as @xmath226 grows , the net thermocurrent and , consequently , the thermovoltage increases . in accordance with our results in sec . [ seccompvol ] , the maximum thermovoltage is largest in the small polaron picture due to the strong renormalization of the dot - lead coupling . when @xmath226 grows further and the zero phonon peak shifts away from the fermi surface , the dot density of states near @xmath202 decreases , and the thermoelectric effect vanishes again . in fig . [ fig4](a ) , the small polaron result is shifted from the @xmath181 curve by the value of the polaron binding energy @xmath153 , which can be understood from setting @xmath180 in eq . ( [ equdefeta ] ) . in both cases , the thermovoltage signal runs linearly through zero . we note that for growing @xmath16 ( not shown here ) its slope reduces , while the position of the positive ( negative ) resonance shifts to lower ( higher ) @xmath6 . in this regard , our result resembles the sawtooth - like thermopower signal that was predicted in ref . and experimentally measured in ref . . there , the periodicity of the thermopower oscillations was determined by the difference in the ground - state energies for different numbers of electrons on the dot . in our model however , we only account for a single dot electron . that is why we only observe a single `` tooth '' , i.e. , only two thermovoltage resonances in fig . [ fig4](a ) . moreover , we find no side peaks in the thermoelectric signal that could be attributed to the phonon side bands in @xmath208 . this is a consequence of the floating effect discussed in sec . [ secstate ] ; since the thermovoltages in fig . [ fig4](a ) are small , the expression @xmath227 in eq . ( [ equcurrent ] ) differs from zero only within the region @xmath228 $ ] . in this region , the phonon side bands in @xmath206 are suppressed and do not contribute to the thermoelectric transport . that is why the resonance signal in fig . [ fig4](a ) is given by the varying position of the zero - phonon peak only . consequently , in fig . [ fig4](a ) the variational calculation with its moderate renormalization of the dot - lead coupling predicts a weaker resonance signal than the small polaron picture , which is also shifted from the result for the rigid dot by less than @xmath153 . we find that in contrast to the two limiting cases , the strength of the resonance now depends on the sign of @xmath213 . this can be understood from fig . [ fig4](b ) , where we show the optimal variational parameter as a function of the dot level . for @xmath205 , we have @xmath229 like in the previous sections . when the renormalized dot level crosses the fermi surface , @xmath143 approaches @xmath230 and a small transient polaron forms at the dot . then the thermoelectric effect increases . figure [ fig4](c ) shows the thermovoltage signals after having raised the temperature difference to @xmath211 . as expected , for all three cases , the heights and widths of the resonances grow with respect to fig . [ fig4](a ) . more importantly , for finite ep coupling the curves now feature side bands at a distance of @xmath3 from their maximum resonances . this becomes possible because for large temperature differences the floating condition for the phonon bands is relaxed : since @xmath231 , the fermi surface of the hot left lead softens considerably . now the third and fourth lines in eq . ( [ equelectronic ] ) contribute to the spectral function even for @xmath228 $ ] . for example , fig . [ fig4](d ) shows the electronic spectral function in the variational calculation for @xmath222 and several temperature differences . since @xmath232 , we have @xmath233 . for growing @xmath144 , a small peak appears near the fermi - surface that is related to the polaronic state with one phonon at @xmath234 . when , for varying @xmath6 , this peak crosses the fermi surface , the side peak in the thermovoltage signal in fig . [ fig4](c ) appears . note , however , that the floating condition still holds for the cold right lead , as can be seen from the shoulder appearing near @xmath235 in fig . [ fig4](d ) . as the thermovoltage grows , the window @xmath228 $ ] closes and the shoulder shifts towards @xmath202 . in this work , we investigated the steady - state thermoelectric transport through a vibrating molecular quantum dot in the crossover regime far from the antiadiabatic limit . within a kadanoff - baym formalism that is generalized to account for different lead temperatures , the nonequilibrium dot self - energy was calculated to second order in the dot - lead interaction coefficient . in order to account for the polaronic character of the dot state , we applied a variational small - polaron transformation and determined the degree of transformation by minimizing the relevant thermodynamic potential . in essence , we calculated the current induced by a finite temperature difference between the leads . for small temperature differences , the influence of the electron - phonon ( ep ) interaction strongly depends on the specific system parameters ; for a near - resonant dot level , the interacting quantum dot acts as a more efficient energy filter and the thermocurrent increases . in the tunneling regime , however , we found a reduction of the thermocurrent due to the decreasing density of states near the fermi surface . in order to relate our results to experiment , we determined the thermovoltage required to compensate the thermally induced particle current at a given temperature difference . we found that in principle the franck - condon blockade boosts the thermovoltage through the reduction of the systems electrical conductance . for intermediate ep coupling , the small polaron picture overestimates this effect . finally , we determined the thermovoltage as a function of the dots energy level . because of pauli blocking , we found no phonon features in the low - temperature thermoelectric signal . however , our variational calculation predicts an asymmetrical line shape due to the formation of a small polaron as the dot level drops beneath the equilibrium fermi energy . for large temperature differences of the order of the phonon energy , pauli blocking is relaxed and the thermovoltage signal features multiple resonances that can be attributed to resonant transport through vibrational dot states . the present study should be considered a first step in applying our variational ansatz to the thermoelectric transport through molecular junctions . although it captures the essential physics , it must be extended in several directions . most importantly , we have yet to consider energy transport through the junction . the efficiency of energy deposition by the excitation of local phonons will strongly depend on the effective ep interaction , i.e. , on the polaronic character of the dot state . to investigate the subsequent heating or cooling of the molecule , the effective temperature of the dot electrons has to be determined numerically , for which several methods have been proposed.@xcite this should have some influence on the variational parameter @xmath34 as well . in our previous works , a voltage - dependent variational parameter was responsible for the junctions negative differential conductance . therefore , it would be highly desirable to carry this nonlinear behavior over to the discussion of thermoelectric transport . another worthwhile extension of our work concerns coulomb interaction effects . recently , andergassen _ et al._@xcite argued for the negative@xmath236 anderson model ( which neglects the coupling to the phonon degrees of freedom however ) that the resulting charge kondo effect leads to a large enhancement of the linear response thermopower due to the highly asymmetric dot spectral function . this effect was shown to be tunable applying a gate voltage . it would be interesting to reexamine this problem for a model with additional ep interaction , using our variational scheme , since as we have demonstrated in sec . [ seclevel]the effective ep interaction ( described by the variational parameter ) is strongly influenced by the gate voltage . in this connection we like to stress that for a combined holstein - hubbard quantum dot model it has been shown that strong ep coupling may result in a net attractive coulomb interaction.@xcite then , depending on the energy of the dot level , a variational ansatz might be able to interpolate between the positive@xmath236 holstein - hubbard dot model and the effective anisotropic kondo model regime investigated in ref . . 46ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) , @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( )

the origin of graph theory dates back to euler s solution of the puzzle of knigsberg s bridges in 1736 @xcite . since then a lot has been learned about graphs and their mathematical properties @xcite . in the 20th century they have also become extremely useful as representation of a wide variety of systems in different areas . biological , social , technological , and information networks can be studied as graphs , and graph analysis has become crucial to understand the features of these systems . for instance , social network analysis started in the 1930 s and has become one of the most important topics in sociology @xcite . in recent times , the computer revolution has provided scholars with a huge amount of data and computational resources to process and analyze these data . the size of real networks one can potentially handle has also grown considerably , reaching millions or even billions of vertices . the need to deal with such a large number of units has produced a deep change in the way graphs are approached @xcite . graphs representing real systems are not regular like , e. g. , lattices . they are objects where order coexists with disorder . the paradigm of disordered graph is the random graph , introduced by p. erds and a. rnyi @xcite . in it , the probability of having an edge between a pair of vertices is equal for all possible pairs ( see appendix ) . in a random graph , the distribution of edges among the vertices is highly homogeneous . for instance , the distribution of the number of neighbours of a vertex , or _ degree _ , is binomial , so most vertices have equal or similar degree . real networks are not random graphs , as they display big inhomogeneities , revealing a high level of order and organization . the degree distribution is broad , with a tail that often follows a power law : therefore , many vertices with low degree coexist with some vertices with large degree . furthermore , the distribution of edges is not only globally , but also locally inhomogeneous , with high concentrations of edges within special groups of vertices , and low concentrations between these groups . this feature of real networks is called _ community structure _ @xcite , or _ clustering _ , and is the topic of this review ( for earlier reviews see refs . communities , also called _ clusters _ or _ modules _ , are groups of vertices which probably share common properties and/or play similar roles within the graph . in fig . [ figure1 ] a schematic example of a graph with communities is shown . a simple graph with three communities , enclosed by the dashed circles . reprinted figure with permission from ref . @xcite . 2009 by springer . ] society offers a wide variety of possible group organizations : families , working and friendship circles , villages , towns , nations . the diffusion of internet has also led to the creation of virtual groups , that live on the web , like online communities . indeed , social communities have been studied for a long time @xcite . communities also occur in many networked systems from biology , computer science , engineering , economics , politics , etc . in protein - protein interaction networks , communities are likely to group proteins having the same specific function within the cell @xcite , in the graph of the world wide web they may correspond to groups of pages dealing with the same or related topics @xcite , in metabolic networks they may be related to functional modules such as cycles and pathways @xcite , in food webs they may identify compartments @xcite , and so on . communities can have concrete applications . clustering web clients who have similar interests and are geografically near to each other may improve the performance of services provided on the world wide web , in that each cluster of clients could be served by a dedicated mirror server @xcite . identifying clusters of customers with similar interests in the network of purchase relationships between customers and products of online retailers ( like , e. g. , www.amazon.com ) enables to set up efficient recommendation systems @xcite , that better guide customers through the list of items of the retailer and enhance the business opportunities . clusters of large graphs can be used to create data structures in order to efficiently store the graph data and to handle navigational queries , like path searches @xcite . _ ad hoc networks _ @xcite , i. e. self - configuring networks formed by communication nodes acting in the same region and rapidly changing ( because the devices move , for instance ) , usually have no centrally maintained routing tables that specify how nodes have to communicate to other nodes . grouping the nodes into clusters enables one to generate compact routing tables while the choice of the communication paths is still efficient @xcite . community detection is important for other reasons , too . identifying modules and their boundaries allows for a classification of vertices , according to their structural position in the modules . so , vertices with a central position in their clusters , i. e. sharing a large number of edges with the other group partners , may have an important function of control and stability within the group ; vertices lying at the boundaries between modules play an important role of mediation and lead the relationships and exchanges between different communities ( alike to csermely s `` creative elements '' @xcite ) . such classification seems to be meaningful in social @xcite and metabolic networks @xcite . finally , one can study the graph where vertices are the communities and edges are set between clusters if there are connections between some of their vertices in the original graph and/or if the modules overlap . in this way one attains a coarse - grained description of the original graph , which unveils the relationships between modules . recent studies indicate that networks of communities have a different degree distribution with respect to the full graphs @xcite ; however , the origin of their structures can be explained by the same mechanism @xcite . another important aspect related to community structure is the hierarchical organization displayed by most networked systems in the real world . real networks are usually composed by communities including smaller communities , which in turn include smaller communities , etc . the human body offers a paradigmatic example of hierarchical organization : it is composed by organs , organs are composed by tissues , tissues by cells , etc . another example is represented by business firms , who are characterized by a pyramidal organization , going from the workers to the president , with intermediate levels corresponding to work groups , departments and management . herbert a. simon has emphasized the crucial role played by hierarchy in the structure and evolution of complex systems @xcite . the generation and evolution of a system organized in interrelated stable subsystems are much quicker than if the system were unstructured , because it is much easier to assemble the smallest subparts first and use them as building blocks to get larger structures , until the whole system is assembled . in this way it is also far more difficult that errors ( mutations ) occur along the process . the aim of community detection in graphs is to identify the modules and , possibly , their hierarchical organization , by only using the information encoded in the graph topology . the problem has a long tradition and it has appeared in various forms in several disciplines . the first analysis of community structure was carried out by weiss and jacobson @xcite , who searched for work groups within a government agency . the authors studied the matrix of working relationships between members of the agency , which were identified by means of private interviews . work groups were separated by removing the members working with people of different groups , which act as connectors between them . this idea of cutting the bridges between groups is at the basis of several modern algorithms of community detection ( section [ sec5 ] ) . research on communities actually started even earlier than the paper by weiss and jacobson . already in 1927 , stuart rice looked for clusters of people in small political bodies , based on the similarity of their voting patterns @xcite . two decades later , george homans showed that social groups could be revealed by suitably rearranging the rows and the columns of matrices describing social ties , until they take an approximate block - diagonal form @xcite . this procedure is now standard . meanwhile , traditional techniques to find communities in social networks are hierarchical clustering and partitional clustering ( sections [ sec4_2 ] and [ sec4_3 ] ) , where vertices are joined into groups according to their mutual similarity . identifying graph communities is a popular topic in computer science , too . in parallel computing , for instance , it is crucial to know what is the best way to allocate tasks to processors so as to minimize the communications between them and enable a rapid performance of the calculation . this can be accomplished by splitting the computer cluster into groups with roughly the same number of processors , such that the number of physical connections between processors of different groups is minimal . the mathematical formalization of this problem is called _ graph partitioning _ ( section [ sec4_1 ] ) . the first algorithms for graph partitioning were proposed in the early 1970 s . in a seminal paper appeared in 2002 , girvan and newman proposed a new algorithm , aiming at the identification of edges lying between communities and their successive removal , a procedure that after some iterations leads to the isolation of the communities @xcite . the intercommunity edges are detected according to the values of a centrality measure , the edge betweenness , that expresses the importance of the role of the edges in processes where signals are transmitted across the graph following paths of minimal length . the paper triggered a big activity in the field , and many new methods have been proposed in the last years . in particular , physicists entered the game , bringing in their tools and techniques : spin models , optimization , percolation , random walks , synchronization , etc . , became ingredients of new original algorithms . the field has also taken advantage of concepts and methods from computer science , nonlinear dynamics , sociology , discrete mathematics . in this manuscript we try to cover in some detail the work done in this area . we shall pay a special attention to the contributions made by physicists , but we shall also give proper credit to important results obtained by scholars of other disciplines . section [ sec2 ] introduces communities in real networks , and is supposed to make the reader acquainted with the problem and its relevance . in section [ sec3 ] we define the basic elements of community detection , i. e. the concepts of community and partition . traditional clustering methods in computer and social sciences , i. e. graph partitioning , hierarchical , partitional and spectral clustering are reviewed in section [ sec4 ] . modern methods , divided into categories based on the type of approach , are presented in sections [ sec5 ] to [ sub_sec6_2 ] . algorithms to find overlapping communities , multiresolution and hierarchical techniques , are separately described in sections [ sec45 ] and [ sub_sec6_20 ] , respectively , whereas section [ sec7_1_2 ] is devoted to the detection of communities evolving in time . we stress that our categorization of the algorithms is not sharp , because many algorithms may enter more categories : we tried to classify them based on what we believe is their main feature / purpose , even if other aspects may be present . sections [ sec6_4 ] and [ sec6 ] are devoted to the issues of defining when community structure is significant , and deciding about the quality of algorithms performances . in sections [ sec7_1 ] and [ sec7_2 ] we describe general properties of clusters found in real networks , and specific applications of clustering algorithms . section [ sec8 ] contains the summary of the review , along with a discussion about future research directions in this area . the review makes use of several concepts of graph theory , that are defined and explained in the appendix . readers not acquainted with these concepts are urged to read the appendix first . in this section we shall present some striking examples of real networks with community structure . in this way we shall see what communities look like and why they are important . social networks are paradigmatic examples of graphs with communities . the word community itself refers to a social context . people naturally tend to form groups , within their work environment , family , friends . in fig . [ fig3 ] we show some examples of social networks . the first example ( fig . [ fig3]a ) is zachary s network of karate club members @xcite , a well - known graph regularly used as a benchmark to test community detection algorithms ( section [ sec6_1 ] ) . it consists of @xmath0 vertices , the members of a karate club in the united states , who were observed during a period of three years . edges connect individuals who were observed to interact outside the activities of the club . at some point , a conflict between the club president and the instructor led to the fission of the club in two separate groups , supporting the instructor and the president , respectively ( indicated by squares and circles ) . the question is whether from the original network structure it is possible to infer the composition of the two groups . indeed , by looking at fig . [ fig3]a one can distinguish two aggregations , one around vertices @xmath1 and @xmath0 ( @xmath0 is the president ) , the other around vertex @xmath2 ( the instructor ) . one can also identify several vertices lying between the two main structures , like @xmath3 , @xmath4 , @xmath5 ; such vertices are often misclassified by community detection methods . [ fig3]b displays the largest connected component of a network of collaborations of scientists working at the santa fe institute ( sfi ) . there are @xmath6 vertices , representing resident scientists at sfi and their collaborators . edges are placed between scientists that have published at least one paper together . the visualization layout allows to distinguish disciplinary groups . in this network one observes many cliques , as authors of the same paper are all linked to each other . there are but a few connections between most groups . in fig . [ fig3]c we show the network of bottlenose dolphins living in doubtful sound ( new zealand ) analyzed by lusseau @xcite . there are @xmath7 dolphins and edges were set between animals that were seen together more often than expected by chance . the dolphins separated in two groups after a dolphin left the place for some time ( squares and circles in the figure ) . such groups are quite cohesive , with several internal cliques , and easily identifiable : only six edges join vertices of different groups . due to this natural classification lusseau s dolphins network , like zachary s karate club , is often used to test algorithms for community detection ( section [ sec6_1 ] ) . protein - protein interaction ( ppi ) networks are subject of intense investigations in biology and bioinformatics , as the interactions between proteins are fundamental for each process in the cell @xcite . [ fig4 ] illustrates a ppi network of the rat proteome @xcite . each interaction is derived by homology from experimentally observed interactions in other organisms . in our example , the proteins interact very frequently with each other , as they belong to metastatic cells , which have a high motility and invasiveness with respect to normal cells . communities correspond to functional groups , i. e. to proteins having the same or similar functions , which are expected to be involved in the same processes . the modules are labeled by the overall function or the dominating protein class . most communities are associated to cancer and metastasis , which indirectly shows how important detecting modules in ppi networks is . community structure in technological networks . sample of the web graph consisting of the pages of a web site and their mutual hyperlinks , which are directed . communities , indicated by the colors , were detected with the algorithm of girvan and newman ( section [ subsec5_1 ] ) , by neglecting the directedness of the edges . reprinted figure with permission from ref . @xcite . 2004 by the american physical society . ] relationships / interactions between elements of a system need not be reciprocal . in many cases they have a precise direction , that needs to be taken into account to understand the system as a whole . as an example we can cite predator - prey relationships in food webs . in fig . [ fig5 ] we see another example , taken from technology . the system is the world wide web , which can be seen as a graph by representing web pages as vertices and the hyperlinks that make users move from one page to another as edges @xcite . hyperlinks are directed : if one can move from page a to page b by clicking on a hyperlink of a , one usually does not find on b a hyperlink taking back to a. in fact , very few hyperlinks ( less than @xmath8 ) are reciprocal . communities of the web graph are groups of pages having topical similarities . detecting communities in the web graph may help to identify the artificial clusters created by link farms in order to enhance the pagerank @xcite value of web sites and grant them a higher google ranking . in this way one could discourage this unfair practice . one usually assumes that the existence of a hyperlink between two pages implies that they are content - related , and that this similarity is independent of the hyperlink direction . therefore it is customary to neglect the directedness of the hyperlinks and to consider the graph as undirected , for the purpose of community detection . on the other hand , taking properly into account the directedness of the edges can considerably improve the quality of the partition(s ) , as one can handle a lot of precious information about the system . moreover , in some instances neglecting edge directedness may lead to strange results @xcite . developing methods of community detection for directed graphs is a hard task . for instance , a directed graph is characterized by asymmetrical matrices ( adjacency matrix , laplacian , etc . ) , so spectral analysis is much more complex . only a few techniques can be easily extended from the undirected to the directed case . otherwise , the problem must be formulated from scratch . edge directedness is not the only complication to deal with when facing the problem of graph clustering . in many real networks vertices may belong to more than one group . in this case one speaks of _ overlapping communities _ and uses the term _ cover _ , rather than partition , whose standard definition forbids multiple memberships of vertices . classical examples are social networks , where an individual usually belongs to different circles at the same time , from that of work colleagues to family , sport associations , etc .. traditional algorithms of community detection assign each vertex to a single module . in so doing , they neglect potentially relevant information . vertices belonging to more communities are likely to play an important role of intermediation between different compartments of the graph . in fig . [ fig6 ] we show a network of word association derived starting from the word `` bright '' . the network builds on the university of south florida free association norms @xcite . an edge between words @xmath9 and @xmath10 indicates that some people associate @xmath10 to the word @xmath9 . the graph clearly displays four communities , corresponding to the categories _ intelligence _ , _ astronomy _ , _ light _ and _ colors_. the word `` bright '' is related to all of them by construction . other words belong to more categories , e.g. `` dark '' ( colors and light ) . accounting for overlapping communities introduces a further variable , the membership of vertices in different communities , which enormously increases the number of possible covers with respect to standard partitions . therefore , searching for overlapping communities is usually much more computationally demanding than detecting standard partitions . overlapping communities in a network of word association . the groups , labeled by the colors , were detected with the clique percolation method by palla et al . ( section [ sec45_1 ] ) . reprinted figure with permission from ref . 2005 by the nature publishing group . ] so far we have discussed examples of unipartite graphs . however , it is not uncommon to find real networks with different classes of vertices , and edges joining only vertices of different classes . an example is a network of scientists and papers , where edges join scientists and the papers they have authored . here there is no edge between any pair of scientists or papers , so the graph is bipartite . for a multipartite network the concept of community does not change much with respect to the case of unipartite graphs , as it remains related to a large density of edges between members of the same group , with the only difference that the elements of each group belong to different vertex classes . multipartite graphs are usually reduced to unipartite projections of each vertex class . for instance , from the bipartite network of scientists and papers one can extract a network of scientists only , who are related by coauthorship . in this way one can adopt standard techniques of network analysis , in particular standard clustering methods , but a lot of information gets lost . detecting communities in multipartite networks can have interesting applications in , e.g. , marketing . large shopping networks , in which customers are linked to the products they have bought , allow to classify customers based on the types of product they purchase more often : this could be used both to organize targeted advertising , as well as to give recommendations about future purchases @xcite . the problem of community detection in multipartite networks is not trivial , and usually requires _ ad hoc _ methodologies . [ fig7 ] illustrates the famous bipartite network of southern women studied by davis et al . there are @xmath11 vertices , representing @xmath12 women from the area of natchez , mississippi , and @xmath13 social events . edges represent the participation of the women in the events . from the figure one can see that the network has a clear community structure . community structure in multipartite networks . this bipartite graph refers to the southern women event participation data set . women are represented as open symbols with black labels , events as filled symbols with white labels . the illustrated vertex partition has been obtained by maximizing a modified version of the modularity by newman and girvan , tailored on bipartite graphs @xcite ( section [ sub_sec6_01 ] ) . reprinted figure with permission from ref . 2007 by the american physical society . ] in some of the previous examples , edges have ( or can have ) weights . for instance , the edges of the collaboration network of fig . [ fig3]b could be weighted by the number of papers coauthored by pairs of scientists . similarly , the edges of the word association network of fig . [ fig6 ] are weighted by the number of times pairs of words have been associated by people . weights are precious additional information on a graph , and should be considered in the analysis . in many cases methods working on unweighted graphs can be simply extended to the weighted case . the problem of graph clustering , intuitive at first sight , is actually not well defined . the main elements of the problem themselves , i. e. the concepts of community and partition , are not rigorously defined , and require some degree of arbitrariness and/or common sense . indeed , some ambiguities are hidden and there are often many equally legitimate ways of resolving them . therefore , it is not surprising that there are plenty of recipes in the literature and that people do not even try to ground the problem on shared definitions . it is important to stress that the identification of structural clusters is possible only if graphs are _ sparse _ , i. e. if the number of edges @xmath14 is of the order of the number of nodes @xmath15 of the graph . if @xmath16 , the distribution of edges among the nodes is too homogeneous for communities to make sense . in this case the problem turns into something rather different , close to data clustering @xcite , which requires concepts and methods of a different nature . the main difference is that , while communities in graphs are related , explicitly or implicitly , to the concept of edge density ( inside versus outside the community ) , in data clustering communities are sets of points which are `` close '' to each other , with respect to a measure of distance or similarity , defined for each pair of points . some classical techniques for data clustering , like _ hierarchical _ , _ partitional _ and _ spectral clustering _ will be discussed later in the review ( sections [ sec4_2 ] , [ sec4_3 ] and [ sec4_4 ] ) , as they are sometimes adopted for graph clustering too . other standard procedures for data clustering include _ neural network clustering _ techniques like , e. g. , _ self - organizing maps _ and _ multi - dimensional scaling _ techniques like , e. g. , _ singular value decomposition _ and _ principal component analysis _ @xcite . in this section we shall attempt an ordered exposition of the fundamental concepts of community detection . after a brief discussion of the issue of computational complexity for the algorithms , we shall review the notions of community and partition . the massive amount of data on real networks currently available makes the issue of the efficiency of clustering algorithms essential . the _ computational complexity _ of an algorithm is the estimate of the amount of resources required by the algorithm to perform a task . this involves both the number of computation steps needed and the number of memory units that need to be simultaneously allocated to run the computation . such demands are usually expressed by their scalability with the size of the system at study . in the case of a graph , the size is typically indicated by the number of vertices @xmath15 and/or the number of edges @xmath14 . the computational complexity of an algorithm can not always be calculated . in fact , sometimes this is a very hard task , or even impossible . in these cases , it is however important to have at least an estimate of the _ worst - case _ complexity of the algorithm , which is the amount of computational resources needed to run the algorithm in the most unfavorable case for a given system size . the notation @xmath17 indicates that the computer time grows as a power of both the number of vertices and edges , with exponents @xmath18 and @xmath19 , respectively . ideally , one would like to have the lowest possible values for the exponents , which would correspond to the lowest possible computational demands . samples of the web graph , with millions of vertices and billions of edges , can not be tackled by algorithms whose running time grows faster than @xmath20 or @xmath21 . algorithms with polynomial complexity form the class * p*. for some important decision and optimization problems , there are no known polynomial algorithms . finding solutions of such problems in the worst - case scenario may demand an exhaustive search , which takes a time growing faster than any polynomial function of the system size , e.g. exponentially . problems whose solutions can be verified in a polynomial time span the class * np * of _ non - deterministic polynomial time _ problems , which includes * p*. a problem is * np*-hard if a solution for it can be translated into a solution for any * np*-problem . however , a * np*-hard problem needs not be in the class * np*. if it does belong to * np * it is called * np*-complete . the class of * np*-complete problems has drawn a special attention in computer science , as it includes many famous problems like the travelling salesman , boolean satisfiability ( sat ) , linear programming , etc . the fact that * np * problems have a solution which is verifiable in polynomial time does not mean that * np * problems have polynomial complexity , i. e. , that they are in * p*. in fact , the question of whether * np*=*p * is the most important open problem in theoretical computer science . * np*-hard problems need not be in * np * ( in which case they would be * np*-complete ) , but they are at least as hard as * np*-complete problems , so they are unlikely to have polynomial complexity , although a proof of that is still missing . many clustering algorithms or problems related to clustering are * np*-hard . in this case , it is pointless to use exact algorithms , which could be applied only to very small systems . moreover , even if an algorithm has a polynomial complexity , it may still be too slow to tackle large systems of interest . in all such cases it is common to use _ approximation algorithms _ , i. e. methods that do not deliver an exact solution to the problem at hand , but only an approximate solution , with the advantage of a lower complexity . approximation algorithms are often non - deterministic , as they deliver different solutions for the same problem , for different initial conditions and/or parameters of the algorithm . the goal of such algorithms is to deliver a solution which differs by a constant factor from the optimal solution . in any case , one should give provable bounds on the goodness of the approximate solution delivered by the algorithm with respect to the optimal solution . in many cases it is not possible to approximate the solution within any constant , as the goodness of the approximation strongly depends on the specific problem at study . approximation algorithms are commonly used for optimization problems , in which one wants to find the maximum or minimum value of a given cost function over a large set of possible system configurations . the first problem in graph clustering is to look for a quantitative definition of community . no definition is universally accepted . as a matter of fact , the definition often depends on the specific system at hand and/or application one has in mind . from intuition and the examples of section [ sec2 ] we get the notion that there must be more edges `` inside '' the community than edges linking vertices of the community with the rest of the graph . this is the reference guideline at the basis of most community definitions . but many alternative recipes are compatible with it . moreover , in most cases , communities are algorithmically defined , i. e. they are just the final product of the algorithm , without a precise _ a priori _ definition . let us start with a subgraph @xmath22 of a graph @xmath23 , with @xmath24 and @xmath25 vertices , respectively . we define the _ internal _ and _ external _ degree of vertex @xmath26 , @xmath27 and @xmath28 , as the number of edges connecting @xmath29 to other vertices of @xmath22 or to the rest of the graph , respectively . if @xmath30 , the vertex has neighbors only within @xmath22 , which is likely to be a good cluster for @xmath29 ; if @xmath31 , instead , the vertex is disjoint from @xmath22 and it should better be assigned to a different cluster . the _ internal degree _ @xmath32 of @xmath22 is the sum of the internal degrees of its vertices . likewise , the _ external degree _ @xmath33 of @xmath22 is the sum of the external degrees of its vertices . the _ total degree _ @xmath34 is the sum of the degrees of the vertices of @xmath22 . by definition , @xmath35 . we define the _ intra - cluster density _ @xmath36 of the subgraph @xmath22 as the ratio between the number of internal edges of @xmath22 and the number of all possible internal edges , i. e. @xmath37 similarly , the _ inter - cluster density _ @xmath38 is the ratio between the number of edges running from the vertices of @xmath22 to the rest of the graph and the maximum number of inter - cluster edges possible , i. e. @xmath39 for @xmath22 to be a community , we expect @xmath36 to be appreciably larger than the average link density @xmath40 of @xmath23 , which is given by the ratio between the number of edges of @xmath23 and the maximum number of possible edges @xmath41 . on the other hand , @xmath38 has to be much smaller than @xmath40 . searching for the best tradeoff between a large @xmath36 and a small @xmath38 is implicitly or explicitly the goal of most clustering algorithms . a simple way to do that is , e. g. , maximizing the sum of the differences @xmath42 over all clusters of the partition$ ] , but essentially equivalent . ] @xcite . a required property of a community is _ connectedness_. we expect that for @xmath22 to be a community there must be a path between each pair of its vertices , running only through vertices of @xmath22 . this feature simplifies the task of community detection on disconnected graphs , as in this case one just analyzes each connected component separately , unless special constraints are imposed on the resulting clusters . with these basic requirements in mind , we can now introduce the main definitions of community . social network analysts have devised many definitions of subgroups with various degrees of internal cohesion among vertices @xcite . many other definitions have been introduced by computer scientists and physicists . we distinguish three classes of definitions : local , global and based on vertex similarity . other definitions will be given in the context of the algorithms for which they were introduced . communities are parts of the graph with a few ties with the rest of the system . to some extent , they can be considered as separate entities with their own autonomy . so , it makes sense to evaluate them independently of the graph as a whole . local definitions focus on the subgraph under study , including possibly its immediate neighborhood , but neglecting the rest of the graph . we start with a listing of the main definitions adopted in social network analysis , for which we shall closely follow the exposition of ref . @xcite . there , four types of criteria were identified : _ complete mutuality _ , _ reachability _ , _ vertex degree _ and the _ comparison of internal versus external cohesion_. the corresponding communities are mostly _ maximal subgraphs _ , which can not be enlarged with the addition of new vertices and edges without losing the property which defines them . social communities can be defined in a very strict sense as subgroups whose members are all `` friends '' to each other @xcite ( complete mutuality ) . in graph terms , this corresponds to a _ clique _ , i. e. a subset whose vertices are all adjacent to each other . in social network analysis , a clique is a maximal subgraph , whereas in graph theory it is common to call cliques also non - maximal subgraphs . triangles are the simplest cliques , and are frequent in real networks . but larger cliques are less frequent . moreover , the condition is really too strict : a subgraph with all possible internal edges except one would be an extremely cohesive subgroup , but it would not be considered a community under this recipe . another problem is that all vertices of a clique are absolutely symmetric , with no differentiation between them . in many practical examples , instead , we expect that within a community there is a whole hierarchy of roles for the vertices , with core vertices coexisting with peripheral ones . we remark that vertices may belong to more cliques simultaneously , a property which is at the basis of the clique percolation method of palla et al . @xcite ( see section [ sec45_1 ] ) . from a practical point of view , finding cliques in a graph is an * np*-complete problem @xcite . the bron - kerbosch method @xcite runs in a time growing exponentially with the size of the graph . it is however possible to relax the notion of clique , defining subgroups which are still clique - like objects . a possibility is to use properties related to reachability , i. e. to the existence ( and length ) of paths between vertices . an _ @xmath15-clique _ is a maximal subgraph such that the distance of each pair of its vertices is not larger than @xmath15 @xcite . for @xmath43 one recovers the definition of clique , as all vertices are adjacent , so each geodesic path between any pair of vertices has length @xmath2 . this definition , more flexible than that of clique , still has some limitations , deriving from the fact that the geodesic paths need not run on the vertices of the subgraph at study , but may run on vertices outside the subgraph . in this way , there may be two disturbing consequences . first , the diameter of the subgraph may exceed @xmath15 , even if in principle each vertex of the subgraph is less than @xmath15 steps away from any of the others . second , the subgraph may be disconnected , which is not consistent with the notion of cohesion one tries to enforce . to avoid these problems , mokken @xcite has suggested two possible alternatives , the _ @xmath15-clan _ and the _ @xmath15-club_. an @xmath15-clan is an @xmath15-clique whose diameter is not larger than @xmath15 , i. e. a subgraph such that the distance between any two of its vertices , computed over shortest paths within the subgraph , does not exceed @xmath15 . an @xmath15-club , instead , is a maximal subgraph of diameter @xmath15 . the two definitions are quite close : the difference is that an @xmath15-clan is maximal under the constraint of being an @xmath15-clique , whereas an @xmath15-club is maximal under the constraint imposed by the length of the diameter . another criterion for subgraph cohesion relies on the adjacency of its vertices . the idea is that a vertex must be adjacent to some minumum number of other vertices in the subgraph . in the literature on social network analysis there are two complementary ways of expressing this . @xmath44-plex _ is a maximal subgraph in which each vertex is adjacent to all other vertices of the subgraph except at most @xmath44 of them @xcite . similarly , a _ @xmath44-core _ is a maximal subgraph in which each vertex is adjacent to at least @xmath44 other vertices of the subgraph @xcite . so , the two definitions impose conditions on the minimal number of absent or present edges . the corresponding clusters are more cohesive than @xmath15-cliques , just because of the existence of many internal edges . in any graph there is a whole hierarchy of cores of different order , which can be identified by means of a recent efficient algorithm @xcite . a @xmath44-core is essentially the same as a _ @xmath45-quasi complete _ subgraph , which is a subgraph such that the degree of each vertex is larger than @xmath46 , where @xmath45 is a real number in @xmath47 $ ] and @xmath44 the order of the subgraph @xcite . determining whether a graph includes a @xmath48-quasi complete subgraph of order at least @xmath44 is * np*-complete . as cohesive as a subgraph can be , it would hardly be a community if there is a strong cohesion as well between the subgraph and the rest of the graph . therefore , it is important to compare the internal and external cohesion of a subgraph . in fact , this is what is usually done in the most recent definitions of community . the first recipe , however , is not recent and stems from social network analysis . an _ @xmath49-set _ @xcite , or _ strong community _ @xcite , is a subgraph such that the internal degree of each vertex is greater than its external degree . this condition is quite strict and can be relaxed into the so - called _ weak _ definition of community @xcite , for which it suffices that the internal degree of the subgraph exceeds its external degree . an @xmath49-set is also a weak community , while the converse is not generally true . hu et al . @xcite have introduced alternative definitions of strong and weak communities : a community is strong if the internal degree of any vertex of the community exceeds the number of edges that the vertex shares with any other community ; a community is weak if its total internal degree exceeds the number of edges shared by the community with the other communities . these definitions are in the same spirit of the _ planted partition model _ ( section [ sec6 ] ) . an @xmath49-set is a strong community also in the sense of hu et al .. likewise , a weak community according to radicchi et al . is also a weak community for hu et al .. in both cases the converse is not true , however . another definition focuses on the robustness of clusters to edge removal and uses the concept of _ edge connectivity_. the edge connectivity of a pair of vertices in a graph is the minimal number of edges that need to be removed in order to disconnect the two vertices , i. e. such that there is no path between them . lambda set _ is a subgraph such that any pair of vertices of the subgraph has a larger edge connectivity than any pair formed by one vertex of the subgraph and one outside the subgraph @xcite . however , vertices of a lambda - set need not be adjacent and may be quite distant from each other . communities can also be identified by a _ fitness measure _ , expressing to which extent a subraph satisfies a given property related to its cohesion . the larger the fitness , the more definite is the community . this is the same principle behind _ quality functions _ , which give an estimate of the goodness of a graph partition ( see section [ sec3_2_2 ] ) . the simplest fitness measure for a cluster is its intra - cluster density @xmath36 . one could assume that a subgraph @xmath22 with @xmath44 vertices is a cluster if @xmath36 is larger than a threshold , say @xmath50 . finding such subgraphs is an * np*-complete problem , as it coincides with the * np*-complete clique problem when the threshold @xmath51 it is better to fix the size of the subgraph because , without this conditions , any clique would be one of the best possible communities , including trivial two - cliques ( simple edges ) . variants of this problem focus on the number of internal edges of the subgraph @xcite . another measure of interest is the _ relative density _ @xmath52 of a subgraph @xmath22 , defined as the ratio between the internal and the total degree of @xmath22 . finding subgraphs of a given size with @xmath52 larger than a threshold is * np*-complete @xcite . fitness measures can also be associated to the connectivity of the subgraph at study to the other vertices of the graph . a good community is expected to have a small cut size ( see section [ sec1_1 ] ) , i. e. a small number of edges joining it to the rest of the graph . this sets a bridge between community detection and graph partitioning , which we shall discuss in section [ sec4_1 ] . communities can also be defined with respect to the graph as a whole . this is reasonable in those cases in which clusters are essential parts of the graph , which can not be taken apart without seriously affecting the functioning of the system . the literature offers many global criteria to identify communities . in most cases they are indirect definitions , in which some global property of the graph is used in an algorithm that delivers communities at the end . however , there is a class of proper definitions , based on the idea that a graph has community structure if it is different from a random graph . a random graph la erds - rnyi ( section [ sec1_3 ] ) , for instance , is not expected to have community structure , as any two vertices have the same probability to be adjacent , so there should be no preferential linking involving special groups of vertices . therefore , one can define a _ null model _ , i. e. a graph which matches the original in some of its structural features , but which is otherwise a random graph . the null model is used as a term of comparison , to verify whether the graph at study displays community structure or not . the most popular null model is that proposed by newman and girvan and consists of a randomized version of the original graph , where edges are rewired at random , under the constraint that the expected degree of each vertex matches the degree of the vertex in the original graph @xcite . this null model is the basic concept behind the definition of _ modularity _ , a function which evaluates the goodness of partitions of a graph into clusters . modularity will be discussed at length in this review , because it has the unique privilege of being at the same time a global criterion to define a community , a quality function and the key ingredient of the most popular method of graph clustering . in the standard formulation of modularity , a subgraph is a community if the number of edges inside the subgraph exceeds the expected number of internal edges that the same subgraph would have in the null model . this expected number is an average over all possible realizations of the null model . several modifications of modularity have been proposed ( section [ sub_sec6_01 ] ) . a general class of null models , including modularity as a special case , has been designed by reichardt and bornholdt @xcite ( section [ sub_sec6_01 ] ) . it is natural to assume that communities are groups of vertices similar to each other . one can compute the similarity between each pair of vertices with respect to some reference property , local or global , no matter whether they are connected by an edge or not . each vertex ends up in the cluster whose vertices are most similar to it . similarity measures are at the basis of traditional methods , like hierarchical , partitional and spectral clustering , to be discussed in sections [ sec4_2 ] , [ sec4_3 ] and [ sec4_4 ] . here we discuss some popular measures used in the literature . if it is possible to embed the graph vertices in an @xmath15-dimensional euclidean space , by assigning a position to them , one could use the _ distance _ between a pair of vertices as a measure of their similarity ( it is actually a measure of dissimilarity because similar vertices are expected to be close to each other ) . given the two data points @xmath53 and @xmath54 , one could use any norm @xmath55 , like the _ euclidean distance _ ( @xmath56-norm ) , @xmath57 the _ manhattan distance _ ( @xmath58-norm ) @xmath59 and the @xmath60-norm @xmath61}|a_k - b_k| . \label{eq5}\ ] ] another popular spatial measure is the _ cosine similarity _ , defined as @xmath62 where @xmath63 is the dot product of the vectors * a * and * b*. the variable @xmath64 is defined in the range @xmath65 . if the graph can not be embedded in space , the similarity must be necessarily inferred from the adjacency relationships between vertices . a possibility is to define a distance @xcite between vertices like @xmath66 where * a * is the adjacency matrix . this is a dissimilarity measure , based on the concept of structural equivalence @xcite . two vertices are structurally equivalent if they have the same neighbors , even if they are not adjacent themselves . if @xmath67 and @xmath68 are structurally equivalent , @xmath69 . vertices with large degree and different neighbours are considered very `` far '' from each other . alternatively , one could measure the _ overlap _ between the neighborhoods @xmath70 and @xmath71 of vertices @xmath67 and @xmath68 , given by the ratio between the intersection and the union of the neighborhoods , i. e. @xmath72 another measure related to structural equivalence is the pearson correlation between columns or rows of the adjacency matrix , @xmath73 where the averages @xmath74 and the variances @xmath75 . an alternative measure is the number of edge- ( or vertex- ) independent paths between two vertices . independent paths do not share any edge ( vertex ) , and their number is related to the maximum flow that can be conveyed between the two vertices under the constraint that each edge can carry only one unit of flow ( max - flow / min - cut theorem @xcite ) . the maximum flow can be calculated in a time @xmath21 , for a graph with @xmath14 edges , using techniques like the augmenting path algorithm @xcite . similarly , one could consider all paths running between two vertices . in this case , there is the problem that the total number of paths is infinite , but this can be avoided if one performs a weighted sum of the number of paths . for instance , paths of length @xmath76 can be weighted by the factor @xmath77 , with @xmath78 . another possibility , suggested by estrada and hatano @xcite , is to weigh paths of length @xmath76 with the inverse factorial @xmath79 . in both cases , the contribution of long paths is strongly suppressed and the sum converges . another important class of measures of vertex similarity is based on properties of random walks on graphs . one of this properties is the _ commute - time _ between a pair of vertices , which is the average number of steps needed for a random walker , starting at either vertex , to reach the other vertex for the first time and to come back to the starting vertex . saerens and coworkers @xcite have extensively studied and used the commute - time ( and variants thereof ) as ( dis)similarity measure : the larger the time , the farther ( less similar ) the vertices . the commute - time is closely related @xcite to the _ resistance distance _ introduced by klein and randic @xcite , expressing the effective electrical resistance between two vertices if the graph is turned into a resistor network . white and smyth @xcite and zhou @xcite used instead the average first passage time , i. e. the average number of steps needed to reach for the first time the target vertex from the source . harel and koren @xcite proposed to build measures out of quantities like the probability to visit a target vertex in no more than a given number of steps after it leaves a source vertex ) and in a recent analysis by nadler et al . @xcite , one defined a dissimilarity measure called `` diffusion distance '' , which is derived from the probability that the walker visits the target after a fixed number of steps . ] and the probability that a random walker starting at a source visits the target exactly once before hitting the source again . another quantity used to define similarity measures is the _ escape probability _ , defined as the probability that the walker reaches the target vertex before coming back to the source vertex @xcite . the escape probability is related to the effective conductance between the two vertices in the equivalent resistor network . other authors have exploited properties of modified random walks . for instance , the algorithm by gori and pucci @xcite and that by tong et al . @xcite used similarity measures derived from google s pagerank process @xcite . a _ partition _ is a division of a graph in clusters , such that each vertex belongs to one cluster . as we have seen in section [ sec2 ] , in real systems vertices may be shared among different communities . a division of a graph into overlapping ( or _ fuzzy _ ) communities is called _ cover_. the number of possible partitions in @xmath44 clusters of a graph with @xmath15 vertices is the _ stirling number of the second kind _ @xmath80 @xcite . the total number of possible partitions is the _ n - th bell number _ @xmath81 @xcite . in the limit of large @xmath15 , @xmath82 has the asymptotic form @xcite @xmath83^{n+1/2}e^{\lambda(n)-n-1 } , \label{eqpart1}\ ] ] where @xmath84 , @xmath85 being the _ lambert @xmath86 function _ @xcite . therefore , @xmath82 grows faster than exponentially with the graph size @xmath15 , which means that an enumeration and/or evaluation of all partitions of a graph is impossible , unless the graph consists of very few vertices . partitions can be _ hierarchically ordered _ , when the graph has different levels of organization / structure at different scales . in this case , clusters display in turn community structure , with smaller communities inside , which may again contain smaller communities , and so on ( fig . [ fig8 ] ) . as an example , in a social network of children living in the same town , one could group the children according to the schools they attend , but within each school one can make a subdivision into classes . hierarchical organization is a common feature of many real networks , where it is revealed by a peculiar scaling of the clustering coefficient for vertices having the same degree @xmath44 , when plotted as a function of @xmath44 @xcite . a natural way to represent the hierarchical structure of a graph is to draw a _ dendrogram _ , like the one illustrated in fig . [ fig9 ] . here , partitions of a graph with twelve vertices are shown . at the bottom , each vertex is its own module ( the `` leaves '' of the tree ) . by moving upwards , groups of vertices are successively aggregated . mergers of communities are represented by horizontal lines . the uppermost level represents the whole graph as a single community . cutting the diagram horizontally at some height , as shown in the figure ( dashed line ) , displays one partition of the graph . the diagram is hierarchical by construction : each community belonging to a level is fully included in a community at a higher level . dendrograms are regularly used in sociology and biology . the technique of hierarchical clustering , described in section [ sec4_2 ] , lends itself naturally to this kind of representation . a dendrogram , or hierarchical tree . horizontal cuts correspond to partitions of the graph in communities . reprinted figure with permission from ref . 2004 by the american physical society . ] reliable algorithms are supposed to identify _ good _ partitions . but what is a good clustering ? in order to distinguish between `` good '' and `` bad '' clusterings , it would be useful to require that partitions satisfy a set of basic properties , intuitive and easy to agree upon . in the wider context of data clustering , this issue has been studied by jon kleinberg @xcite , who has proved an important _ impossibility theorem_. given a set @xmath87 of points , a _ distance function _ @xmath88 is defined , which is positive definite and symmetric ( the triangular inequality is not explicitly required ) . one wishes to find a clustering @xmath89 based on the distances between the points . kleinberg showed that no clustering satisfies at the same time the three following properties : 1 . _ scale - invariance _ : given a constant @xmath18 , multiplying _ any _ distance function @xmath88 by @xmath18 yields the same clustering . richness _ : any possible partition of the given point set can be recovered if one chooses a suitable distance function @xmath88 . consistency _ : given a partition , any modification of the distance function that does not decrease the distance between points of different clusters and that does not increase the distance between points of the same cluster , yields the same clustering . the theorem can not be extended to graph clustering because the distance function can not be in general defined for a graph which is not complete . for weighted complete graphs , like correlation matrices @xcite , it is often possible to define a distance function . on a generic graph , except for the first property , which does not make sense without a distance function , the other two are quite well defined . the property of richness implies that , given a partition , one can set edges between the vertices in such a way that the partition is a natural outcome of the resulting graph ( e.g. , it could be achieved by setting edges only between vertices of the same cluster ) . consistency here implies that deleting inter - cluster edges and adding intra - cluster edges yields the same partition . many algorithms are able to identify a subset of meaningful partitions , ideally one or just a few , whereas some others , like techniques based on hierarchical clustering ( section [ sec4_2 ] ) , deliver a large number of partitions . that does not mean that the partitions found are equally good . therefore it is helpful ( sometimes even necessary ) to have a quantitative _ criterion _ to assess the goodness of a graph partition . quality function _ is a function that assigns a number to each partition of a graph . in this way one can rank partitions based on their score given by the quality function . partitions with high scores are `` good '' , so the one with the largest score is by definition the best . nevertheless , one should keep in mind that the question of when a partition is better than another one is ill - posed , and the answer depends on the specific concept of community and/or quality function adopted . a quality function @xmath90 is _ additive _ if there is an elementary function @xmath91 such that , for any partition @xmath92 of a graph @xmath93 where @xmath22 is a generic cluster of partition @xmath94 . . [ eq10 ] states that the quality of a partition is given by the sum of the qualities of the individual clusters . the function @xmath95 could be any of the cluster fitness functions discussed in section [ sec3_1_2 ] , for instance . most quality functions used in the literature are additive , although it is not a necessary requirement . an example of quality function is the _ performance _ @xmath96 , which counts the number of correctly `` interpreted '' pairs of vertices , i. e. two vertices belonging to the same community and connected by an edge , or two vertices belonging to different communities and not connected by an edge . the definition of performance , for a partition @xmath94 , is @xmath97 by definition , @xmath98 . another example is _ coverage _ , i. e. the ratio of the number of intra - community edges by the total number of edges : by definition , an ideal cluster structure , where the clusters are disconnected from each other , yields a coverage of @xmath2 , as all edges of the graph fall within clusters . the most popular quality function is the modularity of newman and girvan @xcite . it is based on the idea that a random graph is not expected to have a cluster structure , so the possible existence of clusters is revealed by the comparison between the actual density of edges in a subgraph and the density one would expect to have in the subgraph if the vertices of the graph were attached regardless of community structure . this expected edge density depends on the chosen _ null model _ , i. e. a copy of the original graph keeping some of its structural properties but without community structure . modularity can then be written as follows @xmath99 where the sum runs over all pairs of vertices , @xmath9 is the adjacency matrix , @xmath14 the total number of edges of the graph , and @xmath100 represents the expected number of edges between vertices @xmath67 and @xmath68 in the null model . the @xmath101-function yields one if vertices @xmath67 and @xmath68 are in the same community ( @xmath102 ) , zero otherwise . the choice of the null model graph is in principle arbitrary , and several possibilities exist . for instance , one could simply demand that the graph keeps the same number of edges as the original graph , and that edges are placed with the same probability between any pair of vertices . in this case ( bernoulli random graph ) , the null model term in eq . [ eq : mod_0 ] would be a constant ( i. e. @xmath103 $ ] , @xmath104 ) . however this null model is not a good descriptor of real networks , as it has a poissonian degree distribution which is very different from the skewed distributions found in real networks . due to the important implications that broad degree distributions have for the structure and function of real networks @xcite , it is preferable to go for a null model with the same degree distribution of the original graph . the standard null model of modularity imposes that the expected degree sequence ( after averaging over all possible configurations of the model ) matches the actual degree sequence of the graph . this is a stricter constraint than merely requiring the match of the degree distributions , and is essentially equivalent to the _ configuration model _ , which has been subject of intense investigations in the recent literature on networks @xcite . in this null model , a vertex could be attached to any other vertex of the graph and the probability that vertices @xmath67 and @xmath68 , with degrees @xmath105 and @xmath106 , are connected , can be calculated without problems . in fact , in order to form an edge between @xmath67 and @xmath68 one needs to join two _ stubs _ ( i. e. half - edges ) , incident with @xmath67 and @xmath68 . the probability @xmath107 to pick at random a stub incident with @xmath67 is @xmath108 , as there are @xmath105 stubs incident with @xmath67 out of a total of @xmath109 . the probability of a connection between @xmath67 and @xmath68 is then given by the product @xmath110 , since edges are placed independently of each other . the result is @xmath111 , which yields an expected number @xmath112 of edges between @xmath67 and @xmath68 . so , the final expression of modularity reads @xmath113 since the only contributions to the sum come from vertex pairs belonging to the same cluster , we can group these contributions together and rewrite the sum over the vertex pairs as a sum over the clusters @xmath114 . \label{eq : mod1}\ ] ] here , @xmath115 is the number of clusters , @xmath116 the total number of edges joining vertices of module @xmath117 and @xmath118 the sum of the degrees of the vertices of @xmath117 . in eq . [ eq : mod1 ] , the first term of each summand is the fraction of edges of the graph inside the module , whereas the second term represents the expected fraction of edges that would be there if the graph were a random graph with the same expected degree for each vertex . a nice feature of modularity is that it can be equivalently expressed both in terms of the intra - cluster edges , as in eq . [ eq : mod1 ] , and in terms of the inter - cluster edges @xcite . in fact , the maximum of modularity can be expressed as @xmath119\right\}\\ & = & \nonumber \frac{1}{m}\mbox{max}_{\cal p}\left\{\sum_{c=1}^{n_c}\big[l_c-\mbox{ex}(l_c)\big]\right\}\\ & = & -\frac{1}{m}\mbox{min}_{\cal p}\left\{-\sum_{c=1}^{n_c}\big[l_c-\mbox{ex}(l_c)\big]\right\ } , \label{eqr11}\end{aligned}\ ] ] where @xmath120 and @xmath121 indicates the maximum and the minimum over all possible graph partitions @xmath94 and @xmath122 indicates the expected number of links in cluster @xmath117 in the null model of modularity . by adding and subtracting the total number of edges @xmath14 of the graph one finally gets @xmath123\\ & = & -\frac{1}{m}\mbox{min}_{\cal p}(|\mbox{cut}_{\cal p}|-\mbox{excut}_{\cal p } ) . \label{eqr12}\end{aligned}\ ] ] in the last expression @xmath124 is the number of inter - cluster edges of partition @xmath94 , and @xmath125 is the expected number of inter - cluster edges of the partition in modularity s null model . according to eq . [ eq : mod1 ] , a subgraph is a module if the corresponding contribution to modularity in the sum is positive . the more the number of internal edges of the cluster exceeds the expected number , the better defined the community . so , large positive values of the modularity indicate good partitions . ] . the maximum modularity of a graph generally grows if the size of the graph and/or the number of ( well - separated ) clusters increase @xcite . therefore , modularity should not be used to compare the quality of the community structure of graphs which are very different in size . the modularity of the whole graph , taken as a single community , is zero , as the two terms of the only summand in this case are equal and opposite . modularity is always smaller than one , and can be negative as well . for instance , the partition in which each vertex is a community is always negative : in this case the sum runs over @xmath15 terms , which are all negative as the first term of each summand is zero . this is a nice feature of the measure , implying that , if there are no partitions with positive modularity , the graph has no community structure . on the contrary , the existence of partitions with large negative modularity values may hint to the existence of subgroups with very few internal edges and many edges lying between them ( _ multipartite structure _ ) @xcite . holmstrm et al . @xcite have shown that the distribution of modularity values across the partitions of various graphs , real and artificial ( including random graphs with no apparent community structure ) , has some stable features , and that the most likely modularity values correspond to partitions in clusters of approximately equal size . modularity has been employed as quality function in many algorithms , like some of the divisive algorithms of section [ sec5 ] . in addition , modularity optimization is itself a popular method for community detection ( see section [ sub_sec6_0 ] ) . modularity also allows to assess the stability of partitions @xcite ( section [ sec6_4 ] ) , it can be used to design layouts for graph visualization @xcite and to perform a sort of renormalization of a graph , by transforming a graph into a smaller one with the same community structure @xcite . the problem of graph partitioning consists in dividing the vertices in @xmath126 groups of predefined size , such that the number of edges lying between the groups is minimal . the number of edges running between clusters is called _ cut size_. fig . [ figure5 ] presents the solution of the problem for a graph with fourteen vertices , for @xmath127 and clusters of equal size . graph partitioning . the dashed line shows the solution of the minimum bisection problem for the graph illustrated , i. e. the partition in two groups of equal size with minimal number of edges running between the groups . reprinted figure with permission from ref . 2009 by springer.,width=264 ] specifying the number of clusters of the partition is necessary . if one simply imposed a partition with the minimal cut size , and left the number of clusters free , the solution would be trivial , corresponding to all vertices ending up in the same cluster , as this would yield a vanishing cut size . specifying the size is also necessary , as otherwise the most likely solution of the problem would consist in separating the lowest degree vertex from the rest of the graph , which is quite uninteresting . this problem can be actually avoided by choosing a different measure to optimize for the partitioning , which accounts for the size of the clusters . some of these measures will be briefly introduced at the end of this section . graph partitioning is a fundamental issue in parallel computing , circuit partitioning and layout , and in the design of many serial algorithms , including techniques to solve partial differential equations and sparse linear systems of equations . most variants of the graph partitioning problem are np - hard . there are however several algorithms that can do a good job , even if their solutions are not necessarily optimal @xcite . many algorithms perform a bisection of the graph . partitions into more than two clusters are usually attained by iterative bisectioning . moreover , in most cases one imposes the constraint that the clusters have equal size . this problem is called _ minimum bisection _ and is np - hard . the _ kernighan - lin algorithm _ @xcite is one of the earliest methods proposed and is still frequently used , often in combination with other techniques . the authors were motivated by the problem of partitioning electronic circuits onto boards : the nodes contained in different boards need to be linked to each other with the least number of connections . the procedure is an optimization of a benefit function @xmath90 , which represents the difference between the number of edges inside the modules and the number of edges lying between them . the starting point is an initial partition of the graph in two clusters of the predefined size : such initial partition can be random or suggested by some information on the graph structure . then , subsets consisting of equal numbers of vertices are swapped between the two groups , so that @xmath90 has the maximal increase . the subsets can consist of single vertices . to reduce the risk to be trapped in local maxima of @xmath90 , the procedure includes some swaps that decrease the function @xmath90 . after a series of swaps with positive and negative gains , the partition with the largest value of @xmath90 is selected and used as starting point of a new series of iterations . the kernighan - lin algorithm is quite fast , scaling as @xmath128 ( @xmath15 being as usual the number of vertices ) , if only a constant number of swaps are performed at each iteration . the most expensive part is the identification of the subsets to swap , which requires the computation of the gains / losses for any pair of candidate subsets . on sparse graphs , a slightly different heuristic allows to lower the complexity to @xmath129 . the partitions found by the procedure are strongly dependent on the initial configuration and other algorithms can do better . it is preferable to start with a good guess about the sought partition , otherwise the results are quite poor . therefore the method is typically used to improve on the partitions found through other techniques , by using them as starting configurations for the algorithm . the kernighan - lin algorithm has been extended to extract partitions in any number of parts @xcite , however the run - time and storage costs increase rapidly with the number of clusters . another popular technique is the _ spectral bisection method _ @xcite , which is based on the properties of the spectrum of the laplacian matrix . spectral clustering will be discussed more thoroughly in section [ sec4_4 ] , here we focus on its application to graph partitioning . every partition of a graph with @xmath15 vertices in two groups can be represented by an index vector @xmath130 , whose component @xmath131 is @xmath132 if vertex @xmath67 is in one group and @xmath133 if it is in the other group . the cut size @xmath134 of the partition of the graph in the two groups can be written as @xmath135 where @xmath136 is the laplacian matrix and @xmath137 the transpose of vector @xmath130 . vector @xmath130 can be written as @xmath138 , where @xmath139 , @xmath140 are the eigenvectors of the laplacian . if @xmath130 is properly normalized , then @xmath141 where @xmath142 is the laplacian eigenvalue corresponding to eigenvector @xmath139 . it is worth remarking that the sum contains at most @xmath143 terms , as the laplacian has at least one zero eigenvalue . minimizing @xmath134 equals to the minimization of the sum on the right - hand side of eq . this task is still very hard . however , if the second lowest eigenvector @xmath144 is close enough to zero , a good approximation of the minimum can be attained by choosing @xmath130 parallel to the corresponding eigenvector @xmath145 , which is called _ fiedler vector _ @xcite : this would reduce the sum to @xmath144 , which is a small number . but the index vector can not be perfectly parallel to @xmath145 by construction , because all its components are equal in modulus , whereas the components of @xmath145 are not . the best choice is to match the signs of the components . so , one can set @xmath146 ( @xmath133 ) if @xmath147 ( @xmath148 ) . it may happen that the sizes of the two corresponding groups do not match the predefined sizes one wishes to have . in this case , if one aims at a split in @xmath149 and @xmath150 vertices , the best strategy is to order the components of the fiedler vector from the lowest to the largest values and to put in one group the vertices corresponding to the first @xmath149 components from the top or the bottom , and the remaining vertices in the second group . this procedure yields two partitions : the better solution is naturally the one that gives the smaller cut size . the spectral bisection method is quite fast . the first eigenvectors of the laplacian can be computed by using the lanczos method @xcite . the time required to compute the first @xmath44 eigenvectors of a matrix with the lanczos method depends on the size of the eigengap @xmath151 @xcite . if the eigenvalues @xmath152 and @xmath153 are well separated , the running time of the algorithm is much shorter than the time required to calculate the complete set of eigenvectors , which scales as @xmath154 . the method gives in general good partitions , that can be further improved by applying the kernighan - lin algorithm . the well known max - flow min - cut theorem by ford and fulkerson @xcite states that the minimum cut between any two vertices @xmath155 and @xmath156 of a graph , i. e. any minimal subset of edges whose deletion would topologically separate @xmath155 from @xmath156 , carries the maximum flow that can be transported from @xmath155 to @xmath156 across the graph . in this context edges play the role of water pipes , with a given carrying capacity ( e.g. their weights ) , and vertices the role of pipe junctions . this theorem has been used to determine minimal cuts from maximal flows in clustering algorithms . there are several efficient routines to compute maximum flows in graphs , like the algorithm of goldberg and tarjan @xcite . flake et al . @xcite have recently used maximum flows to identify communities in the graph of the world wide web . the web graph is directed but for the purposes of the calculation flake et at . treated the edges as undirected . web communities are defined to be `` strong '' ( ls - sets ) , i. e. the internal degree of each vertex must not be smaller than its external degree @xcite . an artificial sink @xmath156 is added to the graph and one calculates the maximum flows from a source vertex @xmath155 to the sink @xmath156 : the corresponding minimum cut identifies the community of vertex @xmath155 , provided @xmath155 shares a sufficiently large number of edges with the other vertices of its community , otherwise one could get trivial separations and meaningless clusters . other popular methods for graph partitioning include level - structure partitioning , the geometric algorithm , multilevel algorithms , etc . a good description of these algorithms can be found in ref . @xcite . graphs can be also partitioned by minimizing measures that are affine to the cut size , like _ conductance _ the conductance @xmath157 of the subgraph @xmath22 of a graph @xmath23 is defined as @xmath158 where @xmath159 is the cut size of @xmath22 , and @xmath160 , @xmath161 are the total degrees of @xmath22 and of the rest of the graph @xmath162 , respectively . cuts are defined only between non - empty sets , otherwise the measure would not be defined ( as the denominator in eq . [ eq3000 ] would vanish ) . the minimum of the conductance is obtained in correspondence of low values of the cut size and of large values for the denominator in eq . [ eq3000 ] , which peaks when the total degrees of the cluster and its complement are equal . in practical applications , especially on large graphs , close values of the total degrees correspond to clusters of approximately equal size . the problem of finding a cut with minimal conductance is * np*-hard @xcite . similar measures are the _ ratio cut _ @xcite and the _ normalized cut _ @xcite . the ratio cut of a cluster @xmath22 is defined as @xmath163 where @xmath164 and @xmath165 are the number of vertices of the two subgraphs . the normalized cut of a cluster @xmath22 is @xmath166 where @xmath160 is again the total degree of @xmath22 . as for the conductance , minimizing the ratio cut and the normalized cut favors partitions into clusters of approximately equal size , measured in terms of the number of vertices or edges , respectively . on the other hand , graph partitioning requires preliminary assumptions on the cluster sizes , whereas the minimization of conductance , ratio cut and normalized cut does not . the ratio cut was introduced for circuit partitioning @xcite and its optimization is an * np*-hard problem @xcite . the normalized cut is frequently used in image segmentation @xcite and its optimization is * np*-complete @xcite . the cut ratio and the normalized cut can be quite well minimized via spectral clustering @xcite ( section [ sec4_4 ] ) . algorithms for graph partitioning are not good for community detection , because it is necessary to provide as input the number of groups and in some cases even their sizes , about which in principle one knows nothing . instead , one would like an algorithm capable to produce this information in its output . besides , from the methodological point of view , using iterative bisectioning to split the graph in more pieces is not a reliable procedure . for instance , a split into three clusters is necessarily obtained by breaking either cluster of the original bipartition in two parts , whereas in many cases a minimum cut partition is obtained if the third cluster is a merger of parts of both initial clusters . in general , very little is known about the community structure of a graph . it is uncommon to know the number of clusters in which the graph is split , or other indications about the membership of the vertices . in such cases clustering procedures like graph partitioning methods can hardly be of help , and one is forced to make some reasonable assumptions about the number and size of the clusters , which are often unjustified . on the other hand , the graph may have a hierarchical structure , i. e. may display several levels of grouping of the vertices , with small clusters included within large clusters , which are in turn included in larger clusters , and so on . social networks , for instance , often have a hierarchical structure ( section [ sec3_2_1 ] ) . in such cases , one may use _ hierarchical clustering algorithms _ @xcite , i. e. clustering techniques that reveal the multilevel structure of the graph . hierarchical clustering is very common in social network analysis , biology , engineering , marketing , etc . the starting point of any hierarchical clustering method is the definition of a similarity measure between vertices . after a measure is chosen , one computes the similarity for each pair of vertices , no matter if they are connected or not . at the end of this process , one is left with a new @xmath167 matrix @xmath168 , the similarity matrix . in section [ sec3_1_4 ] we have listed several possible definitions of similarity . hierarchical clustering techniques aim at identifying groups of vertices with high similarity , and can be classified in two categories : 1 . _ agglomerative algorithms _ , in which clusters are iteratively merged if their similarity is sufficiently high ; 2 . _ divisive algorithms _ , in which clusters are iteratively split by removing edges connecting vertices with low similarity . the two classes refer to opposite processes : agglomerative algorithms are bottom - up , as one starts from the vertices as separate clusters ( singletons ) and ends up with the graph as a unique cluster ; divisive algorithms are top - down as they follow the opposite direction . divisive techniques have been rarely used in the past ( meanwhile they have become more popular , see section [ sec5 ] ) , so we shall concentrate here on agglomerative algorithms . since clusters are merged based on their mutual similarity , it is essential to define a measure that estimates how similar clusters are , out of the matrix @xmath168 . this involves some arbitrariness and several prescriptions exist . in _ single linkage clustering _ , the similarity between two groups is the minimum element @xmath169 , with @xmath67 in one group and @xmath68 in the other . on the contrary , the maximum element @xmath169 for vertices of different groups is used in the procedure of _ complete linkage clustering_. in _ average linkage clustering _ one has to compute the average of the @xmath169 . the procedure can be better illustrated by means of dendrograms ( section [ sec3_2_1 ] ) , like the one in fig . [ fig9 ] . sometimes , stopping conditions are imposed to select a partition or a group of partitions that satisfy a special criterion , like a given number of clusters or the optimization of a quality function ( e.g. modularity ) . hierarchical clustering has the advantage that it does not require a preliminary knowledge on the number and size of the clusters . however , it does not provide a way to discriminate between the many partitions obtained by the procedure , and to choose that or those that better represent the community structure of the graph . the results of the method depend on the specific similarity measure adopted . the procedure also yields a hierarchical structure by construction , which is rather artificial in most cases , since the graph at hand may not have a hierarchical structure at all . moreover , vertices of a community may not be correctly classified , and in many cases some vertices are missed even if they have a central role in their clusters @xcite . another problem is that vertices with just one neighbor are often classified as separated clusters , which in most cases does not make sense . finally , a major weakness of agglomerative hierarchical clustering is that it does not scale well . if points are embedded in space , so that one can use the distance as dissimilarity measure , the computational complexity is @xmath129 for single linkage , @xmath128 for the complete and average linkage schemes . for graph clustering , where a distance is not trivially defined , the complexity can become much heavier if the calculation of the chosen similarity measure is costly . _ partitional clustering _ indicates another popular class of methods to find clusters in a set of data points . here , the number of clusters is preassigned , say @xmath44 . the points are embedded in a metric space , so that each vertex is a point and a distance measure is defined between pairs of points in the space . the distance is a measure of dissimilarity between vertices . the goal is to separate the points in @xmath44 clusters such to maximize / minimize a given cost function based on distances between points and/or from points to _ centroids _ , i. e. suitably defined positions in space . some of the most used functions are listed below : * _ minimum k - clustering_. the cost function here is the _ diameter _ of a cluster , which is the largest distance between two points of a cluster . the points are classified such that the largest of the @xmath44 cluster diameters is the smallest possible . the idea is to keep the clusters very `` compact '' . * _ k - clustering sum_. same as minimum @xmath44-clustering , but the diameter is replaced by the average distance between all pairs of points of a cluster . * _ k - center_. for each cluster @xmath67 one defines a reference point @xmath170 , the centroid , and computes the maximum @xmath171 of the distances of each cluster point from the centroid . the clusters and centroids are self - consistently chosen such to minimize the largest value of @xmath171 . * _ k - median_. same as k - center , but the maximum distance from the centroid is replaced by the average distance . the most popular partitional technique in the literature is _ @xmath44-means clustering _ @xcite . here the cost function is the total intra - cluster distance , or squared error function @xmath172 where @xmath173 indicates the subset of points of the @xmath67-th cluster and @xmath174 its centroid . the @xmath44-means problem can be simply solved with the lloyd s algorithm @xcite . one starts from an initial distribution of centroids such that they are as far as possible from each other . in the first iteration , each vertex is assigned to the nearest centroid . next , the centers of mass of the @xmath44 clusters are estimated and become a new set of centroids , which allows for a new classification of the vertices , and so on . after a small number of iterations , the positions of the centroids are stable , and the clusters do not change any more . the solution found is not optimal , and it strongly depends on the initial choice of the centroids . nevertheless , lloyd s heuristic has remained popular due to its quick convergence , which makes it suitable for the analysis of large data sets . the result can be improved by performing more runs starting from different initial conditions , and picking the solution which yields the minimum value of the total intra - cluster distance . extensions of @xmath44-means clustering to graphs have been proposed by some authors @xcite . another popular technique , similar in spirit to @xmath44-means clustering , is _ fuzzy k - means clustering _ @xcite . this method accounts for the fact that a point may belong to two or more clusters at the same time and is widely used in pattern recognition . the associated cost function is @xmath175 where @xmath176 is the _ membership matrix _ , which measures the degree of membership of point @xmath67 ( with position @xmath177 ) in cluster @xmath68 , @xmath14 is a real number greater than @xmath2 and @xmath178 is the center of cluster @xmath68 @xmath179 the matrix @xmath176 is normalized so that the sum of the memberships of every point in all clusters yields @xmath2 . the membership @xmath176 is related to the distance of point @xmath67 from the center of cluster @xmath68 , as it is reasonable to assume that the larger this distance , the lower @xmath176 . this can be expressed by the following relation @xmath180 the cost function @xmath181 can be minimized by iterating eqs . [ eq3004 ] and [ eq3005 ] . one starts from some initial guess for @xmath176 and uses eq . [ eq3004 ] to compute the centers , which are then plugged back into eqs . [ eq3005 ] , and so on . the process stops when the corresponding elements of the membership matrix in consecutive iterations differ from each other by less than a predefined tolerance . it can be shown that this procedure indeed delivers a local minimum of the cost function @xmath181 of eq . [ eq3003 ] . this procedure has the same problems of lloyd s algorithm for @xmath44-means clustering , i. e. the minimum is a local minimum , and depends on the initial choice of the matrix @xmath176 . the limitation of partitional clustering is the same as that of the graph partitioning algorithms : the number of clusters must be specified at the beginning , the method is not able to derive it . in addition , the embedding in a metric space can be natural for some graphs , but rather artificial for others . let us suppose to have a set of @xmath15 objects @xmath182 with a pairwise similarity function @xmath87 defined between them , which is symmetric and non - negative ( i. e. , @xmath183 , @xmath184 ) . _ spectral clustering _ includes all methods and techniques that partition the set into clusters by using the eigenvectors of matrices , like @xmath87 itself or other matrices derived from it . in particular , the objects could be points in some metric space , or the vertices of a graph . spectral clustering consists of a transformation of the initial set of objects into a set of points in space , whose coordinates are elements of eigenvectors : the set of points is then clustered via standard techniques , like @xmath44-means clustering ( section [ sec4_3 ] ) . one may wonder why it is necessary to cluster the points obtained through the eigenvectors , when one can directly cluster the initial set of objects , based on the similarity matrix . the reason is that the change of representation induced by the eigenvectors makes the cluster properties of the initial data set much more evident . in this way , spectral clustering is able to separate data points that could not be resolved by applying directly @xmath44-means clustering , for instance , as the latter tends to deliver convex sets of points . the first contribution on spectral clustering was a paper by donath and hoffmann @xcite , who used the eigenvectors of the adjacency matrix for graph partitions . in the same year , fiedler @xcite realized that from the eigenvector of the second smallest eigenvalue of the laplacian matrix it was possible to obtain a bipartition of the graph with very low cut size , as we have explained in section [ sec4_1 ] . for a historical survey see ref . @xcite . in this section we shall follow the nice tutorial by von luxburg @xcite , with a focus on spectral graph clustering . the concepts and methods discussed below apply to both unweighted and weighted graphs . the laplacian is by far the most used matrix in spectral clustering . in section [ sec1_2 ] we see that the unnormalized laplacian of a graph with @xmath44 connected components has @xmath44 zero eigenvalues . in this case the laplacian can be written in block - diagonal form , i. e. the vertices can be ordered in such a way that the laplacian displays @xmath44 square blocks along the diagonal , with ( some ) entries different from zero , whereas all other elements vanish . each block is the laplacian of the corresponding subgraph , so it has the trivial eigenvector with components @xmath185 . therefore , there are @xmath44 degenerate eigenvectors with equal non - vanishing components in correspondence of the vertices of a block , whereas all other components are zero . in this way , from the components of the eigenvectors one can identify the connected components of the graph . for instance , let us consider the @xmath186 matrix , whose columns are the @xmath44 eigenvectors above mentioned . the @xmath67-th row of this matrix is a vector with @xmath44 components representing vertex @xmath67 of the graph . vectors representing vertices in the same connected component of the graph coincide , and their tip lies on one of the axes of a @xmath44-dimensional system of coordinates ( i. e. they are all vectors of the form @xmath187 ) . so , by drawing the vertex vectors one would see @xmath44 distinct points , each on a different axis , corresponding to the graph components . if the graph is connected , but consists of @xmath44 subgraphs which are weakly linked to each other , the spectrum of the unnormalized laplacian will have one zero eigenvalue , all others being positive . now the laplacian can not be put in block - diagonal form : even if one enumerates the vertices in the order of their cluster memberships ( by listing first the vertices of one cluster , then the vertices of another cluster , etc . ) there will always be some non - vanishing entries outside of the blocks . however , the lowest @xmath188 non - vanishing eigenvalues are still close to zero , and the vertex vectors of the first @xmath44 eigenvectors should still enable one to clearly distinguish the clusters in a @xmath44-dimensional space . vertex vectors corresponding to the same cluster are now not coincident , in general , but still rather close to each other . so , instead of @xmath44 points , one would observe @xmath44 groups of points , with the points of each group localized close to each other and far from the other groups . techniques like @xmath44-means clustering ( section [ sec4_3 ] ) can then easily recover the clusters . the scenario we have described is expected from perturbation theory @xcite . in principle all symmetric matrices that can be put in block - diagonal form have a set of eigenvectors ( as many as the blocks ) , such that the elements of each eigenvector are different from zero on the vertices of a block and zero otherwise , just like the laplacian . the adjacency matrix itself has the same property , for example . this is a necessary condition for the eigenvectors to be successfully used for graph clustering , but it is not sufficient . in the case of the laplacian , for a graph with @xmath44 connected components , we know that the eigenvectors corresponding to the @xmath44 lowest eigenvalues come each from one of the components . in the case of the adjacency matrix @xmath189 ( or of its weighted counterpart @xmath190 ) , instead , it may happen that large eigenvalues refer to the same component . so , if one takes the eigenvectors corresponding to the @xmath44 largest eigenvalues , where @xmath191 is the diagonal matrix whose elements are the vertex degrees . ] , some components will be overrepresented , while others will be absent . therefore , using the eigenvectors of @xmath189 ( or @xmath190 ) in spectral graph clustering is in general not reliable . moreover , the elements of the eigenvectors corresponding to the components should be sufficiently far from zero . to understand why , suppose that we take a ( symmetric , block - diagonal ) matrix , and that one or more elements of one of the eigenvectors corresponding to the connected components are very close to zero . if one perturbs the graph by adding edges between different components , all entries of the perturbed eigenvectors will become non - zero and some may have comparable values as the lowest elements of the eigenvectors on the blocks . therefore distinguishing vertices of different components may become a problem , even when the perturbation is fairly small , and misclassifications are likely . on the other hand , the non - vanishing elements of the ( normalized ) eigenvectors of the unnormalized laplacian , for instance , are all equal to @xmath192 , where @xmath193 is the number of vertices in the @xmath67-th component . in this way , there is a gap between the lowest element ( here they are all equal for the same eigenvector ) and zero . this holds as well for the normalized laplacian @xmath194 ( section [ sec1_2 ] ) . for the other normalized laplacian @xmath195 ( section [ sec1_2 ] ) , the non - zero elements of the eigenvectors corresponding to the connected components are proportional to the square root of the degree of the corresponding vertex so , if degrees are very different from each other , and especially if there are vertices with very low degree , some eigenvector elements may be quite small . as we shall see below , in the context of the technique by ng et al . @xcite , a suitable normalization procedure is introduced to alleviate this problem . now that we have explained why the laplacian matrix is particularly suitable for spectral clustering , we proceed with the description of three popular methods : _ unnormalized spectral clustering _ and two _ normalized spectral clustering _ techniques , proposed by shi and malik @xcite and by ng et al . @xcite , respectively . unnormalized spectral clustering uses the unnormalized laplacian @xmath136 . the inputs are the adjacency matrix @xmath189 ( @xmath190 for weighted graphs ) and the number @xmath44 of clusters to be recovered . the first step consists in computing the eigenvectors corresponding to the lowest @xmath44 eigenvalues of @xmath136 . then , one builds the @xmath186 matrix @xmath196 , whose columns are the @xmath44 eigenvectors . the @xmath15 rows of @xmath196 are used to represent the graph vertices in a @xmath44-dimensional euclidean space , through a cartesian system of coordinates . the points are then grouped in @xmath44 clusters by using @xmath44-means clustering or similar techniques ( section [ sec4_3 ] ) . normalized spectral clustering works in the same way . in the version by shi and malik @xcite , one uses the eigenvectors of the normalized laplacian @xmath194 ( section [ sec1_2 ] ) . in the algorithm by ng et al . @xcite one adopts the normalized laplacian @xmath195 ( section [ sec1_2 ] ) . here , however , the matrix @xmath196 is normalized by dividing the elements of each row by their sum , obtaining a new matrix @xmath197 , whose rows are then used to represent the vertices in space , as in the other methods . by doing so , it is much more unlikely that eigenvector components for a well - separated cluster are close to zero , a scenario which would make the classification of the corresponding vertices problematic , as we have said above . however , if the graph has some vertices with low degree , they may still be misclassified . spectral clustering is closely related to graph partitioning . relaxed versions of the minimization of ratio cut and normalized cut ( see section [ sec4_1 ] ) can be turned into spectral clustering problems , by following similar procedures as in spectral graph partitioning . the measure to minimize can be expressed in matrix form , obtaining similar expressions as for the cut size ( see eq . [ eqr ] ) , with index vectors defining the partition of the graph in groups through the values of their entries . for instance , for the minimum cut bipartition of section [ sec4_1 ] , there is only one index vector @xmath130 , whose components equal @xmath198 , where the signs indicate the two clusters . the relaxation consists in performing the minimization over all possible vectors @xmath130 , allowing for real - valued components as well . this version of the problem is exactly equivalent to spectral clustering . the relaxed minimization of ratio cut for a partition in @xmath44 clusters yields the @xmath15 @xmath44-dimensional vertex vectors of unnormalized spectral clustering @xcite ; for normalized cut one obtains the @xmath15 @xmath44-dimensional vertex vectors of normalized spectral clustering , with the normalized laplacian @xmath194 @xcite . the problem is then to turn the resulting vectors into a partition of the graph , which can be done by using techniques like @xmath44-means clustering , as we have seen above . however , it is still unclear what is the relation between the original minimum cut problem over actual graph partitions and the relaxed version of it , in particular how close one can come to the real solution via spectral clustering . random walks on graphs are also related to spectral clustering . in fact , by minimizing the number of edges between clusters ( properly normalized for measures like , e. g. , ratio cut and normalized cut ) one forces random walkers to spend more time within clusters and to move more rarely from one cluster to another . in particular , unnormalized spectral clustering with the laplacian @xmath194 has a natural link with random walks , because @xmath199 ( section [ sec1_2 ] ) , where @xmath200 is the transfer matrix @xmath201 . this has interesting consequences . for instance , meil and shi have proven that the normalized cut for a bipartition equals the total probability that a random walker moves from one of the clusters to the other in either sense @xcite . in this way , minimizing the normalized cut means looking for a partition minimizing the probability of transitions between clusters . spectral clustering requires the computation of the first @xmath44 eigenvectors of a laplacian matrix . if the graph is large , an exact computation of the eigenvectors is impossible , as it would require a time @xmath154 . fortunately there are approximate techniques , like the power method or krylov subspace techniques like the lanczos method @xcite , whose speed depends on the size of the eigengap @xmath151 , where @xmath202 and @xmath152 are the @xmath44-th and @xmath203-th smallest eigenvalue of the matrix . the larger the eigengap , the faster the convergence . in fact , the existence of large gaps between pairs of consecutive eigenvalues could suggest the number of clusters of the graph , an information which is not delivered by spectral clustering and which has to be given as input . we know that , for a disconnected graph with @xmath44 components , the first @xmath44 eigenvalues of the laplacian matrix ( normalized or not ) are zero , whether the @xmath203-th is non - zero . if the clusters are weakly connected to each other , one expects that the first @xmath44 eigenvalues remain close to zero , and that the @xmath203-th is clearly different from zero . by reversing this argument , the number of clusters of a graph could be derived by checking whether there is an integer @xmath44 such that the first @xmath44 eigenvalues are small and the @xmath203-th is relatively large . however , when the clusters are very mixed with each other , it may be hard to identify significant gaps between the eigenvalues . the last issue we want to point out concerns the choice of the laplacian matrix to use in the applications . if the graph vertices have the same or similar degrees , there is no substantial difference between the unnormalized and the normalized laplacians . if there are big inhomogeneities among the vertex degrees , instead , the choice of the laplacian considerably affects the results . in general , normalized laplacians are more promising because the corresponding spectral clustering techniques implicitly impose a double optimization on the set of partitions , such that the intracluster edge density is high and , at the same time , the intercluster edge density is low . on the contrary , the unnormalized laplacian is related to the intercluster edge density only . moreover , unnormalized spectral clustering does not always converge , and sometimes yields trivial partitions in which one or more clusters consist of a single vertex . of the normalized laplacians , @xmath194 is more reliable than @xmath195 because the eigenvectors of @xmath194 corresponding to the lowest eigenvalues are cluster indicator vectors , i. e. , they have equal non - vanishing entries in correspondence of the vertices of each cluster , and zero elsewhere , if the clusters are disconnected . the eigenvectors of @xmath195 , instead , are obtained by ( left- ) multiplying those of @xmath194 by the matrix @xmath204 : in this way , eigenvector components corresponding to vertices of the same cluster are no longer equal , in general , a complication that may induce artefacts in the spectral clustering procedure . a simple way to identify communities in a graph is to detect the edges that connect vertices of different communities and remove them , so that the clusters get disconnected from each other . this is the philosophy of divisive algorithms . the crucial point is to find a property of intercommunity edges that could allow for their identification . divisive methods do not introduce substantial conceptual advances with respect to traditional techniques , as they just perform hierarchical clustering on the graph at study ( section [ sec4_2 ] ) . the main difference with divisive hierarchical clustering is that here one removes inter - cluster edges instead of edges between pairs of vertices with low similarity and there is no guarantee _ a priori _ that inter - cluster edges connect vertices with low similarity . in some cases vertices ( with all their adjacent edges ) or whole subgraphs may be removed , instead of single edges . being hierarchical clustering techniques , it is customary to represent the resulting partitions by means of dendrograms . the most popular algorithm is that proposed by girvan and newman @xcite . the method is historically important , because it marked the beginning of a new era in the field of community detection and opened this topic to physicists . here edges are selected according to the values of measures of _ edge centrality _ , estimating the importance of edges according to some property or process running on the graph . the steps of the algorithm are : 1 . computation of the centrality for all edges ; 2 . removal of edge with largest centrality : in case of ties with other edges , one of them is picked at random ; 3 . recalculation of centralities on the running graph ; 4 . iteration of the cycle from step 2 . girvan and newman focused on the concept of _ betweenness _ , which is a variable expressing the frequency of the participation of edges to a process . they considered three alternative definitions : geodesic edge betweenness , random - walk edge betweenness and current - flow edge betweenness . in the following we shall refer to them as edge betweenness , random - walk betweenness and current - flow betweenness , respectively . edge betweenness is the number of shortest paths between all vertex pairs that run along the edge . it is an extension to edges of the popular concept of site betweenness , introduced by freeman in 1977 @xcite and expresses the importance of edges in processes like information spreading , where information usually flows through shortest paths . historically edge betweenness was introduced before site betweenness in a never published technical report by anthonisse @xcite . it is intuitive that intercommunity edges have a large value of the edge betweenness , because many shortest paths connecting vertices of different communities will pass through them ( fig . [ figure6 ] ) . as in the calculation of site betweenness , if there are two or more geodesic paths with the same endpoints that run through an edge , the contribution of each of them to the betweenness of the edge must be divided by the multiplicity of the paths , as one assumes that the signal / information propagates equally along each geodesic path . edge betweenness is highest for edges connecting communities . in the figure , the edge in the middle has a much higher betweenness than all other edges , because all shortest paths connecting vertices of the two communities run through it . reprinted figure with permission from ref . 2009 by springer . ] the betweenness of all edges of the graph can be calculated in a time that scales as @xmath205 , or @xmath129 on a sparse graph , with techniques based on breadth - first - search @xcite . in the context of information spreading , one could imagine that signals flow across random rather than geodesic paths . in this case the betweenness of an edge is given by the frequency of the passages across the edge of a random walker running on the graph ( random - walk betweenness ) . a random walker moving from a vertex follows each adjacent edge with equal probability . a pair of vertices is chosen at random , @xmath155 and @xmath156 . the walker starts at @xmath155 and keeps moving until it hits @xmath156 , where it stops . one computes the probability that each edge was crossed by the walker , and averages over all possible choices for the vertices @xmath155 and @xmath156 . it is meaningful to compute the _ net _ crossing probability , which is proportional to the number of times the walk crossed the edge in one direction . in this way one neglects back and forth passages that are accidents of the random walk and tell nothing about the centrality of the edge . calculation of random - walk betweenness requires the inversion of an @xmath167 matrix ( once ) , followed by obtaining and averaging the flows for all pairs of nodes . the first task requires a time @xmath154 , the second @xmath206 , for a total complexity @xmath207 $ ] , or @xmath154 for a sparse matrix . the complete calculation requires a time @xmath154 on a sparse graph . current - flow betweenness is defined by considering the graph a resistor network , with edges having unit resistance . if a voltage difference is applied between any two vertices , each edge carries some amount of current , that can be calculated by solving kirchoff s equations . the procedure is repeated for all possible vertex pairs : the current - flow betweenness of an edge is the average value of the current carried by the edge . it is possible to show that this measure is equivalent to random - walk betweenness , as the voltage differences and the random walks net flows across the edges satisfy the same equations @xcite . therefore , the calculation of current - flow betweenness has the same complexity @xmath207 $ ] , or @xmath154 for a sparse graph . calculating edge betweenness is much faster than current - flow or random walk betweenness [ @xmath129 versus @xmath154 on sparse graphs ] . in addition , in practical applications the girvan - newman algorithm with edge betweenness gives better results than adopting the other centrality measures @xcite . numerical studies show that the recalculation step 3 of girvan - newman algorithm is essential to detect meaningful communities . this introduces an additional factor @xmath14 in the running time of the algorithm : consequently , the edge betweenness version scales as @xmath208 , or @xmath154 on a sparse graph . on graphs with strong community structure , that quickly break into communities , the recalculation step needs to be performed only within the connected component including the last removed edge ( or the two components bridged by it if the removal of the edge splits a subgraph ) , as the edge betweenness of all other edges remains the same . this can help saving some computer time , although it is impossible to give estimates of the gain since it depends on the specific graph at hand . nevertheless , the algorithm is quite slow , and applicable to sparse graphs with up to @xmath209 vertices , with current computational resources . in the original version of girvan - newman s algorithm @xcite , the authors had to deal with the whole hierarchy of partitions , as they had no procedure to say which partition is the best . in a successive refinement @xcite , they selected the partition with the largest value of modularity ( see section [ sec3_2_2 ] ) , a criterion that has been frequently used ever since . the method can be simply extended to the case of weighted graphs , by suitably generalizing the edge betweenness . the betweenness of a weighted edge equals the betweenness of the edge in the corresponding unweighted graph , divided by the weight of the edge @xcite . there have been countless applications of the girvan - newman method : the algorithm is now integrated in well known libraries of network analysis programs . tyler et al . proposed a modification of the girvan - newman algorithm , to improve the speed of the calculation @xcite . the gain in speed was required by the analysis of graphs of gene co - occurrences , which are too large to be analyzed by the algorithm of girvan and newman . algorithms computing site / edge betweenness start from any vertex , taken as center , and compute the contribution to betweenness from all paths originating at that vertex ; the procedure is then repeated for all vertices @xcite . tyler et al . proposed to calculate the contribution to edge betweenness only from a limited number of centers , chosen at random , deriving a sort of monte carlo estimate . numerical tests indicate that , for each connected subgraph , it suffices to pick a number of centers growing as the logarithm of the number of vertices of the component . for a given choice of the centers , the algorithm proceeds just like that of girvan and newman . the stopping criterion is different , though , as it does not require the calculation of modularity on the resulting partitions , but relies on a particular definition of community . according to such definition , a connected subgraph with @xmath210 vertices is a community if the edge betweenness of any of its edges does not exceed @xmath211 . indeed , if the subgraph consists of two parts connected by a single edge , the betweenness value of that edge would be greater than or equal to @xmath211 , with the equality holding only if one of the two parts consists of a single vertex . therefore , the condition on the betweenness of the edges would exclude such situations , although other types of cluster structures might still be compatible with it . in this way , in the method of tyler et al . , edges are removed until all connected components of the partition are `` communities '' in the sense explained above . the monte carlo sampling of the edge betweenness necessarily induces statistical errors . as a consequence , the partitions are in general different for different choices of the set of center vertices . however , the authors showed that , by repeating the calculation many times , the method gives good results on a network of gene co - occurrences @xcite , with a substantial gain of computer time . the technique has been also applied to a network of people corresponding via email @xcite . in practical examples , only vertices lying at the boundary between communities may not be clearly classified , and be assigned sometimes to a group , sometimes to another . this is actually a nice feature of the method , as it allows to identify overlaps between communities , as well as the degree of membership of overlapping vertices in the clusters they belong to . the algorithm of girvan and newman , which is deterministic , is unable to accomplish this . another fast version of the girvan - newman algorithm has been proposed by rattigan et al . @xcite . here , a quick approximation of the edge betweenness values is carried out by using a _ network structure index _ , which consists of a set of vertex annotations combined with a distance measure @xcite . basically one divides the graph into regions and computes the distances of every vertex from each region . in this way rattigan et al . showed that it is possible to lower the complexity of the algorithm to @xmath21 , by keeping a fair accuracy in the estimate of the edge betweenness values . this version of the girvan - newman algorithm gives good results on the benchmark graphs proposed by brandes et al . @xcite ( see also section [ sec6_1 ] ) , as well as on a collaboration network of actors and on a citation network . chen and yuan have pointed out that counting all possible shortest paths in the calculation of the edge betweenness may lead to unbalanced partitions , with communities of very different size , and proposed to count only _ non - redundant _ paths , i. e. paths whose endpoints are all different from each other : the resulting betweenness yields better results than standard edge betweenness for mixed clusters on the benchmark graphs of girvan and newman @xcite . holme et al . have used a modified version of the algorithm in which vertices , rather than edges , are removed @xcite . a centrality measure for the vertices , proportional to their site betweenness , and inversely proportional to their indegree , is chosen to identify boundary vertices , which are then iteratively removed with all their edges . this modification , applied to study the hierarchical organization of biochemical networks , is motivated by the need to account for reaction kinetic information , that simple site betweenness does not include . the indegree of a vertex is solely used because it indicates the number of substrates to a metabolic reaction involving that vertex ; for the purpose of clustering the graph is considered undirected , as usual . the algorithm of girvan and newman is unable to find overlapping communities , as each vertex is assigned to a single cluster . pinney and westhead have proposed a modification of the algorithm in which vertices can be split between communities @xcite . to do that , they also compute the betweenness of all vertices of the graph . unfortunately the values of edge and site betweenness can not be simply compared , due to their different normalization , but the authors remarked that the two endvertices of an inter - cluster edge should have similar betweenness values , as the shortest paths crossing one of them are likely to reach the other one as well through the edge . so they take the edge with largest betweenness and remove it only if the ratio of the betweenness values of its endvertices is between @xmath18 and @xmath212 , with @xmath213 . otherwise , the vertex with highest betweenness ( with all its adjacent edges ) is temporarily removed . when a subgraph is split by vertex or edge removal , all deleted vertices belonging to that subgraph are `` copied '' in each subcomponent , along with all their edges . gregory @xcite has proposed a similar approach , named conga ( cluster overlap newman - girvan algorithm ) , in which vertices are split among clusters if their site betweenness exceeds the maximum value of the betweenness of the edges . a vertex is split by assigning some of its edges to one of its duplicates , and the rest to the other . there are several possibilities to do that , gregory proposed to go for the split that yields the maximum of a new centrality measure , called _ split betweenness _ , which is the number of shortest paths that would run between two parts of a vertex if the latter were split . the method has a worst - case complexity @xmath214 , or @xmath154 on a sparse graph , like the algorithm of girvan and newman . the code can be found at http://www.cs.bris.ac.uk/@xmath215steve/networks/index . + html . another promising track to detect inter - cluster edges is related to the presence of cycles , i. e. closed non - intersecting paths , in the graph . communities are characterized by a high density of edges , so it is reasonable to expect that such edges form cycles . on the contrary , edges lying between communities will hardly be part of cycles . based on this intuitive idea , radicchi et al . proposed a new measure , the edge clustering coefficient , such that low values of the measure are likely to correspond to intercommunity edges @xcite . the edge clustering coefficient generalizes to edges the notion of clustering coefficient introduced by watts and strogatz for vertices @xcite ( fig . [ edgeclus ] ) . schematic illustration of the edge clustering coefficient introduced by radicchi et al . the two grey vertices have five and six other neighbors , respectively . of the five possible triangles based on the edge connecting the grey vertices , three are actually there , yielding an edge clustering coefficient @xmath216 . courtesy by f. radicchi . ] the clustering coefficient of a vertex is the number of triangles including the vertex divided by the number of possible triangles that can be formed ( section [ sec1_1 ] ) . the edge clustering coefficient is defined as @xmath217 where @xmath67 and @xmath68 are the extremes of the edge , @xmath218 the number of cycles of length @xmath126 built upon edge @xmath219 and @xmath220 the possible number of cycles of length @xmath126 that one could build based on the existing edges of @xmath67 , @xmath68 and their neighbors . the number of actual cycles in the numerator is augmented by @xmath2 to enable a ranking among edges without cycles , which would all yield a coefficient @xmath221 equal to zero , independently of the degrees of the extremes @xmath67 and @xmath68 and their neighbors . usually , cycles of length @xmath222 ( triangles ) or @xmath223 are considered . the measure is ( anti)correlated with edge betweenness : edges with low edge clustering coefficient usually have high betweenness and vice versa , although the correlation is not perfect . the method works as the algorithm by girvan and newman . at each iteration , the edge with smallest clustering coefficient is removed , the measure is recalculated again , and so on . if the removal of an edge leads to a split of a subgraph in two parts , the split is accepted only if both clusters are ls - sets ( `` strong '' ) or `` weak '' communities ( see section [ sec3_1_2 ] ) . the verification of the community condition on the clusters is performed on the full adjacency matrix of the initial graph . if the condition were satisfied only for one of the two clusters , the initial subgraph may be a random graph , as it can be easily seen that by cutting a random graph la erds and rnyi in two parts , the larger of them is a strong ( or weak ) community with very high probability , whereas the smaller part is not . enforcing the community condition on both clusters , it is more likely that the subgraph to be split indeed has a cluster structure . therefore , the algorithm stops when all clusters produced by the edge removals are communities in the strong or weak sense , and further splits would violate this condition . the authors suggested to use the same stopping criterion for the algorithm of girvan and newman , to get structurally well - defined clusters . since the edge clustering coefficient is a local measure , involving at most an extended neighborhood of the edge , it can be calculated very quickly . the running time of the algorithm to completion is @xmath224 , or @xmath129 on a sparse graph , if @xmath126 is small , so it is much shorter than the running time of the girvan - newman method . the recalculation step becomes slow if @xmath126 is not so small , as in this case the number of edges whose coefficient needs to be recalculated may reach a sizeable fraction of the edges of the graph ; likewise , counting the number of cycles based on one edge becomes lengthier . if @xmath225 , where @xmath88 is the diameter of the graph ( which is usually a small number for real networks ) , the cycles span the whole graph and the measure becomes global and no more local . the computational complexity in this case exceeds that of the algorithm of girvan and newman , but it can come close to it for practical purposes even at lower values of @xmath126 . so , by tuning @xmath126 one can smoothly interpolate between a local and a global centrality measure . the software of the algorithm can be found in http://filrad.homelinux.org / data/. in a successive paper @xcite the authors extended the method to the case of weighted networks , by modifying the edge clustering coefficient of eq . [ eq3010 ] , in that the number of cycles @xmath218 is multiplied by the weight of the edge @xmath219 . the definitions of strong and weak communities can be trivially extended to weighted graphs by replacing the internal / external degrees of the vertices / clusters with the corresponding strengths . more recently , the method has been extended to bipartite networks @xcite , where only cycles of even length are possible ( @xmath226 , @xmath227 , @xmath228 , etc . ) . the algorithm by radicchi et al . may give poor results when the graph has few cycles , as it happens in some social and many non - social networks . in this case , in fact , the edge clustering coefficient is small and fairly similar for most edges , and the algorithm may fail to identify the bridges between communities . an alternative measure of centrality for edges is information centrality . it is based on the concept of efficiency @xcite , which estimates how easily information travels on a graph according to the length of shortest paths between vertices . the efficiency of a network is defined as the average of the inverse distances between all pairs of vertices . if the vertices are `` close '' to each other , the efficiency is high . the information centrality of an edge is the relative variation of the efficiency of the graph if the edge is removed . in the algorithm by fortunato et al . @xcite , edges are removed according to decreasing values of information centrality . the method is analogous to that of girvan and newman . computing the information centrality of an edge requires the calculation of the distances between all pairs of vertices , which can be done with breadth - first - search in a time @xmath205 . so , in order to compute the information centrality of all edges one requires a time @xmath208 . at this point one removes the edge with the largest value of information centrality and recalculates the information centrality of all remaining edges with respect to the running graph . since the procedure is iterated until there are no more edges in the network , the final complexity is @xmath229 , or @xmath230 on a sparse graph . the partition with the largest value of modularity is chosen as most representative of the community structure of the graph . the method is much slower than the algorithm of girvan and newman . partitions obtained with both techniques are rather consistent , mainly because information centrality has a strong correlation with edge betweenness . the algorithm by fortunato et al . gives better results when communities are mixed , i. e. with a high degree of interconnectedness , but it tends to isolate leaf vertices and small loosely bound subgraphs . a measure of vertex centrality based on loops , similar to the clustering coefficient by watts and strogatz @xcite , has been introduced by vragovi and louis @xcite . the idea is that neighbors of a vertex well inside a community are `` close '' to each other , even in the absence of the vertex , due to the high density of intra - cluster edges . suppose that @xmath68 and @xmath44 are neighbors of a vertex @xmath67 : @xmath231 is the length of a shortest path between @xmath68 and @xmath44 , if @xmath67 is removed from the graph . naturally , the existence of alternative paths to @xmath232 implies the existence of loops in the graph . vragovi and louis defined the _ loop coefficient _ of @xmath67 as the average of @xmath233 over all pairs of neighbors of @xmath67 , somewhat reminding of the concept of information centrality used in the method by fortunato et al . high values of the loop coefficient are likely to identify core vertices of communities , whereas low values correspond to vertices lying at the boundary between communities . clusters are built around the vertices with highest values of the loop coefficient . the method has time complexity @xmath234 ; its results are not so accurate , as compared to popular clustering techniques . newman - girvan modularity @xmath90 ( section [ sec3_2_2 ] ) , originally introduced to define a stopping criterion for the algorithm of girvan and newman , has rapidly become an essential element of many clustering methods . modularity is by far the most used and best known quality function . it represented one of the first attempts to achieve a first principle understanding of the clustering problem , and it embeds in its compact form all essential ingredients and questions , from the definition of community , to the choice of a null model , to the expression of the `` strength '' of communities and partitions . in this section we shall focus on all clustering techniques that require modularity , directly and/or indirectly . we will examine fast techniques that can be used on large graphs , but which do not find good optima for the measure @xcite ; more accurate methods , which are computationally demanding @xcite ; algorithms giving a good tradeoff between high accuracy and low complexity @xcite . we shall also point out other properties of modularity , discuss some extensions / modifications of it , as well as highlight its limits . by assumption , high values of modularity indicate good partitions . ] . so , the partition corresponding to its maximum value on a given graph should be the best , or at least a very good one . this is the main motivation for modularity maximization , by far the most popular class of methods to detect communities in graphs . an exhaustive optimization of @xmath90 is impossible , due to the huge number of ways in which it is possible to partition a graph , even when the latter is small . besides , the true maximum is out of reach , as it has been recently proved that modularity optimization is an np - complete problem @xcite , so it is probably impossible to find the solution in a time growing polynomially with the size of the graph . however , there are currently several algorithms able to find fairly good approximations of the modularity maximum in a reasonable time . the first algorithm devised to maximize modularity was a greedy method of newman @xcite . it is an agglomerative hierarchical clustering method , where groups of vertices are successively joined to form larger communities such that modularity increases after the merging . one starts from @xmath15 clusters , each containing a single vertex . edges are not initially present , they are added one by one during the procedure . however , the modularity of partitions explored during the procedure is always calculated from the full topology of the graph , as we want to find the modularity maximum on the space of partitions of the full graph . adding a first edge to the set of disconnected vertices reduces the number of groups from @xmath15 to @xmath143 , so it delivers a new partition of the graph . the edge is chosen such that this partition gives the maximum increase ( minimum decrease ) of modularity with respect to the previous configuration . all other edges are added based on the same principle . if the insertion of an edge does not change the partition , i. e. the edge is internal to one of the clusters previously formed , modularity stays the same . the number of partitions found during the procedure is @xmath15 , each with a different number of clusters , from @xmath15 to @xmath2 . the largest value of modularity in this subset of partitions is the approximation of the modularity maximum given by the algorithm . at each iteration step , one needs to compute the variation @xmath235 of modularity given by the merger of any two communities of the running partition , so that one can choose the best merger . however , merging communities between which there are no edges can never lead to an increase of @xmath90 , so one has to check only the pairs of communities which are connected by edges , of which there can not be more than @xmath14 . since the calculation of each @xmath235 can be done in constant time , this part of the calculation requires a time @xmath21 . after deciding which communities are to be merged , one needs to update the matrix @xmath236 expressing the fraction of edges between clusters @xmath67 and @xmath68 of the running partition ( necessary to compute @xmath90 ) , which can be done in a worst - case time @xmath20 . since the algorithm requires @xmath143 iterations ( community mergers ) to run to completion , its complexity is @xmath237 , or @xmath129 on a sparse graph , so it enables one to perform a clustering analysis on much larger networks than the algorithm of girvan and newman ( up to an order of @xmath238 vertices with current computers ) . in a later paper @xcite , clauset et al . pointed out that the update of the matrix @xmath236 in newman s algorithm involves a large number of useless operations , due to the sparsity of the adjacency matrix . this operation can be performed more efficiently by using data structures for sparse matrices , like _ max - heaps _ , which rearrange the data in the form of binary trees . clauset et al . maintained the matrix of modularity variations @xmath239 , which is also sparse , a max - heap containing the largest elements of each row of the matrix @xmath239 as well as the labels of the corresponding communities , and a simple array whose elements are the sums of the elements of each row of the old matrix @xmath236 . the optimization of modularity can be carried out using these three data structures , whose update is much quicker than in newman s technique . the complexity of the algorithm is @xmath240 , where @xmath88 is the depth of the dendrogram describing the successive partitions found during the execution of the algorithm , which grows as @xmath241 for graphs with a strong hierarchical structure . for those graphs , the running time of the method is then @xmath242 , which allows to analyse the community structure of very large graphs , up to @xmath243 vertices . the greedy optimization of clauset et al . is currently one of the few algorithms that can be used to estimate the modularity maximum on such large graphs . the code can be freely downloaded from http://cs.unm.edu/@xmath215aaron/research/fastmodulari + ty.htm . this greedy optimization of modularity tends to form quickly large communities at the expenses of small ones , which often yields poor values of the modularity maxima . danon et al . suggested to normalize the modularity variation @xmath235 produced by the merger of two communities by the fraction of edges incident to one of the two communities , in order to favor small clusters @xcite . this trick leads to better modularity optima as compared to the original recipe of newman , especially when communities are very different in size . wakita and tsurumi @xcite have noticed that , due to the bias towards large communities , the fast algorithm by clauset et al . is inefficient , because it yields very unbalanced dendrograms , for which the relation @xmath244 does not hold , and as a consequence the method often runs at its worst - case complexity . to improve the situation they proposed a modification in which , at each step , one seeks the community merger delivering the largest value of the product of the modularity variation @xmath235 times a factor ( _ consolidation ratio _ ) , that peaks for communities of equal size . in this way there is a tradeoff between the gain in modularity and the balance of the communities to merge , with a big gain in the speed of the procedure , that enables the analysis of systems with up to @xmath245 vertices . interestingly , this modification often leads to better modularity maxima than those found with the version of clauset et al . , at least on large social networks . the code can be found at http://www.is.titech.ac.jp/@xmath215wakita/en/software + /community - analysis - software/. another trick to avoid the formation of large communities was proposed by schuetz and caflisch and consists in allowing for the merger of more community pairs , instead of one , at each iteration @xcite . this generates several `` centers '' around which communities are formed , which grow simultaneously so that a condensation into a few large clusters is unlikely . this modified version of the greedy algorithm is combined with a simple refinement procedure in which single vertices are moved to the neighboring community that yields the maximum increase of modularity . the method has the same complexity of the fast optimization by clauset et al . , but comes closer to the modularity maximum . the software is available at http://www.biochem-caflisch.uzh.ch/public/5/net + work-clusterization-algorithm.html . the accuracy of the greedy optimization can be significantly improved if the hierarchical agglomeration is started from some reasonable intermediate configuration , rather than from the individual vertices @xcite . xiang et al . suggested to start from a configuration obtained by merging the original isolated vertices into larger subgraphs , according to the values of a measure of topological similarity between subgraphs @xcite . a similar approach has been described by ye et al . @xcite : here the initial partition is such that no single vertex can be moved from its cluster to another without decreasing @xmath90 . higher - quality modularities can be also achieved by applying refinement strategies based on local search at various steps of the greedy agglomeration @xcite . such refinement procedures are similar to the technique proposed by newman to improve the results of his spectral optimization of modularity ( @xcite and section [ sub_sec6_0_4 ] ) . another good strategy consists in alternating greedy optimization with stochastic perturbations of the partitions @xcite . a different greedy approach has been introduced by blondel et al . @xcite , for the general case of weighted graphs . initially , all vertices of the graph are put in different communities . the first step consists of a sequential sweep over all vertices . given a vertex @xmath67 , one computes the gain in weighted modularity ( eq . [ eq : mod2 ] ) coming from putting @xmath67 in the community of its neighbor @xmath68 and picks the community of the neighbor that yields the largest increase of @xmath90 , as long as it is positive . at the end of the sweep , one obtains the first level partition . in the second step communities are replaced by supervertices , and two supervertices are connected if there is at least an edge between vertices of the corresponding communities . in this case , the weight of the edge between the supervertices is the sum of the weights of the edges between the represented communities at the lower level . the two steps of the algorithm are then repeated , yielding new hierarchical levels and supergraphs ( fig . [ blondelmethod ] ) . we remark that modularity is always computed from the initial graph topology : operating on supergraphs enables one to consider the variations of modularity for partitions of the original graph after merging and/or splitting of groups of vertices . therefore , at some iteration , modularity can not increase anymore , and the algorithm stops . the technique is more limited by storage demands than by computational time . the latter grows like @xmath21 , so the algorithm is extremely fast and graphs with up to @xmath246 edges can be analyzed in a reasonable time on current computational resources . the software can be found at http://findcommunities.googlepages.com/. the modularity maxima found by the method are better than those found with the greedy techniques by clauset et al . @xcite and wakita and tsurumi @xcite . however , closing communities within the immediate neighborhood of vertices may be inaccurate and yield spurious partitions in practical cases . so , it is not clear whether some of the intermediate partitions could correspond to meaningful hierarchical levels of the graph . moreover , the results of the algorithm depend on the order of the sequential sweep over the vertices . we conclude by stressing that , despite the improvements and refinements of the last years , the accuracy of greedy optimization is not that good , as compared with other techniques . simulated annealing @xcite is a probabilistic procedure for global optimization used in different fields and problems . it consists in performing an exploration of the space of possible states , looking for the global optimum of a function @xmath247 , say its maximum . transitions from one state to another occur with probability @xmath2 if @xmath247 increases after the change , otherwise with a probability @xmath248 , where @xmath249 is the decrease of the function and @xmath19 is an index of stochastic noise , a sort of inverse temperature , which increases after each iteration . the noise reduces the risk that the system gets trapped in local optima . at some stage , the system converges to a stable state , which can be an arbitrarily good approximation of the maximum of @xmath247 , depending on how many states were explored and how slowly @xmath19 is varied . simulated annealing was first employed for modularity optimization by guimer et al . its standard implementation @xcite combines two types of `` moves '' : local moves , where a single vertex is shifted from one cluster to another , taken at random ; global moves , consisting of mergers and splits of communities . splits can be carried out in several distinct ways . the best performance is achieved if one optimizes the modularity of a bipartition of the cluster , taken as an isolated graph . this is done again with simulated annealing , where one considers only individual vertex movements , and the temperature is decreased until it reaches the running value for the global optimization . global moves reduce the risk of getting trapped in local minima and they have proven to lead to much better optima than using simply local moves @xcite . in practical applications , one typically combines @xmath250 local moves with @xmath15 global ones in one iteration . the method can potentially come very close to the true modularity maximum , but it is slow . the actual complexity can not be estimated , as it heavily depends on the parameters chosen for the optimization ( initial temperature , cooling factor ) , not only on the graph size . simulated annealing can be used for small graphs , with up to about @xmath251 vertices . extremal optimization ( eo ) is a heuristic search procedure proposed by boettcher and percus @xcite , in order to achieve an accuracy comparable with simulated annealing , but with a substantial gain in computer time . it is based on the optimization of local variables , expressing the contribution of each unit of the system to the global function at study . this technique was used for modularity optimization by duch and arenas @xcite . modularity can be indeed written as a sum over the vertices : the local modularity of a vertex is the value of the corresponding term in this sum . a fitness measure for each vertex is obtained by dividing the local modularity of the vertex by its degree , as in this case the measure does not depend on the degree of the vertex and is suitably normalized . one starts from a random partition of the graph in two groups with the same number of vertices . at each iteration , the vertex with the lowest fitness is shifted to the other cluster . the move changes the partition , so the local fitnesses of many vertices need to be recalculated . the process continues until the global modularity @xmath90 can not be improved any more by the procedure . this technique reminds one of the kernighan - lin @xcite algorithm for graph partitioning ( section [ sec4_1 ] ) , but here the sizes of the communities are determined by the process itself , whereas in graph partitioning they are fixed from the beginning . after the bipartition , each cluster is considered as a graph on its own and the procedure is repeated , as long as @xmath90 increases for the partitions found . the procedure , as described , proceeds deterministically from the given initial partition , as one shifts systematically the vertex with lowest fitness , and is likely to get trapped in local optima . better results can be obtained if one introduces a probabilistic selection , in which vertices are ranked based on their fitness values and one picks the vertex of rank @xmath91 with the probability @xmath252 ( @xmath253-eo , @xcite ) . the algorithm finds very good estimates of the modularity maximum , and performs very well on the benchmark of girvan and newman @xcite ( section [ sec6_1 ] ) . ranking the fitness values has a cost @xmath254 , which can be reduced to @xmath20 if heap data structures are used . choosing the vertex to be shifted can be done with a binary search , which amounts to an additional factor @xmath255 . finally , the number of steps needed to verify whether the running modularity maximum can be improved or not is also @xmath20 . the total complexity of the method is then @xmath128 . we conclude that eo represents a good tradeoff between accuracy and speed , although the use of recursive bisectioning may lead to poor results on large networks with many communities . modularity can be optimized using the eigenvalues and eigenvectors of a special matrix , the modularity matrix @xmath256 , whose elements are @xmath257 here the notation is the same used in eq . [ eq : mod ] . let @xmath130 be the vector representing any partition of the graph in two clusters @xmath258 and @xmath259 : @xmath260 if vertex @xmath67 belongs to @xmath258 , @xmath261 if @xmath67 belongs to @xmath259 . modularity can be written as @xmath262 the last expression indicates standard matrix products . the vector @xmath130 can be decomposed on the basis of eigenvectors @xmath263 ( @xmath140 ) of the modularity matrix @xmath256 : @xmath264 , with @xmath265 . by plugging this expression of @xmath130 into eq . [ eqr5 ] one finally gets @xmath266 where @xmath267 is the eigenvalue of @xmath256 corresponding to the eigenvector @xmath263 . [ eqr6 ] is analogous to eq . [ eqr1 ] for the cut size of the graph partitioning problem . this suggests that one can optimize modularity on bipartitions via spectral bisection ( section [ sec4_1 ] ) , by replacing the laplacian matrix with the modularity matrix @xcite . like the laplacian matrix , @xmath256 has always the trivial eigenvector @xmath268 with eigenvalue zero , because the sum of the elements of each row / column of the matrix vanishes . from eq . [ eqr6 ] we see that , if @xmath256 has no positive eigenvalues , the maximum coincides with the trivial partition consisting of the graph as a single cluster ( for which @xmath269 ) , i. e. it has no community structure . otherwise , one has to look for the eigenvector of @xmath10 with largest ( positive ) eigenvalue , @xmath270 , and group the vertices according to the signs of the components of @xmath270 , just like in section [ sec4_1 ] . here , however , one does not need to specify the sizes of the two groups : the vertices with positive components are all in one group , the others in the other group . if , for example , the component of @xmath270 corresponding to vertex @xmath67 is positive , but we set @xmath261 , the modularity is lower than by setting @xmath260 . the values of the components of @xmath270 are also informative , as they indicate the level of the participation of the vertices to their communities . in particular , components whose values are close to zero lie at the border between the two clusters and can be well considered as belonging to both of them . the result obtained from the spectral bipartition can be further improved by shifting single vertices from one community to the other , such to have the highest increase ( or lowest decrease ) of the modularity of the resulting graph partition . this refinement technique , similar to the kernighan - lin algorithm ( section [ sec4_1 ] ) , can be also applied to improve the results of other optimization techniques ( e.g. greedy algorithms , extremal optimization , etc . ) . the procedure is repeated for each of the clusters separately , and the number of communities increases as long as modularity does . at variance with graph partitioning , where one needs to fix the number of clusters and their size beforehand , here there is a clear - cut stopping criterion , represented by the fact that cluster subdivisions are admitted only if they lead to a modularity increase . we stress that modularity needs to be always computed from the full adjacency matrix of the original graph . the drawback of the method is similar as for spectral bisection , i. e. the algorithm gives the best results for bisections , whereas it is less accurate when the number of communities is larger than two . recently , sun et al . @xcite have added a step after each bipartition of a cluster , in that single vertices can be moved from one cluster to another and even form the seeds of new clusters . we remark that the procedure is different from the kernighan - lin - like refining steps , as here the number of clusters can change . this variant , which does not increase the complexity of the original spectral optimization , leads to better modularity maxima . moreover , one does not need to stick to bisectioning , if other eigenvectors with positive eigenvalues of the modularity matrix are used . given the first @xmath45 eigenvectors , one can construct @xmath15 @xmath45-dimensional vectors , each corresponding to a vertex , just like in spectral partitioning ( section [ sec4_4 ] ) . the components of the vector of vertex @xmath67 are proportional to the @xmath45 entries of the eigenvectors in position @xmath67 . then one can define _ community vectors _ , by summing the vectors of vertices in the same community . it is possible to show that , if the vectors of two communities form an angle larger that @xmath271 , keeping the communities separate yields larger modularity than if they are merged ( fig . [ qspectra_plane ] ) . in this way , in a @xmath45-dimensional space the modularity maximum corresponds to a partition in at most @xmath272 clusters . community vectors were used by wang et al . to obtain high - modularity partitions into a number of communities smaller than a given maximum @xcite . in particular , if one takes the eigenvectors corresponding to the two largest eigenvalues , one can obtain a split of the graph in three clusters : in a recent work , richardson et al . presented a fast technique to obtain graph tripartitions with large modularity along these lines @xcite . the eigenvectors with the most negative eigenvalues can also be used to extract useful information , like the presence of a possible multipartite structure of the graph , as they give the most relevant contribution to the modularity minimum . the spectral optimization of modularity is quite fast . the leading eigenvector of the modularity matrix can be computed with the power method , by repeatedly multiplying @xmath256 by an arbitrary vector ( not orthogonal to @xmath270 ) . the number of required iterations to reach convergence is @xmath20 . each multiplication seems to require a time @xmath129 , as @xmath256 is a complete matrix , but the peculiar form of @xmath256 allows for a much quicker calculation , taking time @xmath273 . so , a graph bipartition requires a time @xmath274 $ ] , or @xmath129 on a sparse graph . to find the modularity optimum one needs a number of subsequent bipartitions that equals the depth @xmath88 of the resulting hierarchical tree . in the worst - case scenario , @xmath275 , but in practical cases the procedure usually stops much before reaching the leaves of the dendrogram , so one could go for the average value @xmath276 , for a total complexity of @xmath128 . the algorithm is faster than extremal optimization and it is also slightly more accurate , especially for large graphs . the modularity matrix and the corresponding spectral optimization can be trivially extended to weighted graphs . spectral optimization of modularity by newman @xcite . by using the first two eigenvectors of the modularity matrix , vertices can be represented as points on a plane . by cutting the plane with a line passing through the origin ( like the dashed line in the figure ) one obtains bipartitions of the graph with possibly high modularity values . reprinted figure with permission from ref . 2006 by the american physical society . ] a different spectral approach had been previously proposed by white and smyth @xcite . let @xmath190 indicate the weighted adjancency matrix of a graph @xmath23 . a partition of @xmath23 in @xmath44 clusters can be described through an @xmath186 assignment matrix @xmath277 , where @xmath278 if vertex @xmath67 belongs to cluster @xmath117 , otherwise @xmath279 . it can be easily shown that , up to a multiplicative constant , modularity can be rewritten in terms of the matrix @xmath277 as @xmath280= -\mbox{tr}[{\bf x}^t{\bf { l}_q}{\bf x } ] , \label{eqr7}\ ] ] where @xmath281 is a diagonal matrix with identical elements , equal to the sum of all edge weights , and the entries of @xmath282 are @xmath283 , where @xmath105 is the degree of vertex @xmath67 . the matrix @xmath284 is called the @xmath90-laplacian . finding the assignment matrix @xmath277 that maximizes @xmath90 is an @xmath285-complete problem , but one can get a good approximation by relaxing the constraint that the elements of @xmath277 have to be discrete . by doing so @xmath90 becomes a sort of continuous functional of @xmath277 and one can determine the extremes of @xmath90 by setting its first derivative ( with respect to @xmath277 ) to zero . this leads to the eigenvalue problem @xmath286 here @xmath287 is a diagonal matrix . [ eqr8 ] turns modularity maximization into a spectral graph partitioning problem ( section [ sec4_4 ] ) , using the @xmath90-laplacian matrix . a nice feature of the @xmath90-laplacian is that , for graphs which are not too small , it can be approximated ( up to constant factors ) by the transition matrix @xmath288 , obtained by normalizing @xmath190 such that the sum of the elements of each row equals one . [ eqr8 ] is at the basis of the algorithms developed by white and smyth , which search for partitions with at most @xmath289 clusters , where @xmath289 is a predefined input parameter that may be suggested by preliminary information on the graph cluster structure . the first @xmath290 eigenvectors of the transition matrix @xmath288 ( excluding the trivial eigenvector with all equal components ) can be computed with a variant of the lanczos method @xcite . since the eigenvector components are not integer , the eigenvectors do not correspond directly to a partition of the graph in clusters . however , as usual in spectral graph partitioning , the components of the eigenvectors can be used as coordinates of the graph vertices in an euclidean space and @xmath44-means clustering is applied to obtain the desired partition . white and smyth proposed two methods to derive the clustering after embedding the graph in space . both methods have a worst - case complexity @xmath291 , which is essentially linear in the number of vertices of the graph if the latter is sparse and @xmath292 . however , on large and sparse graphs , @xmath289 could scale with the number of vertices @xmath15 , so the procedure might become quite slow . in order to speed up calculations without losing much accuracy in the final estimates of the maximum modularity , ruan and zhang @xcite have proposed an algorithm , called _ kcut _ , that applies recursively the method by white and smyth , in a slightly modified form : after a first application to the full graph , in the next iteration the method is applied to all clusters of the first partition , treated as independent networks , and so on . the procedure goes on as long as the modularity of the resulting partitions keeps growing . the advantage of _ kcut _ is that one can play with low values for the ( maximal ) number of clusters @xmath293 at each iteration ; if partitions are balanced , after @xmath294 levels of recursions , the number of clusters of the partition is approximately @xmath295 . therefore the complexity of _ kcut _ is @xmath296 $ ] for a final partition in ( at most ) @xmath289 clusters , which is much lower than the complexity of the algorithm by white and smyth . ruan and zhang tested _ kcut _ on artificial graphs generated with the planted @xmath293-partition model ( section [ sec6 ] ) , and on real networks including zachary s karate club @xcite , the american college football network @xcite and two collaboration networks of jazz musicians @xcite and physicists @xcite : the accuracy of _ kcut _ is comparable to that of the algorithm by white and smyth , though generally lower . agarwal and kempe have suggested to maximize modularity within the framework of mathematical programming @xcite . in fact , modularity optimization can be formulated both as a linear and as a quadratic program . in the first case , the variables are defined on the links : @xmath297 if @xmath67 and @xmath68 are in the same cluster , otherwise @xmath298 . the modularity of a partition , up to a multiplicative constant , can then be written as @xmath299 where @xmath256 is the modularity matrix defined by newman ( see section [ sub_sec6_0_4 ] ) . [ eqr9 ] is linear in the variables @xmath300 , which must obey the constraint @xmath301 , because , if @xmath67 and @xmath68 are in the same cluster , and so are @xmath67 and @xmath44 , then @xmath68 and @xmath44 must be in that cluster too . maximizing the expression in eq . [ eqr9 ] under the above constraint is @xmath285-hard , if the variables have to be integer as required . however , if one relaxes this condition by using real - valued @xmath300 , the problem can be solved in polynomial time @xcite . on the other hand , the solution does not correspond to an actual partition , as the @xmath302 variables are fractional . to get clusters out of the @xmath300 one needs a rounding step . the values of the @xmath302 variables are used as sort of distances in a metric space ( the triangular inequality is satisfied by construction ) : clusters of vertices `` close '' enough to each other ( i. e. whose mutual @xmath302 variables are close to zero ) are formed and removed until each vertex is assigned to a cluster . the resulting partition is further refined with the same post - processing technique used by newman for the spectral optimization of modularity , i. e. by a sequence of steps similar to those of the algorithm by kernighan and lin ( see section [ sub_sec6_0_4 ] ) . quadratic programming can be used to get bisections of graphs with high modularity , that can be iterated to get a whole hierarchy of partitions as in newman s spectral optimization . one starts from one of the identities in eq . [ eqr5 ] @xmath303 where @xmath304 , depending on whether the vertex belongs to the first or the second cluster . since the optimization of the expression in eq . [ eqr10 ] is @xmath285-complete , one must relax again the constraint on the variables @xmath155 being integer . a possibility is to transform each @xmath155 into an @xmath15-dimensional vector @xmath305 and each product in the scalar product between vectors . the vectors are normalized so that their tips lie on the unit - sphere of the @xmath15-dimensional space . this vector problem is polynomially solvable , but one needs a method to associate a bipartition to the set of @xmath15 vectors of the solution . any @xmath306-dimensional hyperplane centered at the origin cuts the space in two halves , separating the vectors in two subsets . one can then choose multiple random hyperplanes and pick the one which delivers the partition with highest modularity . as in the linear program , a post - processing technique la newman ( see section [ sub_sec6_0_4 ] ) is used to improve the results of the procedure . the two methods proposed by agarwal and kempe are strongly limited by their high computational complexity , due mostly to the large storage demands , making graphs with more than @xmath251 vertices intractable . on the other hand , the idea of applying mathematical programming to graph clustering is promising . the code of the algorithms can be downloaded from http://www - scf.usc.edu/@xmath215gaurava/. in a recent work @xcite , xu et al . have optimized modularity using mixed - integer mathematical programming , with both integer and continuous variables , obtaining very good approximations of the modularity optimum , at the price of high computational costs . chen et al . have used integer linear programming to transform the initial graph into an optimal target graph consisting of disjoint cliques , which effectively yields a partition @xcite . berry et al . have formulated the problem of graph clustering as a _ facility location problem _ @xcite , consisting in the minimization of a cost function based on a local variation of modularity @xcite . lehmann and hansen @xcite optimized modularity via mean field annealing @xcite , a deterministic alternative to simulated annealing @xcite . the method uses gibbs probabilities to compute the conditional mean value for the variable of a vertex , which indicates its community membership . by making a mean field approximation on the variables of the other vertices in the gibbs probabilities one derives a self - consistent set of non - linear equations , that can be solved by iteration in a time @xmath307 $ ] . the method yields better modularity maxima than the spectral optimization by newman ( section [ sub_sec6_0_4 ] ) , at least on artificial graphs with built - in community structure , similar to the benchmark graphs by girvan and newman ( section [ sec6_1 ] ) . genetic algorithms @xcite have also been used to optimize modularity . in a standard genetic algorithm one has a set of candidate solutions to a problem , which are numerically encoded as chromosomes , and an objective function to be optimized on the space of solutions . the objective function plays the role of biological fitness for the chromosomes . one usually starts from a random set of candidate solutions , which are progressively changed through manipulations inspired by biological processes regarding real chromosomes , like point mutation ( random variations of some parts of the chromosome ) and crossing over ( generating new chromosomes by merging parts of existing chromosomes ) . then , the fitness of the new pool of candidates is computed and the chromosomes with the highest fitness have the greatest chances to survive in the next generation . after several iterations only solutions with large fitness survive . in a work by tasgin et al . @xcite , partitions are the chromosomes and modularity is the fitness function . with a suitable choice of the algorithm parameters , like the number of chromosomes and the rates of mutation and crossing over , tasgin et al . could obtain results of comparative quality as greedy modularity optimization on zachary s karate club @xcite , the college football network @xcite and the benchmark by girvan and newman ( section [ sec6_1 ] ) . genetic algorithms were also adopted by liu et al . @xcite . here the maximum modularity partition is obtained via successive bipartitions of the graph , where each bipartition is determined by applying a genetic algorithm to each subgraph ( starting from the original graph itself ) , which is considered isolated from the rest of the graph . a bipartition is accepted only if it increases the total modularity of the graph . in section [ sec3_2_2 ] we have seen that the modularity maximum is obtained for the partition that minimizes the difference between the cut size and the expected cut size of the partition ( eq . [ eqr12 ] ) . in the complete weighted graph @xmath308 such that the weight @xmath309 of an edge is @xmath310 , if @xmath67 and @xmath68 are connected in @xmath23 , and @xmath311 if they are not , the difference @xmath312 is simply the cut size of partition @xmath94 . so , maximizing modularity for @xmath313 is equivalent to the problem of finding the partition with minimal cut size of the weighted graph @xmath308 , i. e. to a graph partitioning problem . the problem can then be efficiently solved by using existing software for graph partitioning @xcite . in the most recent literature on graph clustering several modifications and extensions of modularity can be found . they are usually motivated by specific classes of clustering problems and/or graphs that one may want to analyze . modularity can be easily extended to graphs with weighted edges @xcite . one needs to replace the degrees @xmath105 and @xmath106 in eq . [ eq : mod ] with the strengths @xmath314 and @xmath315 of vertices @xmath67 and @xmath68 . we remind that the strength of a vertex is the sum of the weights of edges adjacent to the vertex ( section [ sec1_1 ] ) . for a proper normalization , the number of edges @xmath14 in eq . [ eq : mod ] has to be replaced by the sum @xmath86 of the weights of all edges . the product @xmath316 is now the expected weight of the edge @xmath219 in the null model of modularity , which has to be compared with the actual weight @xmath317 of that edge in the original graph . this can be understood if we consider the case in which all weights are multiples of a unit weight , so they can be rewritten as integers . the weight of the connection between two nodes can then be replaced by as many edges between the nodes as expressed by the number of weight units . for the resulting multigraph we can use the same procedure as in the case of unweighted graphs , which leads to the formally identical expression @xmath318 which can be also written as a sum over the modules @xmath319 , \label{eq : mod20}\ ] ] where @xmath320 is the sum of the weights of the internal edges of module @xmath117 and @xmath321 is the sum of the strengths of the vertices of @xmath117 . if edge weights are not mutually commensurable , one can always represent them as integers with good approximation , provided a sufficiently small weight unit is adopted , so the expressions for weighted modularity of eqs . [ eq : mod2 ] , [ eq : mod20 ] are generally valid . in principle , weights can be assigned to the edges of an undirected graph , by using any measure of similarity / correlation between the vertices ( like , e. g. , the measures introduced in section [ sec3_1_4 ] ) . in this way , one could derive the corresponding weighted modularity and use it to detect communities , with a potentially better exploitation of the structural information of the graph as compared to standard modularity @xcite . modularity has also a straightforward extension to the case of directed graphs @xcite . if an edge is directed , the probability that it will be oriented in either of the two possible directions depends on the in- and out - degrees of the end - vertices . for instance , taken two vertices @xmath9 and @xmath10 , where @xmath9 ( @xmath10 ) has a high ( low ) indegree and low ( high ) outdegree , in the null model of modularity an edge will be much more likely to point from @xmath10 to @xmath9 than from @xmath9 to @xmath10 . therefore , the expression of modularity for directed graphs reads @xmath322 where the factor @xmath323 in the denominator of the second summand has been dropped because the sum of the indegrees ( outdegrees ) equals @xmath14 , whereas the sum of the degrees of the vertices of an undirected graph equals @xmath109 ; the factor @xmath323 in the denominator of the prefactor has been dropped because the number of non - vanishing elements of the adjacency matrix is @xmath14 , not @xmath109 as in the symmetric case of an undirected graph . if a graph is both directed and weighted , formulas [ eq : mod2 ] and [ eq : mod3 ] can be combined as @xmath324 which is the most general ( available ) expression of modularity @xcite . kim et al . @xcite have remarked that the directed modularity of eq . [ eq : mod3 ] may not properly account for the directedness of the edges ( fig . [ directedq ] ) , and proposed a different definition based on diffusion on directed graphs , inspired by google s pagerank algorithm @xcite . rosvall and bergstrom raised similar objections @xcite . problem of the directed modularity introduced by arenas et al . the two situations illustrated are equivalent for modularity , as vertices @xmath9 and @xmath325 , as well as @xmath10 and @xmath326 , have identical indegrees and outdegrees . in this way , the optimization of directed modularity is not able to distinguish a situation in which there is directed flow ( top ) or not ( bottom ) . reprinted figure with permission from ref . @xcite.,width=340 ] if vertices may belong to more clusters , it is not obvious how to find a proper generalization of modularity . in fact , there is no unique recipe . shen et al . @xcite , for instance , suggested the simple definition @xmath327 here @xmath328 is the number of communities including vertex @xmath67 . the contribution of each edge to modularity is then the smaller , the larger the number of communities including its endvertices . nicosia et al . @xcite have made some more general considerations on the problem of extending modularity to the case of overlapping communities . they considered the case of directed unweighted networks , starting from the following general expression @xmath329 , \label{eq : mod8}\ ] ] where @xmath330 and @xmath331 are the indegree and outdegree of vertices @xmath67 and @xmath68 , the index @xmath117 labels the communities and @xmath332 , @xmath333 express the contributions to the sum corresponding to the edge @xmath219 in the network and in the null model , due to the multiple memberships of @xmath67 and @xmath68 . if there is no overlap between the communities , @xmath334 , where @xmath335 and @xmath336 correspond to the communities of @xmath67 and @xmath68 . in this case , the edge @xmath219 contributes to the sum only if @xmath337 , as in the original definition of modularity . for overlapping communities , the coefficients @xmath332 , @xmath333 must depend on the membership coefficients @xmath338 , @xmath339 of vertices @xmath67 and @xmath68 . one can assume that @xmath340 , where @xmath341 is some function . the term @xmath333 is related to the null model of modularity , and it must be handled with care . in modularity s original null model edges are formed by joining two random stubs , so one needs to define the membership of a random stub in the various communities . if we assume that there is no correlation _ a priori _ between the membership coefficients of any two vertices , we can assign to a stub originating from a vertex @xmath67 in community @xmath117 the average membership corresponding to all edges which can be formed with @xmath67 . on a directed graph we have to distinguish between outgoing and incoming stubs , so one has @xmath342 @xmath343 and one can write the following general expression for modularity @xmath344 . \label{eq : mod11}\ ] ] the question now concerns the choice of the function @xmath345 . if the formula of eq . [ eq : mod11 ] is to be an extension of modularity to the case of overlapping communities , it has to satisfy some general properties of classical modularity . for instance , the modularity value of a cover consisting of the whole network as a single cluster should be zero . it turns out that a large class of functions yield an expression for modularity that fulfills this requirement . otherwise , the choice of @xmath341 is rather arbitrary and good choices can be only tested _ a posteriori _ , based on the results of the optimization . membership coefficients are also present in an extension of modularity to overlapping communities proposed by shen et al . @xcite . here the membership coefficient of vertex @xmath29 in community @xmath117 is a sum over the edges of @xmath29 belonging to @xmath117 , where each edge has a weight proportional to the number of maximal cliques of @xmath117 containing the edge . gaertler et al . have introduced quality measures based on modularity s principle of the comparison between a variable relative to the original graph and the corresponding variable of a null model @xcite . they remark that modularity is just the difference between the coverage of a partition and the expected coverage of the partition in the null model . we remind that the coverage of a partition is the ratio between the number of edges within clusters and the total number of edges ( section [ sec3_2_2 ] ) . based on this observation , gaertler et al . suggest that the comparison between the two terms can be done with other binary operations as well . for instance , one could consider the ratio @xmath346 where the notation is the same as in eq . [ eq : mod1 ] . this can be done as well for any variable other than coverage . by using performance , for instance , ( section [ sec3_2_2 ] ) one obtains two new quality functions @xmath347 and @xmath348 , corresponding to taking the difference or the ratio between performance and its null model expectation value , respectively . gaertler et al . compared the results obtained with the four functions @xmath349 , @xmath350 , @xmath347 and @xmath348 , on a class of benchmark graphs with built - in cluster structure ( section [ sec6_1 ] ) and social networks . they found that the `` absolute '' variants @xmath351 and @xmath347 are better than the `` relative '' variants @xmath350 and @xmath348 on the artificial benchmarks , whereas @xmath348 is better on social networks . furthermore @xmath347 is better than the standard modularity @xmath351 . modifications of modularity s null model have been introduced by massen and doye @xcite and muff et al . massen and doye s null model is still a graph with the same expected degree sequence as the original , and with edges rewired at random among the vertices , but one imposes the additional constraint that there can be neither multiple edges between a pair of vertices nor edges joining a vertex with itself ( loops or self - edges ) . this null model is more realistic , as multiple edges and loops are usually absent in real graphs . the maximization of the corresponding modified modularity yields partitions with smaller average cluster size than standard modularity . the latter tends to disfavor small communities , because the actual densities of edges inside small communities hardly exceed the null model densities , which are appreciably enhanced by the contributions from multiple connections and loops . muff et al . proposed a local version of modularity , in which the expected number of edges within a module is not calculated with respect to the full graph , but considering just a portion of it , namely the subgraph including the module and its neighbouring modules . their motivation is the fact that modularity s null model implicitly assumes that each vertex could be attached to any other , whereas in real cases a cluster is usually connected to few other clusters . on a directed graph , their _ localized modularity _ @xmath352 reads @xmath353 . \label{eqr16}\ ] ] in eq . [ eqr16 ] @xmath116 is the number of edges inside cluster @xmath117 , @xmath354 ( @xmath355 ) the total internal ( external ) degree of cluster @xmath117 and @xmath356 the total number of edges in the subgraph comprising cluster @xmath117 and its neighbor clusters . the localized modularity is not bounded by @xmath2 , but can take any value . its maximization delivers more accurate partitions than standard modularity optimization on a model network describing the social interactions between children in a school ( school network ) and on the metabolic and protein - protein interaction networks of _ e. coli_. reichardt and bornholdt have shown that it is possible to reformulate the problem of community detection as the problem of finding the ground state of a spin glass model @xcite . each vertex @xmath67 is labeled by a potts spin variable @xmath357 , which indicates the cluster including the vertex . the basic principle of the model is that edges should connect vertices of the same class ( i. e. same spin state ) , whereas vertices of different classes ( i. e. different spin states ) should be disconnected ( ideally ) . so , one has to energetically favor edges between vertices in the same class , as well as non - edges between vertices in different classes , and penalize edges between vertices of different classes , along with non - edges between vertices in the same class . the resulting hamiltonian of the spin model is @xmath358 where @xmath359 is a constant expressing the coupling strength , @xmath360 are the elements of the adjacency matrix of the graph , @xmath361 a parameter expressing the relative contribution to the energy from existing and missing edges , and @xmath362 is the expected number of links connecting @xmath67 and @xmath68 for a null model graph with the same total number of edges @xmath14 of the graph considered . the system is a spin glass @xcite , as the couplings @xmath363 between spins are both ferromagnetic ( on the edges of the graph , provided @xmath364 ) and antiferromagnetic ( between disconnected vertices , as @xmath365 and @xmath366 ) . the multiplicative costant @xmath359 is irrelevant for practical purposes , so in the following we set @xmath367 . the range of the spin - spin interaction is infinite , as there is a non - zero coupling between any pair of spins . [ eqr13 ] bears a strong resemblance with the expression of modularity of eq . [ eq : mod ] . in fact , if @xmath368 and @xmath369 we recover exactly modularity , up to a factor @xmath370 . in this case , finding the spin configuration for which the hamiltonian is minimal is equivalent to maximizing modularity . [ eqr13 ] is much more general than modularity , though , as both the null model and the parameter @xmath371 can be arbitrarily chosen . in particular , the value of @xmath371 determines the importance of the null model term @xmath362 in the quality function . . [ eqr13 ] can be rewritten as @xmath372=-\sum_{s=1}c_{ss}\\ & = & \sum_{s < r}\left[l_{rs}-\gamma(l_{rs})_{p_{ij}}\right]=\sum_{s < r}a_{rs}. \label{eqr14}\end{aligned}\ ] ] here , the sums run over the clusters : @xmath373 and @xmath374 indicate the number of edges within cluster @xmath155 and between clusters @xmath375 and @xmath155 , respectively ; @xmath376 and @xmath377 are the corresponding null model expectation values . [ eqr14 ] defines the coefficients @xmath378 of _ cohesion _ and @xmath379 of _ adhesion_. if a subset of a cluster @xmath155 has a larger coefficient of adhesion with another cluster @xmath375 than with its complement in @xmath155 , the energy can be reduced by merging the subset with cluster @xmath375 . in the particular case in which the coefficient of adhesion of a subset @xmath380 of a cluster @xmath155 with its complement in the cluster exactly matches the coefficient of adhesion of @xmath380 with another cluster @xmath375 , the partitions in which @xmath380 stays within @xmath155 or is merged with @xmath375 have the same energy . in this case one can say that clusters @xmath375 and @xmath155 are overlapping . in general , the partition with minimum energy has the following properties : 1 ) every subset of each cluster has a coefficient of adhesion with its complement in the cluster not smaller than with any other cluster ; 2 ) every cluster has non - negative coefficient of cohesion ; 3 ) the coefficient of adhesion between any two clusters is non - positive . by tuning the parameter @xmath371 one can vary the number of clusters in the partition with minimum energy , going from a single cluster comprising all vertices ( @xmath381 ) , to @xmath15 clusters with a single vertex ( @xmath382 ) . so , @xmath371 is a resolution parameter that allows to explore the cluster structure of a graph at different scales ( see section [ sub_sec6_1 ] ) . the authors used single spin heatbath simulated annealing algorithms to find the ground state of the hamiltonian of eq . [ eqr13 ] . another generalization of modularity was recently suggested by arenas et al . they remarked that the fundamental unit to define modularity is the edge , but that high edge densities inside clusters usually imply the existence of long - range topological correlations between vertices , which are revealed by the presence of _ motifs _ @xcite , i. e. connected undirected subgraphs , like cycles ( section [ sec1_1 ] ) . for instance , a high edge density inside a cluster usually means that there are also several triangles in the cluster , and comparatively few between clusters , a criterion that has inspired on its own popular graph clustering algorithms @xcite . modularity can then be simply generalized by comparing the density of motifs inside clusters with the expected density in modularity s null model ( _ motif modularity _ ) . as a particular case , the _ triangle modularity _ of a partition @xmath22 reads @xmath383 where @xmath384 ( @xmath385 is the label of the cluster @xmath67 belongs to ) , @xmath386 ( @xmath105 is the degree of vertex @xmath67 ) and @xmath387 . if one chooses as motifs paths with even length , and removes the constraint that all vertices of the motif / path should stay inside the same cluster , maximizing motif modularity could reveal the existence of multipartite structure . for example , if a graph is bipartite , one expects to see many @xmath323-paths starting from one vertex class and returning to it from the other class . motif modularity can be trivially extended to the case of weighted graphs . several graphs representing real systems are built out of correlation data between elements . correlation matrices are very common in the study of complex systems : well - known examples are the correlations of price returns , which are intensively studied by economists and econophysicists @xcite . correlations may be positive as well as negative , so the corresponding weighted edges indicate both attraction and repulsion between pairs of vertices . usually the correlation values are filtered or otherwise transformed such to eliminate the weakest correlations and anticorrelations and to maintain strictly positive weights for the edges , yielding graphs that can be treated with standard techniques . however , ignoring negative correlations means to give up useful information on the relationships between vertices . finding clusters in a graph with both positive and negative weights is called _ correlation clustering problem _ @xcite . according to intuition , one expects that vertices of the same cluster are linked by positive edges , whereas vertices of different clusters are linked by negative edges . the best cluster structure is the partition that maximizes the sum of the strengths ( in absolute value ) of positive edges within clusters and negative edges between clusters , or , equivalently , the partition that minimizes the sum of the strengths ( in absolute value ) of positive edges between clusters and negative edges within clusters . this can be formulated by means of modularity , if one accounts for the contribution of the negative edges . a natural way to proceed is to create two copies of the graph at study : in one copy only the weights of the positive edges are kept , in the other only the weights of the negative edges ( in absolute value ) . by applying eq . [ eq : mod2 ] to the same partition of both graphs , one derives the contributions @xmath388 and @xmath389 to the modularity of that partition for the original graph . gmez et al . define the global modularity as a linear combination of @xmath388 and @xmath389 , that accounts for the relative total strengths of positive and negative edge weights @xcite . kaplan and forrest @xcite have proposed a similar expression , with two important differences . first , they have used the total strength of the graph , i. e. the sum of the absolute values of all weights , to normalize @xmath388 and @xmath389 ; gmez et al . instead have used the positive and the negative strengths , for @xmath388 and @xmath389 , respectively , which seems to be the more natural choice looking at eq . [ eq : mod2 ] . second , kaplan and forrest have given equal weight to the contributions of @xmath388 and @xmath389 to their final expression of modularity , which is just the difference @xmath390 . in another work , traag and bruggeman @xcite have introduced negative links in the general spin glass formulation of modularity of reichardt and bornholdt @xcite . here the relative importance of the contribution of positive and negative edge weights is a free parameter , the tuning of which allows to detect communities of various sizes and densities of positive / negative edges . some authors have pointed out that the original expression of modularity is not ideal to detect communities in bipartite graphs , which describe several real systems , like food webs @xcite , scientific @xcite and artistic @xcite collaboration networks , etc .. expressions of modularity for bipartite graphs were suggested by guimer et al . @xcite and barber @xcite . guimer et al . call the two classes of vertices actors and teams , and indicate with @xmath391 the degree of actor @xmath67 and @xmath392 the degree of team @xmath294 . the null model graphs are random graphs with the same expected degrees for the vertices , as usual . the bipartite modularity @xmath393 for a partition @xmath94 ( of the actors ) has the following expression @xmath394 . \label{eqr17}\ ] ] here , @xmath395 is the number of teams in which actors @xmath67 and @xmath68 are together and the sum @xmath396 gives the number of ordered pairs of actors in the same team . the second ratio of each summand is the null model term , indicating the expected ( normalized ) number of teams for pairs of actors in cluster @xmath117 . the bipartite modularity can also be applied to ( unipartite ) directed graphs : each vertex can be duplicated and assigned to both classes , based on its twofold role of source and target for the edges . another interesting alternative was introduced by barber @xcite and is a simple extension of eq . [ eq : mod ] . let us suppose that the two vertex classes ( red and blue ) are made out of @xmath45 and @xmath91 vertices , respectively . the degree of a red vertex @xmath67 is indicated with @xmath105 , that of a blue vertex @xmath68 with @xmath397 . the adjacency matrix @xmath189 of the graph is in block off - diagonal form , as there are edges only between red and blue vertices . because of that , barber assumes that the null model matrix @xmath398 , whose element @xmath100 indicates as usual the expected number of edges between vertices @xmath67 and @xmath68 in the null model , also has the block off - diagonal form @xmath399 where the @xmath400 are square matrices with all zero elements and @xmath401 , as in the null model of standard modularity ( though other choices are possible ) . the modularity maximum can be computed through the modularity matrix @xmath402 , as we have seen in section [ sub_sec6_0_4 ] . however , spectral optimization of modularity gives excellent results for bipartitions , while its performance worsens when the number of clusters is unknown , as it is usually the case . barber has proposed a different optimization technique , called bipartite recursively induced modules ( brim ) , based on the bipartite nature of the graph . the algorithm is based on the special expression of modularity for the bipartite case , for which once the partition of the red or the blue vertices is known , it is easy to get the partition of the other vertex class that yields the maximum modularity . therefore , one starts from an arbitrary partition in @xmath117 clusters of , say , the blue vertices , and recovers the partition of the red vertices , which is in turn used as input to get a better partition of the blue vertices , and so on until modularity converges . brim does not predict the number of clusters @xmath117 of the graph , but one can obtain good estimates for it by exploring different values with a simple bisection approach . typically , for a given @xmath117 the algorithm needs a few steps to converge , each step having a complexity @xmath21 . an expression of the number of convergence steps in terms of @xmath15 and/or @xmath14 still needs to be derived . in this section we shall discuss some features of modularity , which are crucial to identify the domain of its applicability and ultimately to assess the issue of the reliability of the measure for the problem of graph clustering . an important question concerns the value of the maximum modularity @xmath403 for a graph . we know that it must be non - negative , as there is always at least a partition with zero modularity , consisting in a single cluster with all vertices ( section [ sec3_2_2 ] ) . however , a large value for the modularity maximum does not necessarily mean that a graph has community structure . random graphs are supposed to have no community structure , as the linking probability between vertices is either constant or a function of the vertex degrees , so there is no bias _ a priori _ towards special groups of vertices . still , random graphs may have partitions with large modularity values @xcite . this is due to fluctuations in the distribution of edges in the graph , which in many graph realizations is not homogeneous even if the linking probability is constant , like in erds - rnyi graphs . the fluctuations determine concentrations of links in some subsets of the graph , which then appear like communities . according to the definition of modularity , a graph has community structure with respect to a random graph with equal size and expected degree sequence . therefore , the modularity maximum of a graph reveals a significant community structure only if it is appreciably larger than the modularity maximum of random graphs of the same size and expected degree sequence . the significance of the modularity maximum @xmath403 for a graph can be estimated by calculating the maximum modularity for many realizations of the null model , obtained from the original graph by randomly rewiring its edges . one then computes the average @xmath404 and the standard deviation @xmath405 of the results . the statistical significance of @xmath403 is indicated by the distance of @xmath403 from the null model average @xmath404 in units of the standard deviation @xmath405 , i. e. by the @xmath406-score @xmath407 if @xmath408 , @xmath403 indicates strong community structure . cutoff values of @xmath409 for the @xmath406-scores are customary . this approach has problems , though . it can generate both false positives and false negatives : a few graphs that most people would consider without a significant community structure have a large @xmath406-score ; on the other hand , some graphs that are agreed to display cluster structure have very low values for the @xmath406-score . besides , the distribution of the maximum modularity values of the null model , though peaked , is not gaussian . therefore , one can not attribute to the values of the @xmath406-score the significance corresponding to a gaussian distribution , and one would need instead to compute the statistical significance for the right distribution . reichardt and bornholdt have studied the issue of the modularity values for random graphs in some depth @xcite , using their general spin glass formulation of the clustering problem ( section [ sub_sec6_01 ] ) . they considered the general case of a random graph with arbitrary degree distribution @xmath410 and without degree - degree correlations . they set @xmath368 , so that the energy of the ground state coincides with modularity ( up to a constant factor ) . for modularity s null model graphs , the modularity maximum corresponds to an equipartition of the graph , i. e. the magnetization of the ground state of the spin glass is zero , a result confirmed by numerical simulations @xcite . this is because the distribution of the couplings has zero mean , and the mean is only coupled to magnetization @xcite . for a partition of any graph with @xmath15 vertices and @xmath14 edges in @xmath91 clusters with equal numbers of vertices , there is a simple linear relation between the cut size @xmath411 of the partition and its modularity @xmath412 : @xmath413 $ ] . we remind that the cut size @xmath411 is the total number of inter - cluster edges of the partition ( section [ sec4_1 ] ) . in this way , the partition with maximum modularity is also the one with minimum cut size , and community detection becomes equivalent to graph partitioning . reichardt and bornholdt derived analytically the ground state energy for ising spins ( @xmath414 ) , which corresponds to the following expression of the expected maximum modularity @xmath415 for a bipartition @xcite @xmath416 here @xmath417 and @xmath418 is the ground state energy of the sherrington - kirkpatrick model @xcite . the most interesting feature of eq . [ eqr19 ] is the simple scaling with @xmath419 . numerical calculations show that this scaling holds for both erds - rnyi and scale - free graphs ( section [ sec1_3 ] ) . interestingly , the result is valid for partitions in @xmath91 clusters , where @xmath91 is left free , not only for @xmath414 . the number of clusters of the partition with maximum modularity decreases if the average degree @xmath420 increases , and tends to @xmath421 for large values of @xmath420 , regardless of the degree distribution and the size of the graph . from eq . [ eqr19 ] we also see that the expected maximum modularity for a random graph increases when @xmath420 decreases , i. e. if the graph gets sparser . so it is particularly hard to detect communities in sparse graphs by using modularity optimization . as we shall see in section [ sec6_4 ] , the sparsity of a graph is generally a serious obstacle for graph clustering methods , no matter if one uses modularity or not . a more fundamental issue , raised by fortunato and barthlemy @xcite , concerns the capability of modularity to detect `` good '' partitions . if a graph has a clear cluster structure , one expects that the maximum modularity of the graph reveals it . the null model of modularity assumes that any vertex @xmath67 `` sees '' any other vertex @xmath68 , and the expected number of edges between them is @xmath369 . similarly , the expected number of edges between two clusters @xmath258 and @xmath259 with total degrees @xmath422 and @xmath423 , respectively , is @xmath424 . the variation of modularity determined by the merger of @xmath258 and @xmath259 with respect to the partition in which they are separate clusters is @xmath425 , with @xmath426 number of edges connecting @xmath258 to @xmath259 . if @xmath427 , i. e. there is a single edge joining @xmath258 to @xmath259 , we expect that the two subgraphs will often be kept separated . instead , if @xmath428 , @xmath429 . let us suppose for simplicity that @xmath430 , i. e. that the two subgraphs are of about the same size , measured in terms of edges . we conclude that , if @xmath431 and the two subgraphs @xmath258 and @xmath259 are connected , modularity is greater if they are in the same cluster @xcite . the reason is intuitive : if there are more edges than expected between @xmath258 and @xmath259 , there is a strong topological correlation between the subgraphs . if the subgraphs are sufficiently small ( in degree ) , the expected number of edges for the null model can be smaller than one , so even the weakest possible connection ( a single edge ) suffices to keep the subgraphs together . interestingly , this result holds independently of the structure of the subgraphs . in particular it remains true if the subgraphs are cliques , which are the subgraphs with the largest possible density of internal edges , and represent the strongest possible communities . resolution limit of modularity optimization . the natural community structure of the graph , represented by the individual cliques ( circles ) , is not recognized by optimizing modularity , if the cliques are smaller than a scale depending on the size of the graph . in this case , the maximum modularity corresponds to a partition whose clusters include two or more cliques ( like the groups indicated by the dashed contours ) . reprinted figure with permission from ref . 2007 from the national academy of science of the usa . ] in fig . [ figure7 ] a graph is made out of @xmath115 identical cliques , with @xmath76 vertices each , connected by single edges . it is intuitive to think that the clusters of the best partition are the individual cliques : instead , if @xmath115 is larger than about @xmath432 , modularity would be higher for partitions in which the clusters are groups of cliques ( like the clique pairs indicated by the dashed lines in the figure ) . the conclusion is striking : modularity optimization has a resolution limit that may prevent it from detecting clusters which are comparatively small with respect to the graph as a whole , even when they are well defined communities like cliques . so , if the partition with maximum modularity includes clusters with total degree of the order of @xmath433 or smaller , one can not know _ a priori _ whether the clusters are single communities or combinations of smaller weakly interconnected communities . this resolution problem has a large impact in practical applications . real graphs with community structure usually contain communities which are very diverse in size @xcite , so many ( small ) communities may remain undetected . besides , modularity is extremely sensitive to even individual connections . many real graphs , in biology and in the social sciences , are reconstructed through experiments and surveys , so edges may occasionally be false positives : if two small subgraphs happen to be connected by a few false edges , modularity will put them in the same cluster , inferring a relationship between entities that in reality may have nothing to do with each other . the resolution limit comes from the very definition of modularity , in particular from its null model . the weak point of the null model is the implicit assumption that each vertex can interact with every other vertex , which implies that each part of the graph knows about everything else . this is however questionable , and certainly wrong for large systems like , e.g. , the web graph . it is certainly more reasonable to assume that each vertex has a limited horizon within the graph , and interacts just with a portion of it . however , nobody knows yet how to define such local territories for the graph vertices . the null model of the localized modularity of muff et al . ( section [ sub_sec6_01 ] ) is a possibility , since it limits the horizon of a vertex to a local neighborhood , comprising the cluster of the vertex and the clusters linked to it by at least one edge ( neighboring clusters ) . however , there are many other possible choices . in this respect , the null model of girvan and newman , though unrealistic , is the simplest one can think of , which partly explains its success . quality functions that , like modularity , are based on a null model such that the horizon of vertices is of the order of the size of the whole graph , are likely to be affected by a resolution limit @xcite . the problem is more general , though . for instance , li et al . @xcite have introduced a quality function , called _ modularity density _ , which consists in the sum over the clusters of the ratio between the difference of the internal and external degrees of the cluster and the cluster size . the modularity density does not require a null model , and delivers better results than modularity optimization ( e. g. it correctly recovers the natural partition of the graph in fig . [ figure7 ] for any number / size of the cliques ) . however , it is still affected by a resolution limit . to avoid that , li et al . proposed a more general definition of their measure , including a tunable parameter that allows to explore the graph at different resolutions , in the spirit of the methods of section [ sub_sec6_20 ] . a way to go around the resolution limit problem could be to perform further subdivisions of the clusters obtained from modularity optimization , in order to eliminate possible artificial mergers of communities . for instance , one could recursively optimize modularity for each single cluster , taking the cluster as a separate entity @xcite . however , this is not a reliable procedure , for two reasons : 1 ) the local modularities used to find partitions within the clusters have different null models , as they depend on the cluster sizes , so they are inconsistent with each other ; 2 ) one needs to define a criterion to decide when one has to stop partitioning a cluster , but there is no obvious prescription , so any choice is necessarily based on arbitrary assumptions - score is sufficiently high , the subgraph is supposed to have community structure and one accepts the partition in smaller pieces . the procedure stops when none of the subgraphs of the running partitions has significant community structure , based on modularity . ] . resolution limits arise as well in the more general formulation of community detection by reichardt and bornholt @xcite . here the limit scale for the undetectable clusters is @xmath434 . we remind that @xmath371 weighs the contribution of the null model term in the quality function . for @xmath368 one recovers the resolution limit of modularity . by tuning the parameter @xmath371 it is possible to arbitrarily vary the resolution scale of the corresponding quality function this in principle solves the problem of the resolution limit , as one could adjust the resolution of the method to the actual scale of the communities to detect . the problem is that usually one has no information about the community sizes , so it is not possible to decide _ a priori _ the proper value(s ) of @xmath371 for a specific graph . in the most recent literature on graph clustering quite a few _ multiresolution methods _ have been introduced , addressing this problem in several ways . we will discuss them in some detail in section [ sub_sec6_20 ] . the resolution limit can be easily extended to the case of weighted graphs . in a recent paper @xcite , berry et al . have considered the special case in which intracluster edges have weight @xmath2 , whereas intercluster edges have weight @xmath435 . by repeating the same procedure as in ref . @xcite , they conclude that clusters with internal strength ( i. e. sum of all weights of internal edges ) @xmath436 may remain undetected if @xmath437 , where @xmath86 is the total strength of the graph . so , the resolution limit decreases when @xmath435 decreases . berry et al . use this result to show that , by properly weighting the edges of a given unweighted graph , it becomes possible to detect clusters with very high resolution by still using modularity optimization . low - dimensional visualization of the modularity landscape for the metabolic network of the spirochete _ treponema pallidum_. the big degeneracy of suboptimal high - modularity partitions is revealed by the plateau ( whose shape is detailed in the inset ) , which is large and very irregular . modularity values in the plateau are very close to the absolute maximum , although they may correspond to quite different partitions . reprinted figure with permission from ref . @xcite . ] very recently , good et al . @xcite have made a careful analysis of modularity and its performance . they discovered that the modularity landscape is characterized by an exponential number of distinct states / partitions , whose modularity values are very close to the global maximum ( fig . [ figterr ] ) . this problem is particularly dramatic if a graph has a hierarchical community structure , like most real networks . such enormous number of solutions explains why many heuristics are able to come very close to modularity s global maximum , but it also implies that the global maximum is basically impossible to find . in addition , high - modularity partitions are not necessarily similar to each other , despite the proximity of their modularity scores . the optimal partition from a topological point of view , which usually does not correspond to the modularity maximum due to the resolution limit , may however have a large modularity score . therefore the optimal partition is basically indistinguishable from a huge number of high - modularity partitions , which are in general structurally dissimilar from it . the large structural inhomogeneity of the high - modularity partitions implies that one can not rely on any of them , at least in principle , in the absence of additional information on the particular system at hand and its structure . in sections [ sec4_1 ] and [ sec4_4 ] we have learned that spectral properties of graph matrices are frequently used to find partitions . a paradigmatic example is spectral graph clustering , which makes use of the eigenvectors of laplacian matrices ( section [ sec4_4 ] ) . we have also seen that newman - girvan modularity can be optimized by using the eigenvectors of the modularity matrix ( section [ sub_sec6_0_4 ] ) . most spectral methods have been introduced and developed in computer science and generally focus on data clustering , although applications to graphs are often possible as well . in this section we shall review recent spectral techniques proposed mostly by physicists explicitly for graph clustering . early works have shown that the eigenvectors of the _ transfer matrix _ @xmath438 ( section [ sec1_2 ] ) can be used to extract useful information on community structure . the transfer matrix acts as a time propagator for the process of random walk on a graph . given the eigenvector @xmath439 of the transposed transfer matrix @xmath440 , corresponding to the eigenvalue @xmath441 , @xmath442 is the outgoing current flowing from vertex @xmath67 , corresponding to the eigenmode @xmath18 . the _ participation ratio _ ( pr ) @xmath443^{-1 } \label{eqm19}\ ] ] indicates the effective number of vertices contributing to eigenvector @xmath439 . if @xmath444 receives contributions only from vertices of the same cluster , i. e. eigenvector @xmath439 is `` localized '' , the value of @xmath444 indicates the size of that cluster @xcite . the significance of the cluster can be assessed by comparing @xmath444 with the corresponding participation ratio for a random graph with the same expected degree sequence as the original graph . eigenvectors of the adjacency matrix may be localized as well if the graph has a clear community structure @xcite . a recent comprehensive analysis of spectral properties of modular graphs has been carried out by mitrovi and tadi @xcite . donetti and muoz have devised an elegant method based on the eigenvectors of the laplacian matrix @xcite . the idea is the same as in spectral graph clustering ( section [ sec4_4 ] ) : since the values of the eigenvector components are close for vertices in the same community , one can use them as coordinates , such that vertices turn into points in a metric space . so , if one uses @xmath445 eigenvectors , one can embed the vertices in an @xmath445-dimensional space . communities appear as groups of points well separated from each other , as illustrated in fig . [ figure8 ] . the separation is the more visible , the larger the number of dimensions / eigenvectors @xmath445 . spectral algorithm by donetti and muoz . vertex @xmath67 is represented by the values of the @xmath67th components of laplacian eigenvectors . in this example , the graph has an ad - hoc division in four communities , indicated by the colours . the communities are better separated in two dimensions ( b ) than in one ( a ) . reprinted figure with permission from ref . 2004 by iop publishing and sissa . ] the originality of the method consists in the procedure to group the points and to extract the partition . donetti and muoz used hierarchical clustering ( see section [ sec4_2 ] ) , with the constraint that only pairs of clusters which have at least one interconnecting edge in the original graph are merged . among all partitions of the resulting dendrogram , the one with largest modularity is chosen . for the similarity measure between vertices , donetti and muoz used both the euclidean distance and the angle distance . the angle distance between two points is the angle between the vectors going from the origin of the @xmath445-dimensional space to either point . tests on the benchmark by girvan and newman ( section [ sec6_1 ] ) show that the best results are obtained with complete - linkage clustering . the most computationally expensive part of the algorithm is the calculation of the laplacian eigenvectors . since a few eigenvectors suffice to get good partitions , one can determine them with the lanczos method @xcite . the number @xmath445 of eigenvectors that are needed to have a clean separation of the clusters is not known _ a priori _ , but one can compute a number @xmath446 of them and search for the highest modularity partition among those delivered by the method for all @xmath447 . in a related work , simonsen has embedded graph vertices in space by using as coordinates the components of the eigenvectors of the right stochastic matrix @xcite . eigenvalues and eigenvectors of the laplacian matrix have been used by alves to compute the effective conductances for pairs of vertices in a graph , assuming that the latter is an electric network with edges of unit resistance @xcite . the conductances enable one to compute the transition probabilities for a random walker moving on the graph , and from the transition probabilities one builds a similarity matrix between vertex pairs . hierarchical clustering is applied to join vertices in groups . the method can be trivially extended to the case of weighted graphs . the algorithm by alves is rather slow , as one needs to compute the whole spectrum of the laplacian , which requires a time @xmath154 . moreover , there is no criterion to select which partition(s ) of the dendrogram is ( are ) the best . capocci et al . @xcite used eigenvector components of the right stochastic matrix @xmath448 ( section [ sec1_2 ] ) , that is derived from the adjacency matrix by dividing each row by the sum of its elements . the right stochastic matrix has similar properties as the laplacian . if the graph has @xmath126 connected components , the largest @xmath126 eigenvalues are equal to @xmath2 , with eigenvectors characterized by having equal - valued components for vertices belonging to the same component . in this way , by listing the vertices according to the connected components they belong to , the components of any eigenvector of @xmath448 , corresponding to eigenvalue @xmath2 , display a step - wise profile , with plateaus indicating vertices in the same connected component . for connected graphs with cluster structure , one can still see plateaus , if communities are only loosely connected to each other ( fig . [ capoccifig ] ) . basic principle of the spectral algorithm by capocci et al . the bottom diagram shows the values of the components of the second eigenvector of the right stochastic matrix for the graph drawn on the top . the three plateaus of the eigenvector components correspond to the three evident communities of the graph . reprinted figures with permission from ref . 2005 by elsevier.,title="fig : " ] 0.4 cm basic principle of the spectral algorithm by capocci et al . the bottom diagram shows the values of the components of the second eigenvector of the right stochastic matrix for the graph drawn on the top . the three plateaus of the eigenvector components correspond to the three evident communities of the graph . reprinted figures with permission from ref . 2005 by elsevier.,title="fig : " ] here the communities can be immediately deducted by an inspection of the components of any eigenvector with eigenvalue @xmath2 . in practical cases , plateaus are not clearly visible , and one eigenvector is not enough . however , one expects that there should be a strong correlation between eigenvector components corresponding to vertices in the same cluster . capocci et al . derived a similarity matrix , where the similarity between vertices @xmath67 and @xmath68 is the pearson correlation coefficient between their corresponding eigenvector components , averaged over a small set of eigenvectors . the eigenvectors can be calculated by performing a constrained optimization of a suitable cost function . the method can be extended to weighted and directed graphs . it is useful to estimate vertex similarities , however it does not provide a well - defined partition of the graph . yang and liu @xcite adopted a recursive bisectioning procedure . communities are subgraphs such that the external degree of each vertex does not exceed the internal degree ( _ strong communities _ or _ @xmath49-sets _ , see section [ sec3_1_2 ] ) . in the first step of the algorithm , the adjacency matrix of the graph is put in approximately block - diagonal form . this is done by computing a new centrality measure for the vertices , called _ clustering centrality_. this measure is similar to bonacich s eigenvector centrality @xcite , which is given by the eigenvector of the adjacency matrix corresponding to the largest eigenvalue . the clustering centrality of a vertex basically measures the probability that a random walker starting at that vertex hits a given target . such probability is larger if the origin and the target vertices belong to the same cluster than if they do not . if the graph has well - separated communities , the values of the clustering centrality would be similar for vertices in the same cluster . in this way , one can rearrange the original adjacency matrix by listing the vertices in non - decreasing order of their clustering centralities , and blocks would be visible . the blocks are then identified by iterative bisection : each cluster found at some step is split in two as long as the resulting parts are still communities in the strong sense , otherwise the procedure stops . the worst - case complexity of the method is @xmath449 $ ] , where @xmath289 is the number of clusters of the final partition and @xmath156 the ( average ) number of iterations required to compute the clustering centrality with the power method @xcite . since @xmath156 is fairly independent of the graph size , the method scales quite well on sparse graphs [ @xmath254 ] . the main limit of this technique is the assumption that communities are defined in the strong sense , which is too restrictive . on the other hand , one could think of using alternative definitions . this section describes methods employing processes running on the graph , focusing on spin - spin interactions , random walks and synchronization . the potts model is among the most popular models in statistical mechanics @xcite . it describes a system of spins that can be in @xmath91 different states . the interaction is ferromagnetic , i. e. it favours spin alignment , so at zero temperature all spins are in the same state . if antiferromagnetic interactions are also present , the ground state of the system may not be the one where all spins are aligned , but a state where different spin values coexist , in homogeneous clusters . if potts spin variables are assigned to the vertices of a graph with community structure , and the interactions are between neighbouring spins , it is likely that the structural clusters could be recovered from like - valued spin clusters of the system , as there are many more interactions inside communities than outside . based on this idea , inspired by an earlier paper by blatt et al . @xcite , reichardt and bornholdt proposed a method to detect communities that maps the graph onto a zero - temperature @xmath91-potts model with nearest - neighbour interactions @xcite . the hamiltonian of the model , i. e. its energy , reads @xmath451 where @xmath360 is the element of the adjacency matrix , @xmath101 is kronecker s function , @xmath452 the number of spins in state @xmath155 , @xmath359 and @xmath371 are coupling parameters . the energy @xmath453 is the sum of two competing terms : the first is the classical ferromagnetic potts model energy , and favors spin alignment ; the second term instead peaks when the spins are homogeneously distributed . the ratio @xmath454 expresses the relative importance of the two terms : by tuning @xmath454 one can explore different levels of modularity of the system , from the whole graph seen as a single cluster to clusters consisting of individual vertices . if @xmath454 is set to the value @xmath40 of the average density of edges of the graph @xmath313 , the energy of the system is smaller if spins align within subgraphs such that their internal edge density exceeds @xmath40 , whereas the external edge density is smaller than @xmath40 , i. e. if the subgraphs are clusters ( section [ sec3_1_1 ] ) . the minimization of @xmath453 is carried out via simulated annealing ( @xcite and section [ sub_sec6_0_2 ] ) , starting from a configuration where spins are randomly assigned to the vertices and the number of states @xmath91 is very high . the procedure is quite fast and the results do not depend on @xmath91 ( provided @xmath91 is sufficiently high ) . the method also allows to identify vertices shared between communities , from the comparison of partitions corresponding to global and local energy minima . the hamiltonian @xmath453 can be rewritten as @xmath455 which is the energy of an infinite - range potts spin glass , as all pairs of spins are interacting ( neighboring or not ) and there may be both positive and negative couplings . the method can be simply extended to the analysis of weighted graphs , by introducing spin couplings proportional to the edge weights , which amounts to replacing the adjacency matrix @xmath189 with the weight matrix @xmath190 in eq . [ eqspin1 ] . ispolatov et al . @xcite have adopted a similar hamiltonian as in eq . [ eqspin1 ] , with a tunable antiferromagnetic term interpolating between the corresponding term of eq . [ eqspin1 ] and the entropy term ( proportional to @xmath456 ) of the free energy , whose minimization is equivalent to finding the states of the finite - temperature potts model used by blatt et al . @xcite . [ eqspin2 ] is at the basis of the successive generalization of modularity with arbitrary null models proposed by reichardt and bornholdt , that we have discussed in section [ sub_sec6_01 ] . in another work @xcite , son et al . have presented a clustering technique based on the ferromagnetic random field ising model ( frfim ) . given a weighted graph with weight matrix @xmath190 , the hamiltonian of the frfim on the graph is @xmath457 in eq . [ eqspin3 ] @xmath458 and @xmath459 are the spin and the random magnetic field of vertex @xmath67 , respectively . the frfim has been studied to understand the nature of the spin glass phase transition @xcite and the disorder - driven roughening transition of interfaces in disordered media @xcite . the behavior of the model depends on the choice of the magnetic fields . son et al . set to zero the magnetic fields of all vertices but two , say @xmath155 and @xmath156 , for which the field has infinite strength and opposite signs . this amounts to fix the spins of @xmath155 and @xmath156 to opposite values , introducing frustration in the system . the idea is that , if @xmath155 and @xmath156 are central vertices of different communities , they impose their spin state to the other community members . so , the state of minimum energy is a configuration in which the graph is polarized into a subgraph with all positive spins and a subgraph with all negative spins , coinciding with the communities , if they are well defined . finding the minimum of @xmath453 is equivalent to solving a maximum - flow / minimum - cut problem , which can be done through well known techniques of combinatorial optimization , like the augmenting path algorithm @xcite . for a given choice of @xmath155 and @xmath156 , many ground states can be found . the vertices that end up in the same cluster in all ground states represent the cores of the clusters , which are called _ possible vertices not belonging to the coteries indicate that the two clusters overlap . in the absence of information about the cluster structure of the graph , one needs to repeat the procedure for any pair of vertices @xmath155 and @xmath156 . picking vertices of the same cluster , for instance , would not give meaningful partitions . son et al . distinguish relevant clusters if they are of about the same size . the procedure can be iteratively applied to each of the detected clusters , considered as a separate graph , until all clusters have no community structure any more . on sparse graphs , the algorithm has complexity @xmath460 , where @xmath461 , so it is very slow and can be currently used for graphs of up to few thousands vertices . if one happens to know which are the important vertices of the clusters , e.g. by computing appropriate centrality values ( like degree or site betweenness @xcite ) , the choices for @xmath155 and @xmath156 are constrained and the complexity can become as low as @xmath462 , which enables the analysis of systems with millions of vertices . tests on barabsi - albert graphs ( section [ sec1_3 ] ) show that the latter have no community structure , as expected . random walks @xcite can also be useful to find communities . if a graph has a strong community structure , a random walker spends a long time inside a community due to the high density of internal edges and consequent number of paths that could be followed . here we describe the most popular clustering algorithms based on random walks . all of them can be trivially extended to the case of weighted graphs . zhou used random walks to define a distance between pairs of vertices @xcite : the distance @xmath463 between @xmath67 and @xmath68 is the average number of edges that a random walker has to cross to reach @xmath68 starting from @xmath67 . close vertices are likely to belong to the same community . zhou defines a `` global attractor '' of a vertex @xmath67 to be a closest vertex to @xmath67 ( i. e. any vertex lying at the smallest distance from @xmath67 ) , whereas the `` local attractor '' of @xmath67 is its closest neighbour . two types of communities are defined , according to local or global attractors : a vertex @xmath67 has to be put in the same community of its attractor and of all other vertices for which @xmath67 is an attractor . communities must be minimal subgraphs , i. e. they can not include smaller subgraphs which are communities according to the chosen criterion . applications to real networks , like zachary s karate club @xcite and the college football network compiled by girvan and newman @xcite ( section [ sec6_1 ] ) , along with artificial graphs like the benchmark by girvan and newman @xcite ( section [ sec6_1 ] ) , show that the method can find meaningful partitions . the method can be refined , in that vertex @xmath67 is associated to its attractor @xmath68 only with a probability proportional to @xmath464 , @xmath19 being a sort of inverse temperature . the computation of the distance matrix requires solving @xmath15 linear - algebra equations ( as many as the vertices ) , which requires a time @xmath154 . on the other hand , an exact computation of the distance matrix is not necessary , as the attractors of a vertex can be identified by considering only a localized portion of the graph around the vertex ; therefore the method can be applied to large graphs as well . in a successive paper @xcite , zhou introduced a measure of dissimilarity between vertices based on the distance defined above . the measure resembles the definition of distance based on structural equivalence of eq . [ eq7 ] , where the elements of the adjacency matrix are replaced by the corresponding distances . graph partitions are obtained with a divisive procedure that , starting from the graph as a single community , performs successive splits based on the criterion that vertices in the same cluster must be less dissimilar than a running threshold , which is decreased during the process . the hierarchy of partitions derived by the method is representative of actual community structures for several real and artificial graphs , including zachary s karate club @xcite , the college football network @xcite and the benchmark by girvan and newman @xcite ( section [ sec6_1 ] ) . the time complexity of the procedure is again @xmath154 . the code of the algorithm can be downloaded from http://www.mpikg-golm.mpg.de/theory/people/zhou + /networkcommunity.html . in another work @xcite , zhou and lipowsky adopted biased random walkers , where the bias is due to the fact that walkers move preferentially towards vertices sharing a large number of neighbours with the starting vertex . they defined a proximity index , which indicates how close a pair of vertices is to all other vertices . communities are detected with a procedure called _ netwalk _ , which is an agglomerative hierarchical clustering method ( section [ sec4_2 ] ) , where the similarity between vertices is expressed by their proximity . the method has a time complexity @xmath154 : however , the proximity index of a pair of vertices can be computed with good approximation by considering just a small portion of the graph around the two vertices , with a considerable gain in time . the performance of the method is comparable with that of the algorithm of girvan and newman ( section [ subsec5_1 ] ) . a different distance measure between vertices based on random walks was introduced by latapy and pons @xcite . the distance is calculated from the probabilities that the random walker moves from a vertex to another in a fixed number of steps . the number of steps has to be large enough to explore a significant portion of the graph , but not too long , as otherwise one would approach the stationary limit in which transition probabilities trivially depend on the vertex degrees . vertices are then grouped into communities through an agglomerative hierarchical clustering technique based on ward s method @xcite . modularity ( section [ sec3_2_2 ] ) is used to select the best partition of the resulting dendrogram . the algorithm runs to completion in a time @xmath465 on a sparse graph , where @xmath88 is the depth of the dendrogram . since @xmath88 is often small for real graphs [ @xmath255 ] , the expected complexity in practical computations is @xmath128 . the software of the algorithm can be found at http://www-rp.lip6.fr/@xmath215latapy/pp/walktrap.html . hu et al . @xcite designed a graph clustering technique based on a signaling process between vertices , somewhat resembling diffusion . initially a vertex @xmath155 is assigned one unit of signal , all the others have no signal . in the first step , the source vertex @xmath155 sends one unit of signal to each of its neighbors . next , all vertices send as many units of signals they have to each of their neighbors . the process is continued until a given number of iterations @xmath467 is reached . the intensity of the signal at vertex @xmath67 , normalized by the total amount of signal , is the @xmath67-th entry of a vector @xmath468 , representing the source vertex @xmath155 . the procedure is then repeated by choosing each vertex as source . in this way one can associate an @xmath15-dimensional vector to each vertex , which correspons to a point in an euclidean space . the vector @xmath468 is actually the @xmath155-th column of the matrix @xmath469 , where @xmath470 and @xmath189 are the identity and adjacency matrix , respectively . the idea is that the vector @xmath468 describes the influence that vertex @xmath155 exerts on the graph through signaling . vertices of the same community are expected to have similar influence on the graph and thus to correspond to vectors which are `` close '' in space . the vectors are finally grouped via fuzzy @xmath44-means clustering ( section [ sec4_3 ] ) . the optimal number of clusters corresponds to the partition with the shortest average distance between vectors in the same community and the largest average distance between vectors of different communities . the signaling process is similar to diffusion , but with the important difference that here there is no flow conservation , as the amount of signal at each vertex is not distributed among its neighbors but transferred entirely to each neighbor ( as if the vertex sent multiple copies of the same signal ) . the complexity of the algorithm is @xmath471 $ ] , where @xmath420 is the average degree of the graph . like in the previous algorithm by latapy and pons @xcite , finding an optimal value for the number of iterations @xmath467 is non - trivial . delvenne et al . @xcite have shown that random walks enable one to introduce a general quality function , expressing the persistence of clusters in time . a cluster is persistent with respect to a random walk after @xmath156 time steps if the probability that the walker escapes the cluster before @xmath156 steps is low . such probability is computed via the _ clustered autocovariance matrix _ @xmath472 , which , for a partition of the graph in @xmath117 clusters , is defined as @xmath473 here , @xmath474 is the @xmath475 membership matrix , whose element @xmath476 equals one if vertex @xmath67 is in cluster @xmath68 , zero otherwise ; @xmath477 is the transition matrix of the random walk ; @xmath478 the diagonal matrix whose elements are the stationary probabilities of the random walk , i. e. @xmath479 , @xmath105 being the degree of vertex @xmath67 ; @xmath480 is the vector whose entries are the diagonal elements of @xmath478 . the element @xmath481 expresses the probability for the walk to start in cluster @xmath67 and end up in cluster @xmath68 after @xmath156 steps , minus the stationary probability that two independent random walkers are in @xmath67 and @xmath68 . in this way , the persistence of a cluster @xmath67 is related to the diagonal element @xmath482 . delvenne et al . defined the _ stability of the clustering _ @xmath483 . \label{eqbarahona1}\ ] ] the aim is then , for a given time @xmath156 , finding the partition with the largest value for @xmath484 . for @xmath485 , the most stable partition is that in which all vertices are their own clusters . interestingly , for @xmath486 , maximizing stability is equivalent to maximizing newman - girvan modularity ( section [ sec3_2_2 ] ) . the cut size of the partition ( section [ sec4_1 ] ) equals @xmath487 $ ] , so it is also a one - step measure . in the limit @xmath488 , the most stable partition coincides with the fiedler partition @xcite , i. e. the bipartition where vertices are put in the same class according to the signs of the corresponding component of the fiedler eigenvector ( section [ sec4_1 ] ) . therefore , the measure @xmath484 is very general , and gives a unifying interpretation in the framework of the random walk of several measures that were defined in different contexts . in particular , modularity has a natural interpretation in this dynamic picture @xcite . since the size of stable clusters increases with @xmath156 , time can be considered as a resolution parameter . resolution can be fine tuned by taking time as a continuous variable ( the extension of the formalism is straightforward ) ; the linearization of the stability measure at small ( continuous ) times delivers multiresolution versions of modularity @xcite ( section [ sub_sec6_201 ] ) . in a method by weinan et al . @xcite , the best partition of a graph in @xmath44 clusters is such that the markov chain describing a random walk on the meta - graph , whose vertices are the clusters of the original graph , gives the best approximation of the full random walk dynamics on the whole graph . the quality of the approximation is given by the distance between the left stochastic matrices of the two processes , which thus needs to be minimized . the minimization is performed by using a variant of the @xmath44-means algorithm ( section [ sec4_3 ] ) , and the result is the best obtained out of @xmath76 runs starting from different initial conditions , a strategy that considerably improves the quality of the optimum . the time complexity is @xmath489 $ ] , where @xmath156 is the number of steps required to reach convergence . the optimal number of clusters could in principle be determined by analyzing how the quality of the approximation varies with @xmath44 , but the authors do not give any general recipe . the method is rather accurate on the benchmark by girvan and newman @xcite ( section [ sec6_1 ] ) and on zachary s karate club network . the algorithm by weinan et al . is asymptotically equivalent to spectral graph partitioning ( section [ sec4_4 ] ) when the markov chain describing the random walk presents a sizeable spectral gap between some of the largest eigenvalues of the transfer matrix ( section [ sec1_2 ] ) , approximately equal to one , and the others . we conclude this section by describing the _ markov cluster algorithm ( mcl ) _ , which was invented by van dongen @xcite . this method simulates a peculiar process of flow diffusion in a graph . one starts from the _ transfer matrix _ of the graph @xmath201 ( section [ sec1_2 ] ) . the element @xmath490 of the transfer matrix gives the probability that a random walker , sitting at vertex @xmath68 , moves to @xmath67 . the sum of the elements of each column of @xmath438 is one . each iteration of the algorithm consists of two steps . in the first step , called _ expansion _ , the transfer matrix of the graph is raised to an integer power @xmath45 ( usually @xmath491 ) . the entry @xmath492 of the resulting matrix gives the probability that a random walker , starting from vertex @xmath68 , reaches @xmath67 in @xmath45 steps ( diffusion flow ) . the second step , which has no physical counterpart , consists in raising each single entry of the matrix @xmath477 to some power @xmath18 , where @xmath18 is now real - valued . this operation , called _ inflation _ , enhances the weights between pairs of vertices with large values of the diffusion flow , which are likely to be in the same community . next , the elements of each column must be divided by their sum , such that the sum of the elements of the column equals one and a new transfer matrix is recovered . after some iterations , the process delivers a stable matrix , with some remarkable properties . its elements are either zero or one , so it is a sort of adjacency matrix . most importantly , the graph described by the matrix is disconnected , and its connected components are the communities of the original graph . the method is really simple to implement , which is the main reason of its success : as of now , the mcl is one of the most used clustering algorithms in bioinformatics . the code can be downloaded from http://www.micans.org / mcl/. due to the matrix multiplication of the expansion step , the algorithm should scale as @xmath154 , even if the graph is sparse , as the running matrix becomes quickly dense after a few steps of the algorithm . however , while computing the matrix multiplication , mcl keeps only a maximum number @xmath44 of non - zero elements per column , where @xmath44 is usually much smaller than @xmath15 . so , the actual worst - case running time of the algorithm is @xmath493 on a sparse graph . a problem of the method is the fact that the final partition is sensitive to the parameter @xmath18 used in the inflation step . therefore several different partitions can be obtained , and it is not clear which are the most meaningful or representative . synchronization @xcite is an emergent phenomenon occurring in systems of interacting units and is ubiquitous in nature , society and technology . in a synchronized state , the units of the system are in the same or similar state(s ) at every time . synchronization has also been applied to find communities in graphs . if oscillators are placed at the vertices , with initial random phases , and have nearest - neighbour interactions , oscillators in the same community synchronize first , whereas a full synchronization requires a longer time . so , if one follows the time evolution of the process , states with synchronized clusters of vertices can be quite stable and long - lived , so they can be easily recognized . this was first shown by arenas , daz - guilera and prez - vicente @xcite . they used kuramoto oscillators @xcite , which are coupled two - dimensional vectors endowed with a proper frequency of oscillations . in the kuramoto model , the phase @xmath494 of oscillator @xmath67 evolves according to the following dynamics @xmath495 where @xmath496 is the natural frequency of @xmath67 , @xmath289 the strength of the coupling between oscillators and the sum runs over all oscillators ( mean field regime ) . if the interaction coupling exceeds a threshold , depending on the width of the distribution of natural frequencies , the dynamics leads to synchronization . if the dynamics runs on a graph , each oscillator is coupled only to its nearest neighbors . in order to reveal the effect of local synchronization , arenas et al . introduced the local order parameter @xmath497\rangle , \label{eqsync2}\ ] ] measuring the average correlation between oscillators @xmath67 and @xmath68 . the average is computed over different initial conditions . by visualizing the correlation matrix @xmath498 at a given time @xmath156 , one may distinguish groups of vertices that synchronize together . the groups can be identified by means of the _ dynamic connectivity matrix _ @xmath499 , which is a binary matrix obtained from @xmath498 by thresholding its entries . the dynamic connectivity matrix embodies information about both the synchronization dynamics and the underlying graph topology . from the spectrum of @xmath499 it is possible to derive the number of disconnected components at time @xmath156 . by plotting the number of components as a function of time , plateaus may appear at some characteristic time scales , indicating structural scales of the graph with robust communities ( fig . [ syncarenas ] ) . partitions corresponding to long plateaus are characterized by high values of the modularity of newman and girvan ( section [ sec3_2_2 ] ) on graphs with homogeneous degree distributions , whereas such correlation is poor in the presence of hubs @xcite . indeed , it has been proven that the stability ( eq . [ eqbarahona1 ] ) of the dynamics associated to the standard laplacian matrix , which describes the convergence towards synchronization of the kuramoto model with equal intrinsic frequencies , coincides with modularity only for graphs whose vertices have the same degree @xcite . the appearance of plateaus at different time scales hints to a hierarchical organization of the graph . after a sufficiently long @xmath156 all oscillators are synchronized and the whole system behaves as a single component . interestingly , arenas et al . found that the structural scales revealed by synchronization correspond to groups of eigenvalues of the laplacian matrix of the graph , separated by gaps ( fig . [ syncarenas ] ) . based on the same principle , boccaletti et al . designed a community detection method based on synchronization @xcite . the synchronization dynamics is a variation of kuramoto s model , the opinion changing rate ( ocr ) model @xcite . here the interaction coupling between adjacent vertices is weighted by a term proportional to a ( negative ) power of the betweenness of the edge connecting the vertices ( section [ subsec5_1 ] ) , with exponent @xmath18 . the evolution equations of the model are solved by decreasing the value of @xmath18 during the evolution of the dynamics , starting from a configuration in which the system is fully synchronized ( @xmath500 ) . the graph tends to get split into clusters of synchronized elements , because the interaction strengths across inter - cluster edges get suppressed due to their high betweenness scores . by varying @xmath18 , different partitions are recovered , from the graph as a whole until the vertices as separate communities : the partition with the largest value of modularity is taken as the most relevant . the algorithm scales in a time @xmath205 , or @xmath129 on sparse graphs , and gives good results in practical examples , including zachary s karate club @xcite and the benchmark by girvan and newman @xcite ( section [ sec6_1 ] ) . the method can be refined by homogeneizing the natural frequencies of the oscillators during the evolution of the system . in this way , the system becomes more stable and partitions with higher modularity values can be recovered . in a recent paper by li et al . @xcite , it was shown that synchronized clusters in modular networks are characterized by interfacial vertices , whose oscillation frequency is intermediate between those of two or more clusters , so that they do not belong to a specific community . li et al . used this result to devise a technique able to detect overlapping communities . synchronization - based algorithms may not be reliable when communities are very different in size ; tests in this direction are still missing . _ statistical inference _ @xcite aims at deducing properties of data sets , starting from a set of observations and model hypotheses . if the data set is a graph , the model , based on hypotheses on how vertices are connected to each other , has to _ fit _ the actual graph topology . in this section we review those clustering techniques attempting to find the best fit of a model to the graph , where the model assumes that vertices have some sort of classification , based on their connectivity patterns . we mainly focus on methods adopting _ bayesian inference _ @xcite , in which the best fit is obtained through the maximization of a likelihood ( _ generative models _ ) , but we also discuss related techniques , based on _ blockmodeling _ @xcite , _ model selection _ @xcite and _ information theory _ @xcite . bayesian inference uses observations to estimate the probability that a given hypothesis is true . it consists of two ingredients : the evidence , expressed by the information @xmath501 one has about the system ( e.g. , through measurements ) ; a statistical model with parameters @xmath502 . bayesian inference starts by writing the likelihood @xmath503 that the observed evidence is produced by the model for a given set of parameters @xmath502 . the aim is to determine the choice of @xmath502 that maximizes the posterior distribution @xmath504 of the parameters given the model and the evidence . by using bayes theorem one has @xmath505 where @xmath506 is the prior distribution of the model parameters and @xmath507 unfortunately , computing the integral [ eqstinf01 ] is a major challenge . moreover , the choice of the prior distribution @xmath506 is non - obvious . generative models differ from each other by the choice of the model and the way they address these two issues . bayesian inference is frequently used in the analysis and modeling of real graphs , including social @xcite and biological networks @xcite . graph clustering can be considered a specific example of inference problem . here , the evidence is represented by the graph structure ( adjacency or weight matrix ) and there is an additional ingredient , represented by the classification of the vertices in groups , which is a _ hidden _ ( or _ missing _ ) information that one wishes to infer along with the parameters of the model which is supposed to be responsible for the classification . this idea is at the basis of several recent papers , which we discuss here . in all these works , one essentially maximizes the likelihood @xmath503 that the model is consistent with the observed graph structure , with different constraints . we specify the set of parameters @xmath502 as the triplet @xmath508 , where @xmath509 indicates the community assignment of the vertices , @xmath510 the model parameters , and @xmath44 the number of clusters . in the following we shall stick to the notation of the papers , so the variables above may be indicated by different symbols . however , to better show what each method specifically does we shall refer to our general notation at the end of the section . hastings @xcite chooses as a model of network with communities the _ planted partition model _ ( section [ sec6 ] ) . in it , @xmath15 vertices are assigned to @xmath91 groups : vertices of the same group are linked with a probability @xmath511 , while vertices of different groups are linked with a probability @xmath512 . if @xmath513 , the model graph has a built - in community structure . the vertex classification is indicated by the set of labels @xmath514 . the probability that , given a graph , the classification @xmath514 is the right one according to the model is @xmath515\}^{-1 } , \label{eqstinf1}\ ] ] where @xmath516/[p_{out}(1-p_{in})]\}$ ] , @xmath517 $ ] and the first sum runs over nearest neighboring vertices . maximizing @xmath518 is equivalent to minimizing the argument of the exponential , which is the hamiltonian of a potts model with short- and long - range interactions . for @xmath513 , @xmath519 and @xmath520 , so the model is a spin glass with ferromagnetic nearest - neighbor interactions and antiferromagnetic long - range interactions , similar to the model proposed by reichardt and bornholdt to generalize newman - girvan modularity @xcite ( section [ sub_sec6_01 ] ) . hastings used belief propagation @xcite to find the ground state of the spin model . on sparse graphs , the complexity of the algorithm is expected to be @xmath521 , where @xmath18 needs to be estimated numerically . in principle one needs to input the parameters @xmath511 and @xmath512 , which are usually unknown in practical applications . however , it turns out that they can be chosen rather arbitrarily , and that bad choices can be recognized and corrected . newman and leicht @xcite have recently proposed a similar method based on a mixture model and the expectation - maximization technique @xcite . the method bears some resemblance with an _ a posteriori _ blockmodel previously introduced by snijders and nowicki @xcite . they start from a directed graph with @xmath15 vertices , whose vertices fall into @xmath117 classes . the group of vertex @xmath67 is indicated by @xmath522 , @xmath523 the fraction of vertices in group @xmath375 , and @xmath524 the probability that there is a directed edge from vertices of group @xmath375 to vertex @xmath67 . by definition , the sets @xmath525 and @xmath526 satisfy the normalization conditions @xmath527 and @xmath528 . apart from normalization , the probabilities @xmath526 are assumed to be independent of each other . the best classification of the vertices corresponds to the maximum of the average log - likelihood @xmath529 that the model , described by the values of the parameters @xmath525 and @xmath526 fits the adjacency matrix @xmath189 of the graph . the expression of the average log - likelihood @xmath529 requires the definition of the probability @xmath530 , that vertex @xmath67 belongs to group @xmath375 . by applying bayes theorem the probabilities @xmath531 can be computed in terms of the @xmath525 and the @xmath526 , as @xmath532 while the maximization of the average log - likelihood @xmath529 , under the normalization constraints of the model variables @xmath525 and @xmath526 , yields the relations @xmath533 where @xmath105 is the outdegree of vertex @xmath67 . equations [ eqstinf2 ] and [ eqstinf3 ] are self - consistent , and can be solved by iterating them to convergence , starting from a suitable set of initial conditions . convergence is fast , so the algorithm could be applied to fairly large graphs , with up to about @xmath243 vertices . the method , designed for directed graphs , can be easily extended to the undirected case , whereas an extension to weighted graphs is not straightforwad . a nice feature of the method is that it does not require any preliminary indication on what type of structure to look for ; the resulting structure is the most likely classification based on the connectivity patterns of the vertices . therefore , various types of structures can be detected , not necessarily communities . for instance , multipartite structure could be uncovered , or mixed patterns where multipartite subgraphs coexist with communities , etc .. in this respect , it is more powerful than most methods of community detection , which are bound to focus only on proper communities , i. e. subgraphs with more internal than external edges . in addition , since partitions are defined by assigning probability values to the vertices , expressing the extent of their membership in a group , it is possible that some vertices are not clearly assigned to a group , but to more groups , so the method is able to deal with overlapping communities . the main drawback of the algorithm is the fact that one needs to specify the number of groups @xmath117 at the beginning of the calculation , a number that is typically unknown for real networks . it is possible to derive this information self - consistently by maximizing the probability that the data are reproduced by partitions with a given number of clusters . but this procedure involves some degree of approximation , and the results are often not good . in a recent study it has been shown that the method by newman and leicht enables one to rank vertices based on their degree of influence on other vertices , which allows to identify the vertices responsible for the group structure and its stability @xcite . a very similar technique has also been applied by vzquez @xcite to the problem of population stratification , where animal populations and their attributes are represented as hypergraphs ( section [ sec1_1 ] ) . vzquez also suggested an interesting criterion to decide the optimal number of clusters , namely picking the number @xmath534 whose solution has the greatest similarity with solutions obtained at different values of @xmath117 . the similarity between two partitions can be estimated in various ways , for instance by computing the normalized mutual information ( section [ sec6 ] ) . in a successive paper @xcite , vzquez showed that better results are obtained if the classification likelihood is maximized by using variational bayes @xcite . ramasco and mungan @xcite remarked that the normalization condition on the probabilities @xmath526 implies that each group @xmath375 must have non - zero outdegree and that therefore the method fails to detect the intuitive group structure of ( directed ) bipartite graphs ( fig . [ figramasco ] ) . problem of method by newman and leicht . by applying the method to the illustrated complete bipartite graph ( colors indicate the vertex classes ) the natural group structure c ) is not recovered ; instead , the most likely classifications are a ) and b ) . reprinted figure with permission from ref . @xcite.2008 by the american physical society . ] to avoid this problem , they proposed a modification , that consists in introducing three sets for the edge probabilities @xmath526 , relative to edges going from group @xmath375 to vertex @xmath67 ( as before ) , from @xmath67 to @xmath375 and in both directions , respectively . furthermore , they used the average entropy of the classification @xmath535 , where the @xmath536 are the analogs of the probabilities in eq . [ eqstinf2 ] , to infer the optimal number of groups , that the method of newman and leicht is unable to provide . another technique similar to that by newman and leicht has been designed by ren et al . the model is based on the group fractions @xmath525 , defined as above , and a set of probabilities @xmath537 , expressing the relevance of vertex @xmath67 for group @xmath375 ; the basic assumption is that the probability that two vertices of the same group are connected by an edge is proportional to the product of the relevances of the two vertices . in this way , there is an explicit relation between group membership and edge density , and the method can only detect community structure . the community assignments are recovered through an expectation - maximization procedure that closely follows that by newman and leicht . maximum likelihood estimation has been used by opi et al . to define an axiomatization of the problem of graph clustering and its related concepts @xcite . the starting point is again the planted partition model ( section [ sec6 ] ) , with probabilities @xmath511 and @xmath512 . a novelty of the approach is the introduction of the _ size matrix _ @xmath538 , whose element @xmath539 indicates the maximum strength of interaction between vertices @xmath67 and @xmath68 . for instance , in a graph with unweighted connections , all elements of @xmath538 equal @xmath2 . in this case , the probability that the graph conceals a community structure coincides with the expression ( [ eqstinf1 ] ) by hastings . opi et al . used this probability as a quality function to define rankings between graph partitions ( _ likelihood rankings _ ) . the authors show that the likelihood rankings satisfy a number of general properties , which should be satisfied by any reasonable ranking . they also propose an algorithm to find the maximum likelihood partition , by using the auxiliary concept of _ pseudo - community structure _ , i. e. a grouping of the graph vertices in which it is specified which pairs of vertices stay in the same community and which pairs instead stay in different communities . a pseudo - community may not be a community because the transitive property is not generally valid , as the focus is on pairwise vertex relationships : it may happen that @xmath67 and @xmath68 are classified in the same group , and that @xmath68 and @xmath44 are classified in the same group , but that @xmath67 and @xmath44 are not classified as belonging to the same group . we believe that the work by opi et al . is an important first step towards a more rigorous formalization of the problem of graph clustering . zanghi et al . @xcite have designed a clustering technique that lies somewhat in between the method by hastings and that by newman and leicht . as in ref . @xcite , they use the planted partition model to represent a graph with community structure ; as in ref . @xcite , they maximize the classification likelihood using an expectation - maximization algorithm @xcite . the algorithm runs for a fixed number of clusters @xmath91 , like that by newman and leicht ; however , the optimal number of clusters can be determined by running the algorithm for a range of @xmath91-values and selecting the solution that maximizes the integrated classification likelihood introduced by biernacki et al . the time complexity of the algorithm is @xmath129 . hofman and wiggins have proposed a general bayesian approach to the problem of graph clustering @xcite . like hastings @xcite , they model a graph with community structure as in the planted partition problem ( section [ sec6 ] ) , in that there are two probabilities @xmath540 and @xmath541 that there is an edge between vertices of the same or different clusters , respectively . the unobserved community structure is indicated by the set of labels @xmath542 for the vertices ; @xmath523 is again the fraction of vertices in group @xmath375 . the conjugate prior distributions @xmath543 and @xmath544 are chosen to be beta and dirichlet distributions . the most probable number of clusters @xmath545 maximizes the conditional probability @xmath546 that there are @xmath289 clusters , given the matrix @xmath547 . like hastings , hofman and wiggins assume that the prior probability @xmath548 on the number of clusters is a smooth function , therefore maximizing @xmath546 amounts to maximizing the bayesian evidence @xmath549 , obtained by integrating the joint distribution @xmath550 , which is factorizable , over the model parameters @xmath551 and @xmath552 . the integration can be performed exactly only for small graphs . hofman and wiggins used variational bayes @xcite , in order to compute controlled approximations of @xmath553 . the complexity of the algorithm was estimated numerically on synthetic graphs , yielding @xmath554 , with @xmath555 . in fact , the main limitation comes from high memory requirements . the method is more powerful than the one by hastings @xcite , in that the edge probabilities @xmath551 are inferred by the procedure itself and need not be specified ( or guessed ) at the beginning . it also includes the expectation - maximization approach by newman and leicht @xcite as a special case , with the big advantage that the number of clusters need not be given as an input , but is an output of the method . the software of the algorithm can be found at http://www.columbia.edu/@xmath215chw2/. we conclude with a brief summary of the main techniques described above , coming back to our notation at the beginning of the section . in the method by hastings , one maximizes the likelihood @xmath556 over the set of all possible community assignments @xmath509 , given the number of clusters @xmath44 and the model parameters ( i. e. the linking probabilities @xmath511 and @xmath512 ) . newman and leicht maximize the likelihood @xmath556 for a given number of clusters , over the possible choices for the model parameters and community assignments , by deriving the optimal choices for both variables with a self - consistent procedure . hofman and wiggins maximize the likelihood @xmath557 over the possible choices for the number of clusters . block modeling is a common approach in statistics and social network analysis to decompose a graph in classes of vertices with common properties . in this way , a simpler description of the graph is attained . vertices are usually grouped in classes of equivalence . there are two main definitions of topological equivalence for vertices : _ structural equivalence _ @xcite ( section [ sec3_1_4 ] ) , in which vertices are equivalent if they have the same neighbors ; _ regular equivalence _ @xcite , in which vertices of a class have similar connection patterns to vertices of the other classes ( ex . parents / children ) . regular equivalence does not require that ties / edges are restricted to specific target vertices , so it is a more general concept than structural equivalence . indeed , vertices which are structurally equivalent are also regularly equivalent , but the inverse is not true . the concept of structural equivalence can be generalized to probabilistic models , in which one compares classes of graphs , not single graphs , characterized by a set of linking probabilities between the vertices . in this case , vertices are organized in classes such that the linking probabilities of a vertex with all other vertices of the graph are the same for vertices in the same class , which are called _ stochastically equivalent _ @xcite . a thorough discussion of blockmodeling is beyond the scope of this review : we point the reader to ref . @xcite . here we discuss a recent work by reichardt and white @xcite . let us suppose to have a directed graph with @xmath15 vertices and @xmath14 edges . a classification of the graph is indicated by the set of labels @xmath558 , where @xmath559 is the class of vertex @xmath67 . the corresponding blockmodel , or _ image graph _ , is expressed by a @xmath560 adjacency matrix @xmath256 : @xmath561 if edges between classes @xmath562 and @xmath563 are allowed , otherwise it is zero . the aim is finding the classification @xmath558 and the matrix @xmath10 that best fits the adjacency matrix @xmath189 of the graph . the goodness of the fit is expressed by the quality function @xmath564 , \label{eqstinf4}\ ] ] where @xmath565 ( @xmath566 ) reward the presence ( absence ) of edges between vertices if there are edges ( non - edges ) between the corresponding classes , and @xmath14 is the number of edges of the graph , as usual . eq . [ eqstinf4 ] can be rewritten as a sum over the classes @xmath567)b_{rs } , \label{eqstinf5}\ ] ] by setting @xmath568 and @xmath569=(1/m)\sum_{i\neq j}b_{ij}\delta_{\sigma_ir}\delta_{\sigma_js}$ ] . if one sets @xmath570 and @xmath571 , @xmath362 can be interpreted as the linking probability between @xmath67 and @xmath68 , in some null model . thereof , @xmath572 becomes the number of edges running between vertices of class @xmath375 and @xmath155 , and @xmath569 $ ] the expected number of edges in the null model . reichardt and white set @xmath573 , which defines the same null model of newman - girvan modularity for directed graphs ( section [ sub_sec6_01 ] ) . in fact , if the image graph has only self - edges , i. e. @xmath574 , the quality function @xmath575 exactly matches modularity . other choices for the image graph are possible , however . for instance , a matrix @xmath576 describes the classes of a @xmath91-partite graph ( section [ sec1_1 ] ) . from eq . [ eqstinf5 ] we see that , for a given classification @xmath558 , the image graph that yields the largest value of the quality function @xmath575 is that in which @xmath577 when the term @xmath578 $ ] is non - negative , and @xmath579 when the term @xmath578 $ ] is non - positive . so , the best classification is the one maximizing the quality function @xmath580|| , \label{eqstinf6}\ ] ] where all terms of the sum are taken in absolute value . the function @xmath581 is maximized via simulated annealing . the absolute maximum @xmath403 is obtained by construction when @xmath91 matches the number @xmath582 of structural equivalence classes of the graph . however , the absolute maximum @xmath403 does not have a meaning by itself , as one can achieve fairly high values of @xmath581 also for null model instances of the original graph , i. e. if one randomizes the graph by keeping the same expected indegree and outdegree sequences . in practical applications , the optimal number of classes is determined by comparing the ratio @xmath583 [ @xmath584 is the maximum of @xmath581 for @xmath91 classes ] with the expected ratio for the null model . since classifications for different @xmath91-values are not hierarchically ordered , overlaps between classes may be detected . the method can be trivially extended to the case of weighted graphs . model selection @xcite aims at finding models which are at the same time simple and good at describing a system / process . a basic example of a model selection problem is curve fitting . there is no clear - cut recipe to select a model , but a bunch of heuristics , like akaike information criterion ( aic ) @xcite , bayesian information criterion ( bic ) @xcite , minimum description length ( mdl ) @xcite , minimum message length ( mml ) @xcite , etc .. the modular structure of a graph can be considered as a compressed description of the graph to approximate the whole information contained in its adjacency matrix . based on this idea , rosvall and bergstrom @xcite envisioned a communication process in which a partition of a graph in communities represents a synthesis @xmath585 of the full structure that a signaler sends to a receiver , who tries to infer the original graph topology @xmath168 from it ( fig . [ rosvall1 ] ) . the same idea is at the basis of an earlier method by sun et al . @xcite , which was originally designed for bipartite graphs evolving in time and will be described in section [ sec7_1_2 ] . the best partition corresponds to the signal @xmath585 that contains the most information about @xmath168 . this can be quantitatively assessed by the minimization of the conditional information @xmath586 of @xmath168 given @xmath585 , @xmath587 , \label{eqstinf7}\ ] ] where @xmath91 is the number of clusters , @xmath193 the number of vertices in cluster @xmath67 , @xmath588 the number of edges between clusters @xmath67 and @xmath68 . we remark that , if one imposes no constraints on @xmath91 , @xmath586 is minimal in the trivial case in which @xmath589 ( @xmath590 ) . this solution is not acceptable because it does not correspond to a compression of information with respect to the original data set . one has to look for the ideal tradeoff between a good compression and a small enough information @xmath586 . the minimum description length ( mdl ) principle @xcite provides a solution to this problem , which amounts to the minimization of a function given by @xmath586 plus a function of the number @xmath15 of vertices , @xmath14 of edges and @xmath91 of clusters . the optimization is performed by simulated annealing , so the method is rather slow and can be applied to graphs with up to about @xmath251 vertices . however , faster techniques may in principle be used , even if they imply a loss in accuracy . the method appears superior than modularity optimization , especially when communities are of different sizes . this comes from tests performed on the benchmark of girvan and newman @xcite ( section [ sec6_1 ] ) , both in its original version and in asymmetric versions , proposed by the authors , where the clusters have different sizes or different average degrees . in addition , it can detect other types of vertex classifications than communities , as in eq . [ eqstinf7 ] there are no constraints on the relative importance of the edge densities within communities with respect to the edge densities between communities . the software of the algorithm can be found at http://www.tp.umu.se/@xmath215rosvall/code.html . in a recent paper @xcite , rosvall and bergstrom pursued the same idea of describing a graph by using less information than that encoded in the full adjacency matrix . the goal is to optimally compress the information needed to describe the process of information diffusion across the graph . random walk is chosen as a proxy of information diffusion . a two - level description , in which one gives unique names to important structures of the graph and to vertices within the same structure , but the vertex names are recycled among different structures , leads to a more compact description than by simply coding all vertices with different names . this is similar to the procedure usually adopted in geographic maps , where the structures are cities and one usually chooses the same names for streets of different cities , as long as there is only one street with a given name in the same city . huffman coding @xcite is used to name vertices . for the random walk , the above - mentioned structures are communities , as it is intuitive that walkers will spend a lot of time within them , so they play a crucial role in the process of information diffusion . graph clustering turns then into the following coding problem : finding the partition that yields the minimum description length of an infinite random walk . such description length consists of two terms , expressing the shannon entropy of the random walk within and between clusters . every time the walker steps to a different cluster , one needs to use the codeword of that cluster in the description , to inform the decoder of the transition . clearly , if clusters are well separated from each other , transitions of the random walker between clusters will be unfrequent , so it is advantageous to use the map , with the clusters as regions , because in the description of the random walk the codewords of the clusters will not be repeated many times , while there is a considerable saving in the description due to the limited length of the codewords used to denote the vertices . instead , if there are no well - defined clusters and/or if the partition is not representative of the actual community structure of the graph , transitions between the clusters of the partition will be very frequent and there will be little or no gain by using the two - level description of the map . the minimization of the description length is carried out by combining greedy search with simulated annealing . in a successive paper @xcite , the authors adopted the fast greedy technique designed by blondel et al . for modularity optimization @xcite , with some refinements . the method can be applied to weighted graphs , both undirected and directed . in the latter case , the random walk process is modified by introducing a teleportation probability @xmath253 , to guarantee ergodicity , just like in google s pagerank algorithm @xcite . the partitions of directed graphs obtained by the method differ from those derived by optimizing the directed version of newman - girvan modularity ( section [ sub_sec6_01 ] ) : this is due to the fact that modularity focuses on pairwise relationships between vertices , so it does not capture flows . the code of the method is available at http://www.tp.umu.se/@xmath215rosvall/code.html . chakrabarti @xcite has applied the mdl principle to put the adjacency matrix of a graph into the ( approximately ) block diagonal form representing the best tradeoff between having a limited number of blocks , for a good compression of the graph topology , and having very homogeneous blocks , for a compact description of their structure . the total encoding cost @xmath467 includes the information on the total number of vertices of the graph , on the number of blocks and the number of vertices and edges in each block , along with the adjacency matrices of the blocks . the minimization of @xmath467 is carried out by starting from the partition in which the graph is a single cluster . at each step , one operates a bipartition of the cluster of the partition with the maximum shannon entropy per vertex . the split is carried out in order to remove from the original cluster those vertices carrying the highest contribution to the entropy per vertex of the cluster . then , starting from the resulting partition , which has one more cluster than the previous one , @xmath467 is optimized among those partitions with the same number of clusters . the procedure continues until one reaches a number of clusters @xmath591 , for which @xmath467 can not be further decreased . the method by chakrabarti has complexity @xmath592 $ ] , where @xmath593 is the number of iterations required for the convergence of the optimization for a given number of clusters , which is usually small ( @xmath594 in the experiments performed by the author ) . therefore the algorithm can be applied to fairly large graphs . information theory has also been used to detect communities in graphs . ziv et al . @xcite have designed a method in which the information contained in the graph topology is compressed such to preserve some predefined information . this is the basic principle of the information bottleneck method @xcite . to understand this criterion , we need to introduce an important measure , the _ mutual information _ @xmath595 @xcite of two random variables @xmath168 and @xmath585 . it is defined as @xmath596 where @xmath597 indicates the probability that @xmath598 ( similarly for @xmath599 ) and @xmath600 is the joint probability of @xmath168 and @xmath585 , i. e. @xmath601 . the measure @xmath595 tells how much we learn about @xmath168 if we know @xmath585 , and viceversa . if @xmath168 is the input variable , @xmath602 the variable specifying the partition and @xmath585 the variable encoding the information we want to keep , which is called _ relevant variable _ , the goal is to minimize the mutual information between @xmath168 and @xmath602 ( to achieve the largest possible data compression ) , under the constraint that the information on @xmath585 extractable from @xmath602 be accurate . the optimal tradeoff between the values of @xmath603 and @xmath604 ( i. e. compression versus accuracy ) is expressed by the minimization of a functional , where the relative weight of the two contributions is given by a parameter playing the role of a temperature . in the case of graph clustering , the question is what to choose as relevant information variable . ziv et al . proposed to adopt the structural information encoded in the process of diffusion on the graph . they also introduce the concept of _ network modularity _ , which characterizes the graph as a whole , not a specific partition like the modularity by newman and girvan ( section [ sec3_2_2 ] ) . the network modularity is defined as the area under the _ information curve _ , which essentially represents the relation between the extent of compression and accuracy for all solutions found by the method and all possible numbers of clusters . the software of the algorithm by ziv et al . can be found at http://www.columbia.edu/@xmath215chw2/. in this section we describe some algorithms that do not fit in the previous categories , although some overlap is possible . raghavan et al . @xcite have designed a simple and fast method based on _ label propagation_. vertices are initially given unique labels ( e.g. their vertex labels ) . at each iteration , a sweep over all vertices , in random sequential order , is performed : each vertex takes the label shared by the majority of its neighbors . if there is no unique majority , one of the majority labels is picked at random . in this way , labels propagate across the graph : most labels will disappear , others will dominate . the process reaches convergence when each vertex has the majority label of its neighbors . communities are defined as groups of vertices having identical labels at convergence . by construction , each vertex has more neighbors in its community than in any other community . this resembles the strong definition of community we have discussed in section [ sec3_1_2 ] , although the latter is stricter , in that each vertex must have more neighbors in its community than in the rest of the graph . the algorithm does not deliver a unique solution . due to the many ties encountered along the process it is possible to derive different partitions starting from the same initial condition , with different random seeds . tests on real graphs show that all partitions found are similar to each other , though . the most precise information that one can extract from the method is contained by aggregating the various partitions obtained , which can be done in various ways . the authors proposed to label each vertex with the set of all labels it has in different partitions . aggregating partitions enables one to detect possible overlapping communities . the main advantage of the method is the fact that it does not need any information on the number and the size of the clusters . it does not need any parameter , either . the time complexity of each iteration of the algorithm is @xmath21 , the number of iterations to convergence appears independent of the graph size , or growing very slowly with it . so the technique is really fast and could be used for the analysis of large systems . in a recent paper @xcite , tibly and kertsz showed that the method is equivalent to finding the local energy minima of a simple zero - temperature kinetic potts model , and that the number of such energy minima is considerably larger than the number of vertices of the graph . aggregating partitions as raghavan et al . suggest leads to a fragmentation of the resulting partition in clusters that are the smaller , the larger the number of aggregated partitions . this is potentially a serious problem of the algorithm by raghavan et al . , especially when large graphs are investigated . in order to eliminate undesired solutions , barber and clark introduced some constraints in the optimization process @xcite . this amounts to adding some terms to the objective function @xmath605 whose maximization is equivalent to the original label propagation algorithm , where @xmath189 is the adjacency matrix of the graph and @xmath101 is kronecker s function . it is just the negative of the energy of a zero - temperature potts model , as found by tibly and kertsz @xcite ] . interestingly , if one imposes the constraint that partitions have to be balanced , i. e. that clusters have similar total degrees , the objective function becomes formally equivalent to newman - girvan modularity @xmath90 ( section [ sec3_2_2 ] ) , so the corresponding version of the label propagation algorithm is essentially based on a local optimization of modularity . leung et al . have found that the original algorithm by raghavan et al . , applied on online social networks , often yields partitions with one giant community together with much smaller ones @xcite . in order to avoid this disturbing feature , which is an artefact of the algorithm , leung et al . proposed to modify the method by introducing a score for the labels , which decreases as the label propagates far from the vertex to which the label was originally assigned . when choosing the label of a vertex , the labels of its neighbors are weighted by their scores , therefore a single label can not span too large portions of the graph ( as its weight fades away with the distance from the origin ) , and no giant communities can be recovered . tests of the modified algorithm on the lfr benchmark @xcite ( section [ sub_sec6_201 ] ) give good results and encourage further investigations . bagrow and bollt designed an agglomerative technique , called _ l - shell method _ @xcite . it is a procedure that finds the community of any vertex , although the authors also presented a more general procedure to identify the full community structure of the graph . communities are defined locally , based on a simple criterion involving the number of edges inside and outside a group of vertices . one starts from a vertex - origin and keeps adding vertices lying on successive shells , where a shell is defined as a set of vertices at a fixed geodesic distance from the origin . the first shell includes the nearest neighbours of the origin , the second the next - to - nearest neighbours , and so on . at each iteration , one calculates the number of edges connecting vertices of the new layer to vertices inside and outside the running cluster . if the ratio of these two numbers ( `` emerging degree '' ) exceeds some predefined threshold , the vertices of the new shell are added to the cluster , otherwise the process stops the idea of closing a community by expanding a shell has been previously introduced by costa @xcite , in which shells are centered on hubs . however , in this procedure the number of clusters is preassigned and no cluster can contain more than one hub . because of the local nature of the process , the l - shell method is very fast and can identify communities very quickly . unfortunately the method works well only when the source vertex is approximately equidistant from the boundary of its community . to overcome this problem , bagrow and bollt suggested to repeat the process starting from every vertex and derive a _ membership matrix _ @xmath445 : the element @xmath492 is one if vertex @xmath68 belongs to the community of vertex @xmath67 , otherwise it is zero . the membership matrix can be rewritten by suitably permutating rows and columns based on their mutual distances . the distance between two rows ( or columns ) is defined as the number of entries whose elements differ . if the graph has a clear community structure , the membership matrix takes a block - diagonal form , where the blocks identify the communities . the method enables one to detect overlaps between communities as well @xcite . unfortunately , the rearrangement of the matrix requires a time @xmath154 , so it is quite slow . a variant of the algorithm by bagrow and bollt , in which boundary vertices are examined separately and both first and second nearest neighbors of the running community are simultaneously investigated , was suggested by rodrigues et al . @xcite . a recent methodology introduced by papadopoulos et al . @xcite , called _ bridge bounding _ , is similar to the l - shell algorithm , but here the cluster around a vertex grows until one `` hits '' the boundary edges . such edges can be recognized from the values of various measures , like betweenness @xcite or the edge clustering coefficient @xcite . the problem is that there are often no clear gaps in the distributions of the values of such measures , so one is forced to set a threshold to automatically identify the boundary edges from the others , and there is no obvious way to do it . the best results of the algorithm are obtained by using a measure consisting of a weighted sum of the edge clustering coefficient over a wider neighborhood of the given edge . this version of the method has a time complexity @xmath606 , where @xmath420 is the average degree of the graph . in another algorithm by clauset , local communities are discovered through greedy maximization of a local modularity measure @xcite . given a community @xmath22 , the boundary @xmath259 is the set of vertices of @xmath22 with at least one neighbor outside @xmath22 ( fig . [ clausetloc ] ) . schematic picture of a community @xmath22 used in the definition of local modularity by clauset @xcite . the black area indicates the subgraph of @xmath22 including all vertices of @xmath22 , whose neighbors are also in @xmath22 . the boundary @xmath259 entails the vertices of @xmath22 with at least one neighbor outside the community . reprinted figure with permission from ref . 2005 by the american physical society.,width=340 ] the local modularity @xmath134 by clauset is the ratio of the number of edges having both endpoints in @xmath22 ( but at least one in @xmath259 ) , with the number of edges having at least one endpoint in @xmath259 . it is a measure of the sharpness of the community boundary . its optimization consists of a local exploration of the community starting from a source vertex : at each step the neighboring vertex yielding the largest increase ( smallest decrease ) of @xmath134 is added , until the community has reached a predefinite size @xmath115 . this greedy optimization takes a time @xmath607 , where @xmath420 is the average degree of the graph . the local modularity @xmath134 has been used in a paper by hui et al . @xcite , where methods to find communities in networks of mobile devices are designed . another method , where communities are defined based on a local criterion , was presented by eckmann and moses @xcite . the idea is to use the clustering coefficient @xcite of a vertex as a quantity to distinguish tightly connected groups of vertices . many edges mean many loops inside a community , so the vertices of a community are likely to have a large clustering coefficient . the latter can be related to the average distance between pairs of neighbours of the vertex . the possible values of the distance are @xmath2 ( if neighbors are connected ) or @xmath323 ( if they are not ) , so the average distance lies between @xmath2 and @xmath323 . the more triangles there are in the subgraph , the shorter the average distance . since each vertex always has distance @xmath2 from its neighbours , the fact that the average distance between its neighbours is different from @xmath2 reminds what happens when one measures segments on a curved surface . endowed with a metric , represented by the geodesic distance between vertices / points , and a curvature , the graph can be embedded in a geometric space . communities appear as portions of the graph with a large curvature . the algorithm was applied to the graph representation of the world wide web , where vertices are web pages and edges are the hyperlinks that take users from a page to the other . the authors found that communities correspond to web pages dealing with the same topic . long et al . have devised an interesting technique that is able to detect various types of vertex groups , not necessarily communities @xcite . the method is based on graph approximation , as it tries to match the original graph topology onto a coarse type of graph , the _ community prototype graph _ , which has a clear group structure ( block - diagonal for clusters , block - off - diagonal for classes of multipartite graphs , etc . ) . the goal is to determine the community prototype graph that best approximates the graph at study , where the goodness of the approximation is expressed by the distance between the corresponding matrices . in this way the original problem of finding graph subsets becomes an optimization problem . long et al . called this procedure _ community learning by graph approximation _ sometimes the minimization of the matrix distance can be turned into the maximization of the trace of a matrix . measures like cut size or ratio cut can be also formulated as the trace of matrices ( see for instance eq . [ eqr ] ) . in fact , clga includes traditional graph partitioning as a special case ( section [ sec4_1 ] ) . long et al . designed three algorithms for clga : two of them seek for divisions of the graph into overlapping or non - overlapping groups , respectively ; in the third one an additional constraint is introduced to produce groups of comparable size . the complexity of these algorithms is @xmath608 , where @xmath156 is the number of iterations until the optimization converges and @xmath44 the number of groups . the latter has to be given as an input , which is a serious limit of clga . a fast algorithm by wu and huberman identifies communities based on the properties of resistor networks @xcite . it is essentially a method for partitioning graphs in two parts , similar to spectral bisection , although partitions in an arbitrary number of communities can be obtained by iterative applications . the graph is transformed into a resistor network where each edge has unit resistance . a unit potential difference is set between two randomly chosen vertices . the idea is that , if there is a clear division in two communities of the graph , there will be a visible gap between voltage values for vertices at the borders between the clusters . the voltages are calculated by solving kirchoff s equations : an exact solution would be too time consuming , but it is possible to find a reasonably good approximation in a linear time for a sparse graph with a clear community structure , so the more time consuming part of the algorithm is the sorting of the voltage values , which takes time @xmath254 . any possible vertex pair can be chosen to set the initial potential difference , so the procedure should be repeated for all possible vertex pairs . the authors showed that this is not necessary , and that a limited number of sampling pairs is sufficient to get good results , so the algorithm scales as @xmath254 and is very fast . an interesting feature of the method is that it can quickly find the natural community of any vertex , without determining the complete partition of the graph . for that , one uses the vertex as source voltage and places the sink at an arbitrary vertex . the same feature is present in an older algorithm by flake et al . @xcite , where one uses max - flow instead of current flow ( section [ sec4_1 ] ) . an algorithm by orponen and schaeffer @xcite is based on the same principle , but it does not need the specification of target sources as it is based on diffusion in an unbounded medium . the limit of such methods is the fact that one has to give as input the number of clusters , which is usually not known beforehand . ohkubo and tanaka @xcite pointed out that , since communities are rather compact structures , they should have a small volume , where the volume of a community is defined as the ratio of the number of vertices by the internal edge density of the community . ohkubo and tanaka assumed that the sum @xmath609 of the volumes of the communities of a partition is a reliable index of the goodness of the partition . so , the most relevant partition is the one minimizing @xmath609 . the optimization is carried out with simulated annealing . zarei and samani @xcite remarked that there is a symmetry between community structure and anti - community ( multipartite ) structure , when one considers a graph and its complement , whose edges are the missing edges of the original graph . in fact , if a graph has a well identified communities , the same groups of vertices would be strong anti - communities in the complement graph , i. e. they should have a few intra - cluster edges and many inter - cluster edges . based on this remark , the communities of a graph can be identified by looking for anticommunities in the complement graph , which can sometimes be easier . zarei and samani devised a spectral method using matrices of the complement graph . the results of this technique appear good as compared to other spectral methods on artificial graphs generated with the planted @xmath293-partition model @xcite , as well as on zachary s karate club @xcite , lusseau s dolphins network @xcite and a network of protein - protein interactions . however , the authors have used very small graphs for testing . communities make sense on sparse graphs , but the complements of large sparse graphs would not be sparse , but very dense , and their community ( multipartite ) structure basically invisible . gudkov and montealegre detected communities by means of dynamical simplex evolution @xcite . graph vertices are represented as points in an @xmath306-dimensional space . each point initially sits on the @xmath15 vertices of a simplex , and then moves in space due to forces exerted by the other points . if vertices are neighbors , the mutual force acting on their representative points is attractive , otherwise it is repulsive . if the graph has a clear community structure , the corresponding spatial clusters repel each other because of the few connections between them ( repulsion dominates over attraction ) . if communities are more mixed with each other , clusters are not well separated and they could be mistakenly aggregated in larger structures . to avoid that , gudkov and montealegre defined clusters as groups of points such that the distance between each pair of points does not exceed a given threshold , which can be arbitrarily tuned , to reveal structures at different resolutions ( section [ sub_sec6_201 ] ) . the algorithm consists in solving first - order differential equations , describing the dynamics of mass points moving in a viscous medium . the complexity of the procedure is @xmath129 . differential equations are also at the basis of a recent method designed by krawczyk and kuakowski @xcite . here the equations describe a dynamic process , in which the original graph topology evolves to a disconnected graph , whose components are the clusters of the original graph . despite the significant improvements in computational complexity , it is still problematic to apply clustering algorithms to many large networks available today . therefore narasimhamurthy et al . @xcite proposed a two - step procedure : first , the graph at study is decomposed in smaller pieces by a fast graph partitioning technique ; then , a clustering method is applied to each of the smaller subgraphs obtained [ narasimhamurthy et al . used the clique percolation method ( section [ sec45_1 ] ) ] . the initial decomposition of the graph is carried out through the multilevel method by dhillon et al . it is crucial to verify that the initial partitioning does not split the communities of the graph among the various subgraphs of the decomposition . this can be done by comparing , on artificial graphs , the final clusters obtained with the two - step method with those detected by applying the chosen clustering technique to the entire graph . most of the methods discussed in the previous sections aim at detecting standard partitions , i. e. partitions in which each vertex is assigned to a single community . however , in real graphs vertices are often shared between communities ( section [ sec2 ] ) , and the issue of detecting overlapping communities has become quite popular in the last few years . we devote this section to the main techniques to detect overlapping communities . the most popular technique is the clique percolation method ( cpm ) by palla et al . @xcite . it is based on the concept that the internal edges of a community are likely to form cliques due to their high density . on the other hand , it is unlikely that intercommunity edges form cliques : this idea was already used in the divisive method of radicchi et al . ( section [ subsec5_2 ] ) . palla et al . use the term @xmath44-clique to indicate a complete graph with @xmath44 vertices vertices ( section [ sec1_1 ] ) . ] . notice that a @xmath44-clique is different from the @xmath15-clique ( see section [ sec3_1_2 ] ) used in social science . if it were possible for a clique to move on a graph , in some way , it would probably get trapped inside its original community , as it could not cross the bottleneck formed by the intercommunity edges . palla et al . introduced a number of concepts to implement this idea . two @xmath44-cliques are _ adjacent _ if they share @xmath188 vertices . the union of adjacent @xmath44-cliques is called _ @xmath44-clique chain_. two @xmath44-cliques are connected if they are part of a @xmath44-clique chain . finally , a _ @xmath44-clique community _ is the largest connected subgraph obtained by the union of a @xmath44-clique and of all @xmath44-cliques which are connected to it . examples of @xmath44-clique communities are shown in fig . [ figure10 ] . clique percolation method . the example shows communities spanned by adjacent 4-cliques . overlapping vertices are shown by the bigger dots . reprinted figure with permission from ref . 2005 by the nature publishing group . ] one could say that a @xmath44-clique community is identified by making a @xmath44-clique `` roll '' over adjacent @xmath44-cliques , where rolling means rotating a @xmath44-clique about the @xmath188 vertices it shares with any adjacent @xmath44-clique . by construction , @xmath44-clique communities can share vertices , so they can be overlapping . there may be vertices belonging to non - adjacent @xmath44-cliques , which could be reached by different paths and end up in different clusters . unfortunately , there are also vertices that can not be reached by any @xmath44-clique , like , e. g. vertices with degree one ( `` leaves '' ) . in order to find @xmath44-clique communities , one searches first for maximal cliques . then a clique - clique overlap matrix @xmath400 is built @xcite , which is an @xmath610 matrix , @xmath115 being the number of cliques ; @xmath611 is the number of vertices shared by cliques @xmath67 and @xmath68 . to find @xmath44-cliques , one needs simply to keep the entries of @xmath400 which are larger than or equal to @xmath188 , set the others to zero and find the connected components of the resulting matrix . detecting maximal cliques is known to require a running time that grows exponentially with the size of the graph . however , the authors found that , for the real networks they analyzed , the procedure is quite fast , due to the fairly limited number of cliques , and that ( sparse ) graphs with up to @xmath612 vertices can be analyzed in a reasonably short time . the actual scalability of the algorithm depends on many factors , and can not be expressed in closed form . an interesting aspect of @xmath44-clique communities is that they allow to make a clear distinction between random graphs and graphs with community structure . this is a rather delicate issue : we have seen in section [ sub_sec6_1 ] that newman - girvan modularity can attain large values on random graphs . dernyi et al . @xcite have studied the percolation properties of @xmath44-cliques on random graphs , when the edge probability @xmath45 varies . they found that the threshold @xmath613 for the emergence of a giant @xmath44-clique community , i. e. a community occupying a macroscopic portion of the graph , is @xmath614^{-1/(k-1)}$ ] , @xmath15 being the number of vertices of the graph , as usual . for @xmath615 , for which the @xmath44-cliques reduce to edges , one recovers the known expression for the emergence of a giant connected component in erds - rnyi graphs ( section [ sec1_3 ] ) . this percolation transition is quite sharp : if the edge probability @xmath616 , @xmath44-clique communities are rather small ; if @xmath617 there is a giant component and many small communities . to assess the significance of the clusters found with the cpm , one can compare the detected cover with the cover found on a null model graph , which is random but preserves the expected degree sequence of the original graph . the modularity of newman and girvan is based on the same null model ( section [ sec3_2_2 ] ) . the null models of real graphs seem to display the same two scenarios found for erds - rnyi graphs , characterized by the presence of very small @xmath44-clique communities , with or without a giant cluster . therefore , covers with @xmath44-clique communities of large or appreciable size can hardly be due to random fluctuations . palla and coworkers @xcite have designed a software package implementing the cpm , called _ cfinder _ , which is freely available ( www.cfinder.org ) . the algorithm has been extended to the analysis of weighted , directed and bipartite graphs . for weighted graphs , in principle one can follow the standard procedure of thresholding the weights , and apply the method on the resulting graphs , treating them as unweighted . farkas et al . @xcite proposed instead to threshold the weight of cliques , defined as the geometric mean of the weights of all edges of the clique . the value of the threshold is chosen slightly above the critical value at which a giant @xmath44-clique community emerges , in order to get the richest possible variety of clusters . on directed graphs , palla et al . defined _ directed @xmath44-cliques _ as complete graphs with @xmath44 vertices , such that there is an ordering among the vertices , and each edge goes from a vertex with higher order to one with lower order . the ordering is determined from the _ restricted outdegree _ of the vertex , expressing the fraction of outgoing edges pointing to the other vertices of the clique versus the total outdegree . the method has been extended to bipartite graphs by lehmann et al . @xcite . in this case one uses bipartite cliques , or _ bicliques _ : a subgraph @xmath618 is a biclique if each of @xmath294 vertices of one class are connected with each of @xmath619 vertices of the other class . two cliques @xmath618 are adjacent if they share a clique @xmath620 , and a @xmath618 clique community is the union of all @xmath618 cliques that can be reached from each other through a path of adjacent @xmath618 cliques . finding all @xmath621 bicliques of a graph is an @xmath622-complete problem @xcite , mostly because the number of bicliques tends to grow exponentially with the size of the graph . the algorithm designed by lehmann et al . to find biclique communities is similar to the original cpm , and has a total complexity of @xmath623 . on sparse graphs , @xmath621 often grows linearly with the number of edges @xmath14 , yielding a time complexity @xmath624 . bicliques are also the main ingredients of _ bitector _ , a recent algorithm to detect community structure in bipartite graphs @xcite . kumpula et al . have developed a fast implementation of the cpm , called sequential clique percolation algorithm ( scp ) @xcite . it consists in detecting @xmath44-clique communities by sequentially inserting the edges of the graph at study , one by one , starting from an initial empty graph . whenever a new edge is added , one checks whether new @xmath44-cliques are formed , by searching for @xmath625-cliques in the subset of neighboring vertices of the endpoints of the inserted edge . the procedure requires to build a graph @xmath626 , in which the vertices are @xmath627-cliques and edges are set between vertices corresponding to @xmath627-cliques which are subgraphs of the same @xmath44-clique . at the end of the process , the connected components of @xmath626 correspond to the searched @xmath44-clique communities . the technique has a time complexity which is linear in the number of @xmath44-cliques of the graph , so it can vary a lot in practical applications . nevertheless , it turns out to be much faster than the original implementation of the cpm . the big advantage of the scp , however , consists of its implementation for weighted graphs . by inserting edges in decreasing order of weight , one recovers in a single run the community structure of the graph for all possible weight thresholds , by storing every cover detected after the addition of each edge . the standard cpm , instead , needs to be applied once for each threshold . if , instead of edge weight thresholding , one performs @xmath44-clique weight thresholding , as prescribed by farkas et al . @xcite , the scp remains much faster than the cpm , if one applies a simple modification to it , consisting in detecting and storing all @xmath44-cliques on the full graph , sorting them based on their weights , and finding the communities by sequentially adding the @xmath44-cliques in decreasing order of weight . the cpm has the same limit as the algorithm of radicchi et al . @xcite ( section [ subsec5_2 ] ) : it assumes that the graph has a large number of cliques , so it may fail to give meaningful covers for graphs with just a few cliques , like technological networks and some social networks . on the other hand , if there are many cliques , the method may deliver trivial community structure , like a cover consisting of the whole graph as a single cluster . a more fundamental issue is the fact that the method does not look for actual communities , consistent with the shared notion of dense subgraphs , but for subgraphs `` containing '' many cliques , which may be quite different objects than communities ( for instance , they could be `` chains '' of cliques with low internal edge density ) . another big problem is that on real networks there is a considerable fraction of vertices that are left out of the communities , like leaves . one could think of some post - processing procedure to include them in the communities , but for that it is necessary to introduce a new criterion , outside the framework that inspired the method . furthermore it is not clear _ a priori _ which value of @xmath44 one has to choose to identify meaningful structures . finally , the criterion to choose the threshold for weighted graphs and the definition of directed @xmath44-cliques are rather arbitrary . one of the first methods to find overlapping communities was designed by baumes et al . a community is defined as a subgraph which locally optimizes a given function @xmath86 , typically some measure related to the edge density of the cluster ) . ] . different overlapping subsets may all be locally optimal , so vertices can be shared between communities . detecting the cluster structure of a graph amounts to finding the set of all locally optimal clusters . two efficient heuristics are proposed , called iterative scan ( @xmath628 ) and rank removal ( @xmath629 ) . @xmath628 performs a greedy optimization of the function @xmath86 . one starts from a random seed vertex / edge and adds / deletes vertices one by one as long as @xmath86 increases . then another seed is randomly picked and the procedure is repeated . the algorithm stops when , by picking any seed , one recovers a previously identified cluster . @xmath629 consists in removing important vertices such to disconnect the graphs in small components representing the cores of the clusters . the importance of vertices is determined by their centrality scores ( e.g. degree , betweenness centrality @xcite ) , pagerank @xcite ) . vertices are removed until one fragments the graph into components of a given size . after that , the removed vertices are added again to the graph , and are associated to those clusters for which doing so increases the value of the function @xmath86 . the complexity of @xmath628 and @xmath629 is @xmath129 on sparse graphs . the best performance is achieved by using @xmath628 to refine results obtained from @xmath629 . in a successive paper @xcite , baumes et al . further improved such two - step procedure , in that the removed vertices in @xmath629 are reinserted in decreasing order of their centrality scores , and the optimization of @xmath86 in @xmath628 is only extended to neighboring vertices of the running cluster . the new recipe maintains time complexity @xmath129 , but on sparse graphs it requires a time lower by an order of magnitude than the old one , while the quality of the detected clustering is comparable . a different method , combining spectral mapping , fuzzy clustering and the optimization of a quality function , has been presented by zhang et al . the membership of vertex @xmath67 in cluster @xmath44 is expressed by @xmath630 , which is a number between @xmath631 and @xmath2 . the sum of the @xmath630 over all communities @xmath44 of a cover is @xmath2 , for every vertex . this normalization is suggested by the fact that the entry @xmath630 can be thought of as the probability that @xmath67 belongs to community @xmath44 , so the sum of the @xmath630 represents the probability that the vertex belongs to any community of the cover , which is necessarily @xmath2 . if there were no overlaps , @xmath632 , where @xmath105 represents the unique community of vertex @xmath67 . the algorithm consists of three phases : 1 ) embedding vertices in euclidean space ; 2 ) grouping the corresponding vertex points in a given number @xmath115 of clusters ; 3 ) maximizing a modularity function over the set of covers found in step 2 ) , corresponding to different values of @xmath115 . this scheme has been used in other techniques as well , like in the algorithm of donetti and muoz @xcite ( section [ sec43 ] ) . the first step builds upon a spectral technique introduced by white and smyth @xcite , that we have discussed in section [ sub_sec6_0_4 ] . graph vertices are embedded in a @xmath88-dimensional euclidean space by using the top @xmath88 eigenvectors of the right stochastic matrix @xmath190 ( section [ sec1_2 ] ) , derived from the adjacency matrix @xmath189 by dividing each element by the sum of the elements of the same row . the spatial coordinates of vertex @xmath67 are the @xmath67-th components of the eigenvectors . in the second step , the vertex points are associated to @xmath115 clusters by using fuzzy @xmath44-means clustering @xcite ( section [ sec4_3 ] ) . the number of clusters @xmath115 varies from @xmath323 to a maximum @xmath289 , so one obtains @xmath290 covers . the best cover is the one that yields the largest value of the modularity @xmath633 , defined as @xmath634 , \label{eq : mod5}\ ] ] where @xmath635 and @xmath636 the sets @xmath637 and @xmath638 include the vertices of module @xmath117 and of the whole network , respectively . [ eq : mod5 ] is an extension of the weighted modularity in eq . [ eq : mod20 ] , obtained by weighing the contribution of the edges weights to the sums in @xmath320 and @xmath321 by the ( average ) membership coefficients of the vertices of the edge . the determination of the eigenvectors is the most computationally expensive part of the method , so the time complexity is the same as that of the algorithm by white and smyth ( see section [ sub_sec6_0_4 ] ) , i. e. @xmath291 , which is essentially linear in @xmath15 if the graph is sparse and @xmath639 . nepusz et al . proposed a different approach based on vertex similarity @xcite . one starts from the membership matrix @xmath197 , defined as in the previous method by zhang et al . from @xmath197 a matrix @xmath538 is built , where @xmath640 , expressing the similarity between vertices ( @xmath115 is the number of clusters ) . if one assumes to have information about the actual vertex similarity , corresponding to the matrix @xmath641 , the best cover is obtained by choosing @xmath197 such that @xmath538 approximates as closely as possible @xmath641 . this amounts to minimize the function @xmath642 where the @xmath309 weigh the importance of the approximation for each entry of the similarity matrices . in the absence of any information on the community structure of the graph , one sets @xmath643 , @xmath104 ( equal weights ) and @xmath641 equal to the adjacency matrix @xmath189 , by implicitly assuming that vertices are similar if they are neighbors , dissimilar otherwise . on weighted graphs , one can set the @xmath309 equal to the edge weights . minimizing @xmath644 is a nonlinear constrained optimization problem , that can be solved with a gradient - based iterative optimization method , like simulated annealing . the optimization procedure adopted by nepusz et al . , for a fixed number of clusters @xmath115 , has a time complexity @xmath645 , where @xmath646 is the number of iterations leading to convergence , so the method can only be applied to fairly small graphs . if @xmath115 is unknown , as it usually happens , the best cover is the one corresponding to the largest value of the modularity @xmath647 eq . [ eqoverl2 ] is very similar to the expression of newman - girvan modularity ( eq . [ eq : mod_0 ] ) : the difference is that the kronecker s @xmath101 is replaced by the vertices similarity , to account for overlapping communities . once the best cover is identified , one can use the entries of the partition matrix @xmath197 to evaluate the participation of each vertex in the @xmath115 clusters of the cover . nepusz et al . defined the _ bridgeness _ @xmath648 of a vertex @xmath67 as @xmath649 if @xmath67 belongs to a single cluster , @xmath650 . if , for a vertex @xmath67 , @xmath651 , @xmath652 , @xmath653 and @xmath67 is a perfect _ bridge _ , as it lies exactly between all clusters however , a vertex with low @xmath654 may be simply an outlier , not belonging to any cluster . since real bridges are usually rather central vertices , one can identify them by checking for large values of the _ centrality - corrected bridgeness _ , obtained by multiplying the bridgeness of eq . [ eqoverl3 ] by the centrality of the vertex ( expressed by , e.g. , degree , betweenness @xcite , etc . ) . a variant of the algorithm by nepusz et al can be downloaded from http://www.cs.rhul.ac.uk/home/tamas/assets/file + s / fuzzyclust - static.tar.gz . in real networks it is often easier to discriminate between intercluster and intracluster edges than recognizing overlapping vertices . for instance , in social networks , even though many people may belong to more groups , their social ties within each group can be easily spotted . besides , it may happen that communities are joined to each other through their overlapping vertices ( fig . [ figlink ] ) , without intercluster edges . for these reasons , it has been recently suggested that defining clusters as sets of edges , rather than vertices , may be a promising strategy to analyze graphs with overlapping communities @xcite . one has to focus on the _ line graph _ @xcite , i. e. the graph whose vertices are the edges of the original graph ; vertices of the line graph are linked if the corresponding edges in the original graph are adjacent , i. e. if they share one of their endvertices . partitioning the line graph means grouping the edges of the starting graph . evans and lambiotte @xcite introduced a set of quality functions , similar to newman - girvan modularity ( eq . [ eq : mod_0 ] ) , expressing the stability of partitions against random walks taking place on the graph , following the work of delvenne et al . @xcite [ section [ sec44_2 ] ] . they considered a projection of the traditional random walk on the line graph , along with two other diffusion processes , where walkers move between adjacent edges ( rather than between neighboring vertices ) . evans and lambiotte optimized the three corresponding modularity functions to look for partitions in two real networks , zachary s karate club @xcite ( section [ sec6_1 ] ) and the network of word associations derived from the university of south florida free association norms @xcite ( section [ sec2 ] ) . the optimization was carried out with the hierarchical technique by blondel et al . @xcite and the multi - level algorithm by noack and rotta @xcite . while the results for the word association network are reasonable , the test on the karate club yields partitions in more than two clusters . however , the modularities used by evans et lambiotte can be modified to include longer random walks ( just like in ref . @xcite ) , and the length of the walk represents a resolution parameter that can be tuned to get better results . ahn et al . @xcite proposed to group edges with an agglomerative hierarchical clustering technique , called _ hierarchical link clustering _ ( section [ sec4_2 ] ) . they use a similarity measure for a pair of ( adjacent ) edges that expresses the size of the overlap between the neighborhoods of the non - coincident endvertices , divided by the total number of ( different ) neighbors of such endvertices . groups of edges are merged pairwise in descending order of similarity , until all edges are together in the same cluster . the resulting dendrogram provides the most complete information on the community structure of the graph . however , as usual , most of this information is redundant and is an artefact of the procedure itself . so , ahn et al . introduced a quality function to select the most meaningful partition(s ) , called _ partition density _ , which is essentially the average edge density within the clusters . the method is able to find meaningful clusters in biological networks , like protein - protein and metabolic networks , as well as in a social network of mobile phone communications . it can also be extended to multipartite and weighted graphs . the idea of grouping edges is surely interesting . however it is not _ a priori _ better than grouping vertices . in fact , the two situations are somewhat symmetric . edges connecting vertices of different clusters are `` overlapping '' , but they will be assigned just to one cluster ( or else the clusters would be merged ) . communities as sets of edges . in the figure , the graph has a natural division in two triangles , with the central vertex shared between them . if communities are identified by their internal edges , detecting the triangles and their overlapping vertex becomes easier than by using methods that group vertices . reprinted figure with permission from ref . 2009 by the american physical society.,width=264 ] the possibility of having overlapping communities makes most standard clustering methods inadequate , and enforces the design of new _ ad hoc _ techniques , like the ones we have described so far . on the other hand , if it were possible to identify the overlapping vertices and `` separate '' them among the clusters they belong to , the overlaps would be removed and one could then apply any of the traditional clustering methods to the resulting graph . this idea is at the basis of a recent method proposed by gregory @xcite . it is a three - stages procedure : first , one transforms the graph into a larger graph without overlapping vertices ; second , a clustering technique is applied to the resulting graph ; third , one maps the partition obtained into a cover by replacing the vertices with those of the original graph . the transformation step , called _ peacock _ , is performed by identifying the vertices with highest _ split betweenness _ ( section [ subsec5_1 ] ) and splitting them in multiple parts , connected by edges . this is done as long as the split betweenness of the vertices is sufficiently high , which is determined by a parameter @xmath155 . in this way , most vertices of the resulting graph are exactly the same one had initially , the others are multiple copies of the overlapping vertices of the initial graph . the overlaps of the final cover are obtained by checking if copies of the same initial vertex end up in different disjoint clusters . the complexity is dominated by the peacock algorithm , if one computes the exact values of the split betweenness for the vertices , which requires a time @xmath154 on a sparse graph . gregory proposed an approximate local computation , which scales as @xmath254 : in this way the total complexity of the method becomes competitive , if one chooses a fast algorithm for the identification of the clusters . the goodness of the results depends on the specific method one uses to find the clusters after the graph transformation . the software of the version of the method used by gregory in his applications can be found at http://www.cs.bris.ac.uk/@xmath215steve/networks/peaco + ckpaper/. the idea of gregory is interesting , as it allows to exploit traditional methods even in the presence of overlapping communities . the choice of the parameter @xmath155 , which determines whether a vertex is overlapping or not , does not seem to affect significantly the results , as long as @xmath155 is taken sufficiently small . the existence of a resolution limit for newman - girvan modularity ( section [ sub_sec6_1 ] ) implies that the straight optimization of quality functions yields a coarse description of the cluster structure of the graph , at a scale which has _ a priori _ nothing to do with the actual scale of the clusters . in the absence of information on the cluster sizes of the graph , a method should be able to explore all possible scales , to make sure that it will eventually identify the right communities . multiresolution methods are based on this principle . however , many real graphs display _ hierarchical _ cluster structures , with clusters inside other clusters @xcite . in these cases , there are more levels of organization of vertices in clusters , and more relevant scales . in principle , clustering algorithms should be able to identify them . multiresolution methods can do the trick , in principle , as they scan continuously the range of possible cluster scales . recently other methods have been developed , where partitions are by construction hierarchically nested in each other . in this section we discuss both classes of techniques . in general , multiresolution methods have a freely tunable parameter , that allows to set the characteristic size of the clusters to be detected . the general spin glass framework by reichardt and bornholdt ( @xcite and section [ sub_sec6_01 ] ) is a typical example , where @xmath371 is the resolution parameter . the extension of the method to weighted graphs has been recently discussed @xcite . pons has proposed a method @xcite consisting of the optimization of multiscale quality functions , including the _ multiscale modularity _ @xmath655 , \label{eq : mult0}\ ] ] and two other additive quality functions , derived from the _ performance _ ( eq . [ eq10_perf ] ) and a measure based on the similarity of vertex pairs . in eq . [ eq : mult0 ] @xmath656 is the resolution parameter and the notation is otherwise the same as in eq . [ eq : mod ] . we see that , for @xmath657 , one recovers standard modularity . however , since multiplicative factors in @xmath658 do not change the results of the optimization , we can divide @xmath658 by @xmath18 , recovering the same quality function as in eq . [ eqr13 ] , with @xmath659 , up to an irrelevant multiplicative constant . to evaluate the relevance of the partitions , for any given multiscale quality function , pons suggested that the length of the @xmath18-range @xmath660 $ ] , for which a community @xmath22 `` lives '' in the maximum modularity partition , is a good indicator of the stability of the community . he then defined the _ relevance function _ of a community @xmath22 at scale @xmath18 as @xmath661 the relevance @xmath662 of a partition @xmath94 at scale @xmath18 is the average of the relevances of the clusters of the partition , weighted by the cluster sizes . peaks in @xmath18 of @xmath662 reveal the most meaningful partitions . another interesting technique has been devised by arenas et al . @xcite , and consists of a modification of the original expression of modularity . the idea is to make vertices contribute as well to the computation of the edge density of the clusters , by adding a self - loop of strength @xmath375 to each vertex . arenas et al . remarked that the parameter @xmath375 does not affect the structural properties of the graph in most cases , which are usually determined by an adjacency matrix without diagonal elements . with the introduction of the vertex strength @xmath375 , modularity reads @xmath663 , \label{eq : mult1}\ ] ] for the general case of a weighted graph . the notation is the same as in eq . [ eq : mod20 ] , @xmath621 is the number of vertices in cluster @xmath117 . we see that now the relative importance of the two terms in each summand depends on @xmath375 , which can take any value in @xmath664 - 2w / n , \infty[$ ] . arenas et al . made a sweep in the range of @xmath375 , and determined for each @xmath375 the maximum modularity with extremal optimization ( section [ sub_sec6_0_3 ] ) and _ tabu search _ meaningful cluster structures correspond to plateaus in the plot of the number of clusters versus @xmath375 ( fig . [ figarenasplat ] ) . the length of a plateau gives a measure of the _ stability _ of the partition against the variation of @xmath375 . the procedure is able to disclose the community structure of a number of real benchmark graphs . as expected , the most relevant partitions can be found in intervals of @xmath375 not including the value @xmath665 , which corresponds to the case of standard modularity ( fig . [ figarenasplat ] ) . analysis of zachary s karate club with the multiresolution method by arenas et al . the plot shows the number of clusters obtained in correspondence of the resolution parameter @xmath375 . the longest plateau @xmath666 indicates the most stable partition , which exactly matches the social fission observed by zachary . the partition obtained with straight modularity optimization ( @xmath665 ) consists of four clusters and is much less stable with respect to @xmath666 , as suggested by the much shorter length of its plateau . reprinted figure with permission from ref . 2008 by iop publishing . ] a drawback of the method is that it is very slow , as one has to compute the modularity maximum for many values of @xmath375 in order to discriminate between relevant and irrelevant partitions . if the modularity maximum is computed with precise methods like simulated annealing and/or extremal optimization , as in ref . @xcite , only graphs with a few hundred vertices can be analyzed on a single processor . on the other hand the algorithm can be trivially parallelized by running the optimization for different values of @xmath375 on different processors . this is a common feature of all multiresolution methods discussed in this section . in spite of the different formal expressions of modularity , the methods by arenas et al . and reichardt and bornholdt are somewhat related to each other and yield similar results @xcite on zachary s karate club @xcite ( section [ sec6_1 ] ) , synthetic graphs la ravasz - barabsi @xcite and on a model graph with the properties of real weighted social networks . in fact , their modularities can be both recovered from the continuous - time version of the stability of clustering under random walk , introduced by delvenne et al . @xcite ( section [ sec44_2 ] ) . lancichinetti et al . have designed a multiresolution method which is capable of detecting both the hierarchical structure of graphs and overlapping communities @xcite . it is based on the optimization of a fitness function , which estimates the strength of a cluster and entails a resolution parameter @xmath18 . the function could in principle be arbitrary , in their applications the authors chose a simple ansatz based on the tradeoff between the internal and the total degree of the cluster . the optimization procedure starts from a cluster with a single vertex , arbitrarily selected . given a cluster core , one keeps adding and removing neighboring vertices of the cluster as long as its fitness increases . the fitness is recalculated after each addition / removal of a vertex . at some point one reaches a local maximum and the cluster is `` closed '' . then , another vertex is chosen at random , among those not yet assigned to a cluster , a new cluster is built , and so on , until all vertices have been assigned to clusters . during the buildup of a cluster , vertices already assigned to other clusters may be included , i. e. communities may overlap . the computational complexity of the algorithm , estimated on sparse erds - rnyi random graphs , is @xmath667 , with @xmath668 for small values of the resolution parameter @xmath18 , and @xmath669 if @xmath18 is large . for a complete analysis , the worst - case computational complexity is @xmath128 , where the factor @xmath241 comes from the minimum number of different @xmath18-values which are needed to resolve the actual community structure of the graph . relevant partitions are revealed by pronounced spikes in the histogram of the fitness values of covers obtained for different @xmath18-values , where the fitness of a cover is defined as the average fitness of its clusters . a technique based on the potts model , similar to that of reichardt and bornholdt @xcite , has been suggested by ronhovde and nussinov @xcite . the energy of their spin model is @xmath670\delta(\sigma_i,\sigma_j ) . \label{eq : mult2}\ ] ] the big difference with eq . [ eqr13 ] is the absence of a null model term . the model considers pairs of vertices in the same community : edges between vertices are energetically rewarded , whereas missing edges are penalized . the parameter @xmath371 fixes the tradeoff between the two contributions . the energy is minimized by sequentially shifting single vertices / spins to the communities which yield the largest decrease of the system s energy , until convergence . if , for each vertex , one just examines the communities of its neighbors , the energy is minimized in a time @xmath671 , where @xmath19 turns out to be slightly above @xmath2 in most applications , allowing for the analysis of large graphs . this essentially eliminates the problem of limited resolution , as the criterion to decide about the merger or the split of clusters only depends on local parameters . still , for the detection of possible hierarchical levels tuning @xmath371 is mandatory . in a successive paper @xcite , the authors have introduced a new stability criterion for the partitions , consisting of the computation of the similarity of partitions obtained for the same @xmath371 and different initial conditions . the idea is that , if a partition is robust in a given range of @xmath371-values , most replicas delivered by the algorithm will be very similar . on the other hand , if one explores a region of resolutions in between two strong partitions , the algorithm will deliver the one or the other partition and the individual replicas will be , on average , not so similar to each other . so , by plotting the similarity as a function of the resolution parameter @xmath371 , stable communities are revealed by peaks . ronhovde and nussinov adopted similarity measures borrowed from information theory ( section [ sec6_2 ] ) . their criterion of stability can be adopted to determine the relevance of partitions obtained with any multiresolution algorithm . a general problem of multiresolution methods is how to assess the stability of partitions for large graphs . the rapidly increasing number of partitions , obtained by minimal shifts of vertices between clusters , introduces a large amount of noise , that blurs signatures of stable partitions like plateaus , spikes , etc . that one can observe in small systems . in this respect , it seems far more reliable focusing on correlations between partitions ( like the average similarity used by ronhovde and nussinov @xcite ) than on properties of the individual partitions ( like the measures of occurrence used by arenas et al . @xcite and by lancichinetti et al . @xcite ) . the natural procedure to detect the hierarchical structure of a graph is hierarchical clustering , that we have discussed in section [ sec4_2 ] . there we have emphasized the main weakness of the procedure , which consists of the necessity to introduce a criterion to identify relevant partitions ( hierarchical levels ) out of the full dendrogram produced by the given algorithm . furthermore , there is no guarantee that the results indeed reflect the actual hierarchical structure of the graph , and that they are not mere artefacts of the algorithm itself . scholars have just started to deal with these problems . sales - pardo et al . have proposed a top - down approach @xcite . their method consists of two steps : 1 ) measuring the similarity between vertices ; 2 ) deriving the hierarchical structure of the graph from the similarity matrix . the similarity measure , named _ node affinity _ , is based on newman - girvan modularity . basically the affinity between two vertices is the frequency with which they coexist in the same community in partitions corresponding to local optima of modularity . the latter are configurations for which modularity is stable , i. e. it can not increase if one shifts one vertex from one cluster to another or by merging or splitting clusters . the set of these partitions is called @xmath672 . before proceeding with the next step , one verifies whether the graph has a significant community structure or not . this is done by calculating the @xmath406-score ( eq . [ eq : zscore ] ) for the average modularity of the partitions in @xmath672 with respect to the average modularity of partitions with local modularity optima of the equivalent ensemble of null model graphs , obtained as usual by randomly rewiring the edges of the original graph under the condition that the expected degree sequence is the same as the degree sequence of the graph . large @xmath406-scores indicate meaningful cluster structure : sales - pardo et al . used a threshold corresponding to the @xmath673 significance level - score has to be computed with respect to the actual distribution of the maximum modularity for the null model graphs , as the latter is not gaussian ( section [ sub_sec6_1 ] ) . ] . if the graph has a relevant cluster structure , one proceeds with the second step , which consists in putting the affinity matrix in a form as close as possible to block - diagonal , by minimizing a cost function expressing the average distance of connected vertices from the diagonal . the blocks correspond to the communities and the recovered partition represents the uppermost organization level . to determine lower levels , one iterates the procedure for each subgraph identified at the previous level , which is treated as an independent graph . the procedure stops when all blocks found do not have a relevant cluster structure , i. e. their @xmath406-scores are lower than the threshold . the partitions delivered by the method are hierarchical by construction , as communities at each level are nested within communities at higher levels . however , the method may find no relevant partition ( no community structure ) , a single partition ( community structure but no hierarchy ) or more ( hierarchy ) and in this respect it is better than most existing methods . the algorithm is not fast , as both the search of local optima for modularity and the rearrangement of the similarity matrix are performed with simulated annealing , but delivers good results for computer generated networks , and meaningful partitions for some real networks , like the world airport network @xcite , an email exchange network of a catalan university @xcite , a network of electronic circuits @xcite and metabolic networks of _ e. coli _ @xcite . clauset et al . @xcite described the hierarchical organization of a graph by introducing a class of _ hierarchical random graphs_. a hierarchical random graph is defined by a dendrogram @xmath674 , which is the natural representation of the hierarchy , and by a set of probabilities @xmath675 associated to the @xmath143 internal nodes of the dendrogram . ancestor _ of a vertex @xmath67 is any internal node of the dendrogram that is encountered by starting from the `` leaf '' vertex @xmath67 and going all the way up to the top of the dendrogram . the probability that vertices @xmath67 and @xmath68 are linked to each other is given by the probability @xmath676 of the lowest common ancestor of @xmath67 and @xmath68 . clauset et al . searched for the model @xmath677 that best fits the observed graph topology , by using bayesian inference ( section [ subsec_statinf1 ] ) . the probability that the model fits the graph is proportional to the likelihood @xmath678 here , @xmath679 is the number of edges connecting vertices whose lowest common ancestor is @xmath375 , @xmath680 and @xmath681 are the numbers of graph vertices in the left and right subtrees descending from the dendrogram node @xmath375 , and the product runs over all internal dendrogram nodes . for a given dendrogram @xmath674 , the maximum likelihood @xmath682 corresponds to the set of probabilities @xmath683 , where @xmath684 equals the actual density of edges @xmath685 between the two subtrees of @xmath375 ( fig . [ clausethier ] ) . hierarchical random graphs by clauset et al . the picture shows two possible dendrograms for the simple graph on the top . the linking probabilities on the internal nodes of the dendrograms yield the best fit of the model graphs to the graph at study . reprinted figure with permission from ref . 2008 by the nature publishing group . ] one can define the statistical ensemble of hierarchical random graphs describing a given graph @xmath23 , by assigning to each model graph @xmath686 a probability proportional to the maximum likelihood @xmath682 . the ensemble can be sampled by a markov chain monte carlo method @xcite . the procedure suggested by clauset et al . seems to converge to equilibrium roughly in a time @xmath129 , although the actual complexity may be much higher . still , the authors were able to investigate graphs with a few thousand vertices . from sufficiently large sets of model configurations sampled at equilibrium , one can compute average properties of the model , e. g. degree distributions , clustering coefficients . etc . , and compare them with the corresponding properties of the original graph . tests on real graphs reveal that the model is indeed capable to describe closely the graph properties . furthermore , the model enables one to predict missing connections between vertices of the original graph . this is a very important problem @xcite : edges of real graphs are the result of observations / experiments , that may fail to discover some relationships between the units of the system . from the ensemble of the hierarchical random graphs one can derive the average linking probability between all pairs of graph vertices . by ranking the probabilities corresponding to vertex pairs which are disconnected in the original graph , one may expect that the pairs with highest probabilities are likely to be connected in the system , even if such connections are not observed . clauset et al . pointed out that their method does not deliver a sharp hierarchical organization for a given graph , but a class of possible organizations , with well - defined probabilities . it is certainly reasonable to assume that many structures are compatible with a given graph topology . in the case of community structure , it is not clear which information one can extract from averaging over the ensemble of hierarchical random graphs . moreover , since the hierarchical structure is represented by a dendrogram , it is impossible to rank partitions according to their relevance . in fact , the work by clauset et al . questions the concept of `` relevant partition '' , and opens a debate in the scientific community about the meaning itself of graph clustering . the software of the method can be found at http://www.santafe.edu/@xmath215aaronc / hierarchy/. the analysis of dynamic communities is still in its infancy . studies in this direction have been mostly hindered by the fact that the problem of graph clustering is already controversial on single graph realizations , so it is understandable that most efforts still concentrate on the `` static '' version of the problem . another difficulty is represented by the dearth of timestamped data on real graphs . recently , several data sets have become available , enabling to monitor the evolution in time of real systems @xcite . so it has become possible to investigate how communities form , evolve and die . the main phenomena occurring in the lifetime of a community are ( fig . [ commevol ] ) : birth , growth , contraction , merger with other communities , split , death . possible scenarios in the evolution of communities . reprinted figure with permission from ref . 2007 by the nature publishing group . ] the first study was carried out by hopcroft et al . @xcite , who analyzed several snapshots of the citation graph induced by the nec citeseer database @xcite . the snapshots cover the period from @xmath687 to @xmath688 . communities are detected by means of ( agglomerative ) hierarchical clustering ( section [ sec4_2 ] ) , where the similarity between vertices is the _ cosine similarity _ of the vectors describing the corresponding papers , a well known measure used in information retrieval @xcite . in each snapshot hopcroft et al . identified the _ natural communities _ , defined as those communities of the hierarchical tree that are only slightly affected by minor perturbations of the graph , where the perturbation consists in removing a small fraction of the vertices ( and their edges ) . such natural communities are conceptually similar to the _ stable _ communities we will see in section [ sec6_4 ] . hopcroft et al . found the best matching natural communities across different snapshots , and in this way they could follow the history of communities . in particular they could see the emergence of new communities , corresponding to new research topics . the main drawback of the method comes from the use of hierarchical clustering , which is unable to sort out meaningful communities out of the hierarchical tree , which includes many different partitions of the graph . more recently , palla et al . performed a systematic analysis of dynamic communities @xcite . they studied two social systems : 1 ) a graph of phone calls between customers of a mobile phone company in a year s time ; 2 ) a collaboration network between scientists , describing the coauthorship of papers in condensed matter physics from the electronic e - print archive ( cond - mat ) maintained by cornell university library , spanning a period of 142 months . the first problem is identifying the image of a community @xmath689 at time @xmath690 among the communities of the graph at time @xmath156 . a simple criterion , used in other works , is to measure the relative overlap ( eq . [ eqt20 ] ) of @xmath689 with all communities at time @xmath156 , and pick the community which has the largest overlap with @xmath689 . this is intuitive , but in many cases it may miss the actual evolution of the community . for instance , if @xmath691 at time @xmath690 grows considerably and overlaps with another community @xmath692 ( which at the previous time step was disjoint from @xmath691 ) , the relative overlap between @xmath689 and @xmath693 may be larger than the relative overlap between @xmath689 and @xmath691 . it is not clear whether there is a general prescription to avoid this problem . palla et al . solved it by exploiting the features of the clique percolation method ( cpm ) ( section [ sec45_1 ] ) , that they used to detect communities . the idea is to analyze the graph @xmath694 , obtained by merging the two snapshots @xmath695 and @xmath696 of the evolving graph , at times @xmath156 and @xmath690 ( i. e. , by putting together all their vertices and edges ) . any cpm community of @xmath695 and @xmath696 does not get lost , as it is included within one of the cpm communities of @xmath694 . for each cpm community @xmath697 of @xmath694 , one finds the cpm communities @xmath698 and @xmath699 ( of @xmath695 and @xmath696 , respectively ) which are contained in @xmath697 . the image of any community in @xmath699 at time @xmath156 is the community of @xmath698 that has the largest relative overlap with it . the age @xmath253 of a community is the time since its birth . it turns out that the age of a community is positively correlated with its size @xmath700 , i. e. that older communities are also larger ( on average ) . the time evolution of a community @xmath22 can be described by means of the relative overlap @xmath701 between states of the community separated by a time @xmath156 : @xmath702 one finds that , in both data sets , @xmath701 decays faster for larger communities , so the composition of large communities is rather variable in time , whether small communities are essentially static . another important question is whether it is possible to predict the evolution of communities from information on their structure or on their vertices . in fig . [ evolpred]a the probability @xmath703 that a vertex will leave the community in the next step of the evolution is plotted as a function of the relative external strength of the vertex , indicating how much of the vertex strength lies on edges connecting it to vertices outside its community . the plot indicates that there is a clear positive correlation : vertices which are only loosely connected to vertices of their community have a higher chance ( on average ) to leave the community than vertices which are more `` committed '' towards the other community members . the same principle holds at the community level too . relation between structural features and evolution of a community . a ) relation between the probability that a vertex will abandon the community in the next time step and its relative external strength . b ) relation between the probability of disintegration of a community in the next time step and its relative external strength . reprinted figure with permission from ref . 2007 by the nature publishing group . ] fig . [ evolpred]b shows that the probability @xmath704 that a community will disintegrate in the next time step is positively correlated with the relative external strength of the community . finally , palla et al . have found that the probability for two communities to merge increases with the community sizes much more than what one expects from the size distribution , which is consistent with the faster dynamics observed for large communities . palla et al . analyzed two different real systems , a network of mobile phone communications and a coauthorship network , to be able to infer general properties of community evolution . however , communities were only found with the cpm , so their results need to be cross - checked by employing other clustering techniques . asur et al . @xcite explored the dynamic relationship between vertices and communities . communities were found with the mcl method by van dongen @xcite ( section [ sec44_2 ] ) , by analyzing the graph at different timestamps . asur et al . distinguished events involving communities and events involving the vertices . events involving communities are _ continue _ ( the community keeps most of its vertices in consecutive time steps ) , _ @xmath705-merge _ ( two clusters merge into another ) , _ @xmath705-split _ ( two clusters split in two parts ) , _ form _ ( no pair of vertices of the cluster at time @xmath690 were in the same cluster at time @xmath156 ) and _ dissolve _ ( opposite of _ form _ ) . events involving vertices are _ appear _ ( if a vertex joins the graph for the first time ) , _ disappear _ ( if a vertex of the graph at time @xmath156 is no longer there at time @xmath690 ) , _ join _ ( if a vertex of a cluster at time @xmath690 was not in that cluster at time @xmath156 ) and _ leave _ ( if a vertex which was in a cluster at time @xmath156 is not in that cluster at time @xmath690 ) . based on such events , four measures are defined in order to catch the behavioral tendencies of vertices contributing to the evolution of the graph : the _ stability index _ ( measuring the tendency of a vertex to interact with the same vertices over time ) , the _ sociability index _ ( measuring the number of different interactions of a vertex , basically the number of _ join _ and _ leave _ events ) , the _ popularity index _ ( measuring the number of vertices attracted by a cluster in a given time interval ) and the _ influence index _ ( measuring the influence a vertex has on the others , which is computed from the number of vertices that leave or join a cluster together with the vertex ) . applications on a coauthorship network of computer scientists and on a network of subjects for clinical trials show that the behavioral measures above enable one to make reliable predictions about the time evolution of such graphs ( including , e. g. , the inference of missing links @xcite ) . dynamic communities can be as well detected with methods of information compression , such as some of those we have seen in section [ subsec_statinf2 ] . sun et al . @xcite applied the minimum description length ( mdl ) principle @xcite to find the minimum encoding cost for the description of a time sequence of graphs and their partitions in communities . the method is quite similar to that successively developed by rosvall and bergstrom @xcite , which is however defined only for static graphs ( section [ subsec_statinf2 ] ) . here one considers bipartite graphs evolving in time . the time sequence of graphs can be separated in segments , each containing some number of consecutive snapshots of the system . the graphs of each segment are supposed to have the same modular structure ( i. e. they represent the same phase in the history of the system ) , so they are characterized by the same partition of the two vertex classes . for each graph segment it is possible to define an encoding cost , which combines the encoding cost of the partition of the graphs of the segment with the entropy of compression of the segment in the subgraph segments induced by the partition . the total encoding cost @xmath706 of the graph series is given by the sum of the encoding costs of its segments . minimizing @xmath706 enables one to find not only the most modular partition for each graph segment ( high modularity corresponds to low encoding costs for a partition ) , but also the most compact subdivision of the snapshots into segments , such that graphs in the same segment are strongly correlated with each other . the latter feature allows to identify _ change points _ in the time history of the system , i. e. short periods in which the dynamics produces big changes in the graph structure ( corresponding to , e.g. , extreme events ) . the minimization of @xmath706 is @xmath622-hard , so the authors propose an approximation method called _ graphscope _ , which consists of two steps : first , one looks for the best partition of each graph segment ; second , one looks for the best division in segments . in both cases the `` best '' result corresponds to the minimal encoding cost . the best partition within a graph segment is found by local search . graphscope has the big advantage not to require any input , like the number and sizes of the clusters . it is also suitable to operate in a streaming environment , in which new graph configurations are added in time , following the evolution of the system : the computational complexity required to process a snapshot ( on average ) is stable over time . tests on real evolving data sets show that graphscope is able to find meaningful communities and change points . since keeping track of communities in different time steps is not a trivial problem , as we have seen above , it is perhaps easier to adopt a vertex - centric perspective , in which one monitors the community of a given vertex at different times . for any method , given a vertex @xmath67 and a time @xmath156 , the community to which @xmath67 belongs at time @xmath156 is well defined . fenn et al . @xcite used the multiresolution method by reichardt et al . @xcite ( section [ sub_sec6_01 ] ) and investigated a fully connected graph with time - dependent weights , representing the correlations of time series of hourly exchange rate returns . the resolution parameter @xmath371 is fixed to the value that occurs in most stability plateaus of the system at different time steps . motivated by the work of guimer and amaral @xcite ( section [ sec7_1 ] ) , fenn et al . identify the role of individual vertices in their community through the pair @xmath707 , where @xmath708 is the @xmath406-score of the internal strength ( weighted degree , section [ sec1_1 ] ) , defined in eq . [ eqtrol1 ] , and @xmath709 the @xmath406-score of the site betweenness , defined by replacing the internal degree with the site betweenness of freeman @xcite in eq . [ eqtrol1 ] . we remind that the site betweenness is a measure of the number of shortest paths running through a vertex . the variable @xmath709 expresses the importance of a vertex in processes of information diffusion with respect to the other members of its community . another important issue regards the _ persistence _ of communities in time , i. e. how stable they are during the evolution . as a measure of persistence , fenn et al . introduced a vertex - centric version of the relative overlap of eq . [ eqtrol3 ] @xmath710 where @xmath67 is the vertex and @xmath711 , @xmath712 the communities of @xmath67 at times @xmath156 , @xmath713 , respectively . the decay of @xmath714 depends on the type of vertex . in particular , if the vertex is strongly connected to its community ( @xmath708 large ) , @xmath714 decays quite slowly , meaning that it tends to stay attached to a stable core of vertices . the methods described above are basically two - stage approaches , in which clusters are detected at each timestamp of the graph evolution , independently of the results at different times , and relationships between partitions at different times are inferred successively . however , this often produces significant variations between partitions close in time , especially when , as it usually happens , the datasets are noisy . in this case one would be forced to introduce _ ad hoc _ hypotheses on the graph evolution to justify the variability of the modular structure , whereas such variability is mostly an artefact of the approach . it would be desirable instead to have a unified framework , in which clusters are deduced both from the current structure of the graph and from the knowledge of the cluster structure at previous times . this is the principle of _ evolutionary clustering _ , a framework introduced by chakrabarti et al . @xcite and successively adopted and refined by other authors . let @xmath715 be the partition of the graph at time @xmath156 . the _ snapshot quality _ of @xmath715 measures the goodness of the partition with respect to the graph structure at time @xmath156 . the _ history cost _ is a measure of the distance / dissimilarity of @xmath715 with respect to the partition @xmath716 at the previous time step . the overall quality of @xmath715 is given by a combination of the snapshot quality and the history cost at each time step . ideally , a good partition should have high snapshot quality ( i. e. it should cluster well the data at time @xmath156 ) and low history cost ( i. e. it should be similar to the partition at the previous time step ) . in order to find @xmath715 from @xmath716 and the relational data at time @xmath156 chakrabarti et al . suggested to minimize the difference between the snapshot quality and the history cost , with a relative weight @xmath717 that is a tunable parameter . the input of the procedure consists in the sequence of adjacency / similarity matrices at different time steps . in practice , one could use modified versions of such matrices , obtained by performing ( weighted ) averages of the data over some time window , in order to make the relational data more robust against noise and the results of the clustering procedure more reliable . one can adopt arbitrary measures to compute the snapshot quality and the historical cost . besides , several known clustering techniques used for static graphs can be reformulated within this evolutionary framework . chakrabarti et al . derived evolutionary versions of hierarchical clustering ( section [ sec4_2 ] ) and @xmath44-means clustering ( section [ sec4_3 ] ) , whereas chi et al . @xcite designed two implementations for spectral clustering ( section [ sec4_4 ] ) . based on evolutionary clustering , lin et al . @xcite introduced a framework , called _ facetnet _ , that allows vertices to belong to more communities at the same time . here the snapshot cost is the kullback - leibler ( kl ) divergence @xcite between the adjacency / similarity matrix at time @xmath156 and the matrix describing the community structure of the graph at time @xmath156 ; the history cost is the kl divergence between the matrices describing the community structure of the graph at times @xmath718 and @xmath156 . facetnet can be extended to handle adding and removing of vertices as well as variations of the number of clusters in consecutive time steps . however , it is not able to account for the creation and the disintegration of communities and not scalable to large systems due to the high number of iterations necessary for the matrix computations to reach convergence . these issues have been addressed in a recent approach by kim and han @xcite . naturally , what one hopes to achieve at the end of the day is to see how real groups form and evolve in time . backstrom et al . @xcite have carried out an analysis of group dynamics in the free online community of livejournal ( http://www.livejournal.com/ ) and in a coauthorship network of computer scientists . here the groups are identified through the declared memberships of users ( for livejournal ) and conferences attended by computer scientists , respectively . backstrom and coworkers have found that the probability that an individual joins a community grows with the number of friends / coauthors who are already in the community and ( for livejournal ) with their degree of interconnectedness . moreover , the probability of growth of livejournal communities is positively correlated to a combination of factors including the community size , the number of friends of community members which are not in the community and the ratio of these two numbers . a high density of triads within a community appears instead to hinder its growth . given a network , many partitions could represent meaningful clusterings in some sense , and it could be difficult for some methods to discriminate between them . quality functions evaluate the goodness of a partition ( section [ sec3_2_2 ] ) , so one could say that high quality corresponds to meaningful partitions . but this is not necessarily true . in section [ sub_sec6_1 ] we have seen that high values of the modularity of newman and girvan do not necessarily indicate that a graph has a definite cluster structure . in particular we have seen that partitions of random graphs may also achieve considerably large values of @xmath90 , although we do not expect them to have community structure , due to the lack of correlations between the linking probabilities of the vertices . the optimization of quality functions , like modularity , delivers the best partition according to the criterion underlying the quality function . but is the optimal clustering also _ significant _ , i. e. a relevant feature of the graph , or is it just a byproduct of randomness and basic structural properties like , e. g. , the degree sequence ? little effort has been devoted to this crucial issue , that we discuss here . in some works the concept of significance has been related to that of _ robustness _ or _ stability _ of a partition against random perturbations of the graph structure . the basic idea is that , if a partition is significant , it will be recovered even if the structure of the graph is modified , as long as the modification is not too extensive . instead , if a partition is not significant , one expects that minimal modifications of the graph will suffice to disrupt the partition , so other clusterings are recovered . a nice feature of this approach is the fact that it can be applied for any clustering technique . gfeller et al . @xcite considered the general case of weighted graphs . a graph is modified , in that its edge weights are increased or decreased by a relative amount @xmath719 . this choice also allows to account for the possible effects of uncertainties in the values of the edge weights , resulting from measurements / experiments carried out on a given system . after fixing @xmath720 ( usually to @xmath721 ) , multiple realizations of the original graph are generated . the best partition for each realization is identified and , for each pair of adjacent vertices @xmath67 and @xmath68 , the _ in - cluster _ probability @xmath362 is computed , i. e. the fraction of realizations in which @xmath67 and @xmath68 were classified in the same cluster . edges with in - cluster probability smaller than a threshold @xmath722 ( usually @xmath723 ) are called _ external edges_. the stability of a partition is estimated through the _ clustering entropy _ @xmath724 , \label{sign1}\ ] ] where @xmath14 is , as usual , the number of graph edges , and the sum runs over all edges . the most stable partition has @xmath725 along inter - cluster edges and @xmath726 along intra - cluster edges , which yields @xmath727 ; the most unstable partition has @xmath728 on all edges , yielding @xmath729 . the absolute value of @xmath87 is not meaningful , though , and needs to be compared with the corresponding value for a null model graph , similar to the original graph , but with supposedly no cluster structure . gfeller et al . adopted the same null model of newman - girvan modularity , i. e. the class of graphs with expected degree sequence coinciding with that of the original graph . since the null model is defined on unweighted graphs , the significance of @xmath87 can be assessed only in this case , although it would not be hard to think of a generalization to weighted graphs . the approach enables one as well to identify _ unstable vertices _ , i. e. vertices lying at the boundary between clusters . in order to do that , the external edges are removed and the connected components of the resulting disconnected graph are associated with the clusters detected in the original graph , based on their relative overlap ( computed through eq . [ eqt20 ] ) . unstable vertices end up in components that are not associated to any of the initial clusters . a weakness of the method by gfeller et al . is represented by the two parameters @xmath720 and @xmath722 , whose values are in principle arbitrary . more recently , karrer et al . @xcite adopted a similar strategy to unweighted graphs . here one performs a sweep over all edges : the perturbation consists in removing each edge with a probability @xmath18 and replacing it with another edge between a pair of vertices @xmath730 , chosen at random with probability @xmath369 , where @xmath105 and @xmath106 are the degrees of @xmath67 and @xmath68 . we recognize the probability of the null model of newman - girvan modularity . indeed , by varying the probability @xmath18 from @xmath631 to @xmath2 one smoothly interpolates between the original graph ( no perturbation ) and the null model ( maximal perturbation ) . the degree sequence of the graph remains invariant ( on average ) along the whole process , by construction . the idea is that the perturbation affects solely the organization of the vertices , keeping the basic structural properties . for a given value of @xmath18 , many realizations of the perturbed graph are generated , their cluster structures are identified with some method ( karrer et al . used modularity optimization ) and compared with the partition obtained from the original unperturbed graph . the partitions are compared by computing the variation of information @xmath638 ( section [ sec6_2 ] ) . from the plot of the average @xmath731 versus @xmath18 one can assess the stability of the cluster structure of the graph . if @xmath732 changes rapidly for small values of @xmath18 the partition is likely to be unstable . as in the approach by gfeller et al . the behaviour of the function @xmath732 does not have an absolute meaning , but needs to be compared with the corresponding curve obtained for a null model . for consistency , the natural choice is again the null model of modularity , which is already used in the process of graph perturbation . the approaches by gfeller et al . and karrer et al . , with suitable modifications , can also be used to check for the stability of the cluster structure in parts of a graph , up to the level of individual communities . this is potentially important as it may happen that some parts of the graph display a strong community structure and other parts weak or no community structure at all . rosvall and bergstrom @xcite defined the significance of clusters with the bootstrap method @xcite , which is a standard procedure to check for the accuracy of a measurement / estimate based on resampling from the empirical data . the graph at study is supposed to be generated by a parametric model , which is used to create many samples . this is done by assigning to each edge a weight taken by a poisson distribution with mean equal to the original edge weight . for the initial graph and each sample one identifies the community structure with some method , that can be arbitrary . for each cluster of the partition of the original graph one determines the largest subset of vertices that are classified in the same cluster in at least @xmath733 of all bootstrap samples . identifying such cluster cores enables one to track the evolution in time of the community structure , as we explained in section [ sec7_1_2 ] . a different approach has been proposed by massen and doye @xcite . they analyzed an equilibrium canonical ensemble of partitions , with @xmath734 playing the role of the energy , @xmath90 being newman - girvan modularity . this means that the probability of occurrence of a partition at temperature @xmath467 is proportional to @xmath735 . the idea is that , if a graph has a significant cluster structure , at low temperatures one would recover essentially the same partition , corresponding to the modularity maximum , which is separated by an appreciable gap from the modularity values of the other partitions . on the contrary , graphs with no community structure , e. g. random graphs , have many competing ( local ) maxima , and the corresponding configurations will emerge already at low temperatures , since their modularity values are close to the absolute maximum , good et al . @xcite have actually shown that the modularity landscape has a huge degeneracy of states with high modularity values , close to the global maximum , especially on graphs with community structure . so the results of the method by massen and doye may be misleading . ] . these distinct behaviors can manifest themselves in various ways . for instance , if one considers the variation of the specific heat @xmath736 with @xmath467 , the gap in the modularity landscape is associated to a sharp peak of @xmath706 around some temperature value , like it happens in a phase transition . if the gap is small and there are many partitions with similar modularity values , the peak of @xmath706 becomes broad . another strategy to assess the significance of the maximum modularity partition consists of the investigation of the similarity between partitions recovered at a given temperature @xmath467 . this similarity can be expressed by the _ frequency matrix _ , whose element @xmath737 indicates the relative number of times vertices @xmath67 and @xmath68 have been classified in the same cluster . if the graph has a clear community structure , at low temperatures the frequency matrix can be put in block - diagonal form , with the blocks corresponding to the communities of the best partition ; if there is no significant community structure , the frequency matrix is rather homogeneous . the fiedler eigenvalue @xcite @xmath144 , the second smallest eigenvalue of the laplacian matrix associated to the frequency matrix , allows to estimate how `` block - diagonal '' the matrix is ( see section [ sec4_1 ] ) . at low temperatures @xmath738 if there is one ( a few ) partitions with maximum or near to maximum modularity ; if there are many ( almost ) degenerate partitions , @xmath144 is appreciably different from zero even when @xmath739 . a sharp transition from low to high values of @xmath144 by varying temperature indicates significant community structure . another clear signature of significant community structure is the observation of a rapid drop of the average community size with @xmath467 , as `` strong '' communities break up in many small pieces for a modest temperature increase , while the disintegration of `` weak '' communities takes place more slowly . in scale - free graphs ( section [ sec1_3 ] ) clusters are often not well separated , due to the presence of the hubs ; in these cases the above - mentioned transitions of ensemble variables are not so sharp and take place over a broader temperature range . the canonical ensemble of partitions is generated through single spin heatbath simulated annealing @xcite , combined with parallel tempering @xcite . the approach by massen and doye could be useful to recognize graphs without cluster structure , if the modularity landscape is characterized by many maxima with close values ( but see footnote ) . however , it can happen that gaps between the absolute modularity maximum and the rest of the modularity values are created by fluctuations , and the method is unable to identify these situations . furthermore , the approach heavily relies on modularity and on a costly technique like simulated annealing : extensions to other quality functions and/or optimization procedures do not appear straightforward . in a recent work by bianconi et al . @xcite the notion of entropy of graph ensembles @xcite is employed to find out how likely it is for a cluster structure to occur on a graph with a given degree sequence . the entropy is computed from the number of graph configurations which are compatible with a given classification of the vertices in @xmath91 groups . the clustering is quantitatively described by fixing the number of edges @xmath740 running between clusters @xmath562 and @xmath563 , for all choices of @xmath741 . bianconi et al . proposed the following indicator of clustering significance @xmath742 where @xmath743 is the entropy of the graph configurations with given degree sequence @xmath744 and clustering @xmath745 ( with fixed numbers of inter - cluster edges @xmath740 ) , and @xmath746 is the average entropy of the configurations with the same degree sequence and a random permutation @xmath747 of the cluster labels . the absolute value of the entropy @xmath743 is not meaningful , so the comparison of @xmath743 and @xmath746 is crucial , as it tells how relevant the actual cluster structure is with respect to a random classification of the vertices . however , different permutations of the assignments @xmath745 yield different values of the entropy , which can fluctuate considerably . therefore one has to compute the standard deviation @xmath748 of the entropy corresponding to all random permutations @xmath747 , to estimate how significant the difference between @xmath743 and @xmath746 is . in this way , if @xmath749 , the entropy of the given cluster structure is of the same order as the entropy of some random permutation of the cluster labels , so it is not relevant . instead , if @xmath750 , the cluster structure is far more likely than a random classification of the vertices , so the clustering is relevant . the indicator @xmath751 can be simply generalized to the case of directed and weighted graphs . lancichinetti et al . @xcite as well addressed the issue by comparing the cluster structure of the graph with that of a random graph with similar properties . an important novelty of this approach is the fact that it estimates the significance of single communities , not of partitions . in fact , not all communities are equally significant , in general , so it makes a lot of sense to check them individually . in particular , it may happen that real networks are not fully modular , due to their particular history or generating mechanisms , and that only portions of them display community structure . the main idea is to verify how likely it is that a community @xmath22 is a subgraph of a random graph with the same degree sequence of the original graph . this likelihood is called @xmath22-score , and is computed by examining the vertex @xmath752 of @xmath22 , with the lowest internal degree @xmath753 in @xmath22 ( the `` worst '' vertex ) . the @xmath22-score is defined as the probability that the internal degree of the worst vertex in the corresponding community of the null model graphs is larger than or equal to @xmath753 . this probability is computed by using tools from extreme and order statistics @xcite . a low value of the @xmath22-score ( @xmath754 ) is a strong indication that the group of vertices at study is a community and not the product of random fluctuations . in addition , the measure can be used to check whether a subgraph has an internal modular structure . for that , one removes the vertices of the subgraph one at a time , starting from the worst and proceeding in increasing order of the internal degree , and observes how the @xmath22-score varies at each vertex removal : sharp drops indicate the presence of dense subgraphs ( fig . [ cscore ] ) . therefore , one could think of using the @xmath22-score as ingredient of new clustering techniques . as we have seen , the @xmath22-score is based on the behavior of the vertex with lowest internal degree of the subgraph . real networks are characterized by noise , which could strongly affect the structural relationships between vertices and clusters . for this reason , relying on the properties of a single vertex to evaluate the significance of a subgraph could be a problem for applications to real networks . lancichinetti et al . have shown that the @xmath22-score can be easily extended to consider the @xmath156 vertices with lowest internal degree , with @xmath755 ( @xmath259-score ) . the main limit of the @xmath22-score is the fact that its null model is the same as that of newman - girvan modularity . according to this null model , each vertex can in principle be connected to any other , no matter how large the system is . this is however not realistic , especially for large graphs , where it is much more reasonable to assume that each vertex has its own `` horizon '' , i.e. a subset of other vertices with which it can interact , which is usually much smaller than the whole system ( see section [ sub_sec6_1 ] ) . how to define such `` horizons '' and , more in general , realistic null models is still an open problem . however , the @xmath22-score could be easily reformulated with any null model , so one could readily derive more reliable definitions . we conclude with a general issue which is related to the significance of community structure . the question is : given a cluster structure in a graph , can it be recovered _ a priori _ by an algorithm ? in a recent paper @xcite , reichardt and leone studied under which conditions a special built - in cluster structure can be recovered . the clusters have equal size and a pair of vertices is connected with probability @xmath45 if they belong to the same cluster , with probability @xmath756 otherwise . in computer science this is known as the _ planted partitioning problem _ the goal is to propose algorithms that recover the planted partition for any choice of @xmath45 and @xmath375 . for dense graphs , i. e. graphs whose average degree grows with the number @xmath15 of vertices , algorithms can be designed that find the solution with a probability which equals @xmath2 minus a term that vanishes in the limit of infinite graph size , regardless of the difference @xmath757 , which can then be chosen arbitrarily small . since many real networks are not dense graphs , as their average degree @xmath758 is usually much smaller than @xmath15 and does not depend on it , reichardt and leone investigated the problem in the case of fixed @xmath758 and infinite graph size . we indicate with @xmath91 the number of clusters and with @xmath511 the probability that a randomly selected edge of the graph lies within any of the @xmath91 clusters . in this way , if @xmath759 , the inter - cluster edge density matches the intra - cluster edge density ( i. e. @xmath760 ) , and the planted partition would not correspond to a recoverable clustering , whereas for @xmath761 , there are no inter - cluster edges and the partition can be trivially recovered . the value of @xmath511 is in principle unknown , so one has to detect the cluster structure ignoring this information . reichardt and leone proposed to look for a minimum cut partition , i. e. for the partition that minimizes the number of inter - cluster edges , as it is usually done in the graph partitioning problem ( discussed in section [ sec4_1 ] ) . clearly , for @xmath761 the minimum cut partition trivially coincides with the planted partition , whereas for @xmath762 there should be some overlap , which is expected to vanish in the limit case @xmath759 . the minimum cut partition corresponds to the minimum of the following ferromagnetic potts model hamiltonian @xmath763 over the set of all spin configurations with zero magnetization . here the spin @xmath357 indicates the cluster vertex @xmath67 belongs to , and the coupling matrix @xmath363 is just the adjacency matrix of the graph . the constraint of zero magnetization ensures that the clusters have all the same size , as required by the planted partitioning problem . the energy of a spin configuration , expressed by eq . [ eq : mod12 ] , is the negative of the number of edges that lie within clusters : the minimum energy corresponds to the maximum number of intra - cluster edges , which is coupled to the minimum number of inter - cluster edges . the minimum energy can be computed with the cavity method , or belief propagation , at zero temperature @xcite . the accuracy of the solution with respect to the planted partition is expressed by the fraction of vertices which are put in the same class in both partitions . the analysis yields a striking result : the planted clustering is accurately recovered for @xmath511 larger than a critical threshold @xmath764 . so , there is a range of values of @xmath511 , @xmath765 , in which the clustering is not recoverable , as the minimum cut partition is uncorrelated with it . the threshold @xmath766 depends on the degree distribution @xmath767 of the graph . when a clustering algorithm is designed , it is necessary to test its performance , and compare it with that of other methods . in the previous sections we have said very little about the performance of the algorithms , other than their computational complexity . indeed , the issue of testing algorithms has received very little attention in the literature on graph clustering . this is a serious limit of the field . because of that , it is still impossible to state which method ( or subset of methods ) is the most reliable in applications , and people rely blindly on some algorithms instead of others for reasons that have nothing to do with the actual performance of the algorithms , like . e.g. popularity ( of the method or of its inventor ) . this lack of control is also the main reason for the proliferation of graph clustering techniques in the last few years . virtually in any paper , where a new method is introduced , the part about testing consists in applying the method to a small set of simple benchmark graphs , whose cluster structure is fairly easy to recover . because of that , the freedom in the design of a clustering algorithm is basically infinite , whereas it is not clear what a new procedure is adding to the field , if anything . in this section we discuss at length the issue of testing . first , we describe the fundamental ingredients of any testing procedure , i. e. benchmark graphs with built - in community structure , that methods have to identify ( section [ sec6_1 ] ) . we proceed by reviewing measures to compare graph partitions with each other ( section [ sec6_2 ] ) . in section [ sec6_3 ] we present the comparative evaluations of different methods that have been performed in the literature . testing an algorithm essentially means applying it to a specific problem whose solution is known and comparing such solution with that delivered by the algorithm . in the case of graph clustering , a problem with a well - defined solution is a graph with a clear community structure . this concept is not trivial , however . many clustering algorithms are based on similar intuitive notions of what a community is , but different implementations . so it is crucial that the scientific community agrees on a set of reliable benchmark graphs . this mostly applies to computer - generated graphs , where the built - in cluster structure can be arbitrarily designed . in the literature real networks are used as well , in those cases in which communities are well defined because of information about the system . we start our survey from computer - generated benchmarks . a special class of graphs has become quite popular in the last years . they are generated with the so - called _ planted @xmath293-partition model _ the model partitions a graph with @xmath768 vertices in @xmath293 groups with @xmath126 vertices each . vertices of the same group are linked with a probability @xmath511 , whereas vertices of different groups are linked with a probability @xmath512 . each subgraph corresponding to a group is then a random graph la erds - rnyi with connection probability @xmath769 ( section [ sec1_3 ] ) . the average degree of a vertex is @xmath770 . if @xmath513 the intra - cluster edge density exceeds the inter - cluster edge density and the graph has a community structure . this idea is quite intuitive and we have encountered it in several occasions in the previous sections . girvan and newman considered a special case of the planted @xmath293-partition model @xcite . they set @xmath771 , @xmath772 ( so the number of graph vertices is @xmath773 ) and fixed the average total degree @xmath420 to @xmath774 . this implies that @xmath775 , so the probabilities @xmath511 and @xmath512 are not independent parameters . in calculation it is common to use as parameters @xmath776 and @xmath777 , indicating the expected internal and external degree of a vertex , respectively . these particular graphs have by now gained the status of standard benchmarks @xcite ( fig . [ figure11 ] ) . in the first applications of the graphs one assumed that communities are well defined when @xmath778 , corresponding to the situation in which the internal degree exceeds the external degree . however , the threshold @xmath779 implies @xmath780 and @xmath781 , so it is not the actual threshold of the model , where @xmath782 , corresponding to @xmath783 . so , one expects to be able to detect the planted partition up until @xmath783 . testing a method against the girvan - newman benchmark consists in calculating the similarity between the partitions determined by the method and the natural partition of the graph in the four equal - sized groups . several measures of partitions similarity may be adopted ; we describe them in section [ sec6_2 ] . one usually builds many graph realizations for a particular value of @xmath784 and computes the average similarity between the solutions of the method and the built - in solution . the procedure is then iterated for different values of @xmath784 . the results are usually represented in a plot , where the average similarity is drawn as a function of @xmath784 . most algorithms usually do a good job for small @xmath784 and start to fail when @xmath784 approaches @xmath228 . fan et al . @xcite have designed a weighted version of the benchmark of girvan and newman , in that one gives different weights to edges inside and between communities . one could pick just two values , one for intra- and the other for inter - community edges , or uniformly distributed values in two different ranges . for this benchmark there are then two parameters that can be varied : @xmath784 and the relative importance of the internal and the external weights . typically one fixes the topological structure and varies the weights . this is particularly insightful when @xmath785 , which delivers graphs without topological cluster structure : in this case , the question whether there are clusters or not depends entirely on the weights . as we have remarked above , the planted @xmath293-partition model generates mutually interconnected random graphs la erds - rnyi . therefore , all vertices have approximately the same degree . moreover , all communities have exactly the same size by construction . these two features are at odds with what is observed in graph representations of real systems . degree distributions are usually skewed , with many vertices with low degree coexisting with a few vertices with high degree . a similar heterogeneity is also observed in the distribution of cluster sizes , as we shall see in section [ sec7_1 ] . so , the planted @xmath293-partition model is not a good description of a real graph with community structure . however , the model can be modified to account for the heterogeneity of degrees and community sizes . a modified version of the model , called _ gaussian random partition generator _ , was designed by brandes et al . @xcite . here the cluster sizes have a gaussian distribution , so they are not the same , although they do not differ much from each other . the heterogeneity of the cluster sizes introduces a heterogeneity in the degree distribution as well , as the expected degree of a vertex depends on the number of vertices of its cluster . still , the variability of degree and cluster size is not appreciable . besides , vertices of the same cluster keep having approximately the same degree . a better job in this direction has been recently done by lancichinetti et al . ( lfr benchmark ) @xcite . they assume that the distributions of degree and community size are power laws , with exponents @xmath786 and @xmath787 , respectively . each vertex shares a fraction @xmath788 of its edges with the other vertices of its community and a fraction @xmath789 with the vertices of the other communities ; @xmath790 is the _ mixing parameter_. the graphs are built as follows : 1 . a sequence of community sizes obeying the prescribed power law distribution is extracted . this is done by picking random numbers from a power law distribution with exponent @xmath787 . each vertex @xmath67 of a community receives an internal degree @xmath791 , where @xmath105 is the degree of vertex @xmath67 , which is taken by a power law distribution with exponent @xmath786 . in this way , each vertex @xmath67 has a number of stubs @xmath791 . all stubs of vertices of the same community are randomly attached to each other , until no more stubs are `` free '' . in this way the sequence of internal degrees of each vertex in its community is maintained . each vertex @xmath67 receives now an additional number of stubs , equal to @xmath792 ( so that the final degree of the vertex is @xmath105 ) , that are randomly attached to vertices of different communities , until no more stub is `` free '' . numerical tests show that this procedure has a complexity @xmath21 , where @xmath14 is as usual the number of edges of the graph , so it can be used to create graphs of sizes spanning several orders of magnitude . [ fignewbench ] shows an example of a lfr benchmark graph . recently the lfr benchmark has been extended to directed and weighted graphs with overlapping communities @xcite . the software to create the lfr benchmark graphs can be freely downloaded at http://santo.fortunato.googlepages.com/inthepre + ss2 . a class of benchmark graphs with power law degree distributions had been previously introduced by bagrow @xcite . the construction process starts from a graph with a power - law degree distribution . bagrow used barabsi - albert scale free graphs @xcite . then , vertices are randomly assigned to one of four equally - sized communities . finally , pairs of edges between two communities are rewired so that either edge ends up within the same community , without altering the degree sequence ( on average ) . this is straightforward : suppose that the edges join the vertex pairs @xmath793 , @xmath794 and @xmath795 , @xmath796 , where @xmath793 , @xmath795 belong to community @xmath9 and @xmath794 , @xmath796 to community @xmath10 . it the edges are replaced by @xmath793-@xmath795 and @xmath794-@xmath796 ( provided they do not exist already ) , all vertices keep their degrees . with this rewiring procedure one can arbitrarily vary the edge density within and , consequently , between clusters . in this class of benchmarks , however , communities are all of the same size by construction , although one can in principle relax this condition . a ( seemingly ) different benchmark is represented by the class of _ relaxed caveman graphs _ , which were originally introduced to explain the clustering properties of social networks @xcite . the starting point is a set of disconnected cliques . with some probability edges are rewired to link different cliques . such model graphs are interesting as they are smooth variations of the ideal graph with `` perfect '' communities , i. e. disconnected cliques . on the other hand the model is equivalent to the planted @xmath293-partition model , where @xmath797 and @xmath512 is proportional to @xmath45 , with coefficient depending on the size of the clusters . benchmark graphs have also been introduced to deal with special types of graphs and/or cluster structures . for instance , arenas et al . @xcite have introduced a class of benchmark graphs with embedded hierarchical structure , which extends the class of graphs by girvan and newman . here there are @xmath798 vertices and two hierarchical levels , corresponding to a partition in @xmath774 groups ( microcommunities ) with @xmath774 vertices and a partition in @xmath223 larger groups of @xmath799 vertices ( macrocommunities ) , comprising each @xmath223 of the smaller groups . the edge densities within and between the clusters are indicated by three parameters @xmath800 , @xmath801 and @xmath784 : @xmath800 is the expected internal degree of a vertex within its microcommunity ; @xmath801 is the expected number of edges that the vertex shares with the vertices of the other microcommunities within its macrocommunity ; @xmath784 is the expected number of edges connecting the vertex with vertices of the other three macrocommunities . the average degree @xmath802 of a vertex is fixed to @xmath12 . [ fig8 ] shows an example of hierarchical graph constructed based on the same principle , with @xmath803 vertices and an average degree of @xmath11 . guimer et al . @xcite have proposed a model of bipartite graphs with built - in communities . they considered a bipartite graph of actors and teams , here we describe how to build the benchmarks for general bipartite graphs . one starts from a bipartite graph whose vertex classes @xmath9 and @xmath10 are partitioned into @xmath115 groups , @xmath804 and @xmath805 ( @xmath806 ) . each cluster @xmath807 comprises all vertices of the subgroups @xmath804 and @xmath805 , respectively . with probability @xmath45 edges are placed between vertices of subgroups @xmath804 and @xmath805 ( @xmath806 ) , i. e. within clusters . with probability @xmath808 , edges are placed between vertices of subgroups @xmath804 and @xmath809 , where @xmath67 and @xmath68 are chosen at random , so they can be equal or different . by construction , a non - zero value of the probability @xmath45 indicates a preference by vertices to share links with vertices of the same cluster , i. e. for @xmath810 the graph has a built - in community structure . for @xmath811 there would be edges only within clusters , i. e. the graph has a perfect cluster structure . finally , sawardecker et al . introduced a general model , that accounts for the possibility that clusters overlap @xcite . the model is based on the reasonable assumption that the probability @xmath362 that two vertices are connected by an edge grows with the number @xmath210 of communities both vertices belong to . for vertices in different clusters , @xmath812 , if they are in the same cluster ( and only in that one ) @xmath813 , if they belong to the same two clusters @xmath814 , etc .. by hypothesis , @xmath815 . the planted @xmath293-partition model is recovered when @xmath816 . as we have seen , nearly all existing benchmark graphs are inspired by the planted @xmath293-partition model , to some extent . however , the model needs to be refined to provide a good description of real graphs with community structure . the hypothesis that the linking probabilities of each vertex with the vertices of its community or of the other communities are constant is not realistic . it is more plausible that each pair of vertices @xmath67 and @xmath68 has its own linking probability @xmath362 , and that such probabilities are correlated for vertices in the same cluster . tests on real networks usually focus on a very limited number of examples , for which one has precise information about the vertices and their properties . in section [ sec2 ] we have introduced two popular real networks with known community structure , i. e. the social network of zachary s karate club and the social network of bottlenose dolphins living in doubtful sound ( new zealand ) , studied by lusseau . here , the question is whether the actual separation in two social groups could be predicted from the graph topology . zachary s karate club is by far the most investigated system . several algorithms are actually able to identify the two classes , modulo a few intermediate vertices , which may be misclassified . other methods are less successful : for instance , the maximum of newman - girvan modularity corresponds to a split of the network in four groups @xcite . another well known example is the network of american college football teams , derived by girvan and newman @xcite . there are @xmath817 vertices , representing the teams , and two vertices are connected if their teams play against each other . the teams are divided into @xmath818 conferences . games between teams in the same conference are more frequent than games between teams of different conferences , so one has a natural partition where the communities correspond to the conferences . when dealing with real networks , it is useful to resolve their community structure with different clustering techniques , to cross - check the results and make sure that they are consistent with each other , as in some cases the answer may strongly depend on the specific algorithm adopted . however , one has to keep in mind that there is no guarantee that `` reasonable '' communities , defined on the basis of non - structural information , must coincide with those detected by methods based only on the graph structure . checking the performance of an algorithm involves defining a criterion to establish how `` similar '' the partition delivered by the algorithm is to the partition one wishes to recover . several measures for the similarity of partitions exist . in this section we present and discuss the most popular measures . a thorough introduction of similarity measures for graph partitions has been given by meil @xcite and we will follow it closely . let us consider two generic partitions @xmath819 and @xmath820 of a graph @xmath313 , with @xmath821 and @xmath822 clusters , respectively . we indicate with @xmath15 the number of graph vertices , with @xmath823 and @xmath824 the number of vertices in clusters @xmath825 and @xmath826 and with @xmath827 the number of vertices shared by clusters @xmath825 and @xmath826 : @xmath828 . in the first tests using the benchmark graphs by girvan and newman ( section [ sec6_1 ] ) scholars used a measure proposed by girvan and newman themselves , the _ fraction of correctly classified vertices_. a vertex is correctly classified if it is in the same cluster with at least half of its `` natural '' partners . if the partition found by the algorithm has clusters given by the merging of two or more natural groups , all vertices of the cluster are considered incorrectly classified . the number of correctly classified vertices is then divided by the total size of the graph , to yield a number between @xmath631 and @xmath2 . the recipe to label vertices as correctly or incorrectly classified is somewhat arbitrary , though . apart from the fraction of correctly classified vertices , which is somewhat _ ad hoc _ and distinguishes the roles of the natural partition and of the algorithm s partition , most similarity measures can be divided in three categories : measures based on _ pair counting _ , _ cluster matching _ and _ information theory_. measures based on pair counting depend on the number of pairs of vertices which are classified in the same ( different ) clusters in the two partitions . in particular @xmath829 indicates the number of pairs of vertices which are in the same community in both partitions , @xmath830 ( @xmath831 ) the number of pairs of elements which are put in the same community in @xmath832 ( @xmath833 ) and in different communities in @xmath833 ( @xmath832 ) and @xmath834 the number of pairs of vertices that are in different communities in both partitions . wallace @xcite proposed the two indices @xmath835 @xmath836 and @xmath837 represent the probability that vertex pairs in the same cluster of @xmath832 are also in the same cluster for @xmath833 , and viceversa . these indices are asymmetrical , as the role of the two partitions is not the same . fowlkes and mallows @xcite suggested to use the geometric mean of @xmath836 and @xmath837 , which is symmetric . the _ rand index _ @xcite is the ratio of the number of vertex pairs correctly classified in both partitions ( i. e. either in the same or in different clusters ) , by the total number of pairs @xmath838 a measure equivalent to the rand index is the _ mirkin metric _ @xcite @xmath839 . \label{eqt03}\ ] ] the _ jaccard index _ is the ratio of the number of vertex pairs classified in the same cluster in both partitions , by the number of vertex pairs which are classified in the same cluster in at least one partition , i. e. @xmath840 adjusted versions of both the rand and the jaccard index exist , in that a null model is introduced , corresponding to the hypothesis of independence of the two partitions @xcite . the null model expectation value of the measure is subtracted from the unadjusted version , and the result is normalized by the range of this difference , yielding @xmath2 for identical partitions and @xmath631 as expected value for independent partitions ( negative values are possible as well ) . unadjusted indices have the drawback that they are not local , i. e. the result depends on how the whole graph is partitioned , even when the partitions differ only in a small region of the graph . similarity measures based on cluster matching aim at finding the largest overlaps between pairs of clusters of different partitions . for instance , the _ classification error _ @xmath841 is defined as @xcite @xmath842 where @xmath843 is an injective mapping from the cluster indices of partition @xmath833 to the cluster indices of partition @xmath832 . the maximum is taken over all possible injections @xmath510 . in this way one recovers the maximum overlap between the clusters of the two partitions . an alternative measure is the _ normalized van dongen metric _ , defined as @xcite @xmath844 . \label{eqt06}\ ] ] a common problem of this type of measures is that some clusters may not be taken into account , if their overlap with clusters of the other partition is not large enough . therefore if we compute the similarity between two partitions @xmath845 and @xmath846 and partition @xmath847 , with @xmath845 and @xmath846 differing from each other by a different subdivision of parts of the graph that are not used to compute the measure , one would obtain the same score . the third class of similarity measures is based on reformulating the problem of comparing partitions as a problem of message decoding within the framework of information theory @xcite . the idea is that , if two partitions are similar , one needs very little information to infer one partition given the other . this extra information can be used as a measure of dissimilarity . to evaluate the shannon information content @xcite of a partition , one starts by considering the community assignments @xmath848 and @xmath849 , where @xmath170 and @xmath850 indicate the cluster labels of vertex @xmath67 in partition @xmath832 and @xmath833 , respectively . one assumes that the labels @xmath302 and @xmath851 are values of two random variables @xmath168 and @xmath585 , with joint distribution @xmath852 , which implies that @xmath853 and @xmath854 . the _ mutual information _ @xmath595 of two random variables has been previously defined [ eq . ( [ eqstinf8 ] ) ] , and can be applied as well to partitions @xmath832 and @xmath833 , since they are described by random variables . actually @xmath855 , where @xmath856 is the shannon entropy of @xmath168 and @xmath857 is the conditional entropy of @xmath168 given @xmath585 . the mutual information is not ideal as a similarity measure : in fact , given a partition @xmath832 , all partitions derived from @xmath832 by further partitioning ( some of ) its clusters would all have the same mutual information with @xmath832 , even though they could be very different from each other . in this case the mutual information would simply equal the entropy @xmath858 , because the conditional entropy would be systematically zero . to avoid that , danon et al . adopted the _ normalized mutual information _ @xcite @xmath859 which is currently very often used in tests of graph clustering algorithms . the normalized mutual information equals @xmath2 if the partitions are identical , whereas it has an expected value of @xmath631 if the partitions are independent . the measure , defined for standard partitions , in which each vertex belongs to only one cluster , has been recently extended to the case of overlapping clusters by lancichinetti et al . the extension is not straightforward as the community assignments of a partition are now specified by a vectorial random variable , since each vertex may belong to more clusters simultaneously . in fact , the definition by lancichinetti et al . is not a proper extension of the normalized mutual information , in the sense that it does not recover exactly the same value of the original measure for the comparison of proper partitions without overlap , even though the values are close . meil @xcite introduced the _ variation of information _ @xmath860 which has some desirable properties with respect to the normalized mutual information and other measures . in particular , it defines a metric in the space of partitions as it has the properties of distance . it is also a local measure , i. e. the similarity of partitions differing only in a small portion of a graph depends on the differences of the clusters in that region , and not on the partition of the rest of the graph . the maximum value of the variation of information is @xmath241 , so similarity values for partitions of graphs with different size can not be compared with each other . for meaningful comparisons one could divide @xmath861 by @xmath241 , as suggested by karrer et al . @xcite . a concept related to similarity is that of _ distance _ , which indicates basically how many operations need to be performed in order to transform a partition to another . gustafsson et al . defined two distance measures for partitions @xcite . they are both based on the concept of _ meet _ of two partitions , which is defined as @xmath862 . \label{eqt1}\ ] ] the distance measures are @xmath863 and @xmath864 . in both cases they are determined by summing the distances of @xmath832 and @xmath833 from the meet @xmath865 . for @xmath863 the distance of @xmath832 ( @xmath833 ) from the meet is the minimum number of elements that must be moved between @xmath832 and @xmath833 so that @xmath832 ( @xmath833 ) and @xmath865 coincide @xcite . for @xmath864 the distance of @xmath832 ( @xmath833 ) from the meet is the minimum number of divisions that must be done in @xmath832 ( @xmath833 ) so that @xmath832 ( @xmath833 ) and @xmath865 coincide @xcite . such distance measures can easily be transformed in similarity measures , like @xmath866 identical partitions have zero mutual distance and similarity @xmath2 based on eqs . [ eqt2 ] . finally an important problem is how to define the similarity between clusters . if two partitions @xmath832 and @xmath833 of a graph are similar , each cluster of @xmath832 will be very similar to one cluster of @xmath833 , and viceversa , and it is important to identify the pairs of corresponding clusters . for instance , if one has information about the time evolution of a graph , one could monitor the dynamics of single clusters as well , by keeping track of each cluster at different time steps @xcite . given clusters @xmath825 and @xmath826 , their similarity can be defined through the _ relative overlap _ @xmath867 @xmath868 in this way , looking for the cluster of @xmath833 corresponding to @xmath825 means finding the cluster @xmath826 that maximizes @xmath867 . the index @xmath867 can be used to define similarity measures for partitions as well @xcite . an interesting discussion on the problem of comparing partitions , along with further definitions of similarity measures not discussed here , can be found in ref . @xcite . the first systematic comparative analysis of graph clustering techniques has been carried out by danon et al . they compared the performances of various algorithms on the benchmark graphs by girvan and newman ( section [ sec6_1 ] ) . the algorithms examined are listed in table [ table_orders ] , along with their complexity . [ cols="^,^,^,^",options="header " , ] fig . [ figdanon ] shows the performance of all algorithms . instead of showing the whole curves of the similarity versus @xmath784 ( section [ sec6_1 ] ) , which would display a fuzzy picture with many strongly overlapping curves , difficult to appreciate , danon et al . considered three values for @xmath784 ( @xmath227 , @xmath869 and @xmath228 ) , and represented the result for each algorithm as a group of three columns , indicating the average value of the similarity between the planted partition and the partition found by the method for each of the three @xmath784-values . the similarity was measured in terms of the fraction of correctly classified vertices ( section [ sec6_1 ] ) . the comparison shows that modularity optimization via simulated annealing ( section [ sub_sec6_0_2 ] ) yields the best results , although it is a rather slow procedure , that can not be applied to graphs of size of the order of @xmath612 vertices or larger . on the other hand , we have already pointed out that the benchmark by girvan and newman is not a good representation of real graphs with community structure , which are characterized by heterogeneous distributions of degree and community sizes . in this respect , the class of graphs designed by lancichinetti et al . ( lfr benchmark ) @xcite ( section [ sec6_1 ] ) poses a far more severe test to clustering techniques . for instance , many methods have problems to detect clusters of very different sizes ( like most methods listed in table [ table_orders ] ) . for this reason , lancichinetti and fortunato have carried out a careful comparative analysis of community detection methods on the much more restrictive lfr benchmark @xcite . the algorithms chosen are listed in table [ table2 ] . in fig . [ companalysis ] the performances of the algorithms on the lfr benchmark are compared . whenever possible , tests on the versions of the lfr benchmark with directed edges , weighted edges and/or overlapping communities @xcite were carried out . lancichinetti and fortunato also tested the methods on random graphs , to check whether they are able to notice the absence of community structure . from the results of all tests , the infomap method by rosvall and bergstrom @xcite appears to be the best , but also the algorithms by blondel et al . @xcite and by ronhovde and nussinov @xcite have a good performance . these three methods are also very fast , with a complexity which is essentially linear in the system size , so they can be applied to large systems . on the other hand , modularity - based methods ( with the exception of the method by blondel et al . ) have a rather poor performance , which worsens for larger systems and smaller communities , due to the well known resolution limit of modularity @xcite . the performance of the remaining methods worsens considerably if one increases the system size ( dm and infomod ) or the community size ( cfinder , mcl and method by radicchi et al . ) . performances of several algorithms on the lfr benchmark @xcite . the plots show the normalized mutual information ( in the version proposed in ref . @xcite ) as a function of the mixing parameter of the benchmark graphs . the different curves for each method refer to different system sizes ( @xmath870 and @xmath871 vertices ) and community size ranges [ ( s)=from @xmath5 to @xmath872 vertices , ( b)=from @xmath873 to @xmath874 vertices ] . for the gn algorithm only the smaller graph size was adopted , due to the high complexity of the method , whereas for the em method there are eight curves instead of four because for each set of benchmark graphs the algorithm was run starting from two different initial conditions . reprinted figure with permission from ref . 2009 by the american physical society.,title="fig:",width=302 ] performances of several algorithms on the lfr benchmark @xcite . the plots show the normalized mutual information ( in the version proposed in ref . @xcite ) as a function of the mixing parameter of the benchmark graphs . the different curves for each method refer to different system sizes ( @xmath870 and @xmath871 vertices ) and community size ranges [ ( s)=from @xmath5 to @xmath872 vertices , ( b)=from @xmath873 to @xmath874 vertices ] . for the gn algorithm only the smaller graph size was adopted , due to the high complexity of the method , whereas for the em method there are eight curves instead of four because for each set of benchmark graphs the algorithm was run starting from two different initial conditions . reprinted figure with permission from ref . 2009 by the american physical society.,title="fig:",width=302 ] performances of several algorithms on the lfr benchmark @xcite . the plots show the normalized mutual information ( in the version proposed in ref . @xcite ) as a function of the mixing parameter of the benchmark graphs . the different curves for each method refer to different system sizes ( @xmath870 and @xmath871 vertices ) and community size ranges [ ( s)=from @xmath5 to @xmath872 vertices , ( b)=from @xmath873 to @xmath874 vertices ] . for the gn algorithm only the smaller graph size was adopted , due to the high complexity of the method , whereas for the em method there are eight curves instead of four because for each set of benchmark graphs the algorithm was run starting from two different initial conditions . reprinted figure with permission from ref . 2009 by the american physical society.,title="fig:",width=302 ] fan et al . have evaluated the performance of some algorithms to detect communities on weighted graphs @xcite . the algorithms are : modularity maximization , carried out with extremal optimization ( weo ) ( section [ sub_sec6_0_3 ] ) ; the girvan - newman algorithm ( wgn ) ( section [ subsec5_1 ] ) ; the potts model algorithm by reichardt and bornholdt ( potts ) ( section [ sec44_1 ] ) . all these techniques have been originally introduced for unweighted graphs , but we have shown that they can easily be extended to weighted graphs . the algorithms were tested on the weighted version of the benchmark of girvan and newman , that we discussed in section [ sec6_1 ] . edge weights have only two values : @xmath875 for inter - cluster edges and @xmath876 for intra - cluster edges . such values are linked by the relation @xmath877 , so they are not independent . for testing one uses realizations of the benchmark with fixed topology ( i. e. fixed @xmath784 ) and variable weights . in fig . [ fantest ] the comparative performance of the three algorithms is illustrated . the topology of the benchmark graphs corresponds to @xmath878 , i. e. to graphs in which each vertex has approximately the same number of neighbors inside and outside its community . by varying @xmath875 from @xmath631 to @xmath323 one goes smoothly from a situation in which most of the weight is concentrated inside the clusters , to a situation in which instead the weight is concentrated between the clusters . from fig . [ fantest ] we see that weo and potts are more reliable methods . comparative evaluation of the performances of algorithms to find communities in weighted graphs . tests are carried out on a weighted version of the benchmark of girvan and newman . the two plots show how good the algorithms are in terms of the precision and accuracy with which they recover the planted partition of the benchmark . precision indicates how close the values of similarity between the planted and the model partition are after repeated experiments with the same set of parameters ; accuracy indicates how close the similarity values are to the ideal result ( 1 ) after repeated experiments with the same set of parameters . the similarity measure adopted here is based on the relative overlap of clusters of eq . [ eqt20 ] . we see that the maximization of modularity with extremal optimization ( weo ) and the potts model algorithm ( potts ) are both precise and accurate as long as the weight of the inter - cluster edges @xmath875 remains lower than the weight of the intra - cluster edges ( @xmath879 ) . reprinted figures with permission from ref . 2007 by elsevier.,title="fig : " ] comparative evaluation of the performances of algorithms to find communities in weighted graphs . tests are carried out on a weighted version of the benchmark of girvan and newman . the two plots show how good the algorithms are in terms of the precision and accuracy with which they recover the planted partition of the benchmark . precision indicates how close the values of similarity between the planted and the model partition are after repeated experiments with the same set of parameters ; accuracy indicates how close the similarity values are to the ideal result ( 1 ) after repeated experiments with the same set of parameters . the similarity measure adopted here is based on the relative overlap of clusters of eq . [ eqt20 ] . we see that the maximization of modularity with extremal optimization ( weo ) and the potts model algorithm ( potts ) are both precise and accurate as long as the weight of the inter - cluster edges @xmath875 remains lower than the weight of the intra - cluster edges ( @xmath879 ) . reprinted figures with permission from ref . 2007 by elsevier.,title="fig : " ] sawardecker et al . have tested methods to detect overlapping communities @xcite . they considered three algorithms : modularity optimization , the clique percolation method ( cpm ) ( section [ sec45_1 ] ) and the modularity landscape surveying method by sales - pardo et al . @xcite ( section [ sub_sec6_202 ] ) . for testing , sawardecker et al . defined a class of benchmark graphs in which the linking probability between vertices is an increasing function of the number of clusters the vertices belong to . we have described this benchmark in section [ sec6_1 ] . it turns out that the modularity landscape surveying method is able to identify overlaps between communities , as long as the fraction of overlapping vertices is small . curiously , the cpm , designed to find overlapping communities , has a poor performance , as the overlapping vertices found by the algorithm are in general different from the overlapping vertices of the planted partition of the benchmark . the authors also remark that , if the overlap between two clusters is not too small , it may be hard ( for any method ) to recognize whether the clusters are overlapping or hierarchically organized , i. e. loosely connected clusters within a large cluster . we close the section with some general remarks concerning testing . we have seen that a testing procedure requires two crucial ingredients : benchmark graphs with built - in community structure and clustering algorithms that try to recover it . such two elements are not independent , however , as they are both based on the concept of community . if the underlying notions of community for the benchmark and the algorithm are very different , one can hardly expect that the algorithm will do a good job on the benchmark . furthermore , there is a third element , i. e. the quality of a partition . all benchmarks start from a situation in which communities are clearly identified , i. e. connected components of the graph , and introduce some amount of noise , that eventually leads to a scenario where clusters are hardly or no longer detectable . it is then important to keep track of how the quality of the natural partition of the benchmark worsens as the amount of noise increases , in order to distinguish configurations in which the graphs have a cluster structure , that an algorithm should then be able to resolve , from configurations in which the noise prevails and the natural clusters are not meaningful . moreover , quality functions are important to evaluate the performance of an algorithm on graphs whose community structure is unknown . quality functions are strongly related to the concept of community as well , as they are supposed to evaluate the goodness of the clusters , so they require a clear quantitative concept of what a cluster is . it is very important for any testing framework to check for the mutual dependencies between the benchmark , the quality function used to evaluate partitions , and the clustering algorithm to be tested . this issue has so far received very little attention @xcite . finally , empirical tests are also very important , as one ultimately wishes to apply clustering techniques to real graphs . therefore , it is crucial to collect more data sets of graphs whose community structure is known or deducible from information on the vertices and their edges . what are the general properties of partitions and clusters of real graphs ? in many papers on graph clustering applications to real systems are presented . in spite of the variety of clustering methods that one could employ , in many cases partitions derived from different techniques are rather similar to each other , so the general properties of clusters do not depend much on the particular algorithm used . the analysis of clusters and their properties delivers a _ mesoscopic description _ of the graph , where the communities , and not the vertices , are the elementary units of the topology . the term mesoscopic is used because the relevant scale here lies between the scale of the vertices and that of the full graph . one of the first issues addressed was whether the communities of a graph are usually about of the same size or whether the community sizes have some special distribution . most clustering techniques consistently find skewed distributions of community sizes , with a tail described with good approximation by a power law ( at least , a sizeable portion of the curve ) with exponents in the range between @xmath2 and @xmath3 @xcite . so , there seems to be no characteristic size for a community : small communities usually coexist with large ones . cumulative distribution of community sizes for the amazon purchasing network . the partition is derived by greedy modularity optimization . reprinted figure with permission from ref . 2004 by the american physical society.,width=302 ] as an example , fig . [ figure13 ] shows the cumulative distribution of community sizes for a recommendation network of the online vendor amazon.com . vertices are products and there is a connection between item @xmath9 and @xmath10 if @xmath10 was frequently purchased by buyers of @xmath9 . recall that the cumulative distribution is the integral of the probability distribution : if the cumulative distribution is a power law @xmath880 , the probability distribution is also a power law with exponent @xmath881 . leskovec et al . @xcite have gone one step further . they carried out a systematic analysis of communities in large real networks , including traditional and on - line social networks , technological , information networks and web graphs . the main goal was to assess the quality of communities at various sizes . as a quality function the _ conductance _ of the cluster was chosen . we remind that the conductance of a cluster is the ratio between the cut size of the cluster and the minimum between the total degree of the cluster and that of the rest of the graph ( section [ sec4_1 ] ) . so , if the cluster is much smaller than the whole graph , the conductance equals the ratio between the cut size and the total degree of the cluster . since a `` good '' cluster is characterized by a low cut size and a large internal density of edges , low values of the conductance indicate good clusters . for each real network leskovec derived the _ network community profile plot _ ( ncpp ) , showing the minimum conductance score among subgraphs of a given size as a function of the size . interestingly , they found that the ncpps of all networks they studied have a characteristic shape : they go downwards up until subgraphs with about @xmath874 vertices , and then they rise monotonically for larger subgraphs ( fig . [ figlesk ] ) . this seems to suggest that communities are well defined only when they are fairly small in size . analysis of communities in large real networks by leskovec et al . ( left ) typical shape of the network community profile plot ( ncpp ) , showing how the minimum conductance of subgraphs of size @xmath15 varies with @xmath15 . the plot indicates that the `` best '' communities have a size of about @xmath874 vertices ( minimum of the curve ) , whereas communities of larger sizes are not well - defined . in the plot two other ncpps are shown : the one labeled _ rewired network _ corresponds to a randomized version of the network , where edges are randomly rewired by keeping the degree distribution ; the one labeled _ bag of whiskers _ gives the minimum conductance scores of clusters composed of disconnected pieces . ( right ) scheme of the core - periphery structure of large social and information networks derived by leskovec et al . based on the results of their empirical analysis . most of the vertices are in a central core , which does not have a clear community structure , whereas the best communities , which are rather small , are weakly connected to the core . reprinted figure with permission from ref . @xcite.,title="fig : " ] 0.5 cm analysis of communities in large real networks by leskovec et al . ( left ) typical shape of the network community profile plot ( ncpp ) , showing how the minimum conductance of subgraphs of size @xmath15 varies with @xmath15 . the plot indicates that the `` best '' communities have a size of about @xmath874 vertices ( minimum of the curve ) , whereas communities of larger sizes are not well - defined . in the plot two other ncpps are shown : the one labeled _ rewired network _ corresponds to a randomized version of the network , where edges are randomly rewired by keeping the degree distribution ; the one labeled _ bag of whiskers _ gives the minimum conductance scores of clusters composed of disconnected pieces . ( right ) scheme of the core - periphery structure of large social and information networks derived by leskovec et al . based on the results of their empirical analysis . most of the vertices are in a central core , which does not have a clear community structure , whereas the best communities , which are rather small , are weakly connected to the core . reprinted figure with permission from ref . @xcite.,title="fig : " ] such small clusters are weakly connected to the rest of the graph , often by a single edge ( in this case they are called _ whiskers _ ) , and form the _ periphery _ of the network . the other vertices form a big _ core _ , in which larger clusters are well connected to each other , and are therefore barely distinguishable ( fig . [ figlesk ] ) . leskovec et al . performed low - conductance cuts with several methods , to ensure that the result is not a simple artefact of a particular chosen technique . moreover , they have also verified that , for large real networks with known community structure ( such as , e.g. , the social network of the on - line blogging site _ livejournal _ , with its user groups ) , the ncpp has the same qualitative shape if one takes the real communities instead of low - conductance subgraphs . the analysis by leskovec et al . may shed new light on our understanding of community structure and its detection in large networks . the fact that the `` best '' communities appear to have a characteristic size of about @xmath874 vertices is consistent with dunbar conjecture that @xmath882 is the upper size limit for a working human community @xcite . on the other hand , if large communities are very mixed with each other , as leskovec et al . claim , they could hardly be considered communities , and the alleged `` community structure '' of large networks would be limited to their peripheral region . the results by leskovec et al . may be affected by the properties of conductance , and need to be validated with alternative approaches . in any case , whatever the value of the quality score of a cluster may be ( low or high ) , it is necessary to estimate the significance of the cluster ( section [ sec6_4 ] ) , before deciding whether it is a meaningful structure or not . if the community structure of a graph is known , it is possible to classify vertices according to their roles within their community , which may allow to infer individual properties of the vertices . a promising classification has been proposed by guimer and amaral @xcite . the role of a vertex depends on the values of two indices , the _ within - module degree _ and the _ participation ratio _ ( though other variables may be chosen , in principle ) . the within - module degree @xmath883 of vertex @xmath67 is defined as @xmath884 where @xmath885 is the internal degree of @xmath67 in its cluster @xmath314 , @xmath886 and @xmath887 the average and standard deviation of the internal degrees for all vertices of cluster @xmath314 . the within - module degree is then defined as the @xmath406-score of the internal degree @xmath885 . large values of @xmath406 indicate that the vertex has many more neighbors within its community than most other vertices of the community . vertices with @xmath888 are classified as _ hubs _ , if @xmath889 they are _ non - hubs_. the participation ratio @xmath890 of vertex @xmath67 is defined as @xmath891 here @xmath892 is the internal degree of @xmath67 in cluster @xmath155 , @xmath105 the degree of @xmath67 . values of @xmath96 close to @xmath2 indicate that the neighbors of the vertex are uniformly distributed among all clusters ; if all neighbors are within the cluster of the vertex , instead , @xmath893 . based on the values of the pair @xmath894 , guimer and amaral distinguished seven roles for the vertices . non - hub vertices can be _ ultra - peripheral _ ( @xmath895 ) , _ peripheral _ ( @xmath896 ) , _ connectors _ ( @xmath897 ) and _ kinless vertices _ ( @xmath898 ) . hub vertices are classified in _ provincial hubs _ ( @xmath899 ) , _ connector hubs _ ( @xmath900 ) and _ kinless hubs _ ( @xmath901 ) . the regions of the @xmath902 plane corresponding to the seven roles are highlighted in fig . [ figroles ] . we stress that the actual boundaries of the regions can be chosen rather arbitrarily . regions of the @xmath902 plane defining the roles of vertices in the modular structure of a graph , according to the scheme of guimer and amaral @xcite . reprinted figure with permission from ref . 2005 by the nature publishing group . ] on graphs without community structure , like erds - rnyi @xcite random graphs and barabsi - albert @xcite graphs ( section [ sec1_3 ] ) , non - hubs are mostly kinless vertices . in addition , if there are hubs , like in barabsi - albert graphs , they are kinless hubs . kinless hubs ( non - hubs ) vertices have less than half ( one third ) of their neighbors inside any cluster , so they are not clearly associated to a cluster . on real graphs , the topological roles can be correlated to functions of vertices : in metabolic networks , for instance , connector hubs , which share most edges with vertices of other clusters than their own , are often metabolites which are more conserved across species than other metabolites , i. e. they have an evolutionary advantage @xcite . if communities are overlapping , one can explore other statistical properties , like the distributions of the overlaps and of the vertex memberships . the overlap is defined as the number of vertices shared by each pair of overlapping clusters ; the membership of a vertex is the number of communities including the vertex . both distributions turn out to be skewed , so there seem to be no characteristic values for the overlap and the membership . moreover , one could derive a network , where the communities are the vertices and pairs of vertices are connected if their corresponding communities overlap @xcite . such networks seem to have some special properties . for instance , the degree distribution is a particular function , with an initial exponential decay followed by a slower power law decay ) and a coauthorship network of physicists . there is no _ a priori _ reason to believe that this result is general . ] . we stress that the above results have been obtained with the clique percolation method by palla et al . ( section [ sec45_1 ] ) and it is not clear whether other techniques would confirm them or not . in a recent analysis it has been shown that the degree distribution of the network of communities can be reproduced by assuming that the graph grows according to a simple preferential attachment mechanism , where communities with large degree have an enhanced chance to interact / overlap with new communities @xcite . the ultimate goal of clustering algorithms is trying to infer properties of and relationships between vertices , that are not available from direct observation / measurement . if the scientific community agrees on a set of reliable techniques , one could then proceed with careful investigations of systems in various domains . so far , most works in the literature on graph clustering focused on the development of new algorithms , and applications were limited to those few benchmark graphs that one typically uses for testing ( section [ sec6_1 ] ) . still , there are also applications aiming at understanding real systems . some results have been actually mentioned in the previous sections . this section is supposed to give a flavor of what can be done by using clustering algorithms . therefore , the list of works presented here is by no means exhaustive . most studies focus on biological and social networks . we mention a few applications to other types of networks as well . the recent abundance of genomic data has allowed us to explore the cell at an unprecedented depth . a wealth of information is available on interactions involving proteins and genes , metabolic processes , etc . in order to study cellular systems , the graph representation is regularly used . protein - protein interaction networks ( pin ) , gene regulatory networks ( grn ) and metabolic networks ( mn ) are meanwhile standard objects of investigation in biology and bioinformatics @xcite . biological networks are characterized by a remarkable modular organization , reflecting functional associations between their components . for instance , proteins tend to be associated in two types of cellular modules : _ protein complexes _ and _ functional modules_. a protein complex is a group of proteins that mutually interact at the same time and space , forming a sort of physical object . examples are transcription factor complexes , protein transport and export complexes , etc . functional modules instead are groups of proteins taking place in the same cellular process , even if the interactions may happen at different times and places . examples are the cdk / cyclin module , responsible for cell - cycle progression , the yeast pheromone response pathway , etc .. identifying cellular modules is fundamental to uncover the organization and dynamics of cell functions . however , the information on cell units ( e. g. proteins , genes ) and their interactions is often incomplete , or even incorrect , due to noise in the data produced by the experiments . therefore , inferring modules from the topology of cellular networks enables one to restrict the set of possible scenarios and can be a safe guide for future experiments . rives and galitski @xcite studied the modular organization of a subset of the pin of the yeast ( _ saccharomyces cerevisiae _ ) , consisting of the ( signaling ) proteins involved in the processes leading the microorganism to a filamentous form . the clusters were detected with a hierarchical clustering technique . proteins mostly interacting with members of their own cluster are often essential proteins ; edges between modules are important points of communication . spirin and mirny @xcite identified protein complexes and functional modules in yeast with different techniques : clique detection , superparamagnetic clustering @xcite and optimization of cluster edge density . they estimated the statistical significance of the clusters by computing the @xmath45-values of seeing those clusters in random graphs with the same expected degree sequence as the original network . from the known functional annotations of yeast genes one can see that the modules usually group proteins with the same or consistent biological functions . indeed , in many cases , the modules exactly coincide with known protein complexes . the results appear robust if noise is introduced in the system , to simulate the noise present in the experimental data . functional modules in yeast were also found by chen and yuan @xcite , who applied the algorithm by girvan and newman with a modified definition of edge betweenness ( section [ subsec5_1 ] ) . the standard girvan - newman algorithm has proved to be reliable to detect functional modules in pins @xcite . the novelty of the work by chen and yuan is its focus on weighted pins , where the weights come from information derived through microarray expression profiles . weights add information about the system and should lead to a more reliable modular structure . by knocking out genes in the same structural cluster similar phenotypes appeared , suggesting that the genes have similar biological roles . moreover , the clusters often contained known protein complexes , either entirely or to a large extent . finally , chen and yuan were able to make predictions of the unknown function of some genes , based on the structural module they belong to : gene function prediction is the most promising outcome deriving from the application of clustering techniques to pins . farutin et al . @xcite have adopted a local concept of community , and derived a hierarchical decomposition of pins , in that the modules identified at some level become the vertices of a network at the higher level . communities are overlapping , to account for the fact that proteins ( and whole modules ) may have diverse biological functions . high level structures detected in a human pin correspond to general biological concepts like signal transduction , regulation of gene expression , intercellular communication . sen et al . @xcite identified protein clusters for yeast from the eigenvectors of the laplacian matrix ( section [ sec1_2 ] ) , computed via singular value decomposition . in a recent analysis , lewis et al . @xcite carefully explored the relationship between structural communities of pins and their biological function . communities were detected with the multiresolution approach by reichardt and bornholt @xcite ( section [ sub_sec6_01 ] ) . a community is considered biologically homogeneous if the functional similarity between protein pairs of the community ( extracted through the gene ontology database @xcite ) is larger than the functional similarity between all protein pairs of the network . lewis et al . also specified the comparison to interacting and non - interacting protein pairs . as a result , many communities turn out to be biologically homogeneous , especially if they are not too small . moreover , some topological attributes of communities , like the within - community clustering coefficient ( i.e. the average value of the clustering coefficients of the vertices of a community , computed by considering just the neighbors belonging to the community ) and link density ( density of internal edges ) , are good indicators of biological homogeneity : the former is strongly correlated with biological homogeneity , independently of the community size , whereas for the latter the correlation is strong for large communities . metabolic networks have also been extensively investigated . we have already discussed the `` functional cartography '' designed by guimer and amaral @xcite , which applies to general types of networks , not necessarily metabolic . a hierarchical decomposition of metabolic networks has been derived by holme et al . @xcite , by using a hierarchical clustering technique inspired by the algorithm by girvan and newman ( section [ subsec5_1 ] ) . here , vertices are removed based on their betweenness values , which are obtained by dividing the standard site betweenness scores @xcite by the indegree of the respective vertices . a picture of metabolic network emerges , in which there are core clusters centered at hub - substances , surrounded by outer shells of less connected substances , and a few other clusters at intermediate scales . in general , clusters at different scales seem to be meaningful , so the whole hierarchy should be taken into account . wilkinson and huberman @xcite analyzed a network of gene co - occurrence to find groups of related genes . the network is built by connecting pairs of genes that are mentioned together in the abstract of articles of the medline database ( http://medline.cos.com/ ) . clusters were found with a modified version of the algorithm by girvan and newman , in which edge betweenness is computed by considering the shortest paths of a small subset of all vertex pairs , to gain computer time ( section [ subsec5_1 ] ) . as a result , genes belonging to the same cluster turn out to be functionally related to each other . co - occurrence of terms is also used to extract associations between genes and diseases , to find out which genes are relevant for a specific disease . communities of genes related to colon cancer can be helpful to identify the function of the genes . networks depicting social interactions between people have been studied for decades @xcite . recently the modern information and communication technology ( ict ) has opened new interaction modes between individuals , like mobile phone communications and online interactions enabled by the internet . such new social exchanges can be accurately monitored for very large systems , including millions of individuals , whose study represents a huge opportunity for social science . communities of social networks can be friendship circles , groups of people sharing common interests and/or activities , etc .. blondel et al . have analyzed a network of mobile phone communications between users of a belgian phone operator @xcite . the vertices of the graph are 2.6 millions and the edges are weighted by the cumulative duration of phone calls between users in the observation time frame . the clustering analysis , performed with a fast hierarchical modularity optimization technique developed by the authors ( discussed in section [ sub_sec6_0_1 ] ) , delivers six hierarchical levels . the highest level consists of @xmath903 groups with more than @xmath874 vertices , which are clearly arranged in two main groups , linguistically homogeneous , reflecting the linguistic split of belgian population ( fig . [ blondphone ] ) . community structure of a social network of mobile phone communication in belgium . dots indicate subcommunities at the lower hierarchical level ( with more than @xmath874 people ) and are colored in a red - green scale to represent the level of representation of the two main languages spoken in belgium ( red for french and green for dutch ) . communities of the two larger groups are linguistically homogeneous , with more than @xmath904 of people speaking the same language . only one community ( zoomed ) , which lies at the border between the two main aggregations , has a more balanced distribution of languages . reprinted figure with permission from ref . 2008 by iop publishing and sissa.,width=302 ] tyler et al . @xcite studied a network of e - mail exchanges between people working at hp labs . they applied the same modified version of girvan - newman algorithm that two of the authors have used to find communities of related genes @xcite ( section [ sec7_2_1 ] ) . the method enables one to measure the degree of membership of each vertex in a community and allows for overlaps between communities . the detected clusters matched quite closely the organization of the labs in departments and project groups , as confirmed by interviews conducted with researchers . communities in social networking sites . ( top ) visualization of a network of friendships between students at caltech , constructed from facebook data ( september 2005 ) . the colors / shapes indicate the dormitories ( houses ) of the students . ( bottom ) topological communities of the network , which are quite homogeneous with respect to house affiliation . reprinted figures with permission from refs . @xcite and @xcite.,title="fig:",width=264 ] communities in social networking sites . ( top ) visualization of a network of friendships between students at caltech , constructed from facebook data ( september 2005 ) . the colors / shapes indicate the dormitories ( houses ) of the students . ( bottom ) topological communities of the network , which are quite homogeneous with respect to house affiliation . reprinted figures with permission from refs . @xcite and @xcite.,title="fig:",width=264 ] social networking sites , like _ myspace _ ( www.myspace.com ) , _ friendster _ ( www.friendster.com ) , _ facebook _ ( www.facebook.com ) , etc . have become extremely popular in the last years . they are online platforms that allow people to communicate with friends , send e - mails , solicit opinions on specific issues , spread ideas and/or fads , etc . traud et al . @xcite used anonymous facebook data to create networks of friendships between students of different american universities , where vertices / students are connected if they are friends on facebook . communities were detected by applying a variant of newman s spectral optimization of modularity ( section [ sub_sec6_0_4 ] ) : the results were further refined through additional steps la kernighan - lin ( section [ sec4_1 ] ) . one of the goals of the study was to infer relationships between the online and offline lives of the students . by using demographic information on the students populations , one finds that communities are organized by class year or by house ( dormitory ) affiliation , depending on the university ( fig . [ figcaltech ] ) . yuta et al . @xcite observed a gap in the community size distribution of a friendship network extracted from _ mixi _ ( mixi.jp ) , the largest social networking site in japan ( as of december 2006 ) . communities were identified with the fast greedy modularity optimization by clauset et al . the gap occurs in the intermediate range of sizes between @xmath873 and @xmath905 , where but a few communities are observed . yuta et al . introduced a model where people form new friendships both by `` closing '' ties with people who are friends of friends , and by setting new links with individuals having similar interests . in this way most groups turn out to be either small or large , and medium size groups are rare . collaboration networks , in which individuals are linked if they are ( were ) involved in a common activity , have been often studied because they embed an implicit objective concept of acquaintance , that is not easy to capture in direct social experiments / interviews . for instance , somebody may consider another individual a friend , while the latter may disagree . a collaboration instead is a proof of a social relationship between individuals . the analysis of the structure of scientific collaboration networks @xcite has exerted a big influence on the development of the modern network science . scientific collaboration is associated to coauthorship : two scientists are linked if they have coauthored at least one paper together . information about coauthorships can be extracted from different databases of research papers . communities indicate groups of people with common research interests , i. e. topical or disciplinary groups . in the seminal paper by girvan and newman @xcite , the authors applied their method on a collaboration network of scientists working at the santa fe institute , and were able to discriminate between research divisions ( fig . [ fig3]b ) . the community structure of scientific collaboration networks has been investigated by many authors @xcite . other types of collaboration networks have been studied too . gleiser and danon @xcite considered a collaboration network of jazz musicians . vertices are either musicians , connected if they played in the same band , or bands , connected if they have a musician in common . by applying the algorithm of girvan and newman they found that communities reflect both racial segregation ( with two main groups comprising only black or white players ) and geographical separation , due to the different recording locations . citation networks @xcite have been regularly used to understand the citation patterns of authors and to disclose relationships between disciplines . rosvall and bergstrom @xcite used a citation network of over @xmath906 scientific journals to derive a map of science . they used a clustering technique based on compressing the information on random walks taking place on the graph ( section [ subsec_statinf2 ] ) . a random walk follows the flow of citations from one field to another , and the fields emerge naturally from the clustering analysis ( fig . [ maprosvall ] ) . the structure of science resembles the letter @xmath197 , with the social sciences and engineering at the terminals , joined through a chain including medicine , molecular biology , chemistry and physics . reichardt and bornholdt @xcite performed a clustering analysis on a network built from bidding data taken from the german version of ebay ( www.ebay.de ) , the most popular online auction site . the vertices are bidders and two vertices are connected if the corresponding bidders have expressed interest for the same item . clusters were detected with the multiresolution modularity optimization developed by the authors themselves @xcite ( section [ sub_sec6_01 ] ) . in spite of the variety of items that it is possible to purchase through ebay , about @xmath904 of bidders were classified into a few major clusters , reflecting bidders broad categories of interests . ebay data were also examined by jin et al . @xcite , who considered bidding networks where the vertices are the individual auctions and edges are placed between auctions having at least one common bidder . communities , detected with greedy modularity optimization @xcite ( section [ sub_sec6_0_1 ] ) , allow to identify substitute goods , i. e. products that have value for the same bidder , so that they can be purchased together or alternatively . legislative networks enable one to deduce associations between politicians through their parliamentary activity , which may be related or not to party affiliation . porter and coworkers have carried out numerous studies on the subject @xcite , by using data on the congress of the united states . in refs . @xcite , they examined the community structure of networks of committees in the us house of representatives . committees sharing common members are connected by edges , which are weighted by dividing the number of common members by the number one would expect to have if committee memberships were randomly assigned . hierarchical clustering ( section [ sec4_2 ] ) reveals close connections between some of the committees . in another work @xcite , zhang et al . analyzed networks of legislation cosponsorship , in which vertices are legislators and two legislators are linked if they support at least one common bill . communities , identified with a modification of newman s spectral optimization of modularity ( section [ sub_sec6_0_4 ] ) , are correlated with party affiliation , but also with geography and committee memberships of the legislators . networks of correlations between time series of stock returns have received a growing attention in the past few years @xcite . in early studies , scholars found clusters of correlated stocks by computing the _ maximum spanning tree _ of the network @xcite ( section [ sec1_1 ] ) , and realized that such clusters match quite well the economic sectors of the stocks . more recently , the community structure of the networks has been investigated by means of proper clustering algorithms . farkas et al . @xcite have applied the weighted version of the clique percolation method ( section [ sec45_1 ] ) and found that the presence of two strong ( i. e. carrying high correlation ) edges in triangles is usually accompanied by the presence of a strong third edge . heimo et al . @xcite used the weighted version of the multiresolution method by reichardt and bornholdt @xcite ( section [ sub_sec6_01 ] ) . clusters correspond to relevant business sectors , as indicated by forbes classification ; moreover , smaller clusters at lower hierarchical levels seem to correspond to ( economically ) meaningful substructures of the main clusters . despite the remote origins and the great popularity of the last years , research on graph clustering has not yet given a satisfactory solution of the problem and leaves us with a number of important open issues . from our exposition it appears that the field has grown in a rather chaotic way , without a precise direction or guidelines . in some cases , interesting new ideas and tools have been presented , in others existing methods have been improved , becoming more accurate and/or faster . what the field lacks the most is a theoretical framework that defines precisely what clustering algorithms are supposed to do . everybody has his / her own idea of what a community is , and most ideas are consistent with each other , but , as long as there is still disagreement , it remains impossible to decide which algorithm does the best job and there will be no control on the creation of new methods . therefore , we believe that the first and foremost task that the scientific community working on graph clustering has to solve in the future is defining a set of reliable benchmark graphs , against which algorithms should be tested ( section [ sec6_1 ] ) . defining a benchmark goes far beyond the issue of testing . it means designing practical examples of graphs with communities , and , in order to do that , one has to agree on the fundamental concepts of community and partition . clustering algorithms have to be devised consistently with such definitions , in order to give the best performance on the set of designated benchmarks , which represent a sort of ground truth . the explosion in the number of algorithms we have witnessed in recent times is due precisely to the present lack of reliable mechanisms of control of their quality and comparison of their performances . if the community agrees on a benchmark , the future development of the field will be more coherent and the progress brought by new methods can be evaluated in an unbiased manner . the planted @xmath293-partition model @xcite is the easiest recipe one can think of when it comes to defining clusters , and is the criterion underlying well - known benchmarks , like that by girvan and newman . we believe that the new benchmarks have to be defined along the same lines . the benchmark graphs recently introduced by lancichinetti et al . @xcite and by sawardecker et al . @xcite are an important step in this direction . defining a benchmark implies specifying the `` natural '' partition of a graph , the one that any algorithm should find . this issue in turn involves the concept of quality of a partition , that has characterized large part of the development of the field , in particular after the introduction of newman - girvan modularity ( section [ sec3_2_2 ] ) . estimating the quality of a partition allows to discriminate among the large number of partitions of a graph . in some cases this is not difficult . for instance , in the benchmark by girvan and newman there is a single meaningful partition , and it is hard to argue with that . but most graphs of the real world have a hierarchical structure , with communities including smaller communities and so on . hence there are several meaningful partitions , corresponding to different hierarchical levels , and discriminating among them is hard , as they may be all relevant , in a sense . if we consider the human body , we can not say that the organization in tissues of the cells is more important than the organization in organs . we have seen that there are recent methods dealing with the problem of finding meaningful hierarchical levels ( section [ sub_sec6_20 ] ) . such methods rank the hierarchical partitions based on some criterion and one can assess their relevance through the ranking . one may wonder whether it makes sense sorting out levels , which means introducing a kind of threshold on the quality index chosen to rank partitions ( to distinguish `` good '' from `` bad '' partitions ) , or whether it is more appropriate to keep the information given by the whole set of hierarchical partitions . the work by clauset et al . on hierarchical random graphs @xcite , discussed in section [ sub_sec6_202 ] , indirectly raises this issue . there it was shown that the ensemble of model graphs , represented by dendrograms , encodes most of the information on the structure of the graph at study , like its degree distribution , transitivity and distribution of shortest path lengths . at the same time , by construction , the model reveals the whole hierarchy of communities , without any distinction between good and bad partitions . the information given by a dendrogram may become redundant and confusing when the graph is large , as then there is a big number of partitions this is actually the reason why quality functions were originally introduced . however , in that case , one was dealing with artificial hierarchies , produced by techniques that systematically yield a dendrogram as a result of the analysis ( like , e. g. , hierarchical clustering ) , regardless of whether the graph actually has a hierarchical structure or not . here instead we speak of real hierarchy , which is a fundamental element of real graphs and , as such , it must be considered in any serious approach to graph clustering . any good clustering method must be able to tell whether a graph has community structure or not , and , in the first case , whether the community structure is hierarchical ( i. e. with two or more levels ) or flat ( one level ) . we expect that the concept of hierarchy will become a key ingredient of future clustering techniques . in particular , assessing the consistence of the concepts of partitions quality and hierarchical structure is a major challenge . a precise definition of null models , i. e. of graphs without community structure , is also missing . this aspect is extremely important , though , as defining communities also implies deciding whether or not they exist in a specific graph . at the moment , it is generally accepted that random graphs have no communities . the null model of modularity ( section [ sec3_2_2 ] ) , by far the most popular , comprises all graphs with the same expected degree sequence of the original graph and random rewiring of edges . this class of graphs is characterized , by construction , by the fact that any vertex can be linked to any other , as long as the constraint on the degree sequence is satisfied . but this is by no means the only possibility . a community can be generically defined as a subgraph whose vertices have a higher probability to be connected to the other vertices of the subgraph than to external vertices . the planted @xmath293-partition model @xcite is based on this principle , as we have seen . however , this does not mean that the linking probabilities of a vertex with respect to the other vertices in its community or in different communities be constant ( or simply proportional to their degrees , as in the configuration model @xcite ) . in fact , in large networks it is reasonable to assume that the probability that a vertex is linked to most vertices is zero , as the vertex `` ignores '' their existence . this does not exclude that the probability that the vertex gets connected to the `` known '' vertices is the same ( or proportional to their degrees ) , in which case the graph would still be random and have no communities . we believe that we are still far from a precise definition and a complete classification of null models . this represents an important research line for the future of the field , for three main reasons : 1 ) to better disentangle `` true '' communities from byproducts of random fluctuations ; 2 ) to pose a stringent test to existing and future clustering algorithms , whose reliability would be questionable if they found `` false positives '' in null model graphs ; 3 ) to handle `` hybrid '' scenarios , where a graph displays community structure on some portions of it , while the rest is essentially random and has no communities . in the previous chapters we have seen a great number of clustering techniques . which one(s ) shall we use ? at the moment the scientific community is unable to tell . modularity optimization is probably the most popular method , but the results of the analysis of large graphs are likely to be unreliable ( section [ sub_sec6_1 ] ) . nevertheless , people have become accustomed to use it , and there have been several attempts to improve the measure . a newcomer , who wishes to find clusters in a given network and is not familiar with clustering techniques , would not know , off - hand , which method to use , and he / she would hardly find indications about good methods in any single paper on graph clustering , except perhaps on the method presented in the paper . so , people keep using algorithms because they have heard of them , or because they know that other people are using them , or because of the reputation of the scientists who designed them . waiting for future reliable benchmarks , that may give an objective assessment of the quality of the algorithms , there are at the moment hardly solid reasons to prefer an algorithm to another : the comparative analyses by danon et al . @xcite and by lancichinetti and fortunato @xcite ( section [ sec6_3 ] ) represent a first serious assessment of this issue . however , we want to stress that there is no such thing as the perfect method , so it is pointless to look for it . among the other things , if one tries to look for a very general method , that should give good results on any type of graphs , one is inevitably forced to make very general assumptions on the structure of the graph and on the properties of communities . in this way one neglects a lot of specific features of the system , that may lead to a more accurate detection of the clusters . informing a method with features characterizing some types of graphs makes it far more reliable to detect the community structure of those graphs than a general method , even if its applicability may be limited . therefore in the future we envision the development of domain - specific clustering techniques . the challenge here is to identify the peculiar features of classes of graphs , which are bound to become crucial ingredients in the design of suitable algorithms . some of the methods available today are actually based on assumptions that hold only for some specific categories of graphs . the clique percolation method by palla et al . @xcite , for instance , may work well for graphs characterized by a large number of cliques , like certain social networks , whereas it may give poor results otherwise . moving one step further , one should learn how to use specific information about a graph , whenever available , e. g. properties of vertices and/or partial information about their classification . for instance , it may be that one has some information on a subset of vertices , like demographic data on people of a social network , and such data may highlight relationships between people that are not obvious from the network of social interactions . in this case , using only the social network may be reductive and ideally one should exploit both the structural and the non - structural information in the search of clusters , as the latter should be consistent with both inputs . how to do this is an open problem . the scientific community has just begun to study this aspect @xcite . most algorithms in the literature deal with the `` classical '' case of a graph with undirected and unweighted edges . this is certainly the simplest case one could think of , and graph clustering is already a complex task on such types of graphs . we know that real networks may be directed , have weighted connections , be bipartite . methods to deal with such systems have been developed , as we have seen , especially in the most recent literature , but they are mostly preliminary attempts and there is room for improvement . another situation that may occur in real systems is the presence of edges with positive and negative weights , indicating attractive and repulsive interactions , respectively . this is the case , for instance , of correlation data @xcite . in this case , ideal partitions would have positively weighted intracluster edges and negatively weighted intercluster edges . we have discussed some studies in this direction @xcite , but we are just at the beginning of this endeavour . instead , there are no algorithms yet which are capable to deal with graphs in which there are edges of several types , indicating different kinds of interactions between the vertices ( _ multigraphs _ ) . agents of social networks , for instance , may be joined by working relationships , friendship , family ties , etc . at the moment there are essentially two ways of proceeding in these instances : 1 ) keeping edges of one type and forgetting the others , repeating the analysis for each type of edges and eventually comparing the results obtained ; 2 ) analyzing a single ( weighted ) graph , obtained by `` combining '' the contributions of the different types of edges in some way . finally , since most real networks are built through the results of experiments , which carry errors in their estimates , it would be useful to consider as well the case in which edges have not only associated weights , but also errors on their values . since the paper by palla et al . @xcite , overlapping communities have received a lot of attention ( section [ sec45 ] ) . however , there is still no consensus about a quantitative definition of the concept of overlapping community , and most definitions depend on the method adopted . intuitively , one would expect that clusters share vertices lying at their borders , and this idea has inspired most algorithms . however , clusters detected with the clique percolation method ( section [ sec45_1 ] ) often share central vertices of the clusters , which makes sense in specific instances , especially in social networks . so , it is still unclear how to characterize overlapping vertices . moreover , the concept of overlapping clusters seems at odds with that of hierarchical structure . no dendrogram can be drawn if there are overlapping vertices , at least in the standard way . due to the relevance of both features in real networks , it is necessary to adapt them to each other in a consistent way . overlapping vertices pose problems as well when it comes to comparing the results of different methods on the same graph . most similarity measures are defined only in the case of partitions , where each vertex is assigned to a single cluster ( section [ sec6_2 ] ) . it is then necessary to extend such definitions to the case of overlapping communities , whenever possible . another issue that is getting increasingly more popular is the study of graphs evolving in time . this is now possible due to the availability of timestamped network data sets . tracking the evolution of community structure in time is very important , to uncover how communities are generated and how they interact with each other . scholars have just begun to study this problem @xcite ( section [ sec7_1_2 ] ) . typically one analyzes separately snapshots at different times and checks what happened at time @xmath690 to the communities at time @xmath156 . it would be probably better to use simultaneously the whole dynamic data set , and future work shall aim at defining proper ways to do that . in this respect , the evolutionary clustering framework by chakrabarti et al . @xcite is a promising starting point . the computational complexity of graph clustering algorithms has improved by at least one power in the graph size ( on average ) in just a couple of years . due to the large size of many systems one wishes to investigate , the ultimate goal would be to design techniques with linear or even sublinear complexity . nowadays partitions in graphs with up to millions of vertices can be found . however , the results are not yet very reliable , as they are usually obtained by greedy optimizations , which yield rough approximations of the desired solution . in this respect the situation could improve by focusing on the development of efficient local methods , for two reasons : 1 ) they enable analyses of portions of the graph , independently of the rest ; 2 ) they are often suitable for parallel implementations , which may speed up considerably the computation . finally , if there has been a tremendous effort in the design of clustering algorithms , basically nothing has been done to make sense of their results . what shall we do with communities ? what can they tell us about a system ? the hope is that they will enable one to disclose `` hidden '' relationships between vertices , due to features that are not known , because they are hard to measure , for instance . it is quite possible that the scientific community will converge sooner or later to a definition _ a posteriori _ of community . already now , most algorithms yield similar results in practical applications . but what is the relationship between the vertex classification given by the algorithms and real classifications ? this is the main question beneath the whole endeavor . 0.5 cm * acknowledgments * i am indebted to these people for giving useful suggestions and advice to improve this manuscript at various stages : a. arenas , j. w. berry , a. clauset , p. csermely , s. gmez , s. gregory , v. gudkov , r. guimer , y. ispolatov , r. lambiotte , a. lancichinetti , j .- onnela , g. palla , m. a. porter , f. radicchi , j. j. ramasco , c. wiggins . i gratefully acknowledge ictecollective , grant number 238597 of the european commission . a _ graph _ @xmath313 is a pair of sets @xmath907 , where @xmath638 is a set of _ vertices _ or _ nodes _ and @xmath908 is a subset of @xmath909 , the set of unordered pairs of elements of @xmath638 . the elements of @xmath908 are called _ edges _ or _ links _ , the two vertices that identify an edge are called _ endpoints_. an edge is _ adjacent _ to each of its endpoints . if each edge is an ordered pair of vertices one has a _ directed _ graph ( or _ digraph _ ) . in this case an ordered pair @xmath910 is an edge directed from @xmath29 to @xmath752 , or an edge beginning at @xmath29 and ending at @xmath752 . a graph is visualized as a set of points connected by lines , as shown in fig . [ fig1 ] . in many real examples , graphs are _ weighted _ , i. e. a real number is associated to each of the edges . graphs do not include _ loops _ , i. e. edges connecting a vertex to itself , nor multiple edges , i. e. several edges joining the same pair of vertices . graphs with loops and multiple edges are called _ multigraphs_. generalizations of graphs admitting edges between any number of vertices ( not necessarily two ) are called _ hypergraphs_. a graph @xmath911 is a _ subgraph _ of @xmath912 if @xmath913 and @xmath914 . if @xmath380 contains all edges of @xmath313 that join vertices of @xmath915 one says that the subgraph @xmath380 is induced or spanned by @xmath915 . a partition of the vertex set @xmath638 in two subsets @xmath87 and @xmath916 is called a _ cut _ ; the _ cut size _ is the number of edges of @xmath313 joining vertices of @xmath87 with vertices of @xmath916 . we indicate the number of vertices and edges of a graph with @xmath15 and @xmath14 , respectively . the number of vertices is the _ order _ of the graph , the number of edges its _ size_. the maximum size of a graph equals the total number of unordered pairs of vertices , @xmath41 . if @xmath917 and @xmath918 , the graph is a _ clique _ ( or _ complete graph _ ) , and is indicated as @xmath919 . two vertices are _ neighbors _ ( or _ adjacent _ ) if they are connected by an edge . the set of neighbors of a vertex @xmath29 is called _ neighborhood _ , and we shall denote it with @xmath920 . the _ degree _ @xmath921 of a vertex @xmath29 is the number of its neighbors . the _ degree sequence _ is the list of the degrees of the graph vertices , @xmath922 . on directed graphs , one distinguishes two types of degree for a vertex @xmath29 : the _ indegree _ , i. e. the number of edges beginning at @xmath29 and the _ outdegree _ , i. e. the number of edges ending at @xmath29 . the analogue of degree on a weighted graph is the _ strength _ , i. e. the sum of the weights of the edges adjacent to the vertex . another useful local property of graphs is _ transitivity _ or _ clustering _ @xcite , which indicates the level of cohesion between the neighbors of a vertex . the clustering coefficient @xmath923 of vertex @xmath29 is the ratio between the number of edges joining pairs of neighbors of @xmath29 and the total number of possible edges , given by @xmath924 , @xmath925 being the degree of @xmath29 . according to this definition , @xmath923 measures the probability that a pair of neighbors of @xmath29 are connected . since all neighbors of @xmath29 are connected to @xmath29 by definition , edges connecting pairs of neighbors of @xmath29 form triangles with @xmath29 . this is why the definition is often given in terms of number of triangles . a _ path _ is a graph @xmath926 , with @xmath927 and @xmath928 . the vertices @xmath929 and @xmath930 are the _ endvertices _ of @xmath92 , whereas @xmath76 is its _ length_. given the notions of vertices , edges and paths , one can define the concept of _ independence_. a set of vertices ( or edges ) of a graph are independent if no two elements of them are adjacent . similarly , two paths are independent if they only share the endvertices . a _ cycle _ is a closed path whose vertices and edges are all distinct . cycles of length @xmath76 are indicated with @xmath931 . the smallest non - trivial cycle is the _ triangle _ , @xmath932 . paths allow to define the concept of connectivity and distance in graphs . a graph is _ connected _ if , given any pair of vertices , there is at least one path going from one vertex to the other . in general , there may be multiple paths connecting two vertices , with different lengths . shortest path _ , or _ geodesic _ , between two vertices of a graph , is a path of minimal length . such minimal length is the _ distance _ between the two vertices . the _ diameter _ of a connected graph is the maximal distance between two vertices . if there is no path between two vertices , the graph is divided in at least two connected subgraphs . each maximal connected subgraph of a graph is called _ connected component_. a graph without cycles is a _ forest_. a connected forest is a _ tree_. trees are very important in graph theory and deserve some attention . in a tree , there can be only one path from a vertex to any other . in fact , if there were at least two paths between the same pair of vertices they would form a cycle , while the tree is an acyclic graph by definition . further , the number of edges of a tree with @xmath15 vertices is @xmath143 . if any edge of a tree is removed , it would get disconnected in two parts ; if a new edge is added , there would be at least one cycle . this is why a tree is a minimally connected , maximally acyclic graph of a given order . every connected graph contains a _ spanning tree _ , i. e. a tree sharing all vertices of the graph . on weighted graphs , one can define a _ minimum _ ( _ maximum _ ) _ spanning tree _ , i. e. a spanning tree such that the sum of the weights on the edges is minimal ( maximal ) . minimum and maximum spanning trees are often used in graph optimization problems , including clustering . a graph @xmath313 is _ bipartite _ if the vertex set @xmath638 is separated in two disjoint subsets @xmath933 and @xmath934 , or _ classes _ , and every edge joins a vertex of @xmath933 with a vertex of @xmath934 . the definition can be extended to that of _ @xmath375-partition _ , where the vertex classes are @xmath375 and no edge joins vertices within the same class . in this case one speaks of _ multipartite _ graphs . the whole information about the topology of a graph of order @xmath15 is entailed in the _ adjacency matrix _ * a * , which is an @xmath167 matrix whose element @xmath360 equals @xmath2 if there is an edge joining vertices @xmath67 and @xmath68 , otherwise it is zero . due to the absence of loops the diagonal elements of the adjacency matrix are all zero . for an undirected graph * a * is a symmetric matrix . the sum of the elements of the @xmath67-th row or column yields the degree of node @xmath67 . if the edges are weighted , one defines the _ weight matrix _ @xmath935 , whose element @xmath317 expresses the weight of the edge between vertices @xmath67 and @xmath68 . the _ spectrum _ of a graph @xmath313 is the set of eigenvalues of its adjacency matrix * a*. spectral properties of graph matrices play an important role in the study of graphs . for instance , the _ stochastic matrices _ rule the process of diffusion ( random walk ) on a graph . right stochastic matrix _ @xmath448 is obtained from * a * by dividing the elements of each row @xmath67 by the degree of vertex @xmath67 . the _ left stochastic matrix _ @xmath201 , or _ transfer matrix _ , is the transpose of @xmath448 . the spectra of stochastic matrices allow to evaluate , for instance , the mixing time of the random walk , i. e. the time it takes to reach the stationary distribution of the process . the latter is obtained by computing the eigenvector of the transfer matrix corresponding to the largest eigenvalue . another important matrix is the _ @xmath936 , where * d * is the diagonal matrix whose element @xmath937 equals the degree of vertex @xmath67 . the matrix @xmath136 is usually referred to as _ unnormalized laplacian_. in the literature one often uses _ normalized laplacians _ @xcite , of which there are two main forms : @xmath938 and @xmath939 . the matrix @xmath195 is symmetric ; @xmath194 is not symmetric and is closely related to a random walk taking place on the graph . all laplacian matrices have a straightforward extension to the case of weighted graphs . the laplacian is one of the most studied matrices and finds application in many different contexts , like graph connectivity @xcite , synchronization @xcite , diffusion @xcite and graph partitioning @xcite . by construction , the sum of the elements of each row of the laplacian ( normalized or unnormalized ) is zero . this implies that * l * always has at least one zero eigenvalue , corresponding to the eigenvector with all equal components , such as @xmath940 . eigenvectors corresponding to different eigenvalues are all orthogonal to each other . interestingly , * l * has as many zero eigenvalues as there are connected components in the graph . so , the laplacian of a connected graph has but one zero eigenvalue , all others being positive . eigenvectors of laplacian matrices are regularly used in _ spectral clustering _ ( section [ sec4_4 ] ) . in particular , the eigenvector corresponding to the second smallest eigenvalue , called _ fiedler vector _ @xcite , is used for graph bipartitioning , as described in section [ sec4_1 ] . in this section we present the most popular models of graphs introduced to describe real systems , at least to some extent . such graphs are useful null models in community detection , as they do not have community structure , so they can be used for negative tests of clustering algorithms . basic models of complex networks . ( top ) erds - rnyi random graph with @xmath874 vertices and a link probability @xmath941 . ( center ) small world graph la watts - strogatz , with @xmath874 vertices and a rewiring probability @xmath942 . ( bottom ) barabsi - albert scale - free network , with @xmath874 vertices and an average degree of @xmath323 . courtesy by j. j. ramasco.,title="fig:",width=340 ] basic models of complex networks . ( top ) erds - rnyi random graph with @xmath874 vertices and a link probability @xmath941 . ( center ) small world graph la watts - strogatz , with @xmath874 vertices and a rewiring probability @xmath942 . ( bottom ) barabsi - albert scale - free network , with @xmath874 vertices and an average degree of @xmath323 . courtesy by j. j. ramasco.,title="fig:",width=340 ] basic models of complex networks . ( top ) erds - rnyi random graph with @xmath874 vertices and a link probability @xmath941 . ( center ) small world graph la watts - strogatz , with @xmath874 vertices and a rewiring probability @xmath942 . ( bottom ) barabsi - albert scale - free network , with @xmath874 vertices and an average degree of @xmath323 . courtesy by j. j. ramasco.,title="fig:",width=340 ] the oldest model is that of _ random graph _ , proposed by solomonoff and rapoport @xcite and independently by erds and rnyi @xcite . there are two parameters : the number of vertices @xmath15 and the connection probability @xmath45 . each pair of vertices is connected with equal probability @xmath45 independently of the other pairs . the expected number of edges of the graph is @xmath943 , and the expected mean degree @xmath944 . the degree distribution of the vertices of a random graph is binomial , and in the limit @xmath945 , @xmath946 for fixed @xmath420 it converges to a poissonian . therefore , the vertices have all about the same degree , close to @xmath420 ( fig . [ fig2 ] , top ) . the most striking property of this class of graphs is the phase transition observed by varying @xmath420 in the limit @xmath945 . for @xmath947 , the graph is separated in connected components , each of them being microscopic , i. e. occupying but a vanishing portion of the system size . for @xmath948 , instead , one of the components becomes macroscopic ( _ giant component _ ) , i. e. it occupies a finite fraction of the graph vertices . the diameter of a random graph with @xmath15 vertices is very small , growing only logarithmically with @xmath15 . this property ( _ small - world effect _ ) is very common in many real graphs . the first evidence that social networks are characterized by paths of small length was provided by a series of famous experiments conducted by the phychologist stanley milgram @xcite . the expected clustering coefficient of a vertex of a random graph is @xmath45 , as the probability for two vertices to be connected is the same whether they are neighbors of the same vertex or not . real graphs , however , are characterized by far higher values of the clustering coefficient as compared to random graphs of the same size . watts and strogatz @xcite showed that the small world property and high clustering coefficient can coexist in the same system . they designed a class of graphs which result from an interpolation between a regular lattice , which has high clustering coefficient , and a random graph , which has the small - world property . one starts from a ring lattice in which each vertex has degree @xmath44 , and with a probability @xmath45 each edge is rewired to a different target vertex ( fig . [ fig2 ] , center ) . it turns out that low values of @xmath45 suffice to reduce considerably the length of shortest paths between vertices , because rewired edges act as shortcuts between initially remote regions of the graph . on the other hand , the clustering coefficient remains high , since few rewired edges do not perturb appreciably the local structure of the graph , which remains similar to the original ring lattice . for @xmath811 all edges are rewired and the resulting structure is a random graph la erds and rnyi . the seminal paper of watts and strogatz triggered a huge interest towards the graph representation of real systems . one of the most important discoveries was that the distribution of the vertex degree of real graphs is very heterogeneous @xcite , with many vertices having few neighbors coexisting with some vertices with many neighbors . in several cases the tail of this distribution can be described as a power law with good approximation , hence the expression _ scale - free networks_. such degree heterogeneity is responsible for a number of remarkable features of real networks , such as resilience to random failures / attacks @xcite , and the absence of a threshold for percolation @xcite and epidemic spreading @xcite . the most popular model of a graph with a power law degree distribution is the model by barabsi and albert @xcite . a version of the model for directed graphs had been proposed much earlier by de solla price @xcite , building up on previous ideas developed by simon @xcite . the graph is created with a dynamic procedure , where vertices are added one by one to an initial core . the probability for a new vertex to set an edge with a preexisting vertex is proportional to the degree of the latter . in this way , vertices with high degree have large probability of being selected as neighbors by new vertices ; if this happens , their degree further increases so they will be even more likely to be chosen in the future . in the asymptotic limit of infinite number of vertices , this _ rich - gets - richer _ strategy generates a graph with a degree distribution characterized by a power - law tail with exponent @xmath3 . in fig . [ fig2 ] ( bottom ) we show an example of barabsi - albert ( ba ) graph . the clustering coefficient of a ba graph decays with the size of the graph , and it is much lower than in real networks . moreover , the power law decays of the degree distributions observed in real networks are characterized by a range of exponents values ( usually between @xmath323 and @xmath3 ) , whereas the ba model yields a fixed value . however , many refinements of the ba model as well as plenty of different models have been later introduced to account more closely for the features observed in real systems ( for details see @xcite ) .

modified newtonian dynamics ( mond ) was proposed by milgrom ( 1983a , b ) as an alternative to dark matter . mond posits that the effective gravitational force deviates from the newtonian force : at accelerations below a critical value , @xmath0 , the gravitational force is proportional to the square root of the newtonian force ; at higher accelerations , the force is newtonian . in principle , mond predicts the shape and amplitude of the observed rotation curves from the observed mass distribution , i.e. , gas and stars , with only the mass - to - light ratio ( m / l ) of the stellar disk as adjustable parameter . the rotation curves of a large number of spiral galaxies have been considered in the context of mond . for example , begeman et al . ( 1991 ) , sanders ( 1996 ) , sanders & verheijen ( 1998 ) , and sanders & noordermeer ( 2007 ) demonstrated that for the spiral galaxies the observed rotation curves were , in most cases , predicted in detail , using the mond prescription , from the observed light and gas distributions . moreover , the m / ls derived from these fits are generally astrophysically plausible and consistent with stellar population synthesis models . in a handful of these galaxies the rotation curves predicted by mond are noticeably different from the observed rotation curves . these were not considered falsifications of mond because for most of these galaxies the authors identified an obvious problem with the observed rotation curve ( e.g. , uncertain inclination or distance ) or with the use of the rotation curve as a tracer of the gravitational force ( e.g. , disturbed velocity fields ) . although mond appears to explain the magnitude of the discrepancy in spiral galaxies , it has has long been known that the theory predicts more matter than is observed in clusters of galaxies ( e.g. , sanders 1999 ) . in this sense the famous `` bullet cluster '' , which has been presented as direct emprical evidence for the existence of dark matter ( e.g. , clowe et al . 2006 ) , does not pose a new problem for mond . additional undetected matter in some dissipationless form does seem to be required in clusters in the context of mond . in any case , additional required matter is not formally a falsification of mond and in no sense detracts from the success of this algorithm on the scale of galaxies . lcrrrrrrrr[th ] ugc 731 & smbb & 8.0 & 23.0 & 1.65 & -16.6 & 7.37 & 57 & 74 & 0.25 + ugc 3371 & svba & 12.8 & 23.3 & 3.09 & -17.7 & 12.2 & 49 & 86 & 0.23 + ugc 4173 & dbb & 18.0 & 24.3 & 4.77 & -18.0 & 24.3 & 40 & 57 & 0.08 + ugc 4325 & smbb & 10.8 & 21.6 & 1.74 & -18.2 & 8.61 & 41 & 93 & 0.44 + ugc 4499 & smbb & 13.9 & 21.5 & 1.59 & -17.9 & 13.6 & 50 & 74 & 0.20 + ugc 5005 & dbb & 56 & 22.9 & 4.71 & -18.8 & 33.2 & 41 & 99 & 0.11 + ugc 5414 & ssah & 10.7 & 21.8 & 1.59 & -17.7 & 7.42 & 55 & 61 & 0.26 + ugc 5721 & smbb & 6.6 & 20.2 & 0.45 & -16.6 & 6.63 & 61 & 79 & 0.28 + ugc 5750 & dbmr & 60 & 22.6 & 5.99 & -19.6 & 11.4 & 64 & 79 & 0.09 + ugc 6446 & ssah & 12.8 & 21.4 & 2.00 & -18.5 & 15.4 & 52 & 80 & 0.20 + ugc 7232 & ssah & 3.5 & 20.2 & 0.33 & -15.3 & 0.71 & 59 & 44 & 0.62 + ugc 7323 & ssah & 8.7 & 21.2 & 2.35 & -19.0 & 8.48 & 47 & 86 & 0.38 + ugc 7399 & ssah & 8.4 & 20.7 & 0.79 & -17.1 & 7.40 & 55 & 109 & 0.35 + ugc 7524 & ssah & 4.6 & 22.2 & 3.37 & -18.7 & 16.7 & 46 & 84 & 0.22 + ugc 7559 & ssah & 4.9 & 23.8 & 1.02 & -14.6 & 1.71 & 61 & 33 & 0.11 + ugc 7577 & ssah & 2.5 & 22.5 & 0.60 & -14.9 & 0.42 & 63 & 18 & 0.06 + ugc 7603 & dbb & 7.3 & 20.8 & 0.96 & -17.0 & 6.13 & 78 & 64 & 0.21 + ugc 8490 & smbb & 4.6 & 20.5 & 0.62 & -17.1 & 7.01 & 50 & 80 & 0.21 + ugc 9211 & dbb & 12.6 & 22.6 & 1.32 & -16.2 & 10.5 & 44 & 66 & 0.17 + ugc 11707 & smbb & 17.0 & 23.1 & 4.61 & -18.8 & 42.7 & 68 & 100 & 0.20 + ugc 11861 & smbb & 27 & 21.4 & 6.48 & -20.9 & 81.7 & 50 & 153 & 0.43 + ugc 12060 & ssah & 16.8 & 21.6 & 1.89 & -18.1 & 20.7 & 40 & 75 & 0.16 + ugc 12632 & ssah & 6.9 & 23.5 & 2.57 & -17.1 & 8.66 & 46 & 76 & 0.22 + f568-v1 & smbb & 86 & 22.8 & 3.42 & -18.8 & 28.1 & 40 & 118 & 0.24 + f574 - 1 & smbb & 103 & 22.6 & 4.60 & -19.3 & 39.7 & 65 & 100 & 0.22 + f583 - 1 & dbmr & 34 & 22.0 & 1.71 & -17.5 & 19.9 & 63 & 84 & 0.15 + f583 - 4 & dbmr & 52 & 22.4 & 2.89 & -17.8 & 6.30 & 55 & 70 & 0.22 + mond is more strongly tested by spiral galaxies with well - determined distances because the @xmath0 depends inversely upon distance . bottema et al . ( 2002 ) considered a sample of 4 spiral galaxies with well - measured rotation curves and cepheid - based distances determined as part of the hst program on the extragalactic distance scale ( sakai et al . two galaxies are in good agreement with mond predictions , but for ngc 2841 and ngc 3198 the rotation curves are not in agreement with mond , unless the distance to ngc 2841 is at least 20% larger , and the distance to ngc 3198 is at least 10% smaller than their cepheid distances indicate . bottema et al . ( 2002 ) argued that ngc 3198 is probably consistent with mond within the uncertainties of the cepheid distance . with respect to ngc 2841 , milgrom and sanders ( 2008 ) have demonstrated that a reasonable match to the observed rotation curve can be achieved with alternative forms of the function that described the transition from the newtonian to the mond regime . dwarf and low surface brightness ( lsb ) galaxies provide an especially good test for mond , because in most of these galaxies the acceleration is below the mond threshold of @xmath3 cm s@xmath2 ( determined empirically from rotation curve fits , see e.g. , sanders & mcgaugh 2002 ) at all radii , whereas in high surface brightness spiral galaxies the acceleration is usually above the threshold in the inner parts . milgrom ( 1983b ) predicted that lsb galaxies should exhibit a large discrepancy between the detectable and newtonian dynamical mass within the optical disk , and this prediction has since been confirmed ( e.g. , de blok & mcgaugh 1997 ; mcgaugh & de blok 1998 ; swaters et al . 2000 ) . apart from this general prediction , the mond prescription is successful in predicting the detailed rotation curves of lsb galaxies as well . de blok & mcgaugh ( 1998 ) found that for their sample of 15 lsb galaxies all rotation curves are well fitted with mond , although for 6 galaxies modest adjustments of the inclination were necessary in order to get an acceptable match to the rotation curve . the sample presented in sanders & verheijen ( 1998 ) contains 12 lsb galaxies , and most of these the rotation curves can be well explained by mond as well . the results for dwarf galaxies show a fairly large spread . milgrom & braun ( 1988 ) concluded that the observed rotation curve of ddo 154 is in very good agreement with the rotation curve predicted by mond . lake ( 1989 ) considered a sample of 6 dwarf galaxies in the context of mond and concluded that acceptable fits require systematically lower values for @xmath0 than were found for spiral galaxies . in addition , lake ( 1989 ) found an apparent correlation between the asymptotic rotation velocity and the fitted value of @xmath0 . milgrom ( 1991 ) , however , pointed out that uncertainties in inclinations and distances can well explain the fact that mond appears to fail for these galaxies . the rotation curves of the dwarf galaxies included in the samples of begeman et al . ( 1991 ) and sanders ( 1996 ) also agree well with those calculated with mond . on the other hand , blais - ouellette et al . ( 2001 ) reported that mond provides poor predictions of the rotation curves of the three dwarf galaxies in their sample , and reported that a higher value of @xmath0 is needed to make their rotation curves derived from fabry - perot observations consistent with mond . more recently , milgrom & sanders ( 2007 ) concluded that the rotation curves of the four low - mass galaxies in their sample are explain well in the context of mond . given this apparent spread in conclusions on whether mond correctly predicts the rotation curves of dwarf galaxies , we here test mond against a sample of rotation curves of dwarf and lsb galaxies that have recently become available in the literature . we included both dwarf and lsb galaxies because the physical properties of these galaxies are similar , although some dwarf galaxies have high surface brightness , and some lsb galaxies are large in size . there are several difficulties inherent to such a sample . for example , because most dwarf galaxies are nearby , different methods to estimate the distance have been used , and this may introduce a dispersion in the derived parameters . inclination is another uncertainty for a class of objects which are known to be irregular with , in some cases , large scale asymmetries . moreover , the assumption that the rotation curve is a tracer of the radial force distribution may , itself , be questionable in cases where asymmetries are present . bearing these considerations in mind , one might not expect , a priori , the agreement between the predicted and observed rotation curves to be as good as noted for earlier samples . a systematic failure , however , would be problematic for mond . the outline of this paper is as follows . in section [ secsample ] we describe the sample used in this paper . in section [ secfitproc ] we describe the fitting procedure , and in section [ secfitres ] the fitting results . next , we comment on individual galaxies in section [ secindiv ] . we describe the uncertainties in section [ secuncert ] , and we discuss the results in section [ secdisc ] . finally , we present our conclusions in section [ secconc ] . the sample we present here is compiled from data from five studies . a large fraction comes from the rotation curves of a sample of 62 dwarf galaxies ( swaters 1999 ; swaters et al . 2009 , hereafter ssah ) . the resolution of the hi data used to derive the rotation curves is @xmath4 . because of this relatively low resolution , the data have been corrected for beam smearing as described in swaters ( 1999 ) and ssah . from this original sample of 62 dwarf galaxies , we selected galaxies with inclination in the range @xmath5 , and we only included rotation curves that were classified by ssah as high quality . galaxies classified by ssah as having lower quality either have a signal - to - noise ratio that is too low , or are too asymmetric to derive a reliable rotation curve . to avoid the uncertainties associated with these low quality rotation curves , we have excluded these from our sample . the resulting sample contains 19 rotation curves . the second source is the sample of h@xmath6 rotation curves presented in swaters et al . ( 2003a ) . they obtained high resolution rotation curves from h@xmath6 long - slit spectroscopy for a sample of 10 dwarf that are part of the ssah sample described above , and 5 lsb galaxies from the sample presented in de blok et al . ( 1996 ) . as described in swaters et al . ( 2003a ) , these h@xmath6 rotation curves have been combined with the data presented ssah and de blok et al . galaxies with @xmath7 or @xmath8 were not included in the sample here . f568 - 3 was also excluded because of its strong bar . the third source is the sample of lsb galaxies presented in de blok , et al . they present long - slit h@xmath6 rotation curves for a sample of 30 lsb galaxies . from their sample we have selected those galaxies for which optical photometry and imaging are available and with inclinations in the range @xmath9 . the fourth source is the sample presented in de blok & bosma ( 2002 ) , consisting of -h@xmath6 hybrid rotation curves . from this sample we have selected the galaxies that overlap with the ssah sample . we excluded ugc 3851 because its rotation curve depends heavily on the uncertain rotation curve as derived by ssah and which they deemed to be of low quality . ugc 7524 was also excluded because this galaxy is well resolved by the observations and the rotation curve derived from the two - dimensional data probably gives a better representation of the gravitational potential than the one derived from long - slit data . the rotation curves presented in de blok & bosma ( 2002 ) have been combined with the rotation curves presented in ssah as described in de blok & bosma ( 2002 ) . the fifth and final source is the rotation curve of ugc 3371 presented in swaters et al . ( 2003b ) . they obtained high - resolution , two - dimensional h@xmath6 data for this galaxy . where the samples are overlapping , we have used them in this order of decreasing priority : swaters et al . ( 2003b ) , swaters et al . ( 2003a ) , de blok & bosma ( 2002 ) , de blok et al . ( 2001 ) , ssah . most rotation curves in our sample have been derived from h@xmath6 data for the central parts and data for the outer parts . however , for the 10 galaxies that came from ssah ( see column 2 in table [ tabsample ] ) the rotation curves are entirely based on data . we have not corrected the rotation curves in these samples for pressure support , because these corrections tend to be small in comparison to the rotation velocities , and probably uncertain ( see also ssah ) . the combined sample does not constitute a complete sample of dwarf and lsb galaxies . nonetheless , this sample contains galaxies with a wide range in luminosity and surface brightness for which high - quality rotation curve are available . this sample is therefore well suited to test mond in late - type dwarf and lsb galaxies . because many of the galaxies in our sample are at small distances , their systemic velocities are a poor indicator of their distance . where possible , we use other distance indicators , based on the compilation of distances found in the literature presented in swaters & balcells ( 2002 ) , supplemented with other recent distance estimates based on the tip of the red giant branch ( mndez et al . 2002 ; karachentsev et al . 2003 ) . in order of decreasing priority , the distances presented in table [ tabsample ] are based on cepheids , the tip of the red giant branch , brightest stars , group membership , and the recession velocities corrected for virgocentric flow as described in swaters & balcells ( 2002 ) . for a more complete discussion on the distances to these dwarf galaxies , see swaters & balcells ( 2002 ) . the galaxies in our sample that are not in the ssah sample are all at large distances , and for these the distances have been calculated based on a hubble constant of 70 km s@xmath10 mpc@xmath10 , following sanders & mcgaugh ( 2002 ) . to calculate the mond rotation curves , we have used the same method as used in sanders ( 1996 ) and sanders & verheijen ( 1998 ) . to calculate the contribution of the stellar component to the disk , we have assumed that the stellar mass resides in an infinitely thin disk , that the @xmath11-band light accurately traces the mass distribution of the stellar component , and that the stellar mass - to - light ratio is constant with radius . the gas mass distribution is also assumed to be in a thin disk . we assumed that the is optically thin , and we corrected for the mass fraction of helium by scaling the mass by a factor of 1.32 . the optical radial profiles were taken from swaters & balcells ( 2002 ) , de blok et al . ( 1995 ) and mcgaugh & bothun ( 1994 ) . the radial profiles were taken from swaters et al . ( 2002 ) , de blok et al . ( 1996 ) , and van der hulst et al . for the galaxies in which the distribution is poorly resolved , we calculated the radial profiles following the algorithm described by warmels ( 1988 ) that uses a iterative deconvolution scheme ( lucy 1974 ) to correct for the effects of beam smearing ( see swaters et al . 2002 for more details ) . for both the stellar and the gaseous disk , we assumed that the mass distribution is well represented by the radial average . to calculate the mond rotation curves of the visible components , ideally the mond field equation of bekenstein & milgrom ( 1984 ) should be used . however , this is computationally expensive , and milgrom ( 1986 ) has shown that the differences between the results derived for the field equation and those from the original mond prescription are usually much smaller than , and practically never larger than 5% . therefore , we apply the usual mond formula : @xmath12 where @xmath0 is the mond acceleration parameter , @xmath13 is the mond acceleration , @xmath14 is the newtonian acceleration , and @xmath15 which is the commonly assumed form having the required asymptotic behaviour . for @xmath16 gravity is newtonian , and for @xmath17 gravity is of the mond form with @xmath18 . a little algebra shows that the rotation curve for mond can be expressed as : @xmath19 here , @xmath20 is the radius and @xmath21 with @xmath22 , @xmath23 and @xmath24 the contribution of the stellar disk , bulge and gas to the rotation curve , respectively , calculated in the newtonian regime as described above , and @xmath25 and @xmath26 are the stellar mass - to - light ratios of the disk and the bulge . eq . [ eqfit ] is fit to the observed rotation curve with a least - squares algorithm . none of the galaxies in our sample have significant bulges , which eliminates @xmath26 as a free parameter . it should be noted that in the context of mond the internal dynamics of a system is affected by the external acceleration field . if the external field becomes comparable to or larger than @xmath0 , a galaxy in this external field will always be in the newtonian regime , even if the internal accelerations are low ( milgrom 1983a , b ) . we made mond fits by fitting eq . [ eqfit ] to the observed rotation curves with a least squares algorithm . because @xmath0 is a universal constant in mond , we first made fits with only @xmath25 as a free parameter . following e.g. , sanders & mcgaugh ( 2002 ) , the acceleration parameter was fixed at @xmath27 cm s@xmath2 ( @xmath28 km@xmath29 s@xmath2 kpc@xmath10 ) . in addition to the one parameter fits , we have also made mond fits with the distance and inclination as free parameters in order to investigate the effects of uncertainties in these parameters . in addition , we made fits with @xmath0 as a free parameter to measure @xmath0 and look for trends . the one - parameter mond fits , with only @xmath25 as a free parameter , are shown in figure [ figfits ] . in each panel of this figure , the dots represent the measured rotation velocities and their uncertainties , the thin solid line shows the contribution of the stellar disk , and the dashed line the contribution of the hi . the best fitting mond rotation curve is indicated by the thick solid line . a first inspection of the mond fits presented in figure [ figfits ] shows that roughly three quarters of the rotation curves are well fit by mond , with only @xmath25 as a free parameter . this is perhaps surprising , given the possible uncertainties due to inclination , distance , and asymmetries . a closer look shows that more than half of the galaxies deviate somewhat from the observed rotation rotation in a systematic way : the mond curve predicts higher rotation velocities in in the outer regions , and lower in the central regions . this is seen clearly in e.g. , ugc 6446 and ugc 12060 , and it is also visible , although less pronounced , in e.g. , ugc 731 and ugc 3371 . lrrrrrrrrrr[th ] ugc 731 & 4.3 & 1460 & 445 & 8.5 & 0.61 & 0.15 & 4.6 & 38 & 5 & 10.2 + ugc 3371 & 2.8 & 2050 & 590 & 4.2 & 0.81 & 0.06 & 2.9 & 41 & 5 & 4.3 + ugc 4173 & 0.1 & 550 & 350 & 0.6 & 0.43 & 0.08 & 0.1 & 24 & 5 & 0.7 + ugc 4325 & 3.8 & 800 & 500 & 6.4 & 0.41 & 0.15 & 2.5 & 24 & 5 & 12.9 + ugc 4499 & 0.3 & 3510 & 480 & 0.2 & 1.07 & 0.07 & 0.3 & 54 & 6 & 0.2 + ugc 5005 & 0.5 & 1890 & 1100 & 1.3 & 0.78 & 0.21 & 0.8 & 35 & 6 & 1.3 + ugc 5414 & 0.5 & 1880 & 1200 & 1.0 & 0.78 & 0.20 & 0.7 & 43 & 20 & 1.1 + ugc 5721 & 3.9 & 9340 & 900 & 1.5 & 1.81 & 0.09 & 3.9 & 69 & 3 & 4.4 + ugc 5750 & 0.3 & 5350 & 1600 & 0.0 & 1.32 & 0.42 & 0.0 & 63 & 30 & 0.3 + ugc 6446 & 1.1 & 1710 & 300 & 2.1 & 0.69 & 0.09 & 1.3 & 39 & 3 & 2.5 + ugc 7232 & 1.3 & 15300 & 7000 & 0.0 & 2.20 & 0.57 & 0.1 & 63 & 22 & 1.2 + ugc 7323 & 0.8 & 4270 & 1740 & 0.5 & 1.16 & 0.25 & 0.8 & 57 & 18 & 0.5 + ugc 7399 & 9.6 & 14200 & 1500 & 1.8 & 2.21 & 0.10 & 7.2 & 72 & 3 & 11.5 + ugc 7524 & 0.9 & 1290 & 350 & 2.4 & 0.61 & 0.11 & 1.2 & 33 & 3 & 3.0 + ugc 7559 & 0.0 & 650 & 680 & 3.2 & 0.48 & 0.27 & 0.8 & 32 & 9 & 3.4 + ugc 7577 & 0.0 & 80 & 325 & 0.8 & 0.22 & 0.21 & 0.0 & 20 & 10 & 0.8 + ugc 7603 & 0.8 & 6080 & 1100 & 0.2 & 1.40 & 0.13 & 0.4 & 63 & 10 & 0.3 + ugc 8490 & 3.0 & 6690 & 700 & 1.3 & 1.51 & 0.08 & 2.6 & 67 & 3 & 2.6 + ugc 9211 & 2.4 & 2110 & 1000 & 4.5 & 0.82 & 0.26 & 3.2 & 38 & 8 & 4.7 + ugc 11707 & 1.2 & 1240 & 190 & 3.8 & 0.58 & 0.07 & 1.9 & 86 & 2 & 8.9 + ugc 11861 & 2.0 & 2310 & 500 & 2.5 & 0.70 & 0.12 & 1.8 & 39 & 5 & 3.2 + ugc 12060 & 1.8 & 690 & 270 & 5.6 & 0.32 & 0.10 & 1.6 & 21 & 4 & 13.7 + ugc 12632 & 3.0 & 1190 & 390 & 8.0 & 0.54 & 0.14 & 3.6 & 30 & 5 & 11.4 + f568-v1 & 5.4 & 1740 & 700 & 8.3 & 0.67 & 0.20 & 5.1 & 31 & 5 & 10.7 + f574 - 1 & 3.0 & 950 & 460 & 6.5 & 0.34 & 0.14 & 2.1 & 90 & 2 & 47.5 + f583 - 1 & 2.4 & 2590 & 990 & 2.9 & 0.91 & 0.20 & 2.5 & 52 & 18 & 2.9 + f583 - 4 & 1.9 & 610 & 3500 & 6.9 & 0.24 & 0.68 & 1.6 & 24 & 11 & 17.2 + the degree to which mond appears to predict higher velocities in the outer rotation curves is shown in figure [ fighistouter ] , in which a histogram of the difference between the observed and model velocities is presented , as well as one in which the differences have been normalized by the uncertainty in the rotation velocity of the last measured point . in 21 out of 27 galaxies , mond predicts higher than observed rotation velocities , whereas it predicts lower velocities in only 6 cases . the discrepancy is more than twice the uncertainty in the last measured point in 11 galaxies . in figure [ fighistmlfix ] we give the histogram of the best fit values for @xmath25 . the mass - to - light ratios span a large range , from 0 up to 10 . the average @xmath11-band @xmath25 as derived from the best fits its @xmath30 @xmath31 . given that some of the adopted distances may be uncertain , it is possible that , for some of the galaxies for which mond does not correctly predict the rotation curves , we have adopted an incorrect distance . we will investigate this by leaving the distance as a free parameter in the fits . with distance as a free parameter , not only does the radial scale of the rotation curve change , but at the same time @xmath32 and @xmath33 , where @xmath34 is the fractional distance change . strictly speaking , these relations are only correct if both the stellar and disks are infinitely thin . for disks with finite thicknesses , the shape of the rotation curves depends on the assumed thickness of the disk , and hence the shape of the rotation will change slightly if a difference distance is assumed . the fits with distance as a free parameter are shown in fig . [ figfits ] as the dotted lines . from this figure it is clear that , with distance as a free parameter , good fits are obtained in virtually all cases . this demonstrates that a different distance can improve the fit . the best fit values for @xmath35 , the fractional distance change , are listed in table [ tabfits ] . the uncertainties on @xmath35 have been derived from the 68% confidence levels . however , because the errors on the rotation curves are non - gaussian , because the points are correlated , and because the rotation curves and the errors on the points may be affected by systematic effects , the confidence levels and the uncertainties derived from them should be considered estimates . in fig . [ figd0hist ] the distribution of @xmath35 is shown . the best fit fractional distance changes tend to be smaller than 1 . as can be seen in the bottom panel of fig . [ figd0hist ] , this is mostly due to galaxies at larger distances and for which the distances were determined from the hubble flow . for these galaxies , the average fractional distance change @xmath36 is @xmath37 , with a dispersion of @xmath38 . given that all these galaxies are at relatively large distances , it is unlikely that the hubble flow distances are uncertain by that much , in particular because small @xmath35 are found even for galaxies at the largest distances . for galaxies that are close by ( @xmath39 mpc ; middle panel of fig . [ figd0hist ] ) , the best fit fractional distance changes appear to scatter around @xmath40 ( i.e. , no distance change ) . such a large scatter is not unexpected because the galaxies are close - by and different methods have been employed to determine the distances to these galaxies . for most galaxies in our sample the inclination can not be determined from the kinematics , because the rotation curves for most of the galaxies in this sample rise slowly without a well - defined flat part . as a result , for most galaxies in our sample the inclinations have been estimated from the morphology . for the galaxies in the ssah sample , the inclinations mostly have been determined from the axis ratios of the outer isophotes ( as described by swaters & balcells 2002 ) , except for those galaxies were the velocity field allowed a reliable estimate ( see ssah for details ) . for the lsb galaxies , the inclinations have been taken from de blok & mcgaugh ( 1997 ) , who used the maps and optical images to estimate the inclinations . given how these inclinations have been derived , the inclinations of some of the galaxies in our sample are likely to be uncertain . to investigate whether some of mond - predicted rotation curves can be improved by changing their inclinations , we have made mond fits with inclination as a free parameter . if the inclination is left free in the fits , the amplitude of the rotation curve changes with @xmath41 , and at the same time , the contribution of the and the stars to the rotation curve change with @xmath42 ( if optically thin ) . the latter is only valid for modest changes in the inclination , especially at high inclinations . when the changes are large , the rotation curves and density distributions may also change in shape . in table [ tabfits ] we list the best fit inclinations . uncertainties on the inclinations have been estimated from the 68% confidence levels in the fit , with the same caveats as described above for the fits with distance free . we do not show the best fits in fig . [ figfits ] , because the fits with inclinations free are almost indistinguishable from those with distance as a free parameter , except for ugcs 5721 , 7323 , 7399 , and 8490 , for which the fits are nearly indistinguishable from the original fit with only @xmath25 as a free parameter . in figure [ figinclhist ] we plot the histogram of inclination changes , defined as @xmath43 . the inclination changes span a wide range , from @xmath44 ( more edge - on in mond ) to @xmath45 ( more face - on in mond ) . for most galaxies the inclination changes are positive , which might be explained through bars . note that in some cases the differences between the mond and the adopted inclinations may be substantial , which means that for these fits the method used to fit the inclination may break down because the change in inclination is large enough to make the rotation curve shape change as well . however , we are mainly interested in investigating the general trends and not to get a measurement of the inclination change in this fit . if the best fit mond inclination change for a particular galaxy is large , this likely indicates that mond can not be made compatible with that rotation curve by changing the galaxy s inclination , because for most galaxies the uncertainty in the inclinations are less than @xmath46 . because @xmath0 is a universal parameter in mond , the mond fits should be made with @xmath0 fixed . however , leaving @xmath0 free provides both a means to measure @xmath0 ( e.g. , begeman et al . 1991 ; sanders & mcgaugh 2002 ) , and a way to test mond . if @xmath0 is truly universal , it should not depend on galaxy properties . to test this , we have made fits with @xmath0 as a free parameter . the best fit @xmath0 for each galaxy is listed in table [ tabfits ] , along with an estimate of the uncertainties , for which the same caveat as mentioned above apply . the best fits are not shown separately in fig . [ figfits ] , because these fits are virtually indistinguishable from the fits with @xmath35 as a free parameter . lake ( 1989 ) reported that the acceleration parameter for his free fits correlated with the amplitude of the rotation curve . [ figa0vsvmax ] we have plotted our best fit @xmath0 versus the maximum rotation velocity @xmath47 . we find no evidence for a correlation between these parameters . interestingly , as can be seen in fig . [ figa0vsmu ] , there does appear to be a correlation between @xmath0 and surface brightness , in the sense that in galaxies with lower surface brightness lower values for @xmath0 are found . taken at face value , the slope of the best fit to these points is -0.32 , i.e. , @xmath0 drops by a factor of two for each magnitude the surface brightness gets fainter . as can be seen in the bottom panels of fig . [ figa0vsmu ] , this trend does not depend on the method with which the distance was determined ( for galaxies with different distance estimates we made multiple fits , but elsewhere only the fit results for the adopted distances as listed in table [ tabfits ] are presented ) . the apparent trend between the acceleration parameter and the central disk surface brightness depends strongly on the galaxies brighter than @xmath48 mag arcsec@xmath2 . we can not rule out that this trend is the result of uncertainties ( see sections [ secuncert ] and [ secdisc ] ) . below we discuss the fits shown in fig . [ figfits ] . unless otherwise indicated , the ` mond curve ' refers to the fit with only @xmath25 as a free parameter . _ ugc 731._ swaters et al . ( 1999 ) showed that this galaxy is lopsided in its kinematics , with a rotation curve that rises more steeply on one side of the galaxy than on the other . the average rotation curve , as presented in figure [ figfits ] is in good agreement with mond , except that in the outer parts mond predicts somewhat higher rotation velocities . _ ugc 3371._ the rotation curve of this galaxy has been derived from a high - resolution h@xmath6 velocity field ( see swaters et al . the mond curve is in good agreement with the observed rotation curve , although it is slightly above the outer few points . _ ugc 4173._ the mond curve based on the alone already predicts significantly higher rotation velocities than are observed . this galaxy has an optical bar , with a faint surrounding disk , making the inclination difficult to determine . an inclination of 25@xmath49 to 30@xmath49 is consistent with the morphology , and would make this galaxy consistent with mond , and also bring it closer to the tully - fisher relation . _ ugc 4325._ the mond curve and the data agree fairly well . the mond curve falls slightly above the observed points in the outer parts . _ ugc 4499._ there is good general agreement between the mond prediction and the observed rotation curve , although there is considerable scatter in derived rotation velocities near the center . _ ugc 5005._ there is good agreement between the data and the mond curve . _ ugc 5414._ the mond curve is in excellent agreement with the rotation curve . _ ugc 5721._ for this compact , high - surface brightness galaxy , the mond curve fails to describe the observed rotation curve . even with the inclination or the distance as a free parameter , no good fit can be found . there is evidence of twisting isophotes in the central regions ( see swaters & balcells 2002 ) , perhaps indicating the presence of a bar . if so , the associated noncircular motions could have affected the derived rotation curve . _ ugc 5750._ the mond curve is in good general agreement with the observed rotation curve . _ ugc 6446_ the mond curve falls below the observed rotation curve in the central parts , and above the rotation curve in the outer parts . if this galaxy is part of the ursa major cluster ( e.g. , tully et al . 1996 ) , the increased distance of 18.6 mpc ( tully & pierce 2000 ) would make this galaxy less compatible with mond . however , this galaxy is on the edge of the boundary taken to define the uma cluster in both velocity and angle on the sky , so it may not be a genuine cluster member . in this paper , we have adopted a distance of 12.8 mpc . _ ugc 7232._ the mond curve and the data agree fairly well . _ ugc 7323._ the mond curve is somewhat higher than the observed curve in the central parts , but there is good general agreement . _ ugc 7399._ for this compact galaxy , the mond curve does not agree with the observed rotation curve , as it predicts a much steeper central gradient than is observed in the rotation curve . the bar in this galaxy may have affected the derived rotation curve . _ ugc 7524._ the mond curve and the observed rotation curve for this well - resolved galaxy agree well , except in the outer points . _ ugc 7559._ the mond curve agrees well with the observed curve in the inner parts , but it predicts significantly higher - than - observed rotation velocities in the outer parts . like ugc 5721 , this galaxy appears to have a bar , and the associated noncircular motions could have affected the derived rotation curve . _ ugc 7577._ the observed rotation curve of this galaxy can be explained by gas and stars alone , even with newtonian gravity . consequently , the mond curve predicts significantly higher rotation velocities in the outer parts . a lower distance and a lower inclination would help to make the mond prediction more compatible with the observed curve . _ ugc 7603._ although the mond curve falls somewhat below the observed rotation curve in the outer parts , there is good general agreement . _ ugc 8490._ the mond prediction for this galaxy falls systematically below the observed flat part of this rotation curve . a larger distance makes this rotation curve more consistent with the mond curve , although the mond curve still falls short of the observed rotation curve near the turnover in the rotation curve . because of the large warp in this galaxy , the shape and amplitude of the rotation curve are uncertain . _ ugc 9211._ the curve predicted by mond and the observed curve are in good agreement . _ ugc 11707._ the rotation curve predicted by mond falls below the inner rotation curve , and above the outer rotation curve . _ ugc 11861._ the mond curve and the observations are in good agreement . _ ugc 12060._ the rotation curve as predicted by mond falls below the observed one in the inner parts of the rotation curve , and is above the observed one in the outer parts . an inclination of @xmath50 in stead of @xmath51 is required for agreement between mond and the observations . such a low inclination is consistent with the optical morphology of this galaxy , but appears inconsistent with the tully - fisher relation . _ ugc 12632._ the rotation curve predicted by mond falls below the observed curve in the inner parts and above the observed curve in the outer parts . _ f568-v1._ there is good agreement between the mond prediction and the observed rotation curve . _ f574 - 1._ except for the one outermost point in the mond prediction , which may well be due to uncertainties in the contribution of the to rotation curve , the mond curve and the observed one are in good agreement . _ f583 - 1._ although the mond curve falls slightly above the observed velocities in the center , and slightly below at intermediate radii , there is good general agreement between the observed rotation curve and the mond prediction . _ f583 - 4._ there is good agreement between the observed and the mond predicted rotation curves . as is clear from the notes on individual galaxies and the fits shown in fig . [ figfits ] , for about a quarter of the galaxies in the sample there are noticeable discrepancies between the observed and the predicted curves , and in a few instances there are significant inconsistencies . on the one hand , these inconsistencies may signal a problem for mond , but , on the other , they might be the result of uncertainties associated with the galaxies in this sample . because mond fits are essentially one - parameter fits , with only the mass - to - light ratio of the stars as a free parameter , it is much more sensitive to observational uncertainties than fits with dark matter halos . for such fits with dark matter halos , any uncertainties in the rotation curve or in the contribution of the stars and gas , for example as a result of uncertainties in distance or inclination , can be accommodated by changing the parameters of the dark matter halo . the mond fits , which are directly tied to the distribution of the gas and stars , can not accommodate these uncertainties . as a result , uncertainties in e.g. , the rotation curves , the distances , or the inclinations can lead to apparently good dark matter fits ( although likely with incorrect halo parameters ) , whereas the same uncertainties can lead to very poor mond predictions of the rotation curves . it is therefore imperative to consider the uncertainties that may play a role for the data presented here before we can discuss the implications of our results for mond . given that the sample presented here consist in large part of nearby dwarf galaxies , uncertainties in distances will likely play a role . for those nearby galaxies , the distances have been determined with different methods : hubble flow , group membership , brightest stars , and tip of the red giant branch . some of these methods can produce uncertain distance measurements . as we have shown in this paper , the mond fits can be improved by adopting different distances . it is likely that in individual cases incorrect distance estimates have contributed to poor mond predictions of the rotation curves . however , it is unlikely that changes in distance alone can explain all the poor predictions , because that would require the galaxies in our sample to be preferentially closer , even galaxies at large distances which presumably have more accurate distances . another important factor that may contribute to uncertainties in the mond fits are the inclinations of the galaxies in our sample . as described in section [ incfits ] , many of the galaxies in our sample have irregular appearances and slowly rising rotation curves , making it difficult to determine their inclinations from either their morphology or their kinematics . moreover , towards more face - on galaxies , intrinsic noncircular shapes may lead to an overestimate of the inclination , and towards higher inclinations the uncertainties in the intrinsic thicknesses of these galaxies may affect the derived inclinations . thus , the inclinations of the galaxies in our sample may be uncertain . we have shown that the mond fits can significantly be improved by adopting different inclinations . moreover , for some galaxies , like ugc 4173 and ugc 12060 , the mond - preferred inclinations not only improve the mond fits , but also are compatible with the galaxies morphology and kinematics . in the case of ugc 4173 , the mond - preferred inclination also brings the galaxy closer to the tully - fisher relation . however , for most galaxies the required changes in inclination are larger than the expected uncertainties . in addition , if changes in inclination would be the main cause of the poor mond predictions of the rotation curves , then the majority of the galaxies in our sample would have to be more face - on than the inclinations reported in table [ tabsample ] . this could be possible if the galaxies in our sample have significant intrinsic ellipticities or strong bars . however , to explain differences of up to @xmath50 even at inclinations of about @xmath52 , we estimate an intrinsic ellipticity of 0.35 is needed . such large intrinsic ellipticities are not seen among late - type disk galaxies ( e.g. , schoenmakers et al . 1997 ; andersen et al . 2001 ) , although they might occur in low luminosity systems ( sung et al . 1998 ) . however , there is little evidence of the perturbations in the velocity fields associated with such strong ellipticities ( schoenmakers et al . 1997 ) in the velocity fields presented in swaters et al . it is therefore unlikely that all the poor mond predictions can be explained by uncertainties in the inclination alone , but it is likely that in some individual cases incorrect adopted inclinations can be the main reason for poor mond predictions . naturally , it is also possible that both the adopted inclination and distance are incorrect . we investigated this in our fitting process but found that in all but a few of our fits , distance and inclination are generally strongly covariant , and in most cases it was not possible to constrain both of these parameters simultaneously . still , when compared to leaving only the distance or the inclination free in the fits , if both are left free , on average smaller changes are required to make the mond - predicted curve more compatible with observed rotation curve . at the same time , the required changes are still systematic , i.e. , on average the galaxies still need to be closer and more face - on . in addition to the uncertainties in distance and inclination , other factors may also play a role . these include beam smearing , noncircular motions , morphological asymmetries , corrections for asymmetric drift , and uncertainties in the photometric calibration . even though the rotation curves presented here have been corrected for beam smearing ( see ssah ) , the inner rotation curve shape may be uncertain simply due to lack of resolution . for many galaxies in our sample high - resolution h@xmath6 rotation curves are available to mitigate this effect , but h@xmath6 rotation curves themselves may be uncertain because they are determined along a one - dimensional slice and not from the velocity field as a whole and hence are more susceptible to noncircular motions . in addition , the rotation curves presented here have not been corrected for asymmetric drift because it was deemed not to be important . another potential factor are morphological and kinematical asymmetries . not only can these affect the derived baryon distributions and rotation curves , they are also incompatible with the underlying assumption of axisymmetry used in calculating the mond - predicted rotation curves . all these factors combined could result in a derived rotation curve that is not representative of the true circular velocity . in addition to observational uncertainties , the quality of the fits may be affected by the fitting process and the associated assumptions . an example is the adopted thickness for the and stellar disks . here , for ease of comparison with other studies , we have assumed that the gaseous and stellar disks are infinitely thin . compared to a stellar disk that is assumed to have a scale height of 0.2 disk scale lengths , as was adopted by swaters ( 1999 ) , the rotation curve calculated for an infinitely thin disk rises more steeply , on average by 15% , but reaching up to 40% for galaxies with central concentrations of light . the amplitude in the thin disk case is about 5 - 10% higher than in the thick disk case . thus , taking a thick disk in stead of a thin disk would results in m / ls that are 10% to 20% higher , and values for @xmath0 that are correspondingly lower . the rotation curve of the is less sensitive to the adopted thickness . the difference between an infinitely thin disk and a disk with a scale height of 0.2 optical disk scale lengths is in general well below 5% , and hence the choice of thickness of the disk has little influence on the fits . another uncertainty on the results is the adopted form of @xmath53 ( see eq . [ eqmond ] ) , that describes the transition between the newtonian and mond regimes . in the results reported in the tables and figures in this paper , we have used the commonly used form given in eq . we have also tried the form : @xmath54 as used by famaey & binney ( 2005 ) and mcgaugh ( 2008 ) . this revised transition formula results in more noticeable deviations from newtonian gravity at higher accelerations . this has an effect on the fits , and we looked in particular in terms of the possible correlation between @xmath0 and surface brightness as shown in fig . [ figa0vsmu ] . when eq . [ eqmu2 ] is used , galaxies that are deeper in the mond regime ( i.e. , the lsb galaxies ) tend to end up with lower values for @xmath0 , whereas the derived values for @xmath0 change little for high surface brightness galaxies . the fact that some of the rotation curves in our sample have been derived from a combination of and h@xmath6 data might have affected our fit results , because the long - slit h@xmath6 data sample only a single slice through the velocity field , making them more sensitive to the effects of noncircular motions . in addition , the higher spatial sampling in the galaxy centers might bias the fits . however , any such bias would , at least partly , be offset by the usually larger uncertainties on the h@xmath6 rotation velocities . even though such a bias could have played a role , we find it does not for the galaxies in our sample . for all galaxies for which we have hybrid -h@xmath6 rotation curves , we have made fits to the -only rotation curves and found no significant differences . this is also illustrated by the fact that there are no systematic differences between the galaxies in our sample for which we have -only rotation curves , and those for which we have hybrid rotation curves . perhaps the biggest uncertainties on the derived parameters are from the fitting process itself . the values reported in table [ tabfits ] are best - fit parameters . however , in some cases , the best - fit values suggest large changes in distance , inclination , or @xmath0 , even though the nominal values provide fits that are nearly as good in a @xmath55 sense ( e.g , ugc 5414 and f583 - 4 ) . in most of these cases , the estimated uncertainties on the derived parameters are large , thereby indicating that a large range in parameter space produces fits of similar quality . in a number of cases , however , the best - fit values for the derived parameters are off from the nominal value by a large amount , even when a visual comparison of the mond fits and the fits with additional free parameters suggests that both fits are comparable in quality . examples of this are ugc 3371 , ugc 11861 , and f574 - 1 . in light of the uncertainties on the rotation velocity uncertainties , as was also mentioned in sec . [ secfitres ] , the best - fit values derived from the fits should not be taken at face value . we have found that the rotation curves as predicted by mond for the sample of dwarf and lsb galaxies presented in this paper are generally in good agreement with the observed rotation curves for roughly three quarters of the sample . this is remarkable , given that mond is a one parameter fit with only m / l as a free parameter . it is even more remarkable , in light of the uncertainties associated with a sample of low surface brightness and dwarf galaxies , as was discussed in more detail in the previous section . given these uncertainties , a fraction of around a quarter of galaxies for which mond does not adequately predict the observed rotation curves does not seem unexpected , and hence the differences between the observed rotation curves and the mond fits may not signal a failure of mond , but rather reflect the uncertainties associated with the galaxies in this sample . despite these uncertainties , there are a number of interesting results from our study . one is that the mond fits appear to deviate in a systematic way from the observed rotation curves . in almost 80% of the galaxies ( 21 out of 27 ) the mond curve predicts higher rotation velocities in the outer parts than are observed . in 40% of the galaxies in our sample , mond predicts rotation velocities that are higher by more than twice the uncertainty in the rotation velocity of the last measured point of the rotation curve . as we have shown , these discrepancies can be explained if these galaxies have different inclinations or distances . to explain the apparent systematic deviations between the mond curves and the observed rotation curves in terms of changes in distance or inclination , the galaxies in our sample need to be preferentially closer , need to have preferentially lower inclinations , or both , which seems unlikely . an alternative explanation for the fact that most mond - predicted curves tend to have higher rotation velocities in the outer parts than is observed , is that the value for @xmath0 is preferentially lower . of course , in mond @xmath0 is a universal constant and should not vary from galaxy to galaxy . in the mond fits presented here , we have adopted @xmath56 km@xmath29 s@xmath2 kpc@xmath10 ( e.g. , sanders & mcgaugh 2002 ) . if we average the @xmath0 derived from our fits with the acceleration parameter free , we find @xmath57 km@xmath29 s@xmath2 kpc@xmath10 , consistent with the value used by sanders & mcgaugh ( 2002 ) . however , as can be seen in [ figa0vsmu ] , most of our fits result in low values for @xmath0 . excluding the three highest values for @xmath0 , we find @xmath58 km@xmath29 s@xmath2 kpc@xmath10 . thus , a somewhat lower value for @xmath0 would , on average , lead to improved fits for the galaxies in our sample . we also have found what appears to be a correlation between @xmath0 and surface brightness . such a trend , if real , would be a problem for mond , because @xmath0 is a universal constant . however , it seems that this correlation is largely caused by a small number of galaxies at the high surface brightness end , for which large values for @xmath0 were found : ugc 5721 , ugc 7232 , ugc 7399 , ugc 7603 , and ugc 8490 . as we noted in sec . [ secindiv ] , ugc 5721 and ugc 7399 are compact galaxies with a central bar , and hence the inner rotation curve could be affected by noncircular motions . ugc 8490 has a strong warp , making the outer rotation curve less certain . ugc 7232 only has four points in its rotation curve and large uncertainties on the fitted value of @xmath0 . considering these factors , it is possible that the apparent correlation is the result of uncertainties , and not of a true underlying relation . previous studies of mond rotation curves that included lsb galaxies in their samples have not reported a trend between surface brightness and @xmath0 . de blok & mcgaugh ( 1998 ) analyzed the rotation curves presented in de blok et al . ( 1996 ) in the context of mond , and found most galaxies to be consistent with mond , although for a few adjustments to distance and inclination had to be made . the sample presented in sanders & verheijen ( 1998 ) contained 12 lsb galaxies , but the authors did not report a trend between @xmath0 and surface brightness . neither does an inspection of the lsb galaxies in their sample show the prediction by mond of systematically faster rotation in the outer rotation curves , like is seen in our sample , except for ugc 6446 , which is a galaxy also included in this sample . however , our sample extends more than a magnitude deeper in surface brightness , making the sample presented here more sensitive to effects that may play a role at low surface brightness . interestingly , it may be possible to explain , within the context of mond , why at least some galaxies would appear to have a low value for @xmath0 when @xmath0 is left as a free parameter in rotation curve fits . in the results presented here , we have only considered the accelerations within the galaxies themselves . however , whether an object is in the mond regime is determined by the overall acceleration , which includes external effects ( milgrom 1983a , b ) . thus , if an object , with internal accelerations that would place it in the mond regime , is placed in a newtonian external acceleration that is larger than @xmath0 , its kinematics would be entirely newtonian . if one were to fit @xmath0 in that case , one would derive a value of zero , because the object would never enter the mond regime . of course , the effects of the external fields are much smaller for the galaxies in our sample . even for the galaxy in our sample with the closest neighbor ( ugc 7577 , near ngc 4449 ) , the external acceleration is estimated to be @xmath59 cm s@xmath2 ( assuming a projected distance of 14.4 mpc ) . even though the amplitude of the external field is difficult to determine because it depends on the details of the galaxy environment , it can only lower the value of @xmath0 in fits where @xmath0 is left free . moreover , any external fields will have the strongest effects in lsb galaxies , where the internal accelerations are lowest . thus , while external fields may contribute to the apparent correlation between surface brightness and @xmath0 , it is difficult to establish this quantitatively . given the uncertainties in the distances , the inclinations , other observational uncertainties , uncertainties in the fitting process , and possible effects of external fields , it seems that the correlation between surface brightness and @xmath0 may not be significant . perhaps more certain is the finding that the galaxies in the sample presented here suggest a lower value for @xmath0 than the adopted value of @xmath60 km@xmath29 s@xmath2 kpc@xmath10 . to thoroughly test mond in the lsb and dwarf galaxy regime , galaxies are needed with reliable distance measurements and inclinations , well - resolved rotation curves and symmetric appearance , both in morphology and kinematics . none of the galaxies in our sample meet all these criteria simultaneously . however , there are four galaxies that meet all but one : ugc 3371 , ugc 6446 , ugc 7524 , and ugc 12632 . ugc 7524 is the only one with a reliable distance measurement from the trgb ( karachentsev et al . 2003 ) , but it has asymmetric kinematics ( swaters et al . we have presented mond fits for a sample of 27 dwarf and lsb galaxies . mond is remarkably successful at predicting the general shape of the rotation curves in this sample . only for approximately a quarter of the galaxies mond does not adequately predict the observed rotation curves . this is remarkable given the uncertainties associated with the galaxies in this sample , especially the uncertainties in distance and inclination . close inspection of the fits indicates that for almost 80% of the galaxies mond predicts higher velocities in the outer rotation curves than is observed . although this could be the result of uncertainties , this systematic deviation can also be explained if the mond acceleration parameter is slightly lower than usually assumed , with a value of @xmath58 km@xmath29 s@xmath2 kpc@xmath10 ( or @xmath61 cm s@xmath2 ) . we find that there appears to be some evidence of a correlation between @xmath0 and the central extrapolated disk surface brightness @xmath62 , but this possible correlation depends heavily on a few galaxies at high surface brightness , whose rotation curves may be uncertain because of bars or warps . at the low surface brightness end , the derived values of @xmath0 could be be biased towards low values if external fields are important . overall , the uncertainties associated with this sample make it difficult to draw strong conclusions . improved distances and inclination estimates for these galaxies could make it possible to test mond more strongly .

there has been much interest in superconductivity with very short coherence length . this interest is due to its possible relevance to high temperature superconductors ( cuprates , doped bismuthates , iron - based systems , fullerenes ) and also to the several other exotic superconducting materials ( for a review , see @xcite and references therein ) . it can also give relevant insight into behaviour of strongly bounded fermion pairs on the optical lattices . the phase separations involving superconducting states have been evidenced in a broad range of currently intensely investigated materials including iron - pnictides , cuprates and organic conductors ( see for example @xcite and references therein ) . in our work we will study a model which is a simple generalization of the standard model of a local pair superconductor with on - site pairing ( i. e. the model of hard core bosons on a lattice @xcite ) to the case of finite pair binding energy . the hamiltonian considered has the following form : @xmath4 where , , ; @xmath5 ( @xmath6 ) denotes the annihilation ( creation ) operator of an electron with spin at the site @xmath7 , which satisfy canonical anticommutation relations @xmath8 where @xmath9 is the kronecker delta . indicates the sum over nearest - neighbour sites @xmath7 and @xmath10 independently . @xmath11 will denote the number of nearest neighbours . @xmath3 is the chemical potential , depending on the concentration of electrons : @xmath12 with and @xmath13 is the total number of lattice sites . there are two interaction parameters of the model : ( i ) the intersite pair hopping interaction @xmath1 , determining the electron pair mobility and responsible for the long - range superconducting order in the system and ( ii ) the on - site density - density interaction @xmath0 , which contributes ( together with @xmath1 ) to the pair binding energy by reducing ( ) or enhancing ( ) its value . to simplify our analysis we do not include in hamiltonian ( [ row : ham1 ] ) the single electron hopping term ( @xmath14 ) as well as other inter - site interaction terms . this assumption corresponds to the situation when the single particle mobility is much smaller than the pair mobility and can be neglected . in the presence of the @xmath14 term , model ( [ row : ham1 ] ) is called the penson - kolb - hubbard ( pkh ) model @xcite and the effects of will be pointed out in section [ sec : conclusions ] . model ( [ row : ham1 ] ) can be considered as a relatively simple , effective model of a superconductor with local electron pairing . moreover , the knowledge of the limit can be used as starting point for a perturbation expansion in powers of the hopping @xmath14 and provides a benchmark for various approximate approaches analyzing the corresponding finite bandwidth models . in terms of pseudospin charge operators hamiltonian ( [ row : ham1 ] ) takes the form of a special xy - model with a single ion anisotropy in a transverse field ( in the @xmath11-direction ) @xmath15 where and corresponds to the external magnetic field . the operators satisfy standard commutation relations for momentum operators : @xmath16 = \pm \hat{\rho}^\pm_i \delta_{ij } , \qquad [ \hat{\rho}_i^{+ } , \hat{\rho}_j^{- } ] = 2\hat{\rho}_i^{z}\delta_{ij},\ ] ] what follows directly from ( [ row : antycomutation ] ) . however , one should stress that takes on four values , as for model with doubly degenerated @xmath17 value , whereas , as in the system . condition ( [ row : condn1 ] ) for the electron concentration reads : @xmath18 it means that magnetic field @xmath19 is determined by a fixed value of magnetization @xmath20 . the interactions @xmath0 and @xmath1 will be treated as the effective ones and will be assumed to include all the possible contributions and renormalizations like those coming from the strong electron - phonon coupling or from the coupling between electrons and other electronic subsystems in solid or chemical complexes @xcite . in such a general case arbitrary values and signs of @xmath0 and @xmath1 are important to consider . it is notable that formally @xmath1 is one of the off - diagonal terms of the coulomb interaction @xcite , describing a part of the so - called bond - charge interaction , and it sign of the _ coulomb - driven _ charge exchange is typically negative ( repulsive , ) . however , the effective attractive of this form ( ) is also possible @xcite and in particular it can originate from the coupling of electrons with intersite ( intermolecular ) vibrations via modulation of the hopping integral @xcite , or from the on - site hybridization term in generalized periodic anderson model @xcite . the ferromagnetic xy order of pseudospins @xmath21 ( for ) corresponds to the ss phase ( @xmath22-pairing superconducting ) , whereas the antiferromagnetic xy order ( for ) to the @xmath23s phase ( @xmath23-pairing superconducting ) . however , in the absence of the external field in the xy - direction conjugated with the order parameter ( , for and , for , @xmath24 is a half of the smallest reciprocal lattice vector ) there is a symmetry between the ferromagnetic and antiferromagnetic cases ( except for an obvious redefinition of the order parameter ) for lattices consisting of two interpenetrating sublattices ( such as for example sc or bcc lattices ) , thus we restrict ourselves to the case in the following . let us stress that the symmetry is broken by the single electron hopping term ( cf . section [ sec : conclusions ] ) . the first analysis of the phase diagram of model ( [ row : ham1 ] ) have been performed by bari @xcite and ho and barry @xcite using the variational method in order to examine the instability of the mott insulator to superconductivity for the special case of the half - filled band ( ) . the effects of diagonal disorder on the critical temperature for and have been also determined @xcite , arriving at a satisfactory qualitative interpretation of quite a number of different experiments in amorphous superconductors . in the analysis we have adopted the variational approach ( va ) which treats the on - site interaction @xmath0 exactly and the intersite interactions @xmath1 within the mean - field approximation ( mfa ) . we have also derived exact ground state results for a linear chain ( ) . moreover , at several results obtained within the random phase approximation ( rpa ) and the spin - wave approximation ( swa ) ( for and lattices ) and within the systematic low density expansions ( lde ) ( for lattices ) are presented . within the va the phase diagrams of model ( [ row : ham1 ] ) have been investigated in @xcite as a function of the electron concentration @xmath2 , but the stability conditions of states with phase separation have not been discussed . in this paper , we investigate the thermodynamic properties of model hamiltonian ( [ row : ham1 ] ) for arbitrary chemical potential and arbitrary electron concentration . we focus on the states with phase separation , which have not been analyzed till now . arbitrary values of on - site interaction @xmath0 are considered and the range of pair hopping interaction @xmath1 is restricted to the nearest neighbours . our investigation of the general case finds that , depending on the values of the interaction parameters and the electron concentration , the system can exhibit the superconducting ordered and nonordered homogeneous phases as well as the phase separation between them . transitions between various states and phases can be continuous and discontinuous , what implies existence of tricritical points on the phase diagrams . we present detailed results concerning the evolution of phase diagrams and thermodynamic properties as a function of interaction parameters , chemical potential @xmath3 and electron concentration @xmath2 . by taking into account the phase separated states and presenting several new results for lattices of various dimensions derived beyond va ( rigorous solution for a chain ( ) , rpa for sq and sc lattices and lde in ) our paper substantially extends and generalizes the results and conclusions of previous papers concerning the model considered @xcite . the paper is organized as follows . in section [ sec : method ] we describe the method used in this work . there are also derived explicit formulas for the free energies of homogeneous phases and states with phase separation as well as equations determining the superconducting order parameter and the chemical potential in homogeneous phases obtained in va . in section [ sec : gs ] we analyze the properties of the system at zero temperature and present exact ground state results for a chain ( section [ sec : gsexact ] ) , diagrams obtained in the va ( section [ sec : gsmfa ] ) as well as results obtained in some approximations going beyond va for lattices ( section [ sec : rpa ] and section [ sec : lde3d ] ) . section [ sec : kt ] is devoted to the study of the finite temperature phase diagrams derived within the va as a function of @xmath3 ( section [ sec : ktmi ] ) and as a function of @xmath2 ( section [ sec : ktn ] ) . some particular temperature dependencies of the thermodynamic parameters are discussed in section [ sec : properties ] . finally , section [ sec : conclusions ] is devoted to conclusions and supplementary discussion . the appendices report some relevant details : explicit expressions of site - dependent self - consistent va equations ( [ app : sitedependentequations ] ) , particle - hole transformations ( [ app : sym ] ) , and selected analytical va results ( [ app : anal ] ) . the free energy of the system as well as the self - consistent equations for the average number of electrons in the system and the average value of the order parameter are derived within site - dependent va in appendix [ app : sitedependentequations ] . restricting our analysis to the nearest neighbours and assuming no spatial variations of the order parameter the grand potential per site obtained in va is given by : @xmath25 where @xmath26 and @xmath27 . the free energy per site : is derived as @xmath28 the grand potential @xmath29 is minimized in the system considered for fixed @xmath3 , whereas the free energy @xmath30 takes the minimum when @xmath2 is fixed . the condition for electron concentration ( [ row : condn1 ] ) and a minimization of @xmath29 ( or @xmath30 ) with respect to the superconducting order parameter @xmath31 lead to the following self - consistent equations ( for homogeneous phases ) : @xmath32.\ ] ] we also introduce the concentration of locally paired electrons and the ratio . notice that @xmath33 is different from the condensate density ( a fraction of pairs in the condensate ) @xmath34 . these quantities differ even in the va at : and in the superconducting phase . equations ( [ row : mfa1])([row : mfa2 ] ) are solved numerically for and we obtain @xmath31 and @xmath2 when @xmath3 is fixed or @xmath31 and @xmath3 when @xmath2 is fixed . the superconducting phase ( ss ) is characterized by non - zero value of @xmath31 , whereas in the non - ordered ( normal ) phase ( no ) . equation ( [ row : mfa2 ] ) is only the necessary condition for an extremum of ( [ row : grandpotential ] ) ( or ( [ row : freeenergy ] ) ) thus the solutions of can correspond to a minimum or a maximum ( or a point of inflection ) of @xmath29 ( or @xmath30 ) . in addition , the number of minima can be larger than one , so it is very important to find the solution which corresponds to the global minimum of ( [ row : grandpotential ] ) ( or ( [ row : freeenergy ] ) ) . we say that the solution of ( [ row : mfa1])([row : mfa2 ] ) corresponds to a metastable phase if it corresponds to a minimum of the free energy @xmath30 or grand potential @xmath29 and the stability condition @xmath35 is fulfilled . otherwise , we say that the phase is unstable . a stable ( homogeneous ) phase is a metastable phase with the lowest free energy ( among all metastable phases and the phase separated states described below ) . phase separation ( ps ) is a state in which two domains with different electron concentration : @xmath36 and @xmath37 exist in the system ( coexistence of two homogeneous phases ) . the free energies of the ps states are calculated from the expression : @xmath38 where @xmath39 are values of a free energy of two separating phases at @xmath40 corresponding to the lowest homogeneous solution for a given phase , @xmath41 is a fraction of the system with a charge density @xmath36 , is a fraction with density @xmath37 ( ) and . the minimization of ( [ row : freeenergyps ] ) with respect to @xmath36 and @xmath37 yields the equality between the chemical potentials in both domains : @xmath42 ( chemical equilibrium ) and the following equation ( so - called maxwell s construction ) : @xmath43 which is equivalent with equality of grand potentials per site in domains : . it implies that the transitions with a discontinuous change of @xmath2 in the system considered for fixed @xmath3 can lead to occurrence of the regions of phase separation in the concentration range on the diagrams obtained as a function of @xmath2 . in these regions the homogeneous phases can be metastable as well as unstable , depending on the @xmath2-dependence of @xmath3 ( cf . ( [ row : stabcondition ] ) ) . in the ps states the chemical potential is independent of the electron concentration , i.e. . for more detailed discussion of maxwell s construction we refer the reader to existing literature , e. g. @xcite . in the model considered only one type of ps states can occur , which is a coexistence of ss and no phases . in the paper we have used the following convention . a second ( first ) order transition is a transition between homogeneous phases with a ( dis-)continuous change of the order parameter at the transition temperature . a transition between a homogeneous phase and the ps state is symbolically named as a `` third order '' transition . at this transition a size of one domain in the ps state decreases continuously to zero at the transition temperature . second order transitions are denoted by solid lines on phase diagrams , whereas dotted and dashed curves denote first order and `` third order '' transitions , respectively . the phase diagrams obtained are symmetric with respect to half - filling because of the particle - hole symmetry of hamiltonian ( [ row : ham1 ] ) , so the diagrams will be presented only in the range and . the transformation formulas of various thermodynamic parameters and quantities such as electron concentration , superconducting order parameter , chemical potential and free energy under the electron - hole transformation are collected in appendix [ app : sym ] . the numerical computations have been done in c / c++ with source codes from @xcite . the minima have been found by using brent s method in one dimension and downhill simplex method in multidimensions . we have also implemented newton - raphson method using derivative to solve non - linear equations . let us consider the ground state ( gs ) of model ( [ row : ham1 ] ) . in the absence of @xmath14 term , the parity of the number of particles at each site is a good quantum number . thus , one can decompose the eigenspace of @xmath44 into sectors specified by the parity of the occupation number at each site ( @xmath45 ) . within any sector , the system is a collection of _ even _ segments , within which pairs are confined ( the subspace with excluded singly occupied sites ) interleaved by _ odd _ segments ( the subspace with excluded doubly occupied sites ) . since , the interfaces between the _ odd _ and _ even _ regions will cost energy , so in the thermodynamic limit the ground state will be either a single infinitely large _ even _ segment or a single _ odd _ segment for fixed @xmath3 . only in the case when the grand potentials per site ( it is the chemical potential ) of the segments are equal there is possibility of coexistence of both types of segments if it appears as a first order line on the gs phase diagram . the situation is different when @xmath2 is fixed . in such case the ps states can occur in definite ranges of @xmath2 . the reason for this can be explained as the following . let the energy of a single boundary ( interface ) between _ odd _ and _ even _ segments be . if the number of intersite boundaries will increase as @xmath46 ( where ) the contribution from them to the total free energy disappears in the thermodynamic limit . within the subspace excluding the double occupancy of sites ( by electrons for or holes for ) the product for any , i. e. the @xmath1 term does not contribute . effective hamiltonian ( [ row : ham1 ] ) for this case takes the form and the calculations of the gs grand potential and the gs energy are quite obvious . as a result in the limit from ( [ row : freeenergy ] ) and ( [ row : mfa1 ] ) one obtains for fixed @xmath2 in the no phase : @xmath47 this gs is infinitely degenerate for . for ( ) , @xmath48 sites are occupied by electrons ( holes ) and the spins of all particles are independent of each other . inclusion of the intersite spin - spin and density - density interactions can remove this degeneracy and generate various magnetic and charge orderings . the magnetic moments are not suppressed in this state . for there is no order pattern in the distribution of the particles , whereas for mott phase with one - electron - per - site configuration ( with random direction of electron spin per site ) occurs . from ( [ row : grandpotential ] ) and ( [ row : mfa1 ] ) for fixed @xmath19 and one obtains : @xmath49 the average occupancy of sites can only be integer for fixed @xmath19 , what can be intuitively understand if one realizes that only and every site is equivalent ( there is no interactions between different sites in @xmath50 ) . the state @xmath51 ( no empty ) is stable with respect to @xmath52 ( no mott ) if ( and ) , whereas if ( and ) . within the subspace excluding the single occupancy of sites the eigenvalues of charge operators are , for each site . thus , from ( [ row : ham2 ] ) , the effective hamiltonian for this case takes the form : @xmath53 with auxiliary condition ( [ row : condn2 ] ) . @xmath54 has the form of the xy quantum spin model ( ) with an effective external field in the @xmath11-direction , such that the average magnetization has a fixed value equal to ( cf . ( [ row : condn2 ] ) ) . the gs energy of hamiltonian ( [ row : ham3xy ] ) is derived as @xmath55 where . the gs grand potential in this case has the form @xmath56 where it is underlined that @xmath19 is an independent variable . as it follows from previous analyzes @xcite the gs of model ( [ row : ham3xy ] ) exhibits the long - range xy order for dimensional lattice in the whole concentration range . for , only the short range xy correlations are developed @xcite , nevertheless even arbitrary small interchain interactions can induce the real long - range order also in this case . the ss phase will be realized if for fixed @xmath2 ( if for fixed @xmath19 ) . for explicit calculation of the phase boundary between the no ( eqs . ( [ row : gsexe1 ] ) and ( [ row : gsexomega1 ] ) ) and the ss ( eqs . ( [ row : gsexe2 ] ) and ( [ row : gsexomega2 ] ) ) phases and determination of the gs properties of the ss phase for various dimensions and the electron concentration or the chemical potential one has to find the solution of model ( [ row : ham3xy ] ) at . let us stress that if @xmath2 is fixed one has to consider also the ps state in which one domain is in the no phase and the other is in the ss phase and which energy is calculated from ( [ row : freeenergyps ] ) . however , the gs energy of the ps ( ) state can be obtained in the most cases only numerically by minimization of the free energy given by ( [ row : freeenergyps ] ) . in the no domain ( mott phase with one - electron - per - site configuration ) , whereas concentration in the ss domain is dependent on @xmath57 ( and approach used ) . + the exact solutions of model ( [ row : ham3xy ] ) for arbitrary @xmath2 can be derived for and . for , by making use the exact results for the ground state energy of xy model in a transverse field @xcite one obtains : @xmath58 where ( because ) . notice , that due to a well known isomorphism between the planar ferromagnet and antiferromagnet for loose packed ( alternating ) lattices with nearest - neighbour @xcite , the results for @xmath59 are the same for both signs of @xmath1 . thus , from ( [ row : gsexe2 ] ) and ( [ row : energyxy1d ] ) we have @xmath60 and the exact expressions for chemical potential @xmath3 and the superfluid density @xmath61 ( the short - range order parameter ) are : @xmath62 where @xmath63 denotes the nearest neighbour of the @xmath7-site . for ( ss case ) ferromagnetic correlations between xy components of neighbouring pseudospins are present , whereas for ( @xmath23s case ) antiferromagnetic correlations occur . from ( [ row : missexgs ] ) one obtains that the ss can exist in the range . notice that stability condition ( [ row : stabcondition ] ) is always fulfilled in the ss phase . the gs phase diagrams obtained are shown in figure [ rys : gsdiagrams ] : as a function of @xmath19 ( panel ( a ) ) , and as a function of @xmath2 without ( panel ( b ) ) and with ( panel ( c ) ) consideration of the ps state . at half - filling there is no phase separation and the no - mott phase is stable for . the boundaries between different states corresponding to the exact results in are denoted by dotted lines ( if they are different from those obtained in va ) . notice that for homogeneous phases in the ranges of the ps state occurrence . for there are no exact solutions for model ( [ row : ham3xy ] ) available ( except of the lde for see @xcite ) and the approximate approaches are necessary to find @xmath64 . the simplest of these approaches is the va which at gives ( the limit of ( [ row : freeenergy])([row : mfa2 ] ) for ) @xmath65 equation ( [ row : missvags ] ) implies that the ss phase can exist in the range . notice that , which means that all electrons in the system are locally paired in the ss phase at , whereas the finite value of the depletion ( ) at is due to the hard - core effect @xcite . the gs energy of the ps state can be obtained by minimization of ( [ row : freeenergyps ] ) . in the no domain ( mott phase ) , whereas concentration in the ss domain is dependent on @xmath57 and ( ) . it corresponds to the first order boundary on the @xmath57 vs. @xmath66 gs diagram determined by equation . the gs phase diagrams obtained in va are shown in figure [ rys : gsdiagrams ] . the va phase boundaries are denoted by solid lines . similarly as in section [ sec : gsexact ] one finds that for homogeneous phases in the ranges of the ps state occurrence . for hypercubic lattice of the dimension the va becomes the exact theory for model ( [ row : ham1 ] ) ( as well as for model ( [ row : ham3xy ] ) ) . going beyond va for a reliable approach is the self - consistent rpa @xcite . this approach has been proven to be a very good approximation scheme in problems of quantum magnetism and it fully takes into account quantum fluctuations , which can be of crucial importance for the considered system for . the gs energy of model ( [ row : ham3xy ] ) is given by ( [ row : gsexe2 ] ) , where . @xmath67 and @xmath68 denote the space - fourier transform of the charge operators and the pair hopping integral , respectively , and the chemical potential is determined by the energy @xmath69 is obtained by means of green s functions using the spectral theorem , and by applying the approximation for the longitudinal correlation functions @xcite . at this yields @xmath70 \\ & - & ( 1+\cos^2\theta)\frac{r}{2n}\sum_{\vec{k } } i_{\vec{k } } \frac{a_{\vec{k}}}{e_{\vec{k } } } , \nonumber\end{aligned}\ ] ] where , , is the collective excitation spectrum , , @xmath71 is given by and the length of pseudospin @xmath72 is given as a solution of the self - consistent equation @xmath73.\end{aligned}\ ] ] the result corresponding to the swa ( being the lowest order expansion of above equation ) is with @xmath74 determined by ( [ row : rpapsi ] ) . for the ss phase the spectrum @xmath75 is gapless and linear in @xmath76 for small @xmath77 . one gets with the sound velocity @xmath78 given by ( ) . the chemical potential in the ss phase in the rpa is given by , which is the same as in va . the quantum corrections to this va result for @xmath79 can be evaluated as in @xcite and one obtains : @xmath80,\ ] ] where , . one should notice that rpa and swa give very similar result for the model ( [ row : ham3xy ] ) in the gs @xcite . the resulting phase boundaries for square ( sq ) lattice and simple cubic ( sc ) lattice are shown in figure [ rys : gsdiagrams ] and denoted by dashed and dashed - dotted lines , respectively . in @xcite several rigorous results concerning the gs properties of the model of hard - core charged bosons have been derived for various lattices in using a systematic lde based on the knowledge of the exact two - body scattering amplitude . transforming the results of section vi b from @xcite into our problem one gets for the following exact expressions . the gs energy of ss phase @xmath81 is given by @xmath82 , \nonumber\end{aligned}\ ] ] where is the exact scattering length for the considered case , is the watson integral for a given lattice , , @xmath83 is the volume of the unit cell , @xmath84 is the lattice constant . in particular , for sc : , , ; for bcc : , , ; and for fcc : , , . the chemical potential has a form @xmath85\ ] ] and the sound velocity @xmath78 is derived by @xmath86.\ ] ] the phase diagrams obtained within lde are in very good agreement with rpa results ( for sc lattice ) in the range . the homogeneous ss phase boundary derived within lde is moved little downwards in comparison with the rpa ones . an analysis of lde results obtained for various cubic lattices in shows that the region of the ss phase stability shrinks with increasing number of the nearest neighbours in the lattice . for increasing values of @xmath57 ( numbers above the curves ) . solid and dotted lines indicate second order and first order transitions , respectively . the dashed - dotted curve ( labeled as tcp ) is a projection of the tricritical point line on the @xmath87 vs. @xmath66 plane.,scaledwidth=49.0% ] and @xmath57 . solid and dotted lines indicate second order and first order transitions , respectively . the tricritical point line is denoted by dashed - dotted line.,scaledwidth=49.0% ] the general structure of the gs phase diagrams derived is similar for all lattice dimensionalities ( ) and approaches used . there are two sequences of phase transitions with increasing @xmath57 for fixed @xmath2 : * ss @xmath88 ps @xmath88 no ( for ) , * ss @xmath88 no ( mott ) ( for ) . the effects of quantum fluctuations increase with decreasing lattice dimensionality and for a linear chain their influence on phase diagram is the most prominent . thus the region of the homogenous ss phase stability decreases with increasing @xmath89 . in the quantum fluctuations are suppressed . one should also notice that the transition ps - no at is independent of @xmath2 and dimensionality of the lattice . the ps region expands with increasing dimensionality . + all phase transition boundaries , necessary to construct the complete phase diagram for fixed @xmath3 , have been obtained numerically by solving the set of equations ( [ row : mfa1])([row : mfa2 ] ) and comparing the grand potential ( [ row : grandpotential ] ) for the solutions found . the resulting @xmath87 vs. @xmath66 phase diagrams obtained for various fixed values of the on - site interaction @xmath57 are shown in figure [ rys : psktvsmi ] . the overall behaviour of the system as a function of @xmath19 is demonstrated in figure [ rys : psktvsmi3d ] . for only the second order transitions occur with increasing temperature . the transition temperature is maximal for , and it decreases monotonically with increasing @xmath57 and @xmath90 . the most interesting is the range . in this range the of the transition temperature becomes non - monotonic and with increasing @xmath57 the maximum of the transition temperature moves towards higher values of @xmath90 . in this range of the on - site interaction , the order of the transition changes from second order to first order with increasing of @xmath66 . a _ tricritical point _ ( tcp ) is connected with the change of transition order . for only the no phase is stable at any . as it will be showed in section [ sec : ktn ] the first order phase boundaries are associated with the existence of the ps states in definite ranges of @xmath2 if the system is considered for fixed @xmath2 . the transition temperatures have been determined by self - consistent solving of and and comparing free energies ( [ row : freeenergy ] ) and ( [ row : freeenergyps ] ) . the concentration dependencies of ( i ) the second order transition temperatures and ( ii ) the `` third order '' transition temperatures can be also easily obtained from the diagrams as a function of @xmath19 , because of a simple relation between the electron concentration and the chemical potential in ss phase . the transitions ( i ) and ( ii ) for fixed @xmath2 correspond to the second order and first order transitions for fixed @xmath3 , respectively . however , there is no simple correspondence for the transitions ( cf . ( [ row : mino ] ) ) . examples of the @xmath87 vs. @xmath2 phase diagrams evaluated for various values of the on - site interaction @xmath57 are shown in figure [ rys : psktvsn ] . for the ps states do not occur and the obtained phase diagrams are the same as those derived in @xcite . the transition between homogeneous ss and no phases taking place with increasing temperature is second order for arbitrary @xmath2 . the transition temperature is maximal for , and it decreases monotonically with increasing @xmath57 and decreasing @xmath2 ( for ) cf . figure [ rys : psktvsn]a . in the range with increasing @xmath57 the maximum of the transition temperature moves towards lower concentrations ( for ) . in a definite range of @xmath57 and @xmath2 the ps state is stable . for the ps state extends from the ground state ( cf . figures [ rys : psktvsn]d , e , f and figure [ rys : gsdiagrams]c ) , whereas for it is stable only at finite temperatures ( cf . figures [ rys : psktvsn]b , c ) . the critical point for the phase separation ( denoted as @xmath91 , which is a _ tricritical point _ , tcp ) lies on the end of the second order line . between and only the no phase is stable at any and there are no transitions . for given value of @xmath2 the @xmath87 vs. @xmath57 phase diagrams of the system are shown in figure [ rys : psktvsu ] . all of the transition temperatures decrease with increasing @xmath57 . in the range the diagrams are rather similar and they consist of three regions ( exemplary diagrams are shown in figures [ rys : psktvsu]a , b ) . the regions of ss and no phases stability are separated by second order boundary or , for larger on - site repulsion , by the region of ps state occurrence . for half - filling ( , ) only homogeneous phases are present on the diagram ( figure [ rys : psktvsu]c ) . the @xmath91-point , connected with the change of the transition order , is located at and ( for ) . the possible sequences of transitions with increasing temperatures and the transition orders of them are listed below : * ss@xmath88no : second order for and second order or first order for ( and ) , * ps@xmath88no : third order , it can take place only for ( and ) , * ss@xmath88ps@xmath88no : both third order , it can take place only for ( and ) . notice that increasing temperature usually suppresses heterogeneity in a system what favours occurrence of homogeneous phases rather than ps states . in the model considered the ps state can develop at higher temperatures than the homogeneous ss phase . this rather unusual behaviour can be connected with the interplay between the pair hopping interaction and the on - site repulsion . in this section we present several representative dependencies of the thermodynamic characteristics for fixed model parameters ( figures [ rys : homoprop][rys : seppropfunkn ] ) . in particular for fixed @xmath19 ( figure [ rys : homoprop ] ) , one can single out two limiting types of thermodynamic behaviour near transition temperature @xmath92 : ( i ) the _ local pair regime _ and ( ii ) the _ pair breaking regime_. in between , there is a crossover between the two regimes ( see e. g. the plots @xmath93 vs. @xmath87 in figure [ rys : homoprop]b ) . for large on - site attraction the concentration of locally paired electrons exhibits no sharp feature as the temperature is lowered through @xmath92 . the number of non - paired electrons at @xmath92 is negligible and the transition is to the state of dynamically disordered pairs ( only for and , the _ local pair regime _ ) . in the second limit , for on - site repulsion , @xmath94 has a sharp break at @xmath92 and a substantial fraction of single particles can exists above @xmath92 . we call this the _ pair breaking regime_. as temperature is lowered , the condensate growths both from a condensation of pre - existing pairs and from binding and condensation of single particles . for small binding energies , if ( @xmath95 ) , there will be essentially no pre - formed pairs at @xmath92 . notice that non - zero value of @xmath94 does not imply that local pairs are in coherent state and even significant values of @xmath93 are possible in the no phase . in the limit @xmath94 increases to ( each of four states at a given site can be occupied with equal probability ) . the condensate density ( which can be approximated as at least for , ) vanishes for , but the doubly occupied states are still thermally excited above @xmath92 ( ) . in figure [ rys : homoprop]a the temperature dependencies of the superconducting order parameter @xmath96 are presented , where one can see clearly the discontinuous change of the order parameter ( lines c and g ) . the other lines correspond to second order transitions . finally , let us briefly summarize the behaviour of the specific heat at constant volume ( figure [ rys : homoprop]c ) . the no phase is characterized by the relatively broad maximum in @xmath97 connected with continuous changes in a short - range electronic ordering ( in higher temperatures , not shown in figure [ rys : homoprop ] ) . the narrow peak in @xmath98 is associated with the first order transition , while the @xmath99-point behaviour is typical for the second order transition . let us concentrate now on the thermodynamic properties of the system in the ps state and their changes at the `` third order '' transitions for fixed electron concentration @xmath2 ( figures [ rys : seppropfunkkt ] and [ rys : seppropfunkn ] ) . as an example , we consider the temperature dependencies of : ( i ) the specific heat , ( ii ) the entropy , ( iii ) the concentration of paired electrons @xmath93 , ( iv ) the chemical potential @xmath66 , and ( v ) the fractions and ( sizes of domains ) with a charge density @xmath36 and @xmath37 ( figure [ rys : seppropfunkkt ] ) as well as the concentration dependencies of : ( a ) @xmath66 , ( b ) @xmath100 and ( c ) @xmath93 ( figure [ rys : seppropfunkn ] ) . thermodynamic characteristics for three different possible sequences of transitions with increasing temperature are shown in figure [ rys : seppropfunkkt ] . the label ( a ) corresponds to the case of transition . at such a transition the specific heat @xmath97 exhibits a finite jump at transition temperature , what is associated with the continuity of the entropy @xmath22 ( cf . figure [ rys : seppropfunkkt]a ) . the lines labelled by ( b ) correspond to case . at the transition there is also a finite jump in @xmath97 , however the @xmath97 in the ps state is larger than that in the ss phase , contrary to the transition , where the jump has an opposite sign . finally , the lines ( c ) correspond to the case of second order ss@xmath88no transition . the concentration ratio of locally paired electrons @xmath94 is presented in figure [ rys : seppropfunkkt]b . on the inset of figure [ rys : seppropfunkkt]b the chemical potential is shown as a function of temperature . one can see that @xmath19 in the ss phase is independent of @xmath101 ( cf . ( [ row : miss ] ) ) . in figure [ rys : seppropfunkkt]c there are shown dependencies of the fractions of the system @xmath102 and @xmath103 with charge density @xmath36 and @xmath37 , respectively . for instance , in the case ( a ) the ss domain with lower concentration ( @xmath37 ) vanishes at `` third order '' transition . the chemical potential @xmath66 and the free energy @xmath100 as a function of concentration for and are presented in figure [ rys : seppropfunkn]a and figure [ rys : seppropfunkn]b , respectively . for the ss phase is stable and its free energy is the lowest one . in the range the ps state has the lowest energy ( solid line ) . the solutions of corresponding to the lowest energy ones ( the homogeneous phases with the lowest energy ) are thermodynamically stable i. e. @xmath104 in their existence ranges ( denoted by dashed lines in figures [ rys : seppropfunkn]a , b ) . at point labelled as @xmath105 the transition between metastable phases occurs , eg . for the free energy @xmath106 of the ss phase is lower than the energy @xmath107 of the no phase , whereas for . the dotted lines in figures [ rys : seppropfunkn]a , b correspond to the homogeneous phases with the highest energy . one should notice that in the described case for and in the whole range of the ps state occurrence two homogeneous phases are metastable , whereas in the ranges of the homogeneous phases stability there is one metastable phase at the most . the linear dependence is characteristic for the ss phase , whereas in the ps state @xmath66 is independent of @xmath2 . this is consistent with square dependence of @xmath106 and linear dependence of @xmath108 as a function of @xmath2 . the anomalous decrease of the chemical potential with doping in the lowest energy metastable homogeneous phases near @xmath105-point ( i. e. an abrupt drop at @xmath105-point ) is a signal that it is thermodynamically more convenient to phase separate into two subsystems with the different electron densities @xmath37 and @xmath36 ( figure [ rys : seppropfunkn]a ) . the density range for which the ps occurs can be readily found by the maxwell s construction described in section [ sec : method ] . the @xmath2-dependencies of the chemical potential @xmath66 , the free energy @xmath100 and the concentration of paired electrons are shown on the insets of figure [ rys : seppropfunkn]a and figure [ rys : seppropfunkn]b and in figure [ rys : seppropfunkn]c , respectively , for at several fixed temperatures . they are obtained for : ( a ) , ( b ) , ( c ) , and ( d ) . at `` third order '' transitions ( ss - ps and ps - no ) the thermodynamic parameters such as @xmath19 , @xmath93 , @xmath22 , @xmath30 are continuous , whereas the @xmath97 exhibits a finite jump as at the second order transitions . however , one should notice that the order parameter for `` third order '' transitions is the concentration difference ( not @xmath96 , which is the order parameter in one domain ) and its change is discontinuous at transition temperature . such transitions are present if the system is consider for fixed @xmath2 and they are associated with first order transitions at fixed @xmath19 . we have studied a simple model of a superconductor with very short coherence length ( i. e. with the pair size being of the order of the radius of an effective lattice site ) and considered the situation where the single particle mobility is much smaller than the pair mobility and can be neglected . we have evaluated the phase diagrams and thermodynamic characteristics of the model at fixed @xmath3 and at fixed @xmath2 and determined the ranges of the ps occurrence in the system . one should stress that within the va the on - site @xmath0 term is treated exactly . thus , the major conclusions of our paper concerning the evolution of the properties of the system with @xmath0 are reliable for arbitrary @xmath0 . the mean - field used for the intersite term is best justified if the @xmath109 interactions are long - ranged or if the number of nearest neighbours is relatively large . the derived va results are exact in the limit of infinite dimensions , where the mfa treatment of the intersite interaction @xmath1 term becomes the rigorous one . moreover , the va yields exact results ( in the thermodynamic limit ) for @xmath109 of infinite range ( for any @xmath110 ) regardless of the dimensionality of the system @xcite . the properties of the system are strongly dependent on the ratio of @xmath111 , which is related to the relative values of the pair binding energy @xmath112 and the effective pair mobility @xmath113 . for the case of a single pair in a system , i. e. for , we have , ( the pair effective mass is ) @xcite and increasing @xmath2 can modify @xmath112 . there are two well defined limits of the model : ( i ) the _ local pair regime _ and ( ii ) the _ pair breaking limit_. in between there is a crossover between the two regimes for fixed @xmath3 . the _ local pair limit _ is realized for large pair binding energy ( i. e. for ) . in such a case , the two interactions : the on - site interaction @xmath0 and the intersite pair hopping @xmath1 cooperate and transition temperature is determined by pair mobility ( i. e. by @xmath1 ) . the transition with increasing temperature is second order and is to the no phase being a state of dynamically disordered local pairs . the opposite regime i. e. the _ pair breaking limit _ is realized for substantial values of . repulsive @xmath0 destabilizes on - site pairing and competes with @xmath1 . as we have shown above , the increasing @xmath0 : ( i ) for fixed @xmath3 it changes first the nature of phase transitions from a continuous to a discontinuous type and then it suppresses superconductivity for , ( ii ) it stabilizes the phase separation state ( ) in definite ranges of @xmath2 and temperature . this behaviour is associated with the presence of the tricritical point on the phase diagrams . with increasing @xmath57 the maximum of the transition temperature moves towards lower concentrations ( for ) . in definite range of @xmath57 and @xmath2 the ps state is stable . for the ps state extends from the ground state , whereas for it is stable only at finite temperatures . finally , for large @xmath0 ( ) , the system remains in a normal state at all temperatures for any chemical potential or concentration . comparing the va results at with the other ones ( exact ones for chain , rpa ones for square lattice and cubic lattices ) , one can see that the general structure of the diagrams is similar . it implies that the va can give qualitatively reasonable results also for lattices of finite dimensionality . however , one should stress that in finite dimensions due to quantum fluctuations connected with @xmath109 interactions the regions of the homogeneous ss phase occurrence are extended in comparison with the va results . moreover , in a linear chain only the short - range order occurs at in contrary to the va and rpa ( swa ) in , where the long - range order is present . one should emphasize that for the va overestimates transition temperatures and gives incorrect concentration dependence of them ( for , cf . ( [ row:2ndordertssno ] ) ) . for ideal bose gas with @xmath114 spectrum ( the and limit of the model considered ) the @xmath2-dependence of the temperature is , what can be obtained , for example in the rpa , even for finite values of @xmath57 @xcite . the rpa predicts also the existence of tricritical point in finite temperatures @xcite in , what confirms that the va can give qualitatively reasonable results also for real lattices in finite temperatures . the rpa results for at @xcite are in agreement with the mermin - wagner theorem @xcite , which proofs the absence of the long - range order for at . in one has the kosterlitz - thouless - type transition with increasing temperature . the model considered can describe the systems in which the single electron mobility ( single electron hopping ) is much smaller than the pair mobility ( kinetic energy of pairs ) and it can be neglected in first approximation . let us comment on the finite electron bandwidth effects . the presence of the hopping term breaks a symmetry between ( favouring ss ) and ( favouring @xmath23s ) cases . the phase diagrams of the penson kolb hubbard model are more complicated than those obtained for hamiltonian ( [ row : ham1 ] ) @xcite , even in the ground state . we can suppose that small but finite single electron hopping @xmath14 will not qualitatively alter the phase diagrams , at least for the case . the main effect of @xmath14 ( for , ) is a renormalization of the pair hopping term and an introduction of an effective intersite density - density repulsion . for and the charge density wave state can also occur @xcite . for and both signs of @xmath1 the @xmath14 term generates magnetic correlations competing with superconducting ones and its effects essentially modify the phase diagrams and the properties of normal state ( particularly for larger values of @xmath14 ) . in such a case it is necessary to consider also various magnetic orderings . moreover , several phase separation states involving superconducting , charge and ( or ) magnetic orderings could also be stable for . for larger values of @xmath14 the studies of gs phase diagrams of the half - filled pkh model for @xmath89-dimensional lattices have been determined by means of the ( broken symmetry ) hartree - fock approximation ( hfa ) and by the slave - boson mean - field method ( sbmfa ) in @xcite . for the diagrams supported also by the results of the continuum - limit field theory and exact diagonalisation studies @xcite are shown to consist of at least nine different phases including superconducting states , bond located antiferromagnetic and charge - density - wave states as well as mixed phases with coexisting site and bond orderings . the stability range of the bond - type orderings shrinks with increasing lattice dimensionality and for the phase diagrams involve exclusively site - located orderings . one should notice that the phase boundary at and determined by the hfa and sbmfa ( ) are in agreement with our va result , even for relatively large @xmath14 ( especially for ) @xcite . obviously , for the transition is to the antiferromagnetic mott phase . between nearest neighbours for half - filling obtained within va ( solid lines ) . the dashed lines denote the rigorous boundaries for 1d - chain . all transitions between homogeneous phases are discontinuous . .,scaledwidth=49.0% ] the intersite density - density interactions @xmath115 can introduce mixed phases and ( or ) phase separations between charge orderings and superconductivity @xcite . the ground state phase diagram in presence of interaction @xmath116 between nearest neighbours ( added a term to hamiltonian ( [ row : ham1 ] ) ) calculated for half - filling is presented in figure [ rys : uwi ] ( cf . with @xcite , where the case has been analyzed ) . for sufficiently strong intersite repulsion the homogeneous charge ordered ( co ) phase occurs ( with ) , whereas intersite attraction introduce the ps state in which two no phases coexists ( with and , i. e. electron droplets , ps[no / no ] ) . for @xmath117 the regions of the ss ( or @xmath23s , depending on the sign of @xmath1 ) and co phases occurrence are extended , whereas a region of the no ( mott ) phase is reduced . the superconducting order ( in @xmath117 ) , similarly as in section [ sec : gsexact ] , is a short - range order whereas the co is a long - range order ( cf . also @xcite ) . the electron concentration @xmath2 and chemical potential @xmath3 are ( thermodynamically ) conjugated variables in the bulk systems . however , one can fit the concentration rather than the potential in a controlled way experimentally . in such a case @xmath3 is a dependent internal parameter , which is determined by the temperature , the value of @xmath2 , and other model parameters ( cf . ( [ row : condn1 ] ) ) . thus the obtained phase diagrams as a function of the concentration are quite important because in real systems @xmath2 can vary in a large range . although our model is ( in several aspects ) oversimplified , it can be useful in qualitative analysis of experimental data for real narrow - band materials and it can be used for better understanding of various systems mentioned below and in section [ sec : intro ] . in particular , our results predict the existence of the phase separation ( ) near the mott state and describe its possible evolution and phase transitions with increasing temperature and a change of @xmath2 ( @xmath19 ) . of course the ps instability is specific to the short - range nature of the interactions in the model . when the ( unscreened ) long - range coulomb interactions are included the large - scale ps of charged particles is prevented and only a frustrated ps can occur ( mesoscale , nanoscale ) with the formation of various possible textures @xcite . the electron phase separation involving ss is shown experimentally in several systems . in particular , for special cases of la@xmath118cuo@xmath119 and la@xmath120sr@xmath121cuo@xmath119 , muon and superconducting quantum interference measurements suggest that the electron inhomogeneities move beyond local variations to form fully phase separated regions @xcite . organic compounds also exhibit the superconductor - insulator phase separations as a result of the external pressure ( e. g. quasi - one dimensional ( tmtsf)@xmath118pf@xmath122 @xcite , ( tmtsf)@xmath118reo@xmath123 @xcite ) and fast cooling rate through the glass - like structure transition ( e. g. @xmath124-(et)@xmath118cu[n(co)@xmath118]br @xcite ) . finally we mention the family of iron - pnictides , e. g. in ba@xmath126k@xmath121fe@xmath127se@xmath118 , where mesoscopic ps between ss and insulating ( magnetic ) phases have been observed @xcite . the detailed analysis of rpa solutions will be explored further in subsequent publication @xcite . it is of interest to analyze the effects of external magnetic field as well as the impact of density - density @xcite and magnetic @xcite interactions on the phase diagrams of model ( [ row : ham1 ] ) . the authors wish to thank t. kostyrko and g. pawowski for helpful discussions and a careful reading of the manuscript . the work ( k. k. ) has been financed by national science center ( ncn , poland ) as a research project in years 2011 - 2013 , grant no . dec-2011/01/n / st3/00413 . k. k. would like to thank the european commission and ministry of science and higher education ( poland ) for the partial financial support from european social fund operational programme `` human capital '' pokl.04.01.01 - 00 - 133/09 - 00 `` _ _ proinnowacyjne ksztacenie , kompetentna kadra , absolwenci przyszoci _ _ '' . within the va the on - site interaction term is treated exactly and the intersite interactions are decoupled within the mfa ( site - dependent ) : @xmath128 a variational hamiltonian has the following form : @xmath129,\end{aligned}\ ] ] where , and . @xmath130 can be diagonalized easily and a general expression for the grand potential @xmath131 in the grand canonical ensemble in the va is @xmath132\right\},\ ] ] where . the average value of operator @xmath133 is defined as @xmath134}{\textrm{tr}\left[\exp(-\beta\hat{h}_{0})\right]}.\ ] ] @xmath135 means a trace of operator @xmath136 calculated in the fock space . the explicit formula for the grand potential obtained in the va has the following form @xmath137 where @xmath138 and . the expression for the average number of electrons at @xmath7-site is given by @xmath139 and the average of the charge exchange operator at @xmath7-site is derived as @xmath140 so one has a set of self - consistent equations to solve consisting of : ( i ) @xmath13 equations in form of ( [ row : nonsite ] ) and ( ii ) @xmath13 equations in form of ( [ row : rhoonsite ] ) ( two equations for each site from @xmath13 sites ) . the free energy is obtained as , where , is the number of electrons in the system ( @xmath2 is defined by ( [ row : condn1 ] ) ) . the double occupancy @xmath141 of the site @xmath7 is determined by the following equation : @xmath142=\frac{\partial \omega_i}{\partial u},\ ] ] where @xmath45 is defined by ( [ row : nonsite ] ) . the solutions of the set can correspond to a minimum , a maximum or a point of inflection of the grand potential ( [ row : sitedependentgrandpotential ] ) . in other words , the solutions of the set fulfil the necessary condition for a minimum of @xmath131 with respect to . to find the solutions corresponding to stable ( or metastable ) states of the system , one should find a minimum of with respect to all @xmath143 . one can show easily that in the case of neglected spatial variations of the superconducting order parameter ( i. e. for every site @xmath7 ) reduce to eqs . ( [ row : grandpotential ] ) and obtained in section [ sec : method ] . let us introduce the particle - hole transformation on the alternate lattice defined as @xmath144 where if and if ( @xmath145 , @xmath146 label two sublattices in the alternating lattice ) . then the operators appearing in hamiltonian ( [ row : ham1 ] ) are transformed to @xmath147 whereas the thermodynamic parameters and averages : chemical potential , free energy , specific heat , superconducting - order parameter and double occupancy transform as follows @xmath148 respectively . from any characteristics of the considered model in the range we can find the behaviour of the system in the range . in particular , the phase diagrams are symmetric with respect to half - filling . from the set one can obtain the relation between the chemical potential and the electron concentration in both homogeneous phases : @xmath149 let us stress that @xmath92 determined from the above equation can be not only a transition temperature between stable phases but also a transition temperature between metastable phases ( e. g. in the regions of ps states stability ) or unstable phases . however , all second order transition temperatures which were obtained in section [ sec : kt ] by comparison of free energies fulfill ( [ row:2ndordertssno ] ) . from ( [ row:2ndordertssno ] ) one can obtain @xmath92 as a function of @xmath2 ( or @xmath19 cf . ( [ row : miss ] ) ) . in the limiting cases it yields @xmath151

in the last 20 years , the processes involved in the formation of solar type stars have been well characterized . an evolutionary sequence has been developed in which a prestellar core ( i.e. * ? ? ? * ) begins to collapse under its own gravity and forms a class 0 object ( i.e. * ? ? ? * ) with a large disk of material surrounding it . this accreting protostar begins to produce an observable molecular outflow which releases the build up of angular momentum ( i.e. * ? ? ? * ; * ? ? ? * ) and evolves into the class 1 stage . later , as the protostar becomes more revealed , accretion slows and a pre - main - sequence star forms . it then undergoes quasi - static contraction until core hydrogen burning commences and the star reaches the main sequence ( i.e. * ? ? ? these processes have been well studied and characterized based in part on the high resolution observations possible towards the closest examples of low mass star forming regions . the initial mass function ( imf ) of stars ( i.e. * ? ? ? * ) shows that most of the mass contained in stars is within the lower mass stars , and that massive stars do not form nearly as often as their lower mass counterparts . there are a number of nearby low mass star forming regions ( i.e. taurus , ophiucus , chameleon , and serpens are well studied nearby examples ) in which there is no massive star formation . high mass stars do not , in general , form in isolation , but together with low mass stars ( i.e. orion , see * ? ? ? the high densities required for massive star formation also mean that more of them form in the molecular ring . this means that statistically , the closest examples of regions forming high mass stars are further away , and are also forming clusters of stars . both distance and clustering can cause confusion when attempting to determine the dynamics in a region . add to the large average distances and source confusion the higher energies involved in the formation of massive stars and the presence of hii regions surrounding the more evolved massive protostars , and it becomes clear why high mass star formation is not nearly as well understood as lower mass star formation . if we assume that massive stars form in a manner similar to their lower mass counterparts ( i.e. * ? ? ? * ) , it is possible to suggest an evolutionary sequence starting from an infrared dark cloud ( i.e. * ? ? ? * ) with a deeply embedded protostar which then begins to heat its surroundings and creates a hot core ( i.e. * ? ? ? the protostar then begins burning hydrogen brightly enough to ionize its surroundings , forming a hypercompact hii region @xcite , which continues to expand into an ultracompact hii region @xcite as the protostar evolves towards and up the main sequence @xcite . there are many observable phenomena , such as large scale infall and outflows , which appear to be common between both low and high mass star formation , which suggests that perhaps high mass stars may form via the accretion of matter onto a single object in a manner similar to low mass stars . however , for high mass star formation , there are still a number of questions to be answered . for the large scale phenomena observable with a single dish telescope , we still do not know how large an area the outflowing gas shocks , how much energy is required to shock these regions , how large the infall region is , or how much mass from the protostellar core undergoes infall onto the central massive object . here , we present follow - up observations to ( * ? ? ? * hereafter paper 1 ) with which we address the questions posed above for seven high mass star forming regions mapped using the james clerk maxwell telescope . in paper 1 , we obtained single pointing observations towards 23 massive star forming regions previously known to have hii regions @xcite and molecular outflows as traced by cs @xcite or co @xcite . we were able to show that , of our observed high mass star forming regions with ongoing outflow activity ( 14 sources , as traced by sio emission ) , half ( seven ) showed evidence for large scale infall motion as well ( as traced by hco@xmath0 line asymmetries ) . we suggested that the sources with infall and outflow detections were still actively accreting material ( possibly in an ionized form , like that discussed in @xcite , while the other half may have already finished accreting ( like the source discussed in * ? ? ? * ) . however , with single pointing observations , we could not determine the sizes of the infalling or shocked ( outflowing ) regions , and therefore could not discuss properties such as the extents of the infall and outflow regions , whether these phenomena were beam diluted , or how their sizes compared . in this paper , we present maps of seven sources in the same tracers as paper 1 in order to understand better the nature of the large scale infall and outflow towards massive star forming regions . for this paper , we chose to map the sio ( j=8 - 7 ) and hco@xmath0/h@xmath1co@xmath0 ( j=4 - 3 ) emission towards seven massive star forming regions . six sources were chose from paper 1 , and were chosen to be representative of the survey as a whole . two sources ( g29.96 - 0.02 and g5.89 - 0.39 ) showed only outflow signatures , while two showed both infall and outflow signatures ( g10.47 + 0.03 and g19.61 - 0.23 ) , and one source ( k3 - 50a ) showed only a weak outflow signature . it was unclear whether the emission towards the sixth source ( g20.08 - 0.14 ) was due to one cloud ( as suggested by the single peaked sio line profile ) or two ( as suggested by the double peaked hco@xmath0 and h@xmath1co@xmath0 line profiles ) . the g20.08 region was previously known to have two ultracompact hii regions in close proximity ( i.e. * ? ? ? * ) , although the two hii regions are separated by more than a single jcmt beam . to this list of six sources , we added one more source , g10.6 - 0.4 . this source was added to determine whether the ionized and molecular accretion observed on small scales in this region ( i.e. * ? ? ? * ; * ? ? ? * ) has a corresponding larger scale molecular infall signature . g10.6 - 0.4 will act as a good test for determining whether our large scale results reflect the smaller scale dynamics , since the small scale structures are fairly well understood for this region . from this point forward , all source names based on galactic coordinates will be truncated to latitudes only . in section [ sec : observations ] we present our observations and in section [ sec : results ] we determine outflow and infall properties based on these observations . our results are then discussed in section [ sec : discussion ] , and we summarize in section [ sec : conclusions ] observations of sio ( j=8 - 7 ) , hco@xmath0 and h@xmath1co@xmath0 ( j=4 - 3 ) were obtained at the jcmt towards g10.47 , g10.62 , g19.61 , g20.08 , g29.96 and k3 - 50a in april and august 2007 as project m06bc08 along with hco@xmath0 observations of g5.89 . sio ( j=8 - 7 ) and h@xmath1co@xmath0 were observed simultaneously towards g5.89 - 0.39 as part of project m03ac13 ( see * ? ? ? * for a description of these observations ) and taken from the jcmt archive for this study . the archive map is smaller than our new maps ; however , both the sio and h@xmath1co@xmath0 emission falls below the rms limits within the edges of this map . the spectral resolution of the new datasets is 0.42 km s@xmath2 , the observations are spaced at 7@xmath3 to nyquist sample the 14@xmath3 full width of the jcmt beam at these frequencies , and the main beam efficiency was 0.62 . map center positions and local standard of rest velocities are given for our six original sources in paper 1 . for g10.6 , these properties are as follows : , , and . for reasons presented in section [ sec : g1047 ] , in this paper we adopt a distance of 5.8 kpc ( i.e. * ? ? ? * ) for g10.37 , which is much closer than the value of 12 kpc ( i.e. * ? ? ? * ) used in paper 1 . the 2007 observations were obtained using harp - b and the acsis autororrelator system in either grid position switching or jiggle position switching mode . both methods provided similar noise limits in similar times . the noise levels in each map are given in the respective figure captions and were generally consistent across each map , staying within approximately 30% of a mean value when receivers with bad noise levels were removed from the maps . these observations were reduced using the joint astronomy center starlink software . the 2003 observations were obtained with rxb in raster mapping mode using the das autocorrelator system and reduced using specx . first order linear baselines were removed from each pixel within each map , and the reduced maps were exported to the miriad format for analysis . sio and hco@xmath0 were chosen for this study because they trace outflow and infall , respectively , in star forming regions on all mass scales . once each map was properly calibrated and exported into miriad , the peak brightness temperature and line full width at zero power ( fwzp ) were determined from the spectrum at the map pointing center . the uncertainty in the integrated intensity for each line was determined using @xmath4t@xmath5n@xmath6 where t@xmath7 is the single channel rms uncertainty , @xmath8 is the velocity resolution of our observations , and n@xmath9 is the number of channels used to cover the full width of the line ( fwzp ) . note that for sources with self absorbed hco@xmath0 profiles , the peak line temperature is an underestimate of the source brightness temperature . in our maps , we find that the hco@xmath0 emission towards each source is extended , as is the h@xmath1co@xmath0 emission ; however , h@xmath1co@xmath0 does not reach the same spatial scales as the 77 times more abundant ( wilson & rood , 1994 ) hco@xmath0 . in only five of our seven sources is the sio emission at least marginally resolved . the deconvolved source size was determined using @xmath10 where the longest axis of the observed region full width at half maximum diameter ( @xmath11 ) was used to compare to the beam diameter ( @xmath12 ) . towards the two unresolved sources ( k3 - 50a and g19.61 ) , the deconvolved diameter of the emitting region is less than 50% of the jcmt primary beam , while the deconvolved diameter is greater than 70% of the jcmt primary beam towards our other sources . our sio detection is weak in k3 - 50a ( the line peak was only detected at 3.5 - 4@xmath13 ) and was insufficient for detecting sio beyond the central pixel . this was not the case for g19.61 ( the line peak was detected at 7@xmath13 ) , yet the sio emission in this source was also unresolved . thus , if the shocked region in g19.61 were as extended as our jcmt beam , we would have detected the extended sio emission . emission maps for each source are shown in figures [ fig : g5map]-[fig : k3 - 50map ] . in each case , hco@xmath0 emission spectra are shown for pixels with hco@xmath0 integrated intensities which are 10@xmath13 or greater . for each tracer , the intensity scale for the hco@xmath0 spectra is constant for a given source , and the peak hco@xmath0 intensity for each source is listed in table [ tab : hco_prop ] . the dashed contour in each plot represents the 3@xmath13 emission region for h@xmath1co@xmath0 and the solid contours represent sio emission , starting at 3@xmath13 and increasing in intervals of 3@xmath13 . the 3@xmath13 rms uncertainties in the integrated intensities are given in each figure caption . the star in each figure marks the position of the peak in the hco@xmath0 integrated intensity . in all cases except k3 - 50a and g29.96 , the positions of the hco@xmath0 peak is within the same beam as the sio peak . for the two remaining sources , this offset may be due to pointing errors between the two maps . using the fwzp from the central pixel , the integrated intensity for each species across each map was determined . the rms uncertainty in the integrated intensity was determined using the same method as presented in paper 1 . for each tracer , the line is broadest in the map center , and so using the same fwzp across the entire map captures all of the emission from that transition . the integrated intensities averaged over each map in sio and hco@xmath0 are shown in tables [ tab : sio_prop ] and [ tab : hco_prop ] , respectively . the integrated intensities in sio and hco@xmath0 were used to find the column density of gas emitting across the map in that tracer . for optically thin gas , the column density in the higher energy level is proportional to the integrated intensity of the line , with a scaling constant proportional to the frequency of the transition and the einstein coefficient for that transition ( see equation 1 of paper 1 , or * ? ? ? assuming local thermodynamic equilibrium and an ambient temperature of 44 k ( as was done in paper 1 ) , we can use the partition function to determine the column density of optically thin gas in each tracer ( see for instance equation 14.38 of * ? ? ? the average column densities of sio and hco@xmath0 are presented in tables [ tab : sio_prop ] and [ tab : hco_prop ] . the values for the h@xmath1co@xmath0 column density are not presented here , but were found to be only 4 - 22 times less than those for hco@xmath0 . that we do not see a larger difference in the column densities , given the abundance ratio of 77 between the isotopologues , suggests that we are not seeing all of the hco@xmath0 emission because the line is optically thick . thus , our derived column densities are lower limits to the true hco@xmath0 column densities due to optical depth effects . this trend was also seen in paper 1 , and supports the claim that the double peaked hco@xmath0 profiles towards four of our sources are due to self absorption of the optically thick line , and not multiple clouds along the line of sight . in an environment which has recently been shocked , the abundance of sio can be enhanced up to a factor of 10@xmath14 ( from dark cloud abundances of approximately 10@xmath15 ) with respect to molecular hydrogen , due to silicon being liberated from dust grains and combining with oxygen to form sio ( i.e. * ? ? ? * ; * ? ? ? * ) . after approximately 10@xmath16 years , sio depletes out of the gas phase again and the abundance drops ( i.e. * ? ? ? because of the short enhancement lifetime , sio is generally only detectable in regions which have recently been shocked . where we detect sio ( j=4 - 3 ) we suggest there is ongoing outflow activity . we present the column density of sio towards each source in table [ tab : sio_prop ] . we can divide this value by the abundance of sio with respect to h@xmath17 in order to determine the column density of shocked gas in each region . these values are also presented in table [ tab : sio_prop ] . we used the same methods as described in paper 1 to calculate the abundance of sio towards g10.6 and we have assumed a constant sio abundance across each map . we note that the sio column densities listed in paper 1 are over estimated by a factor of 3.8 , and thus the logarithmic sio abundances listed in that paper are corrected in table [ tab : sio_prop ] . because we have adopted a new distance for g10.47 , we have had in addition to modify our original abundance estimate from paper 1 . the cs column density quoted in @xcite appears to be distance independent ; however , the cs abundance presented in @xcite is dependent on the assumed distance to the source ( see their equations 7 and 12 ) . thus , our sio abundance has changed from -9.53 in paper 1 to -10.42 when the new distance and revised column density are taken into account . the sio abundance determined for k3 - 50a in paper 1 is much closer to that found in dark clouds than in outflow regions . the region in which sio is emitting is unresolved by our observations and highly beam diluted . thus , for our analysis , we have assumed an sio abundance of -10.15 , the average abundance from our other six sources . based on the abundance estimates presented in table [ tab : sio_prop ] , we have determined the total column of shocked material towards each source , from which we determined the mass of shocked gas because we know the area of the emitting region ( see below ) . for our sources which are unresolved in sio , we have taken the radius of our primary beam to be the radius of the emitting region . comparing the integrated intensities of our sio emission in the central pixel of each map , where our signal to noise ratio is highest , to the models presented in @xcite , we suggest shock velocities of 20 - 22 km s@xmath2 , and ambient densities of @xmath18 @xmath19 . the ambient densities suggested here are consistent with those of @xcite who determined densities and cs column densities for these regions using large velocity gradient models . for each of our sources , we adopt a shock velocity of @xmath20 km s@xmath2 . because we have mapped the sio emission towards seven sources , we can determine the radius of the region in which sio is being emitted ( r@xmath21 ) . the method used to determine the deconvolved source size was described in section [ sec : observations ] , and the radius of the emitting region is presented in table [ tab : sio_prop ] . from this radius and the shock velocity determined above , we can constrain the ages of our outflows from simple kinematic arguments . these approximate outflow ages are presented in table [ tab : hco_prop ] and are used later to constrain the amount of infalling material . in order for the outflow shocks to reach their observed sizes given the mass of shocked gas , the powering sources for these outflows must produce enough energy to push through the ambient medium . if we assume a steady wind is driving a spherical bubble of gas outward , we can use the model of @xcite to determine the luminosity required to push the ambient gas . equation 6 of @xcite can be re - expressed as : @xmath22 where @xmath23 is the luminosity imparted by the wind , @xmath24 is the radius of the bubble , @xmath25 is the shock velocity , and related to the outflow age by @xmath26 , and @xmath27 is the ambient density , and related to the ambient number density by the mass of molecular hydrogen ( @xmath28 ) . while much of this wind luminosity goes into accelerating the ambient medium , some of it is also transferred into thermal energy . from @xcite , the thermal energy can be expressed as @xmath29 , or , given the value for @xmath30 above , @xmath31 . both the calculated wind luminosity and thermal energy required to power these wind driven shocks are listed in table [ tab : sio_prop ] . these thermal energies are of approximately the same order of magnitude as the kinetic energies calculated for the co outflows in a number of massive star forming regions ( see for instance * ? ? ? * ; * ? ? ? * ) ; similarly the wind luminosities are comparable to the luminosities of their hii regions ( see for instance * ? ? ? many authors suggest that an asymmetric self absorbed line profile of an optically thick line tracer in which the blue emission peak is brighter than the red , and the optically thin tracer has a single line peak , can be indicative of infall ( i.e. * ? ? ? * ; * ? ? ? , our optically thick line is hco@xmath0 ( j=4 - 3 ) and the optically thin isotopologue is h@xmath1co@xmath0 ( j=4 - 3 ) . from paper 1 , we expected to observe infall signatures in g10.47 and g19.61 . infall is detected towards these two sources , as well as towards g20.08 from our original study . in our single pointing observations , g20.08 was shown to have a double peaked line profile in both hco@xmath0 and h@xmath1co@xmath0 , which was suggested to be due to two overlapping sources . however , in the more extended map , we see that h@xmath1co@xmath0 becomes single peaked , suggesting that perhaps it is optically thick towards the center of the source and that mapping is required to resolve the ambiguity with respect to the chance super - position of line of sight clouds and self absorption of multiple isotopologues . g10.6 was not in our original sample , but was added because it was known to have small scale infall motions . it too is seen to have a large scale infall signature . in paper 1 , we suggested that beam dilution may have been a contributing factor as to why we did not detect infall in a number of sources . that we have detected extended infall in two distant ( @xmath32 6 kpc ) sources in this study suggests that if beam dilution of the signal is a problem , it only affects intrinsically smaller sources , and not intrinsically more distant ones , since our unresolved detections are for g19.61 and g20.08 at 4.5 and 4.1 kpc . infall velocities were calculated for each self absorbed hco@xmath0 spectrum towards g10.47 , g10.6 , g19.61 , and g20.08 using the models of @xcite in the same way as described in paper 1 . from our rms uncertainties in line temperatures and the spectral resolution of our observations , we determined the uncertainties associated with our derived infall velocities . we used these uncertainties to limit the regions in which we claim infall detections . only if the derived infall velocity is more than three times our uncertainty for that spectrum do we claim to detect infall in that pixel . towards g19.61 and g20.08 , robust detections were made in one and three pixels ( respectively ) suggesting that the infall in both of these sources is unresolved . towards g10.47 and g10.6 ( both at approximately 6 kpc ) , robust infall detections were made in seven nyquist sampled pixels per map , suggesting that the large scale infall in these regions is extended . along the northern , blue shifted , outflow lobe in g10.47 , the hco@xmath0 line profiles again become double peaked . it is quite possible that there is a second instance of a collapsing region here since sio is marginally detected ( at the 3@xmath13 level ) in the middle of this northern extension . the double peaked profiles do lead to infall velocities ( in three pixels ) which are greater than 3@xmath13 ; however , we do not include this region in our infall analysis for a number of reasons . the infall signature is not continuous from the central region to this northern region , which suggests that there are two distinct regions of infall . this discontinuity would confuse conclusions about the central infall region . similarly , we are interested in the infall activity surrounding hii regions , where we know massive stars are forming . neither the mid infrared nor radio maps of this region ( @xcite and @xcite , respectively ) extend far enough to the north to determine whether there are hii regions in this northern extension . archival scuba maps of this region at the same resolution as our observations @xcite show emission which has the same extent and morphology as our hco@xmath0 observations . for each region in which infall was observed , we determined an average infall velocity by taking the mean of the infall velocities determined for each position in each map using the models of @xcite . this averaging resulted in the velocities shown in table [ tab : hco_prop ] . we chose to use this method over creating an averaged spectrum and determining infall velocities from it . the latter method resulted in unreasonably high infall rates ( i.e. 5 km s@xmath2 in g10.47 ) , which give very large mass infall rates ( 0.22 m@xmath33 yr@xmath2 ) and total infall masses ( @xmath34 m@xmath33 ) assuming constant infall rates over the lifetime of the outflow . averaging the spectra and then determining the infall velocity increased the infall velocity by increasing the full width at half maximum ( fwhm ) of the line . since the infall velocity varies as the fwhm squared ( see equation 9 of * ? ? ? * ) , this broadening leads to an overestimate of the infall velocity . thus , for our unresolved sources , the calculated infall velocities may be upper limits , since there could be unresolved velocity gradients within the jcmt beam . higher resolution imaging of these sources could decrease our calculated infall velocities . using the derived average infall velocities for each region , we determined mass infall rates for each sources using @xmath35 , where @xmath36 is the mass density , @xmath37 is the infall velocity presented in table [ tab : hco_prop ] , and @xmath38 is the area of the infalling area ( which we assume to be circular ) . since the area in which we detected an infall signature was always similar to the area in which sio was detected , we used the radius of the sio emitting region as the radius of the infalling region . from the radii of the sio emitting regions and a shock velocity of 21 km s@xmath2 , we were able to estimate outflow ages in section [ subsec : sio ] . assuming infall and outflow are contemporaneous , these outflow ages can be used to constrain how much mass has undergone infall . these values are also given in table [ tab : hco_prop ] , and represent the total mass of infalling molecular gas , not just the mass of the infalling hco@xmath0 . it is interesting to note that the infall and outflow signatures in g19.61 are both unresolved , and the signatures in g20.08 are both marginally resolved . this comparison suggests that the shocked region , as traced by sio , may have a similar physical size as the infall regions in these smaller sources . towards g10.6 and g10.47 , the regions over which we have robustly detected extended infall , our deconvolved sio emitting regions also appear to be on the same approximate spatial scales . comparing infall mass and outflow mass , we note that our infall masses are lower than our outflow masses . this is contrary to the results of @xcite ; however , their ratio was calculated for the protostellar jet , not the entrained molecular outflow . this comparison to @xcite suggests that more of the ambient mass is entrained into the large scale molecular outflow ( produced by a protostellar jet ) than falls in onto the forming stars ( i.e. * ? ? ? * ) . the large spatial extent of the hco@xmath0 emission in our map of g10.47 places it among the largest molecular outflows due to star formation in our galaxy . some herbig - haro objects have been observed emanating from nearby lower mass star forming regions at distances of up to 12 pc from their powering stars ( i.e. * ? ? ? * ) ; however in those cases , the associated molecular outflow is not nearly as wide as the case of g10.47 , and those jets are not pushing through an ambient medium dense enough for massive stars to form ( i.e. 10@xmath39 @xmath19 ) . in paper 1 , we used a distance estimate of 12 kpc to g10.47 . given the extent of hco@xmath0 gas ( @xmath40 ) in our current maps , this distance would give a linear scale of over 4 pc for a single outflow . other authors have suggested a distance to this source of 5.8 kpc ( i.e. * ? ? ? for the analysis presented here , we adopt this new distance , which changes the length of the outflow from 4 pc to 2 pc . given that this source is along a line of sight reasonably close to both the galactic plane and galactic center , it is not unreasonable to suggest that what appears to be an extremely large outflow is instead the super - position of a number of clouds along the line of sight . however , based on the arguments presented below , we suggest that the gas kinematics and morphology are due to a large outflow . examining the spectra of our hco@xmath0 map , we note that while the hco@xmath0 emission is extended , it is not present in all pixels , suggesting that we have captured all of the emission in this velocity range . figure [ fig : averaged_spectra ] shows the hco@xmath0 line profiles averaged over the blue and red shifted emission regions as well as the hco@xmath0 and h@xmath1co@xmath0 line profiles averaged over the central nine pixels of our map . the red and blue spectra , while having very little overlap between their profiles , do both fit within the width of the source center line profile . the velocities at line peak for the blue and red spectra are -64 and -70 km s@xmath2 respectively , and match the velocities of the two peaks of the self absorbed , averaged hco@xmath0 central profile . the single peak of the optically thin h@xmath1co@xmath0 is at -67 km s@xmath2 as is the minimum of the hco@xmath0 self absorption feature . this self - absorption lies directly between the peaks of the red and blue profiles , and is the same as the source rest velocity . no other sources in the w31 region have rest velocities similar to the ones seen in the g10.47 region . the double peak in the hco@xmath0 spectrum is due to self absorption , not multiple components , since there is a single peak in the h@xmath1co@xmath0 spectrum . the same is true for the co isotopologues in this region . the bima observations of @xcite show one peak in c@xmath41o , while the @xmath1co observations of @xcite with the iram interferometer show two peaks indicative of self absorption . the first moment map of the hco@xmath0 emission ( grayscale in figure [ fig : g1047_mom1 ] ) in this region also suggests a single outflow . the observed smooth transition from blue to red along the source would be more likely to be observed in one source with a velocity gradient than three separate sources with discontinuous cloud boundaries . the second moment map of this region ( contours in figure [ fig : g1047_mom1 ] ) suggest that the largest velocity dispersion is towards the center , and continuously decreases towards the emission edges ( outflow lobe ends ) , suggesting that most of the mass is centrally concentrated . while we favor the outflow interpretation of this morphology , higher resolution , large scale observations of this source are necessary in order to resolve this ambiguity . this region contains four individual uchii regions and what is likely a large scale outflow with a collimation factor ( @xmath42 ) of approximately 1.5 , which may be even more highly collimated at higher resolution . objects like g10.47 , with large scale , fairly collimated outflow emanating from a cluster of hii regions pose interesting challenges for accretion models . as stated previously , g10.6 was added to our source list because it was previously known to show evidence for small scale infall , both in molecular gas @xcite and ionized gas @xcite . these small scale infall signatures appear to be consistent with a single accretion flow which changes from being molecular to ionized as it passes the ionization boundary of the hii region . while we can not determine the orientation of the large scale infall from our observations , we note that our derived mass infall rate ( 0.03 m@xmath43 yr@xmath2 ) is comparable to , but slightly larger than , that of @xcite ( 0.02 @xmath44 yr@xmath2 ) . this similarity may suggest that the large scale infall possibly maps to the small scale accretion flow ; in addition , despite resolving the infall region , we may still be over estimating the infall velocity due to velocity gradients in our larger jcmt beam ( see section [ sec : hco ] ) . none of our basic conclusions drawn from our single pointing observations of sio ( i.e. column densities in the central pixel or outflow lifetimes ) have changed due to mapping save for the infall detection in g20.08 . however , with respect to the outflowing gas , we have been able to place constraints on the energy required to shock these regions . with respect to infalling gas , we have placed better constraints on infall velocities and shown that the lack of infall signatures in some sources is not due to the signature being beam diluted out of observability at large distances . we have found that one of our sources which was previously not thought to have an infall signature ( g20.08 ) in fact does appear to have large scale infall motions . this discovery suggests that more accurate infall results are possible with maps of massive star forming environments than with single pointing observations . returning to the results of paper 1 , we determined that seven out of the 14 sources with ongoing outflow signatures had infall signatures as well . these numbers can now be revised to eight out of 15 outflow sources with infall signatures when g10.6 is added to the source list . thus , we find that approximately half of the sources with ongoing outflows have infall signatures which we take to trace accretion . because our sources were selected based on the presence of hii regions , ongoing infall suggests that there is a phase of massive star formation in which the star is still actively accreting , but is also burning hydrogen brightly enough to ionize its surroundings ( see for instance the discussion in * ? ? ? we have found that just over half of the sources in our sample have large scale infall motions suggesting that the protostars powering the uchii regions and massive outflows are still actively accreting material . where we have detected infall motions , they have been on quite large scales , suggesting that infall , as traced by hco@xmath0 , is in general observable with single dish maps at the distances to massive star forming regions . maps are required in order to capture all of the infalling gas since the infall signature can be extended , even at 6 kpc . in mapping these regions , we have been able to determine how much mass is falling into the central regions of our sample sources , which we would not have been able to determine with our single pointing observations . we find that when the infall is resolved , the infalling mass is greater than 30 m@xmath43 . our derived mass infall rate towards g10.6 is similar to the small scale rate found by @xcite , suggesting that we may be observing the same physical process , but on different spatial scales . in general , we have found that sio is not very extended ; however , since we mapped these regions in sio , we were able to determine the radius of the emitting region , the mass of shocked gas and how much energy is required to shock the gas surrounding a massive protostar . we find these energies and luminosities to be similar to the kinetic energy imparted to the outflowing gas and the luminosities of the hii regions themselves . we also find that the region over which we detect an infall signature and the region in which we detect sio is quite similar , and find that in these regions , the total infall mass is less than the total outflow mass ( m@xmath45/m@xmath46 0.1 to 0.8 ) assuming the two processes are contemporaneous . we would like to acknowledge the support of the national science and engineering research council of canada ( nserc ) . p. d. k would also like to thank j. wadsley for helpful discussions during the preparation of this manuscript . rccccccccccc g5.89 - 0.39 & 2.0 & 2.37 & 105 & -9.48 & 4.38 & 12.81 & 22.29 & 3.7 & 1.9 & 1.1 & 0.18 + g10.47 + 0.03 & 5.8 & 1.24 & 30 & -10.42 & 2.70 & 12.60 & 23.02 & 184 & 5.8 & 10.3 & 5.1 + g10.6 - 0.4 & 6.0 & 1.14 & 20 & -9.92 & 8.74 & 13.11 & 23.03 & 497 & 9.4 & 27.2 & 21.6 + g19.61 - 0.23 & 4.5 & 0.71 & 45 & -10.32 & 4.57 & 12.83 & 23.15 & 162 & @xmath474.7&@xmath476.8&@xmath472.7 + g20.08 - 0.14 & 4.1 & 0.66 & 30 & -11.06 & 2.68 & 12.60 & 23.66 & 374 & 4.0 & 4.8 & 1.6 + g29.96 - 0.02 & 9.0 & 0.63 & 40 & -9.70 & 5.14 & 12.88 & 22.58 & 124 & 7.9 & 19.2 & 12.8 + k3 - 50a & 8.6 & 0.48 & 20 & -10.15 & 3.15 & 12.67 & 22.82 & 276 & @xmath479.0&@xmath4724.7&@xmath4718.7 [ tab : sio_prop ] rccccccccc g5.89 - 0.39 & 31.5 & 140 & 35.6 & 13.17 & & & 0.29 & & + g10.47 + 0.03 & 11.7 & 60 & 11.2 & 12.67 & 1.02@xmath480.1 & 8.89 & 0.88 & 78.0 & 0.42 + g10.6 - 0.4 & 31.2 & 50 & 28.4 & 13.08 & 1.19@xmath480.1 & 27.34 & 1.42 & 389 & 0.78 + g19.61 - 0.23 & 8.3 & 80 & 14.4 & 12.78 & 1.49@xmath480.6 & @xmath478.5&@xmath470.71&@xmath4760.3&@xmath470.37 + g20.08 - 0.14 & 5.9 & 55 & 9.2 & 12.59 & 1.35@xmath480.3 & 5.51 & 0.60 & 33.0 & 0.09 + g29.967 - 0.02 & 16.5 & 30 & 9.2 & 12.59&&&1.19 & & + k3 - 50a & 17.0 & 30 & 13.2 & 12.75&&&@xmath471.36 & & + [ tab : hco_prop ] beuther , h. , & shepherd , d. 2005 , cores to clusters : star formation with next generation telescopes , 105 beuther , h. , schilke , p. , sridharan , t. k. , menten , k. m. , walmsley , c. m. , & wyrowski , f. 2002 , a&a , 383 , 892 castor , j. , mccray , r. , & weaver , r. 1975 , apjl , 200 , l107 caselli , p. , hartquist , t. w. , & havnes , o. 1997 , a&a , 322 , 296 di francesco , j. , johnstone , d. , kirk , h. , mackenzie , t. , & ledwosinska , e. 2008 , , 175 , 277 fuller , g. a. , williams , s. j. , & sridharan , t. k. 2005 , a&a , 442 , 949 gibb , a. g. , wyrowski , f. , & mundy , l. g. 2004 , apj , 616 , 301 gregersen , e. m. , evans , n. j. , ii , zhou , s. , & choi , m. 1997 , apj , 484 , 256 hunter , t. r. 1997 , ph.d . thesis , keto , e. , & klaassen , p. 2008 , , 678 , l109 keto , e. 2007 , apj , 666 , 976 klaassen , p. d. , & wilson , c. d. 2007a , , 663 , 1092 klaassen , p. d. , & wilson , c. d. 2007b , arxiv e - prints , 710 , arxiv:0710.3785 klaassen , p. d. , plume , r. , ouyed , r. , von benda - beckmann , a. m. , & di francesco , j. 2006 , apj , 648 , 1079 kroupa , p. 2001 , mnras , 322 , 231 kurtz , s. , cesaroni , r. , churchwell , e. , hofner , p. , & walmsley , c. m. 2000 , protostars and planets iv , 299 kurtz , s. , churchwell , e. , & wood , d. o. s. 1994 , apjs , 91 , 659 mardones , d. , myers , p. c. , tafalla , m. , wilner , d. j. , bachiller , r. , & garay , g. 1997 , apj , 489 , 719 mckee , c. f. , & tan , j. c. 2003 , , 585 , 850 myers , p. c. , mardones , d. , tafalla , m. , williams , j. p. , & wilner , d. j. 1996 , apjl , 465 , l133 olmi , l. , cesaroni , r. , & walmsley , c. m. 1996 , a&a , 307 , 599 pascucci , i. , apai , d. , henning , t. , stecklum , b. , & brandl , b. 2004 , a&a , 426 , 523 pillai , t. , wyrowski , f. , carey , s. j. , & menten , k. m. 2006 , , 450 , 569 pineau des forets , g. , flower , d. r. , & chieze , j .- 1997 , herbig - haro flows and the birth of stars , 182 , 199 plume , r. , jaffe , d. t. , evans , n. j. , ii , martin - pintado , j. , & gomez - gonzalez , j. 1997 , apj , 476 , 730 riddick , f. c. , roche , p. f. , & lucas , p. w. 2007 , , 381 , 1077 rholfs , k. , wilson , t. l. 2004 , tools of radio astronomy ( 4th ed ; berlin ; springer ) schilke , p. , walmsley , c. m. , pineau des forets , g. , & flower , d. r. 1997 , a&a , 321 , 293 sewilo , m. , churchwell , e. , kurtz , s. , goss , w. m. , & hofner , p. 2004 , apj , 605 , 285 shepherd , d. s. , & churchwell , e. 1996 , , 472 , 225 shirley , y. l. , evans , n. j. , ii , young , k. e. , knez , c. , & jaffe , d. t. 2003 , apjs , 149 , 375 shu , f. h. , adams , f. c. , & lizano , s. 1987 , , 25 , 23 sollins , p. k. , zhang , q. , keto , e. , & ho , p. t. p. 2005 , , 624 , l49 tielens , a. g. g. m. 2005 , the physics and chemistry of the interstellar medium ( cambridge university press ) wood , d. o. s. , & churchwell , e. 1989 , apjs , 69 , 831 wu , y. , wei , y. , zhao , m. , shi , y. , yu , w. , qin , s. , & huang , m. 2004 , a&a , 426 , 503 zinnecker , h. , & yorke , h. w. 2007 , ara&a , 45 , 481

the primordial scenario of globular cluster ( gc ) formation was first proposed by @xcite . their paper considers gcs forming in the early universe out of primordial gas fluctuations on a jeans mass scale . realization that the dominant mass component of the universe is dark matter ( dm ) led to the revision of this idea in @xcite , where gcs are proposed to have formed in the potential wells of dm minihalos in the early universe . in this revised picture gcs can be considered as some of the earliest galaxies in the universe , with a baryons dominated core and extended dm halo . this picture of dm - dominated gcs forming in the early universe was later considered by many authors , including @xcite , @xcite , @xcite , @xcite , and @xcite . primordial gcs could have formed somewhere between a redshift of @xmath1 , when the first stars in the universe are believed to have been born inside minihalos with total masses @xmath2 @xmath3 @xcite , and redshift of @xmath4 , when the reionization of the universe was complete @xcite . it is often assumed that the very first stellar objects produced enough lyman - werner photons ( capable of dissolving h@xmath5 molecules , which are the main coolant in the pristine gas ) to prohibit star formation in most of the low mass ( @xmath6 @xmath3 ) gas - rich halos . cosmological simulations with radiative transfer of @xcite suggest another possibility : in their model , numerous mini - galaxies with masses of @xmath7 @xmath3 form inside relic cosmological h ii regions . the observed positive feedback can be explained by high non - equilibrium fraction of free electrons in defunct h ii regions , which leads to enhanced h@xmath5 production and hence star formation . this is an interesting mechanism for forming gcs , as it naturally explains the observed higher specific frequency of gcs in higher density regions of the universe ( such as around cd galaxies at the center of clusters of galaxies ) , where large cosmological h ii regions should have formed in the early universe . in a widely accepted hierarchical paradigm of structure formation in the universe , small bound objects form first , later merging to form objects of increasingly larger size . in this picture , primordial gcs would first experience a major merger with comparable mass halos , or be accreted by a larger halo of a dwarf galaxy size . in their simulations of the formation of a dwarf galaxy in the early universe ( at @xmath8 ) , @xcite observed a formation of gc - type baryonic objects inside dm minihalos , which later merged to form a dwarf galaxy . by the end of the simulations , baryonic cores , believed to represent proto - gcs , had apparently lost their individual dm halos . @xcite suggested that dm was lost because of the violent relaxation accompanying the major merger . the lack of resolution did not allow the authors to reach more definitive conclusion as to the fate of dm in their proto - gcs . in the next step , dwarf galaxies containing primordial gcs would merge to produce objects of increasingly larger size . this process should continue at the present time . a well - known example is that of sagittarius dwarf galaxy , which is presently in the process of being disrupted by the tidal forces of milky way and depositing its system of gcs into the halo of our galaxy @xcite . in the primordial picture of gc formation , gcs can be considered as small nucleated dwarfs accreted at some point by larger galaxies , which lose their extra - nuclear material ( baryons and dm ) due to stripping by the tidal forces of the accreting galaxy . such mechanism was invoked to explain the formation of a few of the largest gcs : m51 @xcite , @xmath9 centauri @xcite , and g1 @xcite , which is the satellite of the m31 galaxy . in the quoted papers , these gcs are considered to be exceptions . we suggest instead that this method of forming these gcs is a rule rather than exception , and is fully consistent with primordial gc formation scenarios . in this picture , the above largest gcs are examples of the most recent gc - producing accretion events , and represent a small fraction of the total number of proto - gcs accreted directly by a large galaxy ( milky way or m31 ) in the modern universe ( thus skipping the intermediate step of being accreted by a larger dwarf galaxy in the early universe ) . @xcite used numerical simulations of a dm - dominated gc orbiting inside the static potential of milky way halo and observations of tidal tails of m2 of @xcite to claim that primordial scenarios of gc formation are inconsistent with observations of galactic gcs . as discussed by @xcite and @xcite , the model of @xcite was too simplistic to address this issue . most importantly , the author considered all galactic gcs being directly accreted by the milky way , which is definitely not the case in the hierarchical picture of structure formation in the universe . moreover , the dm subhalo model of @xcite is a lowered isothermal sphere with arbitrary ( not cosmologically normalized ) structural parameters , which makes it very different from the dm halos expected to host a gc in the early universe . one also has to be careful with claims on the existence of `` extratidal features '' around galactic satellites , which includes both gcs and dwarf spheroidal galaxies . in the case of @xmath9 centauri , @xcite showed that the apparent `` tidal tails '' observed by @xcite around this gc are most probably caused by spatially variable foreground dust extinction in its outskirts . in the case of the draco dwarf spheroidal galaxy , the detection of apparent `` extratidal features '' around this galaxy by @xcite was later refuted by observations of @xcite which had much higher sensitivity . the `` extratidal '' extension of the radial surface brightness profiles of galactic gcs beyond a `` tidal '' king radius is almost always seen at the levels lower than the inferred background contamination level , and could be an artifact of the background correction procedure . one beautiful example of a gc with obvious tidal tails is that of palomar 5 @xcite . as we will show in this paper , a presence of tidal tails in some gcs is not inconsistent with these gcs having formed with dm halos in the early universe , and hence with primordial scenarios of gc formation . in this series of two papers , we are using high - resolution @xmath0-body simulations of gcs with dm halos to test primordial scenarios of gc formation . in particular , we are trying to answer the following questions : 1 ) are there obvious signatures of dm presence in gcs ? 2 ) will dm in a primordial gc survive hierarchical process of assembling galaxies in the early universe ? in the first paper , ( * ? ? ? * hereafter paper i ) , we considered the initial relaxation of a stellar cluster at the center of a dm minihalo in the early universe ( around @xmath10 ) . the structural parameters of dm halos were fixed by results of cosmological @xmath11cdm simulations . the initial non - equilibrium configuration of stellar clusters was that of ( * ? ? ? * hereafter ms04 ) , where we showed that purely stellar ( no dm ) homogeneous isothermal spheres with the universal values of the initial density @xmath12 @xmath3 pc@xmath13 and velocity dispersion @xmath14 km s@xmath15 collapse to form gc - like clusters , with all bivariate correlations between structural and dynamic parameters of galactic gcs being accurately reproduced . we showed in paper i that many observational features of galactic gcs , used traditionally to argue that these systems are tidally limited purely stellar clusters without dm , can be produced by the presence of significant amounts of dm in their outskirts . in particular , in warm collapse models ( with the initial virial parameter for the stellar core of @xmath16 ) we observed the formation of an apparent `` tidal '' cutoff in surface brightness profiles , with no unusual features in the outer parts of the velocity dispersion profile . in cold collapse models ( with @xmath17 ) , on the contrary , an apparent `` break '' in the outer parts of both surface brightness and velocity dispersion profiles is seen , which can be mistakingly interpreted as presence of `` extratidal '' stars heated by the tidal field of the host galaxy . the results of paper i are directly applicable to dynamically young intergalactic gcs and gcs which have not experienced significant tidal stripping in the potential of the host galaxies . in this second paper , we test the regime of severe tidal stripping of our primordial dm - dominated gcs in the potential of the host dwarf galaxy . as we discussed above , being accreted by a dwarf galaxy would be a typical fate of primordial gcs formed at high redshift . we use warm collapse models of primordial gcs from paper i to set up the initial conditions for the present paper . we consider both simulations suggested ( * ? ? ? * hereafter nfw ) and observationally motivated @xcite density profiles for our dm halos . this paper is organized as follows . in section [ model ] we describe our models . in section [ results ] we discuss the results of the simulations . we conclude with section [ conclusions ] . in our model , we explore the impact of tidal fields on dm - dominated gcs orbiting inside a static potential of host dwarf galaxy with a virial mass of @xmath18 @xmath3 . gcs are assumed to be accreted by the dwarf galaxy at a redshift of @xmath19 . structural parameters ( virial radius @xmath20 , scale radius @xmath21 , and concentration @xmath22 ) of the host galaxy are fixed by the results of cosmological @xmath11cdm simulations , and at @xmath19 are equal to @xmath23 kpc , @xmath24 kpc , and @xmath25 ( see eqs . [ 1 - 9 ] from paper i ) . throughout this paper we assume the following values of the cosmological parameters : @xmath26 and @xmath27 km s@xmath15 mpc@xmath15 @xcite . galaxies with @xmath28 @xmath3 correspond to @xmath29 density fluctuations collapsing at @xmath19 ( * ? ? ? * their fig . 6 ) and hence are very common at that redshift . the initial parameters of our gcs are identical to those of our warm collapse models from paper i. we could not use the cold collapse model of paper i because it would be an impossible task with the present day technology with the method we are using ( @xmath0-body simulations ) . indeed , all the models presented in the present paper for the warm collapse only case required 3.6 cpu - years to run . the cold collapse models from paper i ( models c@xmath30 ) took @xmath31 times longer to run than the warm collapse ones ( models w@xmath30 ) for the same evolution time , which is mainly due to @xmath4 times shorter stellar crossing time in the cold models . in addition , to run the cold collapse models of paper i for the much longer evolution time of the present paper ( 3 gyr ) , we would have to increase significantly the accuracy of the integration , and increase the number of dm particles by a factor of 20 ( up to @xmath32 particles ) to avoid artificial mass segregation effects which should be very significant for such long evolution time . as a result , the time required to simulate the cold collapse models for 3 gyr would be two or more orders of magnitude longer than for the warm collapse case , or a few hundred cpu - years , which is obviously not feasible . the reasons for choosing the warm versus hot collapse model are that the warm one corresponds to a more typical gc ( with the mass of @xmath33 m@xmath34 for the warm model versus @xmath35 m@xmath34 for the hot one ) , and that the warm model of paper i exhibits interesting tidal - like features in the outskirts of the cluster ( namely , a cutoff in the surface brightness profile ) caused by the presence of dm halo , whereas the hot collapse model does not have any such features . an isothermal homogeneous stellar sphere with the universal values of the initial stellar density and velocity dispersion , @xmath36 @xmath3 pc@xmath13 and @xmath14 km s@xmath15 ( ms04 ) , is set at the center of a dm halo with either nfw or burkert density profiles . the virial mass of the dm halo is @xmath37 @xmath3 . the stellar mass is @xmath38 @xmath3 , so the baryonic - to - dm mass ratio is @xmath39 . our fiducial value of @xmath40 , on one hand , is larger than the fraction of baryons in gcs of @xmath41 @xcite , and on the other hand , is smaller than the universal baryonic - to - dm density ratio of @xmath42 @xcite . we assumed gcs have formed at @xmath10 , so the structural parameters of their dm halos are @xmath43 pc , @xmath44 pc , and @xmath45 ( from eqs . [ 1 - 9 ] of paper i ) . the initial virial ratio for our stellar clusters is @xmath46 . the initial stellar radius is 11.2 pc . as we showed in paper i , in the absence of dm our stellar cluster will first collapse ( with the smallest value of the half - mass radius of @xmath47 pc ) , and then bounce to form a relaxed cluster with a flat core , with the radius of @xmath48 pc , and a steep power - law outer density profile . the crossing time at the half - mass radius of the relaxed cluster is @xmath49 myr . the crossing times at the half - mass radius for our dm halos are 52 and 61 myr for nfw and burkert profiles , respectively . for consistency , gcs with nfw halos are assumed to orbit a host galaxy with nfw profile , and gcs with burkert halos orbit a burkert host galaxy . in all models , the pericentric distance is @xmath50 kpc , and the apocentric distance is @xmath51 kpc . we chose such a small value of @xmath52 to explore a regime of strong tidal stripping , potentially approaching a violent relaxation mode of dm stripping in the gc formation simulations of @xcite . our apocentric - to - pericentric distance ratio @xmath53 is similar to orbits of substructure in cosmological @xmath11cdm simulations . the theoretical radial orbital period is @xmath54 myr for the host galaxy with nfw profile , and @xmath55 myr for the burkert case . in our simulations , subhalos experience dynamic friction caused by the motion of the remnant in the halo of tidally stripped dm particles this effect is the most noticeable during first 12 orbits ( for the following orbits the mass of the subhalo becomes too small to be affected by the dynamic friction ) , and leads to slightly smaller values of @xmath52 , @xmath56 , and @xmath57 . all our simulations are run for @xmath58 gyr . ( this corresponds to the range of redshifts of @xmath59 . ) the actual radial orbital periods in our models are @xmath60 myr for nfw potential and 290 myr for burkert potential , so the total number of orbits is 1011 . cccccccc model & @xmath61 & @xmath62 & @xmath63 & @xmath64 & @xmath65 & @xmath66&note + & & pc & & pc & gyr & gyr & + s & @xmath67 & 0.30 & & & 3 & @xmath68 & 1 + d@xmath69 & & & @xmath70 & 1.5 & 3 & @xmath68 & 2 + d@xmath71 & & & @xmath70 & 1.5 & 3 & @xmath68 & 2 + sd@xmath69 & @xmath67 & 0.30 & @xmath70 & 1.5 & 3 & @xmath68 & 3 + sd@xmath71 & @xmath67 & 0.30 & @xmath70 & 1.5 & 3 & @xmath68 & 3 + do@xmath69 & & & @xmath70 & 1.5 & 3 & @xmath68 & 4 + do@xmath71 & & & @xmath70 & 1.5 & 3 & @xmath68 & 4 + sdo@xmath69 & @xmath67 & 0.30 & @xmath70 & 1.5 & 3 & @xmath68 & 5 + sdo@xmath71 & @xmath67 & 0.30 & @xmath70 & 1.5 & 3 & @xmath68 & 5 + in total , we run 9 different simulations ( see table [ tab1 ] ) . models s ( stars only ) , d@xmath30 ( dm only ) , and sd@xmath30 ( dm @xmath72 stars ) are evolved in isolation ( no external static gravitational field ) , and are used as reference cases for the models do@xmath30 ( dm only ) and sdo@xmath30 ( dm @xmath72 stars ) where subhalos orbit the static potential of the host galaxy . ( here subscripts `` _ n _ '' and `` _ b _ '' denote nfw and burkert dm profiles , respectively . ) for models containing dm , we generate initial distribution of dm particles using the rejection method described in paper i. halos are represented by @xmath70 particles with individual masses of @xmath73 @xmath3 and a softening length of @xmath74 pc . @xcite showed that for studies dealing with tidal disruption of substructure one should not use so - called `` local maxwellian approximation '' , where local velocity distribution is assumed to be multivariate gaussian , to set up initial conditions , as such configuration is not in equilibrium for cuspy dm density profiles and can lead to artificially high disruption rate for subhalos . for our models , we explicitly use phase - space distribution functions ( dfs ) to assign the components of velocity vectors to different particles , which guarantees the central part of the halo to be in equilibrium initially . to calculate dfs , for nfw profile we use the analytical fitting formula of @xcite , and for burkert profile we use our own fitting formula ( paper i , appendix a ) . our halos are truncated at a finite radius @xmath75 , which results in the outer parts of the halos being not in equilibrium . as we will see in [ isolated ] , this effect has negligible impact on the results of our simulations . for models containing a stellar core , stars are represented by @xmath67 equal mass particles , with individual masses of @xmath76 @xmath3 . the softening length for stars is @xmath77 pc . stars are set up as a homogeneous isothermal sphere located at the center of the dm halo ( for models containing dm ) . stellar particles initially have a maxwellian distribution of velocity vectors . as we showed in ms04 , such initial non - equilibrium configuration of a stellar cluster leads to formation of core - halo structure , with a radial density profile resembling that of gcs , after the initial relaxation phase . the model of ms04 also successfully reproduces all empiric bivariate correlations between structural and dynamic parameters of galactic gcs , given that all proto - gcs start with the same values of the stellar density @xmath12 @xmath3 pc@xmath13 and velocity dispersion @xmath78 km s@xmath15 . we use these values of @xmath79 and @xmath80 for setting up our models . our models sdo@xmath69 and sdo@xmath71 are first evolved in isolation ( with no static gravitational field ) for 120 and 170 myr , respectively . this allowed stars and dm at the center of the halo to reach a state of equilibrium . we use a parallel version of the multistepping tree code gadget @xcite to run our simulations . the values of the code parameters which control the accuracy of simulations are the same as for our warm collapse models w@xmath30 from paper i. in particular , we use a very conservative ( small ) value of the parameter @xmath81 controlling the values of the individual timesteps , which are equal to @xmath82 , where @xmath83 is the acceleration of a particle . also , the individual timesteps are not allowed to be larger than @xmath84 gyr ( see table [ tab1 ] ) . as a result , we achieved an acceptable level of accuracy in our simulations , with the total energy change @xmath85 being equal to 1.8% after 3 gyr ( or more than 6000 crossing times at the half - mass radius ) for our purely stellar model s. for models with dm ( d@xmath30 and sd@xmath30 ) this number is significantly smaller ( @xmath86% ) , because their total energy budget is dominated by low density parts of the dm halo . we can not estimate @xmath87 for our static field models do@xmath30 and sdo@xmath30 , but we expect the integration accuracy for these models to be comparable with that of the corresponding models without the static field ( d@xmath30 and sd@xmath30 ) . for all models , we output @xmath88 snapshots for different moments of time . in every snapshot we identify a gravitationally bound structure ( if present ) using the same routine as in paper i. this procedure consists of two main steps . 1 ) we use the program addgravity , which is based on the algorithm of @xcite and is part of the nemo software package , to assign local gravitational potential @xmath89 values to individual particles ( both dm and stars ) . we use the softening length values from table [ tab1 ] for purely stellar and purely dm models , and an intermediate value of @xmath90 pc for hybrid models . next , we find the 1000 ( 100 for model s ) particles with lowest potential . for these particles we calculate weighted six phase - space components of the halo center , using a normalized potential @xmath91 as a weight . then we recenter the snapshot both spatially and in velocity to the halo center . 2 ) we remove all unbound particles ( with velocity module @xmath92^{1/2}$ ] ) in a single step and recompute the potential for the remnant using addgravity . we repeat this unbinding procedure in an iterative manner until there are no unbound particles . the above procedure is reasonably fast and accurate . it failed to find a bound subhalo only in the second half of do@xmath69 run . for a few `` failed '' snapshots we had to use the program skid , which is significantly slower than our simple unbinding procedure . ( skid is often used to find gravitationally bound structure in the results of cosmological simulations . ) cccccccccccccccc model & @xmath93 & @xmath94 & @xmath95 & @xmath96 & @xmath97 & @xmath98 & @xmath99 & @xmath100 & @xmath101 & @xmath102 & @xmath103 & @xmath104 & @xmath105 & @xmath106 & @xmath107 + & @xmath3 & pc & km s@xmath15&@xmath3 pc@xmath108 & pc & pc & km s@xmath15&mag arcsec@xmath109 & pc & & @xmath3 @xmath110 & @xmath3 & pc & pc & pc + s & @xmath111 & 4.73 & 4.02 & 360 & 3.65 & 2.55 & 3.82 & 18.66 & 2.75 & 0.211 & 1.47 & & & & + s & @xmath112 & 4.68 & 4.05 & 1300 & 3.56 & 1.33 & 3.96 & 17.91 & 1.43 & 0.105 & 1.53 & & & & + d@xmath69 & & & & & & & & & & & & @xmath113 & 434 & & + d@xmath71 & & & & & & & & & & & & @xmath114 & 497 & & + sd@xmath69 & @xmath111 & 4.31 & 4.46 & 980 & 3.26 & 1.60 & 4.41 & 18.05 & 1.83 & 0.153 & 1.79 & @xmath115 & 424 & 7.7 & 19.1 + sd@xmath71 & @xmath111 & 4.77 & 4.05 & 1400 & 3.59 & 1.31 & 3.98 & 17.91 & 1.42 & 0.105 & 1.56 & @xmath113 & 484 & 17.8 & 52.9 + do@xmath69 & & & & & & & & & & & & @xmath116 & 28.4 & & + sdo@xmath69&@xmath111 & 4.31 & 4.41 & 1200 & 3.27 & 1.44 & 4.37 & 17.97 & 1.67 & 0.138 & 1.81 & @xmath117 & 38.0 & 8.6 & 24.5 + sdo@xmath71&@xmath118 & 4.72 & 4.01 & 1400 & 3.57 & 1.28 & 3.98 & 17.89 & 1.40 & 0.106 & 1.56 & @xmath119 & 65.1 & 25.7 & + in the first two lines of table [ tab2 ] we show the model parameters for an isolated stellar cluster without dm ( model s ) for two moments of time soon after the initial relaxation of the cluster ( @xmath120 gyr ; first line ) and at the end of the simulations ( @xmath121 gyr ; second line ) . as you can see , some stellar cluster parameters ( such as half - mass radius @xmath94 and central velocity dispersion @xmath95 ) stay virtually the same throughout the simulations , whereas others ( central density @xmath96 , central surface brightness @xmath100 , king core radius @xmath101 , and core mass fraction @xmath102 ) undergo significant changes . on a more detailed level , these changes can be followed in figure [ rho_st ] , where we show averaged radial density profiles for model s for four different time intervals . as you can see in this figure , the flat core of the cluster becomes smaller and denser with time . interestingly , the `` dent '' feature , seen outside of the core in the radial density profile of a freshly relaxed cluster , disappears in significantly evolved clusters . the outer part of the profile becomes increasingly more shallow with time , with the following values of the power - law exponent : @xmath122 , @xmath123 , @xmath124 , and @xmath125 . as you can see in table [ tab2 ] , by the end of the simulations , @xmath126% of core stars in model s evaporate . these stars stay gravitationally bound to the cluster , populating its outer parts and making the outer density profile more shallow . the observed secular evolution of our model stellar cluster is very similar to gravothermal instability ( or core collapse ) known to occur in real gcs . the code we use to simulate the cluster is collisionless : it softens gravitational potential of particles separated by less than the softening length @xmath62 ( which is comparable to the average distance between particles in the cluster ) . as a result , close encounters between particles are not treated correctly . moreover , our stellar particles do not represent individual stars , as they are @xmath127 times more massive than stars in gcs . nevertheless , it appears that our collisionless simulations capture the essence of the long - term dynamic evolution of gcs , probably because the main driving mechanism for such evolution is not infrequent very close encounters between stars , but rather the cumulative effect of numerous weak interactions @xcite , which are resolved reasonably well by our code . the similarity between our stellar model s secular evolution and gravothermal instability is also seen on a more quantitative level . the idealized model of core collapse of a gc predicts a correlation between the central density and the radius of the core : @xmath128 @xcite . in our model s , we observe a very similar correlation : @xmath129 for the time interval @xmath130 gyr . both the idealized theory of @xcite and our model exhibit very mild evolution of central velocity dispersion : @xmath131 and @xmath132 , respectively . the fact that our stellar particles are @xmath127 times more massive than stars in gcs results in the pace of secular evolution in our models being significantly faster than in real gcs . to estimate how dynamically old our models are at the end of simulations we use the analytical theory of core collapse of @xcite . the most sensitive dynamic age parameter is the core density @xmath96 , which according to the model of @xcite evolves with time @xmath133 as @xmath134 , where @xmath135 is the core collapse time . for our model s , @xmath136 ( see table [ tab2 ] ) , so @xmath137 at @xmath138 gyr . galactic gcs are known to span the whole spectrum of dynamic ages , ranging from dynamically young with @xmath139 ( such as @xmath9 centauri and ngc 2419 ) to post - core - collapse systems with @xmath140 ( @xmath141% of all galactic gcs ) . as you can see , at @xmath142 gyr our models span a large range of dynamic ages , corresponding to many ( probably most ) galactic gcs . in addition , the ratio of the predicted model core collapse time @xmath143 gyr to the model orbital time @xmath144 gyr is approximately the same as for dynamically old galactic gcs , which have @xmath145 ( assuming that @xmath146 gyr and @xmath147 gyr ) . as a result , our models can correctly describe both secular evolution and tidal stripping of many galactic gcs . one could treat the observed long - term dynamic evolution of stars in our models as a numeric artifact and a nuisance . instead , by arguing that this effect reflects main features of gravothermal instability in real gcs , we can use it to explore the impact of long - term dynamic evolution on the properties of hybrid ( stars @xmath72 dm ) gcs . in paper i we demonstrated that a warm collapse of a homogeneous isothermal stellar sphere inside a live dm halo ( either nfw or burkert ) produces a gc - like cluster with an outer density cutoff in the stellar density distribution resembling a tidal feature in king model . our results confirmed the idea of @xcite that a presence of dm can be an alternative explanation for apparent `` tidal '' radial density cutoffs observed in some gcs . here we use models s and sd@xmath30 to check if this explanation can be extended to stellar clusters which have experienced significant secular evolution . in figure [ rho ] we show the stellar radial density profiles for models s ( short - dashed lines ) and sd@xmath30 ( solid lines ) for the same four time intervals as in figure [ rho_st ] . as you can see , flattening of the outer density profile caused by the dynamic evolution of the cluster does not remove an apparent cutoff feature observed in dm - dominated gcs . as time goes on , the slope of the outer density profile becomes more shallow for both purely stellar and hybrid gcs , but the relative change of the slope caused by the presence of dm stays approximately the same . also , the radius where the two profiles start diverging , stays approximately the same ( @xmath148 pc ) for the nfw profile , and gradually increases ( from @xmath149 to @xmath150 pc ) for the burkert profile . for both types of dm halos , this radius is very close to the radius @xmath107 where the inclosed masses of stars and dm become equal ( see table [ tab2 ] ) . the reason for the persistence of the density cutoff in the course of the secular evolution of our cluster is the same as for the appearance of the cutoff at the end of the initial violent relaxation phase ( see paper i ) . the original cutoff is caused by the fact that at large radii ( @xmath151 ) the potential of the hybrid gc is dominated by dm . as a result , a smaller fraction of stars , ejected from the cluster during the violent relaxation , can populate outer halo , creating a density cutoff at @xmath152 . similarly , during long - term dynamic evolution of the cluster , caused by encounters between individual stars , stars are being ejected from the core ( core evaporation ) , with very few of them populating the halo beyond @xmath107 because of the potential being dominated by dm at such large radii . we use isolated dm - only models d@xmath30 to see how numerical artifacts affect the dm density distribution in our simulations . @xcite discussed the impact of discreteness effects of @xmath0-body simulations on radial density profiles of isotropic nfw halos . according to these authors , for the first @xmath153 crossing times at the virial radius , the density at the center of the halo is decreasing due to heating by faster moving particles , causing the core to expand . at the end of this phase , the central velocity dispersion becomes comparable to the velocity dispersion at the scale radius @xmath154 . ( in nfw models the velocity dispersion is highest around @xmath154 , and becomes smaller for both smaller and larger radii . ) we see a similar trend in our nfw model , d@xmath69 . in our case , the total evolution time is @xmath155 crossing times at the virial radius @xmath156 myr , so at the end of the simulations the model is still in the core - heating regime . as you can see in figure [ dm_rho]a , at @xmath157 gyr the inner dm density profile is reasonably close to the initial one down to a radius of @xmath158 pc @xmath159 . in the case of the burkert profile , the initial central velocity dispersion is also smaller than the dispersion at @xmath154 , though the difference is much smaller than for the case of nfw halo . in our model d@xmath71 , we see a slight decrease in the dm density in the inner part of the halo at the end of the simulations ( figure [ dm_rho]b ) , with a reasonably accurate profile down to a radius of @xmath160 pc @xmath161 . nfw and burkert models would be in equilibrium only if they had infinitive size and mass . as our models are truncated at at a finite radius @xmath75 , the dm density in the outer parts of the halos becomes lower with time . as you can see in figure [ dm_rho ] , even at the end of the simulations the deviation of the outer dm density profile from the theoretical one is not significant within @xmath75 , and is negligible for @xmath162 . most of the changes in the outer density profiles happen within first one or two @xmath163 , which is comparable to the orbital period in models do@xmath30 and sdo@xmath30 . all the above let us conclude that the truncation of our models at the radius of @xmath75 should not have a noticeable impact on the results of our simulations . @xcite showed that an isotropic nfw subhalo , following a circular orbit inside the static potential of a host galaxy with nfw profile , can be completely disrupted by tidal forces after a few orbits if its tidal radius @xmath164 is smaller than @xmath165 . these authors argued that the relative easiness to disrupt an nfw satellite is linked to the fact that when an nfw halo is instantaneously truncated at a radius @xmath166 , it becomes unbound ( with the total energy @xmath167 of the remnant becoming positive ) . the situation with singular isothermal spheres is completely different , as such halos stay gravitationally bound for any @xmath168 , and potentially can survive indefinitely in an external tidal field . in figure [ e_tot ] we compare the binding properties of our nfw and burkert halos ( thick and thin long - dashed lines , respectively ) . as you can see , burkert halos are similar to nfw ones in becoming unbound if truncated below a certain radius . burkert halos appear to be much easier to disrupt tidally than nfw halos : in the case of burkert profile , @xmath169 , which is @xmath170 times larger than for nfw profile . also , the positive total energy of a burkert halo truncated to @xmath171 is @xmath172 times larger than the corresponding quantity for an nfw halo with the same concentration and mass ( see figure [ e_tot ] ) . in figure [ a_tid ] we compare the strength of the tidal force as a function of radius for two types of host galaxies : with nfw profile ( thick line ) and with burkert profile ( thin line ) . more specifically , in this figure we plot the radial dependence of the radial gradient of gravitational acceleration @xmath173 $ ] . a product of this quantity on the linear size of a subhalo gives an estimate of the differential ( tidal ) acceleration between two opposite parts of the subhalo . two vertical dotted lines in figure [ a_tid ] show the range of radial distances covered by our models . as you can see in figure [ a_tid ] , for nfw profile the quantity @xmath174 is always positive ( meaning that the tidal force is always stretching a subhalo in radial direction ) . in the small radii limit , @xmath174 asymptotically approaches a constant , which is equal to @xmath175^{-1}\ , gm_{\rm vir}/r_s^3 $ ] . for burkert halos , the radial tidal force reaches a maximum at @xmath176 , becomes equal to zero at @xmath177 , and asymptotically approaches a negative constant @xmath178^{-1}\ , gm_{\rm vir}/r_s^3 $ ] at small radii . the negative sign for @xmath174 means that at @xmath179 the radial tidal force becomes compressing instead of being stretching . in the interval of radii covered by the orbital motion of our subhalos , the nfw host halo has significantly stronger radial tidal acceleration than the burkert halo ( see figure [ a_tid ] ) . to summarize the above analysis , the burkert halos are easier to disrupt tidally than nfw halos . on the other hand , a tidal field of a burkert host galaxy appears to be less disrupting than that of an nfw host galaxy with the same mass and concentration . in addition , burkert halos have an unusual property of having a compressing radial tidal force within the scale radius @xmath21 . one has thus to resort to numerical simulations to understand differences in substructure evolution for nfw and burkert cases . in our models do@xmath30 , a dm - only ( nfw or burkert ) subhalo is orbiting on eccentric ( @xmath53 ) orbit inside a static potential of the host ( nfw or burkert ) galaxy . we use the same definition of @xmath164 as @xcite : @xmath180 \frac{m(r)}{r^3}. \label{eq_rtid}\ ] ] here @xmath181 and @xmath182 is enclosed mass as a function of radius for the host and satellite halos , respectively . at @xmath183 , the subhalos have the following values of tidal radii : @xmath184 pc @xmath185 for nfw profile , and @xmath186 pc @xmath187 for burkert profile . in figure [ dm_mass]a , we show the evolution of the gravitationally bound dm mass for model do@xmath188 ( solid thin line ) . as you can see , tidal stripping is severe in this model , with only @xmath189% of the total mass surviving as a bound structure after 11 orbits ( see table [ tab2 ] ) . this behavior is similar to the critical case of @xcite with @xmath190 ( see their fig . 7 ) , which separates their models staying relatively intact ( with @xmath191 ) and completely disrupted models ( with @xmath192 ) . in our case , this ratio is slightly smaller than 2.1 : @xmath193 . slightly larger resilience to tidal disruption exhibited by our model do@xmath188 can be explained by the fact that @xcite used `` local maxwellian approximation '' to set up the initial particle distribution in their models , which according to @xcite leads to artificially high disruption rate for nfw subhalos . even for the very last snapshot of model do@xmath188 ( at @xmath138 gyr ) , the stripped - down subhalo appears to be reasonably stable against total disruption by the tidal forces . indeed , the binding radius of the remnant @xmath194 pc is significantly smaller than the current value of the tidal radius @xmath195 pc even at the very end of the simulations . ( to estimate @xmath164 , we use the actual value of the pericentric distance of @xmath196 pc which is smaller than the original value of @xmath197 pc because of the dynamic friction experienced by the subhalo moving in the halo of unbound dm particles . ) we expect thus that our nfw subhalo should survive as a gravitationally bound remnant for a few more orbits ( and perhaps indefinitely ) . the fate of a burkert dm subhalo orbiting inside a burkert host galaxy ( model do@xmath198 ) is completely different , as can be seen in figure [ dm_mass]b ( solid thin line ) . after only 3 pericentric passages , the subhalo ceases to exist as a coherent structure . to make sure that the failure to find a bound structure after @xmath199 gyr is not an artifact of our halo finding algorithm , we measure both binding radius @xmath200 pc and tidal radius @xmath201 pc for the last snapshot containing a bound structure . as you can see , at this point the subhalo is bound to be dispersed by tidal forces as its current tidal radius is significantly ( almost 4 times ) smaller than the binding radius . two main conclusions we arrive at in this section are ( 1 ) our nfw dm - only subhalo is relatively resilient to total tidal disruption , and ( 2 ) a burkert dm subhalo orbiting inside a burkert host galaxy is much easier to disrupt that for the case of an nfw dm subhalo orbiting inside an nfw host . we are now turning to the results of our two principle models : sdo@xmath69 and sdo@xmath71 . in these models , a hybrid gc ( stars @xmath72 dm ) is orbiting inside the same host galaxies and along the same eccentric orbit as in models do@xmath30 . before being placed in the static potential of the host galaxy , we allow stars and dm to relax in isolation for 120 myr ( model sdo@xmath69 ) and 170 myr ( model sdo@xmath71 ) . as you can see in figure [ dm_mass ] , presence of a stellar core , with a mass of mere @xmath202% of the total mass , inside a dm subhalo makes the subhalo much more resilient to tidal forces disruption . in the case of nfw profile ( figure [ dm_mass]a ; see also table [ tab2 ] ) , the dm mass of the tidally stripped remnant at @xmath157 gyr is 7 times larger in the presence of the stellar core than in the dm - only case . it is also 3 times larger than the mass of the stellar core . for burkert profile , the difference is even more dramatic : whereas a starless burkert subhalo is completely disrupted after 3 orbits , the subhalo with a stellar core survives till the end of the simulations , with @xmath203% of the total mass of the remnant being in dm form . figure [ e_tot ] provides an explanation for such a marked difference . as we discussed in [ prelim ] , both nfw and burkert halos become unbound if instantaneously truncated below a certain radius ( @xmath204 for nfw profile , and @xmath205 for burkert profile ) . as you can see in figure [ e_tot ] , the presence of a stellar gc - like cluster at the center of either halo makes it significantly more bound . in the case of nfw profile ( thick solid line ) , the halo becomes bound for virtually any truncation radius @xmath168 ( being similar in this regard to a singular isothermal sphere ) . for burkert halo ( thin solid line ) , there is still a range in truncation radius @xmath168 where the halo is formally unbound , but the maximum positive value of @xmath167 is significantly lower in the presence of a stellar core . moreover , for @xmath206 the halo again becomes bound ( see figure [ e_tot ] ) . after 7 orbits or @xmath207 gyr , the mass of dm gravitationally bound to the remnant in model sdo@xmath71 becomes smaller than the mass of the stellar core . at this point , the rate of the mass loss due to tidal stripping becomes smaller , but still non - negligible ( see figure [ dm_mass]b ) . we tested the possibility that the observed decrease in bound dm mass after @xmath208 gyr is a numerical artifact , caused by a small number ( @xmath209 ) of dm particles attached to the remnant , by resimulating the late evolution of the subhalo in the absence of the static gravitational potential of the host galaxy . for this , we used the gravitationally bound remnant from the @xmath210 gyr snapshot of model sdo@xmath71 . the bound dm mass evolution for this additional run is shown with dotted line in figure [ dm_mass]b . as you can see , in the absence of external tidal field the bound dm mass of the remnant stays virtually constant . it appears thus very unlikely that the late time dm stripping observed in model sdo@xmath71 is caused by numerical artifacts ( such as evaporation of dm particles due to two - body interactions ) . in figure [ column ] we show dm column density maps for models sdo@xmath69 ( left panels ) and sdo@xmath71 ( right panels ) , for two moments of time : after @xmath211 orbits ( top panels ) and after @xmath212 orbits ( bottom panels ) . two concentric circles show the range of radial distances covered by the models . at the earlier moment of time , you can see a long stream of dm particles , stripped during the first pericentric passage , with a relatively massive bound structure in the center of the stream . at the end of the simulations , multiple streams of stripped particles mix together to create a fuzzy thick disk of dm . the gravitationally bound remnant is barely visible at this point ( especially for burkert model ) . in figure [ dm_rho ] we show averaged dm radial density profiles for @xmath213 gyr for models d@xmath30 ( thick solid lines ) , sd@xmath30 ( thin solid lines ) , and sdo@xmath30 ( short - dashed lines ) . as you can see , the dm density in the innermost part of the nfw halo is not modified by external tidal field . outside of the stars - dominated central region , the dm density profile becomes significantly steeper in the presence of tidal field . at this point , in models sdo@xmath30 dm is stripped down to the original scale radius @xmath154 . analysis of table [ tab2 ] shows that the parameters of our stellar clusters at the end of the simulations are very similar for all our models containing stars ( s , sd@xmath30 , and sdo@xmath30 ) . for sd@xmath69 and sdo@xmath69 models , the presence of relatively large amounts of dm at the center of the cluster leads to somewhat larger values of the central dispersion @xmath95 and smaller values of the king core radius @xmath101 . the value of the apparent mass - to - light ratio @xmath103 , defined as ( see paper i ) @xmath214 where @xmath215 is the assumed baryonic mass - to - light ratio in gcs and @xmath216 is the projected stellar surface mass density at the center of the cluster , is larger by @xmath148% for a stellar cluster in nfw halo . overall , the presence of an external tidal field does not seem to change the global structural and dynamic parameters of gcs with dm in a noticeable way . as can be seen in figure [ rho ] , tidal stripping does not modify significantly the stellar radial density profiles of hybrid gcs : at any moment of time , density profiles for sdo@xmath30 models ( dotted lines ) are very close to the density profiles of sd@xmath30 models ( solid lines ) . in figure [ rho2 ] we show the final radial density profiles for both stars and dm for models sdo@xmath69 and sdo@xmath71 . as you can see , dm still dominates stars in the outskirts of the stellar cluster . in the stars - dominated area , the dm density profile is steeper because of the adiabatic contraction of dm in the presence of stars . dm density profiles are dramatically modified both by the presence of a stellar core and by tidal stripping . in model sdo@xmath71 , stellar cluster becomes tidally limited by the end of the simulations . by @xmath138 gyr , around 0.6% of stars ( or @xmath217 stellar particles ) are not gravitationally bound to the cluster , forming distinctive trailing and leading stellar tidal tails . ( conversely , not a single stellar particle has been tidally stripped by @xmath138 gyr in model sdo@xmath69 . ) in model sdo@xmath71 , the shape of the cluster becomes increasingly non - spherical at large radii at @xmath218 gyr . as can be seen in figure [ sigma]a , there is an excess of stars in the outskirts of the cluster along its major axis , and a sharp tidal cutoff around the analytical tidal radius @xmath164 near the plane perpendicular to the major axis . ( to calculate surface brightness and velocity dispersion profiles for sdo@xmath71 model , we used all stellar particles both bound and unbound . ) as figure [ sigma]a shows , the final surface brightness profiles of the stellar clusters in sdo@xmath30 models look remarkably similar to the corresponding profiles of galactic gcs . ( we assumed that the baryonic v - band mass - to - light ratio in gcs is @xmath219 see discussion in paper i. ) the `` dent '' feature observed in the surface brightness profile of a freshly relaxed hybrid ( dm @xmath72 stars ) gc ( see paper i ) has been removed in models sdo@xmath30 by secular evolution of the stellar cluster ( see [ secular ] ) . the apparent `` tidal '' cutoff in the outer surface brightness profile of model sdo@xmath188 is caused by the presence of significant amounts of dm in the outskirts of the cluster ( similarly to paper i ) . in the case of burkert halo ( model sdo@xmath198 ) , the cutoff is of truly tidal nature . in figure [ sigma]b we show the final line - of - sight velocity dispersion profiles for the stellar clusters in sdo@xmath30 models . in the case of nfw halo ( thick solid line ) , the line - of - sight velocity dispersion appears to be uniformly inflated by a small factor ( @xmath127% ) across all radii , which can be misinterpreted as a purely stellar cluster with a somewhat larger value of baryonic mass - to - light ratio . for the burkert halo ( thin solid lines ) , the radial velocity dispersion profile for the stars near the plane perpendicular to the major axis is very close to the profile of a purely stellar cluster , whereas the stars along the major axis show signs of being tidally heated . dm subhalos with burkert density profile are much easier to disrupt tidally in the potential of the host galaxy than subhalos with nfw profile . we link this effect to the difference in the binding properties of both types of halos , with the total energy of the burkert halos becoming positive if truncated at a significantly larger radius than for nfw halos . setting a low - mass ( @xmath220% of the total mass ) dense stellar core at the center of either nfw or burkert halo makes them much more resilient to tidal disruption . primordial gcs with nfw dm halo can survive severe tidal stripping in the host galaxy , with dm still being the dominant mass component ( though not by a large margin ) in the tidally stripped - down remnant . dm is concentrated in the outskirts of the remnant . as a result , an apparent `` tidal '' cutoff in the radial surface brightness profile in isolated warm collapse models ( paper i ) , caused by the presence of dm , is also present in our tidally stripped nfw model . we used warm collapse hybrid models to show that neither secular evolution of the stellar cluster nor severe tidal stripping change noticeably the inferred core mass - to - light ratio @xmath103 . for both flat - core and cuspy dm halo profiles , @xmath103 stays close to the purely baryonic value . tidal stripping can remove almost all dm from primordial gcs with burkert dm halo . the remaining dm is dynamically unimportant , and can not prevent stars being stripped off by tidal forces of the host galaxy . this result makes gcs possessing obvious tidal tails ( the only known example being palomar 5 ) be fully consistent with primordial scenarios of gc formation . secular evolution of a dm - dominated gc does not change the main results of paper i derived for freshly relaxed warm - collapse systems . in particular , in both unevolved and evolved clusters , presence of significant amounts of dm in the outskirts of the cluster manifest itself as a `` tidal '' cutoff in the outer part of the radial surface brightness profile . it is not clear if the `` extratidal '' features of cold collapse models from paper i will be preserved in significantly evolved clusters . as we discussed in [ physical ] , simulating a cold - collapse hybrid gc to address this issue is not feasible with the present day technology . the above results reinforce our conclusion from paper i that a presence of obvious tidal tails is probably the only observational evidence which can reliably rule out a presence of significant amounts of dm in gcs . ( but it can not rule out primordial scenarios of gc formation . ) to summarize , the results presented in both paper i and this paper suggest that the whole range of features seen in milky way gcs ( from apparently truncated profiles of some clusters to the extended tidal tails of palomar 5 ) can be consistent with the primordial picture of gc formation , given that there was a range of the initial virial ratios for stellar clusters ( from `` cold '' to `` hot '' collapses ) , tidal stripping histories , and/or inner dm density profiles . we would like to thank volker springel for his help with introducing an external static potential in gadget . s. m. is partially supported by sharcnet . the simulations reported in this paper were carried out on mckenzie cluster at the canadian institute for theoretical astrophysics . barkana , r. & loeb , a. 2001 , physics reports , 349 ( issue 2 ) , 125 beasley , m. a. , kawata , d. , pearce , f. r. , forbes , d. a. , & gibson , b. k. 2003 , , 596 , l187 becker , r. h. , et al . 2001 , , 122 , 2850 bekki , k. & chiba , m. 2004 , , 417 , 437 bellazzini , m. , ferraro , f. r. , & ibata , r. 2003 , , 125 , 188 bromm , v. & clarke , c. j. 2002 , , 566 , l1 burkert , a. 1995 , , 447 , l25 cen , r. 2001 , , 560 , 592 dehnen , w. 2000 , , 536 , l39 grillmair , c. j. , freeman , k. c. , irwin , m. , & quinn , p. j. 1995 , , 109 , 2553 hayashi , e. , navarro , j. f. , taylor , j. e. , stadel , j. , & quinn , t. 2003 , , 584 , 541 irwin , m. & hatzidimitriou , d. 1995 , , 277 , 1354 kazantzidis , s. , magorrian , j. , & moore , b. 2004 , , 601 , 37 law , d. r. , majewski , s. r. , skrutskie , m. f. , carpenter , j. m. , & ayub , h. f. 2003 , , 126 , 1871 layden , a. c. & sarajedini , a. 2000 , , 119 , 1760 leon , s. , meylan , g. , & combes , f. 2000 , , 359 , 907 mashchenko , s. & sills , a. 2004a , , 605 , l121 ( ms04 ) mashchenko , s. & sills , a. 2004b , , in press ( preprint astro - ph/0409605 ) ( paper i ) mclaughlin , d. e. 1999 , , 117 , 2398 mizutani , a. , chiba , m. , & sakamoto , t. 2003 , , 589 , l89 moore , b. 1996 , , 461 , l13 navarro , j. f. , frenk , c. s. , & white , s. d. m. 1997 , , 490 , 493 odenkirchen , m. , et al . 2001 , , 122 , 2538 odenkirchen , m. , et al . 2003 , , 126 , 2385 padoan , p. , jimenez , r. , & jones , b. 1997 , , 285 , 711 peebles , p. j. e. 1984 , , 277 , 470 peebles , p. j. e. & dicke , r. h. 1968 , , 154 , 891 ricotti , m. , gnedin , n. y. , & shull , j. m. 2002 , , 575 , 49 rosenblatt , e. i. , faber , s. m. , & blumenthal , g. r. 1988 , , 330 , 191 spergel , d. n. , et al . 2003 , , 148 , 175 spitzer , l. 1987 , dynamical evolution of globular clusters ( princeton , nj : princeton university press ) springel , v. , yoshida , n. , & white , s. d. m. 2001 , new astronomy , 6 , 79 tsuchiya , t. , dinescu , d. i. , & korchagin , v. i. 2003 , , 589 , l29 widrow , l. m. 2000 , , 131 , 39 yoshida , n. , abel , t. , hernquist , l. , & sugiyama , n. 2003 , , 592 , 645

there are many observational examples of starbursts occurring in active galaxies ( e.g. mauder et al . 1994 ; maiolino , rieke & keller 1995 ; heckman et al . , 1997 ; gonzalez - delgado et al . , 1998 ; nelson & whittle , 1998 ; kunth & contini , 1999 ) . as a result , possible connections between active galaxies and circumnuclear starbursts are the subject of extensive speculation . such associations might arise just because circumnuclear starbursts have a common cause with agns . for example , mergers , interactions between galaxies , and bars ( shlosman , frank & begelman , 1989 ; shlosman , begelman & frank , 1990 ; maiolino et al . , 1997 ; knapen , shlosman & peletier , 1999 ) have all been invoked as a means to distort the gravitational potential of the host galaxy , allowing the gas to lose angular momentum to the stars , so it can fall inward toward the agn . all of these phenomena appear also to be associated with nuclear starbursting . a more direct connection has been proposed by von linden et al . ( 1993 ) , who suggest that starburst - induced turbulence in the ism is responsible for the final stages of infall of gas into the nuclear engine . weedman ( 1983 ) pioneered an agn model involving a large number of small accretors - the remnants of the massive stars in a compact circumnuclear starburst , preceding the agn stage . bailey ( 1980 ) , david ( 1987a , 1987b ) , and norman & scoville ( 1988 ) refined this idea , pointing out that a compact star cluster can form a super massive black hole . the theoretical understanding of this process was advanced by the efforts of perry & dyson ( 1985 ) , perry ( 1992 ) , and more recently , of williams , baker & perry ( 1999 ) . it has also been proposed that a starburst can mimic an agn ( terlevich & melnick , 1985 ) . terlevich et al . ( 1992 ) showed that at least low luminosity seyfert 2 s can be explained via active circumnuclear star formation alone . however , this model has difficulties explaining the rapid x - ray variability , and the broad iron emission lines of more luminous active galaxies . cid fernandes ( 1997 ) summarized the current status of the model and suggested a dichotomy among agns , in which some are powered by a black hole and others by a nuclear starburst . still another possibility is that circumnuclear starbursts strongly alter the observed properties of agns . for example , ohsuga & umemura ( 1999 ) consider a model in which the wind from the circumnuclear starburst forms an obscuring dust wall . the alternative process , in which an agn disturbs the galactic ism and triggers a starburst was reported by van breugel et al . ( 1985 ) , van breugel & dey ( 1993 ) , and dey et al . this behavior is usually related to the expansion of a radio source into the ambient medium . also , once formed , by whatever means , the black hole can gravitationally disturb the interstellar gas , triggering a starburst . a better understanding of the frequency of agn and starburst coincidences is required to probe the connection between the two phenomena . a conventional method to search for young starbursts is to look for stellar absorption features in the ultraviolet spectra of galaxies ( e.g. heckman et al . , 1997 ; gonzalez - delgado et al . , 1998 ) . current instrumentation allows this approach to be applied in just a few cases ; heckman ( 1999 ) found only four seyfert 2 nuclei bright enough to obtain uv spectra with sufficiently high signal - to - noise to identify young stars . it is possible that they show starburst signatures because of the selection bias toward galaxies with high surface brightness , which could be a result of the active star formation . in addition , application of this technique to agns is restricted by the extinction . since the ism lies around the agn in a complex three - dimensional distribution , a significant component of the observed uv flux may come from the stellar population in front of the active nucleus and not be associated closely with it . in the optical , kunth & contini ( 1999 ) detected emission signatures of wolf - rayet stars in the spectra of two seyfert 2 galaxies out of five observed . caii triplet absorption at @xmath3 is strong in the spectra of massive stars and can also indicate the presence of starbursts ( terlevich , diaz & terlevich , 1989 ; terlevich , diaz & terlevich , 1990 ; forbes , boisson & ward , 1992 ; nelson & whittle , 1998 ) . however , studies of this line in seyfert galaxies were not conclusive because for a fraction of these galaxies the caii index values were inconsistent with a normal galaxy spectrum diluted by a featureless power law continuum ( see also persson , 1988 ; diaz , terlevich & terlevich , 1989 ) . the infrared range has an advantage in probing agn environments because of the lower extinction . since the strengths of the co bands at @xmath4 and @xmath2 are sensitive to the stellar surface gravity , they can indicate the presence of young red supergiants . the co band strength is also a function of metallicity . however , in the bulges of large spirals , the typical haunt of a seyfert nucleus , the surface gravity dependence dominates . empirically , quiescent spirals have a very narrow range of @xmath2 band strengths ( frogel et al . , 1978 ; frogel , 1985 ) and most starburst galaxies have bands substantially deeper than this value ( ridgway , wynn - williams & becklin 1994 ; goldader et al . 1995 ; engelbracht 1997 ) although the @xmath2 strength has proven useful in identifying and studying starbursts , in applying similar techniques to seyfert galaxies one must contend with the possibility that the band can be diluted by radiation from the warm dust surrounding the central engine in the agn . oliva et al . ( 1995 ) therefore used the second overtone at @xmath4 to probe the environments of agns . unfortunately , this band is weak and its dependence on metallicity is strong comparable to @xmath5 m co band . to avoid these issues , oliva et al . ( 1999 ) derived the mass to luminosity of the nuclear region and identified starburst / seyfert combinations where this ratio was low , rather than relying on the co band strengths . both of these approaches require high signal to noise spectra of moderately high resolution . thus , they can only be applied to agns that are bright in the near infrared , which may result in a selection bias toward examples with strong circumnuclear starbursts . to address the seyfert - starburst connection more thoroughly , we obtained @xmath6-band spectra of 46 seyfert galaxies in the cfa sample at the @xmath0 m co bandhead . in addition , we obtained spectra of 5 ellipticals , and used 19 starburst galaxy spectra from engelbracht ( 1997 ) for comparison . with this sample , we hope to alleviate the typical small number statistics of previous studies , and to overcome the bias toward galaxies with high surface brightness . we apply a technique based on @xmath1 aperture photometry and imaging to correct for the dilution of the co band from dust heated by the agn . we then compare the corrected co band strengths with those of the ellipticals and starbursts . the infrared spectra were taken over a period of four years at three different telescopes with fspec ( williams et al . , 1993 ) , a cryogenic long slit near infrared spectrometer utilizing a nicmos3 256x256 array . we used @xmath7 , @xmath8 , and @xmath9 resolution modes , corresponding to gratings of 75 , 300 , and 600 lines / mm . the slit width was 1.2 arcsec at the mmt , 2.4 and 3.6 arcsec at steward observatory 2.3 m and 1.55 m telescopes . the plate scales were 0.4 , 1.2 and 1.8 arcsec / pixel respectively . the observing log is presented in table [ tbl-1 ] . the galaxies are sorted in order of right ascension . the spectra were reduced using iraf tasks written specifically for fspec ( engelbracht , 1999 ) . an average of sky background images taken immediately before and after each object image was subtracted to remove the sky emission lines , and simultaneously , the dark current and offset level . dark - subtracted dome flats were used to correct for pixel - to - pixel sensitivity variations . then the object images were shifted and combined to produce a single two - dimensional spectrum . known bad pixels were masked out . the large number of object images allowed us to exclude any remaining bad pixels and cosmic rays . the one - dimensional spectrum was extracted by fitting a 3 - 5 order polynomial to the continuum in the two - dimensional image . the angular sizes of the extraction apertures are given in table [ tbl-1 ] . the object spectra were divided by spectra of stars close to solar type , observed at the same airmass , and then multiplied by a solar spectrum to remove the effects of the atmospheric absorption ( maiolino , rieke & rieke , 1996 ) . the true shape of the continuum was restored by multiplying the spectra by the ratio of black bodies with standard star and solar effective temperatures . we used the oh airglow lines ( oliva & origlia , 1992 ) , complemented when necessary with ne - kr comparison lamp spectra , for wavelength calibration . special attention was paid to determining the signal - to - noise of our data . the main sources of uncertainty are the photon statistics for faint objects , the strong sky emission lines and sky variation . first , we calculated the photon statistics , taking into account the object , the sky emission and the contribution from the standard star used for atmospheric correction . this is an optimistic estimate since it neglects the sky variations . second , we calculated the noise from a relatively featureless part of the spectrum , which gives a pessimistic estimate because weak absorption lines were included as noise . finally , for a few galaxies we calculated the signal - to - noise in the process of combining one - dimensional spectra extracted from each individual two - dimensional image . this is the most robust method and it accounts for all possible sources of error . the last two determinations gave virtually the same answers . the second technique is easier to implement but it has the disadvantage of estimating the noise level only on a clear continuum part of the spectra . to obtain the noise level at the lines and bands of interest , we rescaled the photon statistics noise spectrum to match the noise determined from the clear part of the spectra . the corresponding averaged signal - to - noise ratio is shown in table [ tbl-1 ] . the line measurements used gaussian fitting . broadband @xmath6 images , kindly provided by mcleod & rieke ( 1995 ) , were used to flux calibrate the spectra using the absolute calibration of campins , rieke & lebofsky ( 1985 ) . in turn , most of the images were re - calibrated using our aperture photometry ( discussed below ) . the spectra in figure [ fig1 ] are divided by a low - order polynomial fit to the continuum , and shifted vertically , for display purposes . the individual absorption feature measurements are listed in tables [ tbl-2 ] , and [ tbl-3 ] . emission lines will be discussed in a separate paper . @xmath10 errors for the flux measurements include uncertainties both in the spectra ( sky variation , continuum placement , shot noise ) and in the flux calibration . @xmath10 errors for the equivalent line widths include only the uncertainties in the spectra which are dominated by the continuum placement error . the latter was estimated by taking measurements at different continuum positions . empty items in the tables denote a lack of spectral coverage , and zero equivalent widths stand for a non - detection . in the latter case , the @xmath10 error has the meaning of a @xmath10 upper limit . our high resolution spectra lack the wavelength coverage to measure the photometric co index of frogel et al . ( 1978 ) or the spectroscopic co index of doyon , joseph & wright ( 1994 ) . we defined a narrower co index to utilize the complete set of data . our index has the advantage of being insensitive to extinction . it follows the recipe of doyon , joseph & wright ( 1994 ) , but the measured regions are narrower and closer : @xmath11 where @xmath12 is flux averaged in the band within a @xmath13 wide window centered at @xmath14 , and @xmath15 is flux averaged on nearby continuum within a @xmath13 wide window centered at @xmath16 , in the rest frame of the object . a narrow index that does not encompass totally the spectral feature is sensitive to velocity dispersion variations . to alleviate this problem , we left out ngc 5990 and ngc 6240 from the comparison starburst sample , since they are known to have exceptionally large velocity dispersions . ngc 6240 is also not a typical starburst galaxy , since its faint @xmath17 is indicative of a relatively old burst of star formation . an additional incentive to use narrower bandpasses is related to possible uncertainties in the continuum shape of the infrared spectra . the co feature is close to the edge of an atmospheric window and we can not measure the continuum level on both sides of the spectral feature . although during the data reduction the original continuum shape was restored after the spectra were multiplied by the solar spectrum , we could not exclude some small deviations in the continuum shape . the narrow co index minimizes the effect of the continuum shape on our analysis . to quantify this effect we used our @xmath18 aperture photometry to calculate the continuum slope at @xmath2 via simple linear interpolation . we imposed this slope on continuum divided spectrum and measured the co index again . the statistical errors in the co index @xmath19 mag for a spectrum with s / n=30 per pixel ) are negligible , compared to the systematic errors due to the uncertain continuum slope . to determine the true uncertainties , we took advantage of the multiple spectra of some galaxies , taken on different nights , and with different spectral resolution . we found the typical total uncertainty in the co index to be 0.02 mag . finally , we compared our co index with that of doyon , joseph & wright ( 1994 ) calculating both indices for the set of stellar spectra by kleinmann & hall ( 1986 ) . for 26 stars , the linear conversion is : @xmath20 with standard deviation of residuals of 0.02 . that is , within the errors the two indices are equivalent . our co index is related to other systems of co indices via transformations derived by doyon , joseph & wright ( 1994 ) . the aperture photometry was carried out at the mmt using the facility single pixel insb aperture photometer . aperture diameters , fluxes and observed colors with @xmath10 errors are shown in table [ tbl-4 ] . the calibration is based on the standard star network of elias et al . ( 1982 ) with zero points based on campins , rieke & lebofsky ( 1985 ) . with data already available in the literature , a full set ( jhkln ) of ground - based photometry is now available for the entire cfa sample . @xmath6-band imaging was taken from the survey of mcleod & rieke ( 1995 ) . all the images were calibrated using our aperture photometry . many of the images had independent calibrations from mcleod & rieke ( 1995 ) , and a comparison with the result from the photometry showed good agreement . the calibration was transferred to the spectra by extracting fluxes with a rectangular aperture that matched the slit size for the corresponding spectrum . to obtain an upper limit to the uncertainties imposed by field rotation at the mmt , we placed this pseudo slit in 4 orientations rotated by 45 degrees on the nucleus of each galaxy . in most cases , the errors due to slit rotation were the dominant source of uncertainty . these uncertainties and the statistical ones were propagated through our co band correction procedure to obtain a final error on the co correction . the goal of this study is to probe the presence of circumnuclear starbursts in the seyfert galaxies by measuring their 2.3@xmath21 m first overtone co bandhead strength . the strength of this band depends on stellar temperature , metallicity , and surface gravity / luminosity . these multiple dependencies suggest that interpretation of the band would be difficult . however , in the case of nuclear starbursts in luminous galaxies , the behavior is dominated by the stellar luminosity . the best evidence for this conclusion is empirical . from the work of ridgway , wynn - williams & becklin ( 1994 ) , we compute that 23 `` normal '' galaxies recalibrated to the spectroscopic scale have a co index of 0.20 . because of the tight distribution , this value is very well determined . in comparison , from this same work we find that the average co band absorption index in 14 nuclear starbursts is more widely distributed , but with an average value of 0.27 ( as suggested by ridgway et al . , we include only starbursts with @xmath22 the study of luminous infrared galaxies by goldader et al . ( 1995 ) includes 31 examples dominated by star formation , for which the average co index corrected to the spectroscopic scale is also 0.27 . the average photometric co index of the higher quality spectra of engelbracht ( 1997 ) is @xmath23 . the data of oliva et al . ( 1995 ) show a similar behavior . for four low metallicity starbursting galaxies with well developed co bands , the band strengths fall close to the average for normal galaxies ( vanzi & rieke , 1997 ) . the metallicities of these galaxies are typically 3 to 5 times lower than solar . thus , metallicity does have the expected effect on starburst co band strengths , but it is only a modest one . because of this modest dependence , it is unlikely that the strong co bands in starbursts could arise from high metallicity , particularly since there is little evidence for higher than solar metallicities in starbursts . the starburst co band strengths measured by ridgway , winn - williams & becklin ( 1994 ) and goldader et al . ( 1995 ) do show a range that overlaps with the average value for normal galaxies , so it would be hazardous to conclude that any given galaxy with a normal co band depth could not harbor a starburst . however , there are few if any normal galaxies with bands as deep as the average for the starbursts , so deep bands are a relatively reliable indication of a recent starburst . this conclusion becomes even stronger in comparing samples of galaxies , since a few outliers in the distributions will have little effect on average values of the co index . in principle , the co bands will be weak if the starburst is so young that hot stars make a significant contribution to the @xmath6-band light . this situation may underly the few starbursts with weak co. however , this phase is short - lived ; the red supergiants start to dominate the near infrared after only a few million years , and continue to do so , until the burst is 50 - 100 myr old . agns often have spectral energy distributions peaking in the mid infrared due to emission by hot dust , and this component could fill in the @xmath2 absorption and weaken it . to use our spectra to estimate intrinsic co band strengths requires that we correct for this emission . the hot dust emission rises steeply between the @xmath6 and @xmath24(3.5@xmath21 m ) bands ( e.g. , mcalary & rieke 1988 ) . in most sy1s , where the hot dust dominates the emission in l , the estimates of hot dust emission at 2.3@xmath21 m are too uncertain to allow an accurate correction to the co band strength . in most sy2s , the excess emission above the stellar contribution at l is relatively weak , however , and we can conclude that the dust emission at 2.3@xmath21 m is small enough that even approximate corrections can yield a useful determination of the intrinsic co band strength . the distribution of observed co indices is shown on the top panel in figure [ fig2 ] . we assume that the observed fluxes at @xmath2 are the sums of similar stellar and non - stellar contributions . we have developed a procedure to disentangle the two based on a priori knowledge of their typical infrared colors . \(i ) our first step is to assume that the @xmath6-band emission is totally of stellar origin . as we shall see , this assumption is usually not accurate but by repeating the procedure iteratively we can correct for the remaining non - stellar contribution . \(ii ) next we estimate the stellar component in @xmath24-band using the typical @xmath25 color for stars . the infrared colors of normal galaxy nuclei are well known and are not a strong function of the stellar population . we adopted the typical colors of ellipticals as determined by aaronson ( 1977 ) : @xmath26 , @xmath27 , and by impey , wynn - williams & becklin ( 1986 ) : @xmath28 . \(iii ) then we subtract the stellar contribution at @xmath24 from the observed flux , and obtain the non - stellar @xmath24-band flux . this step is facilitated by the equal diameter of @xmath6 and @xmath24 band apertures . \(iv ) the next step is to use the typical @xmath25 color for the non - stellar component to estimate the non - stellar contribution in the @xmath6-band . typical colors of the non - stellar component were inferred from seyferts which show no or very weak co absorption , hence excluding galaxies with a significant stellar contribution . we selected eight galaxies with observed co index @xmath29 and obtained an average nuclear color @xmath30 mag , as a difference between the @xmath6 flux measured within the slit aperture and the @xmath24-band flux within the circular aperture . here we assume that the non - stellar component is concentrated in a point - source , and is fully encompassed by both the circular apertures and the slit . \(v ) we then subtract a featureless spectrum , normalized to the non - stellar @xmath6-band flux , from the flux - normalized @xmath6-band spectrum . the resulting spectrum is corrected for dilution . the procedure has to be repeated to correct for the inaccurate initial assumption but it converges after only one or two iterations . the correction can be applied in other ways , such as starting at @xmath31 , or @xmath32 rather than @xmath6-band . the final results converge closely to the similar values . an underlying assumption in this approach is that the nuclear sources in the sample galaxies all have similar @xmath25 colors . it appears that the near infrared nuclear emission is characterized approximately by a blackbody at the typical destruction temperature of dust grains ( rieke & lebofsky , 1981 ; mcalary & rieke , 1988 ; alonso - herrero et al . , 1998 ) . as a test , we combined in various fractions a spectrum of an elliptical galaxy , representing the dominant old stellar population , with a diluting black body spectrum , and measured the colors and the co index of the composite spectra . most of our galaxies are consistent with @xmath33 ( figure [ fig3 ] ) , supporting our assumption for the universal nature of the diluting spectra in seyfert galaxies . in addition , the near infrared continuum appears to be `` uncovered '' abruptly at an appropriate viewing angle rather than showing a succession of galaxy types with differing colors in this continuum ( ruiz , rieke , & schmidt , 1994 ; maiolino , rieke & keller , 1995 ; quillen et al . , 2000b ) . therefore , this underlying assumption is plausible , although differences in nuclear color will limit our approach to galaxies where the corrections are modest . we also assume that the non - stellar nuclear flux at @xmath24-band flux comes from a source smaller than our slit width ; that is , that there is no significant level of extended featureless emission . typical scale sizes for the hot dust source component are @xmath34 10 pc , judged from variability time scales ( rieke & lebofsky , 1979 ; glass , 1992 ; glass , 1997 ; glass , 1998 ; quillen et al . , 2000a ) . the central engine is not expected to heat up the dust at 0.5 - 1 kpc from the center or further , which is the typical linear size of the slit at the distance of most of our targets . the dust can also be heated by an extended starburst in the disk . in that case , the @xmath24-band nuclear component will be overestimated , leading to an overcorrection of the co band , and deduction of a stronger starburst . this effect would only strengthen our conclusions if we do not detect excess starburst activity . we minimized the effect of reddening by starting our correction procedure from the @xmath6-band . however , we can put limits on the reddening by comparing the observed @xmath35 colors of the seyferts with those of a typical stellar population . this comparison is not exact because of both the likely variations in the reddening within our photometric aperture and the emission associated with the central engine . however , the derived @xmath35 color excesses are small , so these potential difficulties are relatively unimportant . we derive that @xmath25 color excesses for the stars are only a couple of percent , smaller than or comparable to the observational uncertainties . hence , in our analysis we neglected the effects of reddening in @xmath25 . we will discuss in more detail the colors of some of the galaxies in the next section . the procedure described above can not be applied to all of our targets . many type 1 seyferts have too strong a non - stellar component . we adopted the criterion for attempting a correction that the @xmath24-to-@xmath6 aperture flux ratio has to be smaller or equal to 1.0 , equivalent to @xmath36 mag . the limit was chosen after an error propagation analyses showed that the co absorption in galaxies with larger @xmath24 band excess can not be reliably corrected for the non - stellar contribution . table [ tbl-5 ] includes all seyferts with corrected co indices . we excluded from further analysis 2237 + 07 because of the insufficient signal - to - noise ratio at the long wavelength edge of the k - band . we also discarded galaxies with corrections larger than 25 % in flux at @xmath6 to keep the additional uncertainty added to the corrected co indices by the correction procedure itself to @xmath37 0.01 mag . this constraints our maximum co index correction to 0.05 mag . for comparison purposes we carried out the same procedure on a sample of starburst galaxies from engelbracht ( 1997 ) . the photometric data for these galaxies were collected mostly from the literature ( table [ tbl-6 ] ) . unlike the seyferts , starburst galaxies display significant reddening . for most of the targets engelbracht ( 1997 ) inferred @xmath38 mag , from infrared spectroscopy . we also used the infrared colors to supplement the reddening estimates . we carried out the co correction for the starbursts twice . first , we neglected the extinction . second , we dereddened the photometry before correcting the co band . the reddening accounts for part of the @xmath25 color excess leading to smaller co corrections this would strengthen a conclusion for the absence of strong starbursts in seyferts . the results of the co correction procedure starting at @xmath6 , as described in the previous section , are shown in table [ tbl-5 ] for seyferts and table [ tbl-7 ] for starburst galaxies . the histograms of the co indices are plotted in figure [ fig2 ] . the top panel shows the observed co indices , the middle panel shows the corrected co indices neglecting the reddening ( @xmath39 ) in starbursts , and the bottom panel shows again the corrected co indices but the colors of the starbursts were dereddened prior to the co correction . ticks on the horizontal axis show the bin borders . the co absorption strength in luminous ellipticals is narrowly distributed within the 0.18 - 0.22 bin ( we excluded abell 770 as a double nucleus galaxy , and m 87 because of its mildly active nucleus ) . see also ridgway et al . as shown in figure [ fig2 ] , our corrections to the co index in the type 2 seyfert galaxies usually recovered the co indices of an underlying old stellar population , similar to that of the elliptical galaxies . we do not find evidence for a significant contribution from young red stars , except in a few of the active galaxies . to determine the significance of this result we ran a monte - carlo simulation . we fitted a gaussian to the distribution of the corrected co indices of starburst galaxies , with reddening taken into account . next , we drew random realizations , and counted the cases that match or have an excess of shallower co indices compared with the corrected distribution of the co indices of type 1.8 - 2 seyfert galaxies . based on @xmath40 realizations , there is a probability of only 0.2% that the two distributions are drawn from the same parent distribution . color - color diagrams ( @xmath35 vs. @xmath41 and @xmath41 vs. @xmath25 ) from our sample of seyfert galaxies are shown in figure [ fig4 ] , with @xmath10 errors along both axes . solid circles represent seyfert types 1.8 and later , and open circles represent earlier seyferts . crosses and stars represent the typical colors of elliptical ( frogel et al . , 1978 ) and starburst ( engelbracht , 1997 ) galaxies , respectively . the circle on the right hand side color - color diagram represents the typical colors of an elliptical galaxy giants lie along the solid lines and supergiants along the dashed lines . stellar colors are from johnson ( 1966 ) and frogel et al . although the photometric systems may not be exactly the same , particularly the @xmath24-band , they are sufficiently close to our system for qualitative comparison . the giants span the range between g5 and m8 on left panel and between g5 and m5 on the right . the supergiants span the range between b0 and m5 on both panels . reddening vectors according to the rieke & lebofsky ( 1985 ) reddening law are also shown . most of our targets are redder then a typical old , red giant dominated stellar population as represented by the elliptical galaxy locus . the spread of colors can not be explained just by reddening , as in pure starburst galaxies which are distributed along a line approximately parallel to the reddening line . there is an additional component , showing the non - stellar nuclear contribution . it is present more strongly in @xmath42 ( glass & moorwood , 1985 ; spinoglio et al . , 1995 ; alonso - herrero , ward & kotilainen , 1996 ; alonso - herrero et al . , 1998 ) . a number of galaxies show peculiar blue colors . in these cases , we have searched the literature for additional measurements , shown in table [ tbl-8 ] . although comparisons are undermined to some extend by possible variability , from them we can get a feeling for the external accuracy of our measurements . in the case of ngc 1144 , we believe the best measurement is @xmath43 from carico et al . ( 1988 ) . 1058 + 45 is a faint source and given the observational uncertainties , we attributed the peculiar @xmath25 color to a statistical error . our literature search for mkn 270 and ngc 7682 yielded measurements consistent with ours . the blue colors ( @xmath41 and @xmath25 ) of ngc 7682 can be partially explained if it has a relatively weak non - stellar component . this explanation is consistent with the observed co index of 0.18 , which is close to the typical purely stellar value . still , the colors are bluer than expected from a giant dominated stellar population , suggesting perhaps the presence of blue young stars of early spectral types . in contrast , mkn 270 has peculiar @xmath35 and @xmath25 colors , while @xmath41 is close to the typical values of other target galaxies . it has a co index of 0.11 , suggesting some dilution . the observed colors can be produced by a heavily reddened @xmath44 blue population , even younger than in ngc 7682 . oliva et al . ( 1999 ) have used the near infrared stellar absorption bands to determine velocity dispersions and mass to luminosity ratios , @xmath45 , for active galaxies . eleven `` obscured '' seyferts ( types 1.8 - 2 ) have a distribution of @xmath45 that is consistent with a combination of those for normal spirals and starbursts . in this comparison , we exclude ngc 1052 and m87 because their nuclear properties are rather different from the other objects in their sample . five of these eleven have sufficiently low @xmath45 to suggest recent circumnuclear starbursts . by comparison , in 14 type 1.8 - 2 seyferts , we find two ( mkn 993 and mkn 461 ) with deep co indicative of starbursts . because of the broad distribution of co band strength in starbursts , we can not exclude other members of the sample from this group . however , from the distributions in figure 3 , we can conclude that such events are present in only a minority of the galaxies we have observed . the oliva et al . ( 1999 ) sample favors galaxies with bright nuclei ; our study is less biased in this direction . nonetheless , the results are consistent in indicating that the near infrared stellar luminosities in the majority of type 1.8 - 2 seyfert galaxies are not dominated by red supergiants from starbursts . oliva et al . show that the @xmath45 distribution for eight seyfert 1 galaxies is indistinguishable from that of normal spirals , because of the relatively strong nuclear emission of type 1 seyfert galaxies , we have been able to measure co band strengths for only two , but neither is strong . it therefore appears that the near infrared luminosities of type 1 seyfert nuclei are seldom dominated by starburst - generated supergiants . we quantified the role of starbursts in our sample by constructing a set of composite spectra . we added increasing portions of starburst spectra to an average spectrum of an elliptical galaxy , and ran the same monte - carlo simulation described above . we took an elliptical galaxy co strength distribution and added the distribution typical of starbursts until there was less than a 5% probability that the co strengths in the synthetic spectra were drawn from the same distribution as the observed distribution of co strengths in the seyfert galaxies . we excluded mkn 993 from the fitting because its extremely strong co index may partially result from the relatively low signal - to - noise ratio of that spectrum . increasing the starburst component in the composite spectrum quickly lowers the probability to draw a matching co index distribution . the simulation suggests that typically less than 1/3 of the @xmath6-band flux originates from a starburst in our sample of type 1.8 - 2 seyfert galaxies , at the 95% confidence level . let us assume that the starburst and the old population are represented by single bursts of ages @xmath46 and @xmath47 years , respectively . then , the upper limit on the k luminosity implies that the starburst would contribute no more than 31% of the bolometric luminosity , 8% of the flux in the v - band and 11% of the flux in the b - band . that is , the blue continuum of a typical type 1.8 - 2 seyfert galaxy in our sample may have a weak contribution from young blue stars , but is unlikely to be dominated by such stars . the starburst masses can not typically exceed 10@xmath48 , or in the most extreme cases , 10@xmath49 m@xmath50 . the corresponding limits on the starburst luminosities are similar to or lower than those of the agns in these galaxies . according to the predictions of the same model , the co strength reaches the typical value of an old stellar population about 50 - 100 myr after the beginning of the starburst . thus , our technique is not sensitive to very old starbursts . however , after 100 myr , the starburst absolute @xmath51 flux decreases by 3 - 4 mag , and its contribution to the total @xmath51 flux from the galaxy nucleus would become negligible . we obtained @xmath6 band spectroscopy of a set of seyfert galaxies , to measure their stellar first overtone co band strengths at @xmath2 . we used @xmath1 aperture photometry to estimate the non - stellar contribution at this wavelength . we successfully corrected the co band for the dilution for 16 galaxies which were not dominated by the non - stellar component . the comparison to co indices measured in elliptical and purely starbursting galaxies yielded no evidence for strong starbursts in the majority of seyfert 1.8 - 2 galaxies . this result suggests the nuclear activity in many seyfert 2 galaxies arises independently of any strong circumnuclear starbursts . we used an evolutionary starburst model to place an upper limit of @xmath52 11% on the contribution to the b band light from hot , young stars in a typical type 1.8 - 2 seyfert . the research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with nasa . this work has been supported by nsf grant ast 95 - 29190 . we thank the anonymous referee for the useful suggestions which helped to improve the paper . rrlllllcrcrrc - & 334 & 00:00:35 + 21:40:54&pec & 1.8&6582@xmath532 & 95/10/09&2.3&1440&2.25&75 & 80 & 2.4x4.8 & 00:48:53 + 29:07:46&(r)sb(s)b & 1 & 10770 & 95/10/10&2.3&720 & 2.25&75 & 30 & 2.4x4.8- & 993 & 01:22:43 + 31:52:36&sab0/a & 2 & 4658@xmath536 & 97/10/12&mmt&960 & 2.33&300&15 & 1.2x1.6- & 573 & 01:41:23 + 02:05:56&(r)sab(rs)0 + & 2 & 5174@xmath5323 & 95/10/09&2.3&720 & 1.60&75 & 30 & 2.4x4.8 & & & & & & 95/10/09&2.3&720 & 2.25&75 & 50 & 2.4x4.8 & & & & & & 97/10/12&mmt&2160&2.33&300&30 & 1.2x1.6 & 0152:45 + 06:22:02&sa(rs)b & 1.9&5208 & 95/10/09&2.3&1440&2.25&75 & 25 & 2.4x4.8863 & 590 & 02:12:00 -00:59:58&sa(s)a : & 1.2&7910@xmath5312 & 95/10/09&2.3&720 & 2.25&75 & 90 & 2.4x4.81144&- & 02:52:39 -00:23:08&ringb & 2 & 8648@xmath5314 & 97/10/18&mmt&5520&2.33&300&45 & 1.2x1.6- & 3 & 06:09:48 + 71:03:11&s0 : & 2 & 4050@xmath535 & 95/03/13&mmt&4320&2.34&600&40 & 1.2x1.62273&620 & 06:45:38 + 60:54:13&sb(r)a : & 2 & 1849 & 95/03/14&mmt&5760&2.32&600&30 & 1.2x2.03081&- & 09:57:10 -22:35:10&(r@xmath54)sab(r)0/a&2 & 2358@xmath537 & 95/03/14&mmt&1920&2.34&600&20 & 1.2x1.63080&1243&09:57:14 + 13:17:03&sa & 1 & 10608 & 96/02/06&mmt&1920&2.25&75 & 35 & 1.2x2.03227&- & 10:20:47 + 20:07:06&sab(s ) pec & 1.5&1157@xmath533 & 96/02/07&mmt&480 & 2.20&75 & 70 & 1.2x2.4 & & & & & & 94/03/23&2.3&960 & 2.31&600&25 & 2.4x6.0 & & & & & & 95/03/13&mmt&1440&2.32&600&40 & 1.2x2.0 & 10:58:43 + 45:55:22&sa ? & 2 & 8778 & 96/04/06&mmt&480 & 2.20&75 & 15 & 1.2x2.03516&- & 11:03:23 + 72:50:20&(r)sb(s)00 : & 1.5&2649@xmath537 & 93/11/25&mmt&2880&2.34&600&50 & 1.2x1.63786&744 & 11:37:05 + 32:11:11&sab(rs)a pec & 1.8&2678@xmath536 & 96/02/05&mmt&960 & 2.25&75 & 80 & 1.2x2.03982&- & 11:53:52 + 55:24:18&sab(r)b : & 2 & 1109@xmath536 & 96/02/10&mmt&1920&2.20&75 & 35 & 1.2x2.44051&- & 12:00:36 + 44:48:35&sab(rs)bc & 1.5&0725@xmath535 & 96/02/10&mmt&1440&2.20&75 & 80 & 1.2x2.0 & & & & & & 95/03/13&mmt&2880&2.32&600&40 & 1.2x1.64151&- & 12:08:01 + 39:41:02&(r)sab(rs)ab : & 1.5&0995@xmath533 & 96/02/10&mmt&1440&2.20&75 & 140&1.2x2.04235&- & 12:14:37 + 07:28:09&sa(s)a & 1 & 2410@xmath5310 & 96/02/10&mmt&1440&2.20&75 & 60 & 1.2x2.0 & & & & & & 95/03/14&mmt&12960&2.34&600&30&1.2x1.64253&766 & 12:15:56 + 30:05:26&(r)sb(s)a : & 1.5&3876@xmath5316 & 96/02/10&mmt&960 & 2.20&75 & 70 & 1.2x2.04258&- & 12:16:29 + 47:34:53&sab(s)bc & 1.9&0448@xmath533 & 95/03/13&mmt&1440&2.31&600&40 & 1.2x1.64388&- & 12:23:15 + 12:56:17&sa(s)b : sp & 2 & 2524@xmath531 & 96/02/06&mmt&480 & 2.25&75 & 30 & 1.2x2.4- & 231 & 12:54:05 + 57:08:38&sa(rs)c ? pec & 1 & 12651@xmath536 & 94/04/02&1.55&360 & 1.60&75 & 110&3.6x7.2 & & & & & & 94/04/02&1.55&1080&2.21&75 & 90 & 3.6x7.25033&- & 13:11:09 + 36:51:31&sa(s)c & 1.9&0875@xmath531 & 96/02/06&mmt&240 & 2.25&75 & 60 & 1.2x2.4 & & & & & & 95/03/13&mmt&2880&2.32&600&30 & 1.2x1.6- & 789 & 13:29:55 + 11:21:44&- & 1 & 9476@xmath5322 & 96/03/29&mmt&2400&2.20&75 & 30 & 1.2x2.0 & 13:35:28 + 39:24:31&s ? & 1.8&6023 & 96/03/29&mmt&2880&2.20&75 & 25 & 1.2x2.05252&- & 13:35:44 + 04:47:47&s0 & 1.9&6926 & 96/03/29&mmt&1440&2.20&75 & 45 & 1.2x2.05256&266 & 13:36:15 + 48:31:48&compact pec & 2 & 8360@xmath5381 & 96/03/27&mmt&1800&2.20&75 & 35 & 1.2x1.6 & & & & & & 96/03/28&mmt&1440&2.20&75 & 40 & 1.2x2.05283&270 & 13:39:41 + 67:55:28&s0 ? & 2 & 2700@xmath5377 & 96/03/29&mmt&960 & 2.20&75 & 70 & 1.2x2.0 & & & & & & 95/03/13&mmt&5280&2.31&600&30 & 1.2x1.65273&- & 13:39:55 + 35:54:21&sa(s)00 & 1.9&1089 & 96/02/09&mmt&1440&2.20&75 & 50 & 1.2x2.4- & 461 & 13:45:04 + 34:23:52&s & 2 & 4856@xmath5332 & 96/03/29&mmt&2880&2.20&75 & 40 & 1.2x2.0- & 279 & 13:51:54 + 69:33:13&s0 & 1.5&8814@xmath5390 & 96/02/09&mmt&480 & 2.20&75 & 75 & 1.2x2.05427&- & 14:00:48 -05:47:19&sa(s)c pec & 2 & 2730@xmath534 & 95/03/14&mmt&960 & 2.33&600&5 & 1.2x1.65548&1509&14:15:44 + 25:22:01&(r)sa(s)0/a & 1.5&5149@xmath537 & 96/02/07&mmt&960 & 2.20&75 & 80 & 1.2x1.65674&- & 14:31:22 + 05:40:38&sabc & 1.9&7442 & 96/03/28&mmt&3360&2.20&75 & 65 & 1.2x2.4 & & & & & & 95/03/14&mmt&3840&2.37&600&20 & 1.2x1.65695&686 & 14:35:20 + 36:47:02&sbb & 2 & 4225@xmath539 & 96/03/29&mmt&1440&2.20&75 & 40 & 1.2x2.0 & & & & & & 95/03/14&mmt&1920&2.34&600&20 & 1.2x1.6- & 841 & 15:01:36 + 10:37:56&e & 1.5&10860@xmath5390&96/03/29&mmt&480 & 2.20&75 & 70 & 1.2x2.05929&- & 15:24:19 + 41:50:41&sab : pec & 2 & 2492@xmath538 & 96/03/29&mmt&960 & 2.20&75 & 45 & 1.2x2.05940&1511&15:28:51 + 07:37:38&sbab & 1 & 10122 & 96/02/06&mmt&1440&2.20&75 & 25 & 1.2x2.46104&(1614 + 35)&16:14:40 + 35:49:50&- & 1.5&8382@xmath5350 & 95/10/10&2.3&2880&2.25&75 & 30 & 2.4x4.8(iizw136)&1513&21:30:01 + 09:55:01&- & 1 & 18630@xmath5390&95/10/10&2.3&4320&2.25&75 & 100&2.4x4.8 & 22:37:46 + 07:47:34&sba & 1.8&7375@xmath5390 & 95/10/07&2.3&1440&2.20&75 & 70 & 2.4x6.0 & & & & & & 95/10/10&2.3&720 & 2.25&75 & 50 & 2.4x4.87469&1514&23:00:44 + 08:36:16&(r)sab(rs)a & 1.2&4892@xmath532 & 94/11/19&mmt&720 & 2.35&600&40 & 1.2x1.67603&530 & 23:16:23 -00:01:47&sa(rs)b : pec & 1.5&8851@xmath5322 & 95/10/07&2.3&720 & 2.20&75 & 60 & 2.4x6.0 & & & & & & 95/10/10&2.3&720 & 2.25&75 & 90 & 2.4x4.87674&533 & 23:25:24 + 08:30:13&sa(r)bc pec & 2 & 8713@xmath5310 & 95/10/10&2.3&720 & 2.25&75 & 70 & 2.4x4.87682&- & 23:26:31 + 03:15:28&sb(r)ab & 2 & 5107 & 95/10/09&2.3&1440&1.60&75 & 30 & 2.4x4.8 & & & & & & 95/10/09&2.3&2160&2.25&75 & 30 & 2.4x4.8 & 09:14:12 + 60:38:00&e & - & - & 94/04/04&1.55&1860&2.21&75 & 90 & 3.6x7.2 & 09:14:12 + 60:38:00&e & - & - & 94/04/04&1.55&1860&2.21&75 & 50 & 3.6x5.43379&(m105)&10:45:11 + 12:50:48&e1 & - & 0920@xmath5310 & 94/04/02&1.55&5520&2.21&75 & 90 & 3.6x7.24472&(m49)&12:27:14 + 08:16:36&e2/s0 & - & 0868@xmath538 & 94/04/02&1.55&4860&2.21&75 & 70 & 3.6x7.24486&(m87)&12:28:18 + 12:39:58&e+0 - 1 pec & sy & 1282@xmath539 & 94/04/04&1.55&3060&2.21&75 & 110&3.6x7.2 lrrrrr mkn334 & @xmath67&@xmath68 & @xmath69&@xmath70&0.070048 + 29 & & @xmath57 & @xmath71 & & 0.18mkn993 & & @xmath72 & @xmath73&@xmath74&0.26mkn573 & @xmath75&@xmath76 & @xmath77&@xmath78&0.150152 + 06 & & @xmath79 & @xmath80 & & 0.16mkn590 & @xmath75&@xmath81 & @xmath82&@xmath70&0.06ngc1144 & & @xmath83 & @xmath84&@xmath70&0.15mkn3 & & & & @xmath85&0.20ngc2273 & & & & @xmath86&0.13ngc3081 & & & & & 0.20mkn1243 & @xmath87&@xmath88 & @xmath89&@xmath90&0.04ngc3227 & @xmath70&@xmath91 & @xmath92&@xmath70&0.101058 + 45 & @xmath85&@xmath93&@xmath94&@xmath60&0.22ngc3516 & & & & & 0.06mkn744 & @xmath95&@xmath96 & @xmath92&@xmath70&0.09ngc3982 & @xmath70&@xmath97 & @xmath98&@xmath60&0.17ngc4051 & @xmath95&@xmath99 & @xmath100&@xmath70&0.06ngc4151 & @xmath85&@xmath101 & @xmath102&@xmath70&0.01ngc4235 & @xmath103&@xmath65 & @xmath69&@xmath104&0.09mkn766 & @xmath70&@xmath101 & @xmath99&@xmath105&0.02ngc4258 & & & & @xmath106&0.18ngc4388 & @xmath107&@xmath108 & @xmath109&@xmath60&0.09mkn231 & @xmath60&@xmath70 & @xmath110&@xmath60&0.01ngc5033 & @xmath60&@xmath111 & @xmath61&@xmath62&0.14mkn789 & @xmath112&@xmath113&@xmath114&@xmath112&0.131335 + 39 & @xmath115&@xmath116 & @xmath117&@xmath112&0.04ngc5252 & @xmath118&@xmath119 & @xmath120&@xmath121&0.10mkn266 & @xmath122&@xmath123&@xmath124&@xmath78&0.17mkn270 & @xmath125&@xmath126 & @xmath127&@xmath104&0.11ngc5273 & @xmath101&@xmath128 & @xmath129&@xmath119&0.25mkn461 & @xmath64&@xmath130 & @xmath60&@xmath131&0.19mkn279 & @xmath132&@xmath133 & @xmath104&@xmath134&0.03ngc5427 & & & & & 0.05ngc5548 & @xmath135&@xmath136 & @xmath70&@xmath70&0.03ngc5674 & @xmath132&@xmath137 & @xmath116&@xmath60&0.14mkn686 & @xmath138&@xmath139 & @xmath140&@xmath60&0.16mkn841 & @xmath101&@xmath56 & @xmath60&@xmath60&0.02ngc5929 & @xmath95&@xmath141 & @xmath91&@xmath60&0.17ngc5940 & @xmath132&@xmath132 & @xmath142&@xmath60&0.031614 + 35 & @xmath112&@xmath143 & @xmath144&@xmath132&0.05iizw136 & @xmath70&@xmath102 & @xmath60&@xmath70&0.012237 + 07 & @xmath135&@xmath145 & @xmath146&@xmath147&0.06ngc7469 & & & & & 0.25mkn530 & @xmath86&@xmath148 & @xmath103&@xmath70&0.06mkn533 & @xmath132&@xmath149 & @xmath62&@xmath60&0.03ngc7682 & @xmath56&@xmath150 & @xmath151&@xmath78&0.18a770a & @xmath101&@xmath152 & @xmath153&@xmath60&0.16a770b & @xmath60&@xmath154&@xmath155&@xmath156&0.23ngc3379 & @xmath86&@xmath157 & @xmath96&@xmath62&0.20ngc4472 & @xmath101&@xmath157 & @xmath158&@xmath104&0.20m87 & @xmath101&@xmath159 & @xmath81&@xmath160&0.16 lcrrrrrrrrl mkn334&8.6&@xmath161&@xmath162&@xmath163&@xmath164&@xmath165&@xmath166&@xmath167&@xmath168&0048 + 29&5.4&@xmath169&@xmath170&@xmath171&@xmath172&@xmath173&@xmath174&@xmath175&@xmath176 & mkn993&8.6&@xmath177&@xmath178&@xmath179&@xmath172&@xmath180&@xmath181&@xmath182&@xmath183 & mkn573&8.6&@xmath184&@xmath185&@xmath186&@xmath187&@xmath188&@xmath174&@xmath189&@xmath190&@xmath191 & & & & & & @xmath192 & & & & @xmath1930152 + 06&5.4&@xmath194&@xmath195&@xmath196&@xmath197&@xmath198&@xmath199&@xmath200&@xmath201 & mkn590&5.4&@xmath202&@xmath203&@xmath204&@xmath205&@xmath206&@xmath207&@xmath208&@xmath209&ngc1144&5.4&@xmath210&@xmath211&@xmath212&@xmath195&@xmath213&@xmath181&@xmath214&@xmath215&mkn1243&5.4&@xmath216&@xmath217&@xmath218&@xmath219&@xmath220&@xmath221&@xmath222&@xmath223&ngc3227&8.6&@xmath224&@xmath225&@xmath225&@xmath226&@xmath227&@xmath228&@xmath229&@xmath230&@xmath231 & & & & & & @xmath232 & & & & @xmath193 & & & & & & @xmath233 & & & & @xmath234ngc3362&5.4&@xmath235&@xmath236&@xmath237&@xmath238 & & @xmath239&@xmath240&@xmath241&1058 + 45&5.4&@xmath242&@xmath243&@xmath244&@xmath245&@xmath246&@xmath199&@xmath240&@xmath247 & mkn744&8.6&@xmath248&@xmath249&@xmath250&@xmath251&@xmath252&@xmath253&@xmath254&@xmath255&ngc3982&5.4&@xmath256&@xmath257&@xmath258&@xmath259 & & @xmath260&@xmath182&@xmath261&ngc4051&8.6&@xmath262&@xmath263&@xmath264&@xmath265&@xmath266&@xmath267&@xmath268&@xmath269&@xmath193 & & & & & & @xmath270 & & & & @xmath231ngc4151&8.6&@xmath271&@xmath272&@xmath273&@xmath274&@xmath275&@xmath228&@xmath276&@xmath277&ngc4235&5.4&@xmath278&@xmath279&@xmath280&@xmath281&@xmath282&@xmath283&@xmath182&@xmath284&@xmath193 & & & & & & @xmath285 & & & & @xmath231 mkn766&5.4&@xmath286&@xmath287&@xmath288&@xmath289&@xmath290&@xmath291&@xmath292&@xmath293 & mkn231&5.8&@xmath294&@xmath265&@xmath295&@xmath296&@xmath297&@xmath298&@xmath299&@xmath300&ngc5033&8.6&@xmath301&@xmath302&@xmath303&@xmath304&@xmath305&@xmath306&@xmath307&@xmath308&@xmath193 & & & & & & @xmath309 & & & & @xmath231 & & & & & & @xmath310 & & & & @xmath3111335 + 39&5.4&@xmath312&@xmath313&@xmath314&@xmath258&@xmath315&@xmath316&@xmath317&@xmath318&ngc5252&5.4&@xmath319&@xmath320&@xmath320&@xmath321&@xmath322&@xmath306&@xmath229&@xmath323 & mkn266&8.6&@xmath324&@xmath325&@xmath326&@xmath195&@xmath327&@xmath328&@xmath329&@xmath330&@xmath193 & & & & & & @xmath331 & & & & @xmath231 mkn270&5.8&@xmath332&@xmath333&@xmath334&@xmath195&@xmath335&@xmath336&@xmath175&@xmath337&@xmath231 & & & & & & @xmath338 & & & & @xmath193 mkn461&5.4&@xmath339&@xmath340&@xmath341&@xmath342&@xmath327&@xmath343&@xmath344&@xmath345 & mkn279&8.6&@xmath346&@xmath347&@xmath187&@xmath348&@xmath349&@xmath174&@xmath350&@xmath351 & ic4397&5.4&@xmath352&@xmath353&@xmath236&@xmath354 & & @xmath355&@xmath240&@xmath356&ngc5548&8.6&@xmath357&@xmath358&@xmath359&@xmath360&@xmath361&@xmath174&@xmath283&@xmath362&ngc5674&5.4&@xmath363&@xmath364&@xmath365&@xmath366&@xmath367&@xmath181&@xmath368&@xmath369&@xmath193 & & & & & & @xmath370 & & & & @xmath311 mkn686&5.4&@xmath371&@xmath372&@xmath170&@xmath373&@xmath374&@xmath375&@xmath376&@xmath377&@xmath193 & & & & & & @xmath378 & & & & @xmath231ngc5929&5.4&@xmath379&@xmath380&@xmath372&@xmath381&@xmath382&@xmath383&@xmath384&@xmath385&ngc5940&5.4&@xmath352&@xmath195&@xmath386&@xmath387 & & @xmath316&@xmath222&@xmath388&1614 + 35&5.4&@xmath389&@xmath390&@xmath390&@xmath391 & & @xmath392&@xmath393&@xmath394&2237 + 07&8.6&@xmath210&@xmath395&@xmath396&@xmath179&@xmath397&@xmath398&@xmath399&@xmath400&@xmath234 & & & & & & @xmath401 & & & & @xmath191ngc7469&8.6&@xmath402&@xmath403&@xmath404&@xmath405&@xmath406&@xmath407&@xmath408&@xmath409 & mkn530&8.6&@xmath410&@xmath203&@xmath203&@xmath203&@xmath411&@xmath167&@xmath229&@xmath412&@xmath234 & & & & & & @xmath413 & & & & @xmath191 mkn533&8.6&@xmath414&@xmath415&@xmath416&@xmath417&@xmath418&@xmath419&@xmath420&@xmath421&ngc7682&8.6&@xmath422&@xmath320&@xmath218&@xmath423&@xmath424&@xmath268&@xmath425&@xmath426 & lccccccccl 0048 + 29&0.76&0.18&0.21&10&1&0.20 & 6&2&mkn993 & 0.62&0.26&0.33&18&1&0.29&10&1&mkn573 & 0.87&0.15&0.21&29&1&0.19&21&1&0152 + 06&0.75&0.16&0.18&11&1&0.18 & 5&2&ngc1144&0.48&0.15&0.15 & 2&1&0.15 & 0&0&ngc3227&0.91&0.10&0.15&40&1&0.13&28&1&1058 + 45&0.40&0.22&0.22 & 0&0&0.22 & 0&0&@xmath427ngc4235&0.74&0.09&0.11&27&1&0.10&19&1&ngc5033&0.90&0.14&0.23&48&1&0.22&42&1&mkn266 & 0.80&0.17&0.22&22&1&0.20&14&1&mkn270 & 0.41&0.11&0.11 & 0&0&0.11 & 0&0&@xmath427mkn461 & 0.93&0.19&0.26&29&1&0.24&23&1&mkn686 & 0.64&0.16&0.18&11&1&0.18 & 7&1&ngc5929&0.80&0.17&0.23&25&1&0.21&18&1&2237 + 07&0.82&0.06&0.08&14&1&0.08 & 8&2&ngc7682&0.48&0.18&0.18 & 0&0&0.18 & 0&0&@xmath427 lccrrrrrrrc ngc253 & @xmath428&6.1 & 6.0 & 115.0 & 226.7 & 280.6 & 278.1 & 390.3 & 7.4 & 1,8 ngc660 & @xmath429&5.4 & 5.0 & 27.9 & 40.9 & 36.7 & 69.2 & 34.7 & 6.4 & 6 maffei2 & @xmath234&4.3&slit & 16.6 & 45.6 & 55.5 & 55.5 & - & 8.5 & 1 ic342 & @xmath234&4.3 & 3.8 & 49.0 & 74.0 & 71.0 & 87.1 & 44.0 & 5.1 & 9 ngc1614 & @xmath234&4.3 & 6.0 & 32.3 & 52.4 & 51.5 & 27.6 & 55.9 & 3.5 & 11 ngc2146 & @xmath234&4.3&17.0 & 132.1 & 253.2 & 270.5 & 94.3 & 285.7 & 6.4 & 2 ngc2782 & @xmath234&4.3&17.0 & 55.1 & 73.7 & 62.5 & 22.1 & 65.6 & 2.9 & 2 m82 & @xmath430&8.9 & 7.8 & 161.0 & 350.0 & 475.0 & - & 436.0&10.2 & 10 ngc3079 & @xmath234&4.3 & 6.0 & 29.2 & 66.6 & 82.4 & 34.7 & 63.9 & 7.8 & 4 ngc3628 & @xmath234&4.3&slit & 06.2 & 14.8 & 19.4 & 19.4 & - & 9.7 & 1 ngc4102 & @xmath234&4.3 & 3.8 & 80.5 & 118.0 & 117.0 & 87.1 & 89.0 & 5.7 & 5 ngc4194 & @xmath234&4.3&13.0 & 73.3 & 78.6 & 67.3 & 24.9 & 69.0 & 3.5 & 3 ngc4339 & @xmath234&4.3&slit & 15.0 & 21.6 & 17.2 & 17.2 & - & 1.0 & 1 ngc5990 & @xmath234&4.3&slit & 16.6 & 23.7 & 27.6 & 27.6 & - & 7.7 & 1 ngc6000 & @xmath234&4.3&slit & 30.3 & 37.3 & 37.3 & 37.3 & - & 5.1 & 1 ngc6240 & @xmath234&4.3 & 9.0 & 27.9 & 50.5 & 53.5 & 31.3 & 43.2 & 9.9 & 3 ngc6946 & @xmath431&4.7 & 3.0 & 23.8 & 35.3 & 36.3 & - & - & 4.3 & 1 ngc7714 & @xmath234&4.3 & 5.0 & 20.4 & 25.6 & 21.5 & 18.5 & 22.4 & 4.3 & 4 mkn331 & @xmath234&4.3 & 5.0 & 26.0 & 36.4 & 34.9 & 23.8 & 29.1 & 4.8 & 7 lrrrcrrrc ngc253 & 0.24&0.31&0.29&1 & 7.4&0.27&0.26&1ngc660 & 0.22&0.23&0.23&1 & 6.4&0.22&0.22&1maffei2&0.22&0.22&0.22&0 & 8.5&0.22&0.22&0ic342 & 0.26&0.27&0.26&1 & 5.1&0.25&0.27&2ngc1614&0.28&0.28&0.28&0 & 3.5&0.28&0.28&0ngc2146&0.18&0.27&0.24&1 & 6.4&0.21&0.20&1ngc2782&0.14&0.19&0.18&1 & 2.9&0.17&0.17&1m82 & 0.27&0.27&0.27&0&10.2&0.27&0.27&0ngc3079&0.17&0.22&0.19&1 & 7.8&0.17&0.17&1ngc3628&0.24&0.24&0.24&0 & 9.7&0.24&0.24&0ngc4102&0.19&0.21&0.22&1 & 5.7&0.19&0.20&2ngc4194&0.15&0.21&0.20&1 & 3.5&0.18&0.18&1ngc4339&0.25&0.25&0.25&0 & 1.0&0.25&0.25&0ngc5990&0.12&0.12&0.12&0 & 7.7&0.12&0.12&1ngc6000&0.25&0.25&0.25&0 & 5.1&0.25&0.25&0ngc6240&0.18&0.22&0.21&2 & 9.9&0.17&0.18&1ngc6946&0.26&0.26&0.26&0 & 4.3&0.26&0.26&0ngc7714&0.26&0.30&0.29&1 & 4.3&0.27&0.28&1mkn331 & 0.22&0.25&0.26&2 & 4.8&0.23&0.23&2 lrcccc ngc1144 & 4.0&0.89&0.38 & 0.55&1 & 5.0&0.72&0.44 & 0.73&2 & 5.4&1.01&0.47 & & 3 & 8.5&1.43&0.58 & & 4 & 12.0&0.75&0.41 & & 5 & 30.0&0.75&0.21 & & 6mkn270 & 5.9&0.50&0.46 & -0.07&7 & 15.0&0.50&0.35&@xmath340.84&8ngc7682 & 8.5&0.80&0.02 & & 4 aaronson , m , 1977 , ph.d . thesis , harvard university alonso - herrero , a. , ward , m.j . , & kotilainen , j.k . , 1996 , , 278 , 902 alonso - herrero , a. , simpson , c. , ward , m.j . , wilson , a.s . , 1998 , , 495 , 196 allen , d.a . , 1976 , , 207 , 367 becklin , e.e . , gatley , i. , matthews , k. , neugebauer , g. , sellgren , k. , werner , m.w . , winn - williams , c.g . , 1980 , , 236 , 441 bailey , m.e . , 1980 , , 191 , 195 brindle , c. , hough , j.h . , bailey , j.a . , axon , d.j . , sparks , w.b . , 1991 , , 252 , 288 bushouse , h.a . , & stanford , s.a . , 1992 , , 79 , 213 campins , h. , rieke , g.h . , & lebofsky , m.j . , 1985 , , 90 , 896 carico , d.p . , sanders , d.b . , soifer , b.t . , elias , j.h . , matthews , k. , & neugebauer , g. , 1988 , , 95 , 365 cid fernandes , r. , 1997 , in starburst activity in galaxies , eds . franco j. , terlevich , r.j . , serrano , a. , rev . a. a. conf . , 6 , 201 cutri , r.m . , & mcalary , c.w . , 1985 , , 296 , 90 david , l.p . , durstein , r.h . , & cohn , h.n . , 1987a , , 313 , 556 david , l.p . , 1987b , , 316 , 505 dey , a. , breugel , w. , vacca , w.d . , & antonucci , r. , 1997 , , 490 , 698 diaz , a.i . , terlevich , e. , & terlevich , r. , 1989 , , 239 , 325 doyon , r. , joseph , r.d . , & wright g.s . , 1994 , , 421 , 101 edelson , r.a . , malkan , m.a . , & rieke , g.h . , 1987 , , 321 , 233 elias , j.h . , frogel , j.a . , matthews , k. , & neugebauer , g. , 1982 , , 87 , 1029 engelbracht , c.,w . , 1997 , ph.d . thesis , the university of arizona engelbracht , c.,w . , 1999 , private communication forbes , d.a . , boisson , c. , ward , m.j . , 1992 , , 259 , 293 frogel , j.a . , persson , s.e . , aaronson , m. , & matthews , k. , 1978 , , 220 , 75 frogel , j. a. , 1985 , , 298 , 528 goldader , j.d . , joseph , r.d . , doyon , r. , sanders , d.b . , 1995 , , 444 , 97 glass , i.s . , & moorwood , a.f.m . , 1985 , , 214 , 429 glass , i.s . , 1992 , , 256 , 23 glass , i.s . , 1997 , , 292 , l50 glass , i.s . , 1998 , , 297 , 18 gonzalez - delgado , r.m . , heckman , t. , leitherer , c. , meurer , g. , krolik , j. , wilson , a.s . , kinney , a. , koratkar , a. , 1998 , , 505 , 174 heckman , t.m . , gonzalez - delgado , r. , leitherer , c. , meurer , g.r . , krolik , j. , wilson , a.s . , koratkar , a. , & kinney , a. , 1997 , , 482 , 114 heckman , t.m . , 1999 , preprint ( astro - ph/9912029 ) hunt , l.k . , & giovanardi , c. , 1992 , , 104 , 1018 impey , c.d . , winn - williams , c.g . , & becklin , e.e . , 1986 , , 309 , 572 johnson , h.l . , , 4 , 193 joy , m. , & ghigo , f.d . , 1988 , , 332 , 179 kleinmann , s.g . , & hall , d.n.b . , , 62 , 501 knapen , j.h . , shlosman , i. , & peletier , r.f . , 1999 , preprint kunth , d. , & contini , t. , 1999 , in wolf - rayet phenomena in massive stars and starburst galaxies , 193th symp . of iau , eds . van der hucht , k. , koenigsberger , g. , & eenens , p.r.j . , p. 725 lawrence , a. , ward , m. , elvis , m. , fabbiano , g. , willner , s.p . , carleton , n.p . , longmore , a. , 1985 , , 291 , 117 maiolino , r. , rieke g.h . , & keller , l.d . , 1995 , , 446 , 561 maiolino , r. , rieke g.h . , & rieke , m.j . , 1996 , , 111 , 537 maiolino , r. , ruiz , m. , rieke g.h . , & papadopoulos , p. , 1997 , , 485 , 552 mauder , w. , weigelt , g. , appenzeller , i. , wagner , s.j . , 1994 , , 285 , 44 mcalary , c.w . , mclaren , r.a . , & crabtree , d.r . , 1979 , , 234 , 471 mcalary , c.w . , & rieke , g.h . , 1988 , , 333 , 1 mcleod , k.k . , & rieke g.h . , 1995 , , 441 , 96 nelson , c. , & whittle , m. , 1998 , in the agn / host galaxy connection as part of the scientific assembly of cospar ( astro - ph/9811035 ) norman , c. , & scoville . , n. , 1988 , , 332 , 124 ohsuga , k. , & umemura , m. , 1999 , , 521 , l13 oliva , e. , & origlia , l. 1992 , , 254 , 466 oliva , e. , origlia , l. , kotilainen , j.k . , & moorwood , a.f.m . , 1995 , , 301 , 55 oliva , e. , origlia , l. , maiolino , r. , & moorwood , a.f.m . , 1999 , , 350 , 9 perry , j.j . , & dyson , j.e . , 1985 , , 235 , 665 perry , j.j . , 1992 , in relationships between active galactic nuclei and starburst galaxies , ed . filippenko , a.v . , asp conf . ser . , 31 , 169 persson , s.e . , 1988 , , 330 , 751 quillen , a.c . , shaked , s. , alonso - herrero , a. , mcdonald , c. , lee , a. , rieke , m.j . , rieke , g.h . , 2000a , apj , 532 , l17 quillen , a.c . et al . , 2000b , in preparation ridgway , s.e . , winn - williams , c.g . , & becklin , e.e . , 1994 , , 428 , 609 rieke , g.h . , & low , f.j . , 1975 , , 197 , 17 rieke , g.h . , 1978 , , 226 , 550 rieke , g.h , & lebofsky , m.j . , 1979 , , 229 , l111 rieke , g.h . , lebofsky , m.j . , thompson , r.i . , low , f.j . , tokunaga , a.t . , 1980 , , 238 , 24 rieke , g.h . , & lebofsky , m.j . , 1981 , , 250 , 87 rieke , g.h . , & lebofsky , m.j . , 1985 , , 288 , 618 roche , p.f . , aitken , d.k . , smith , c.h . , ward , m.j . , 1991 , , 248 , 606 ruiz , m. , rieke , g.h . , & schmidt , g.d . , 1994 , , 423 , 608 shlosman , i. , frank , j. , & begelman , m.c . , 1989 , , 338 , 45 shlosman , i. , begelman , m.c . , & frank , j. , 1990 , , 345 , 679 spinoglio , l. , malkan , m.a . , rush , b. , carrasco , l. , & recillas - cruz , 1995 , , 453 , 616 terlevich , e. , diaz , a.i . , & terlevich , r. , 1989 , , 157 , 15 terlevich , e. , diaz , a.i . , & terlevich , r. , 1990 , , 242 , 284 terlevich , r. , & melnick , j. , 1985 , , 213 , 841 terlevich , r. , tenorio - tagle , g. , franco , j. , & melnick , j. , 1992 , , 255 , 713 van breugel , w. , filippenko , a.v . , heckman , t. , & miley , g. , 1985 , , 293 , 83 van breugel , w.j.m . , & dey , a. , 1993 , , 414 , 563 vanzi , l. , & rieke , g.h . , 1997 , , 479 , 694 von linden , s. , biermann , p.l . , duschl , w.j . , lesch , h. , & schmutzler , t. , 1993 , , 280 , 468 weedman , d.w . , 1983 , , 266 , 479 williams , d.m . , thompson , c.l . , rieke , g.h . , & montgomery , e.f . , 1993 , s.p.i.e . , 1946 , 482 williams , r.j.r . , baker , a.c . , & perry , j.j . , 1999 , , 310 , 913

electronic mach - zehnder interferometer ( mzi ) @xcite can be realized with the edge states of the quantum hall liquids ( qhls ) . together with the quantum antidots @xcite , mzis in the regime of the fractional quantum hall effect ( fqhe ) are expected @xcite to be useful for observation of the fractional statistics of fqhe quasiparticles . in contrast to fractional quasiparticle charge , which has been confirmed in several experiments @xcite , there is no commonly accepted observation of anyonic statistics of the quasiparticles , which remains a challenging experimental problem . currently , this problem attracts interest in the context of solid - state quantum computation , since individual manipulation of anyonic quasiparticles involving their braiding provides an interesting possible basis for implementation of the quantum information processing @xcite . however , in typical interferometer - based experimental set - up the quasiparticles emerge as a continuum of gapless edge excitations that should be described by a 1d field theory @xcite . individual quasiparticles can be realized in this theory only asymptotically in a special limit . in the fractional edge - states mzi , such a limit occurs at _ large _ voltages , when the system is characterized by the hamiltonian dual to the hamiltonian of the initial electron tunneling model of the mzi . the latter is perturbative in electron tunneling at low voltages , and is much better defined , since weak electron tunneling is probably the most basic process in solid - state physics . in the dual description derived from the electronic model by the instanton technique @xcite , mzi acquires @xmath7 different quantum states which differ by the the effective flux @xmath8 through it . this flux contains , in addition to the external flux @xmath9 , a statistical contribution earlier also found @xcite for the antidots . the tunneling of each quasiparticle changes this effective flux by @xmath10 and therefore switches the mzi from one flux state into another . since quasiparticle also carries the charge @xmath5 , the change of flux by @xmath10 results in the change of the interference phase for the quasiparticles by @xmath11 and the corresponding change of the rate of coherent tunneling through the interferometer . summation over the @xmath2 flux states in calculation of the physical quantities restores @xmath12 periodicity of their @xmath9 dependence . at low voltages , this periodicity is guaranteed by @xmath12 periodicity of the electron tunneling amplitude . in this work , we follow the approach to the mzi that allows us to obtain the uniform description of its transport properties in both regimes of electron and quasiparticle tunneling . we consider the standard mzi geometry ( fig . 1 ) with two tunneling contacts between the two effectively parallel edges of qhls , but allow for , in general , different filling factors , @xmath0 , with @xmath1 , of these edges . in the symmetric case of equal propagation times along the two edges between the contacts , the corresponding 1d field theory permits an exact bethe ansatz solution @xcite . making use of this solution , we calculate the zero - temperature full counting statistics @xcite of the charge transferred between the two edges forming mzi . the transferred charge distribution is shown to reflect the anyonic braiding statistics of the tunneling quasiparticles , and decomposition of electrons into quasiparticles with increase of voltage . ( 3.,1.4 ) ( 0.6,-0.05)=1.9 in the paper is organized as follows . in section 2 we introduce the model of symmetric mzi and its bethe ansatz solution . in section 3 , we use this solution to derive the general expression for the generating function @xmath13 of the distribution of the transferred charge . the method employed in this calculation generalizes to the mzi transport through two point contacts the method developed earlier by saleur and weiss @xcite for a single point contact . the charge transfer statistics found for one contact @xcite demonstrated fractionalization of electron charges with increasing bias voltage across the contact . in the case of mzi , one can expect that in addition to charge fractionalization , braiding properties of anyonic quasiparticles should emerge in the charge transfer statistics . to see this , we analyze our result for the transfer statistics in different regimes . we find that the logarithm of the generating function separates into two parts , which for very different absolute values of the electron tunneling amplitudes in the two contacts , @xmath14 , become identical with the cumulant - generating functions of the two separate contacts . in general , however , each `` single - contact '' term accounts also for the interference between the contacts . at low voltages , these single - contact terms are combined in such a way that the charge transfer occurs only in electron units , with the low - voltage asymptotic describing the regular poisson distribution of tunneling electrons . in section 4 , we develop analogous qualitative interpretation of the large - voltage asymptotics of the bethe ansatz result for the generating function . the interpretation is based on the @xmath2-state model of the quasiparticle tunneling described above . following the method of ref . , the cumulant - generating function is calculated from the kinetic equation governing the quasiparticle transitions in the basis of the @xmath2 flux states . direct comparison of this generating function with the leading asymptotics of the bethe ansatz result reveals their coincidence under a special choice of @xmath2 quasiparticle interference phases which are found as @xmath15 , where @xmath16 numbers the flux states of the interferometer . the @xmath17-dependent , statistical , part of these phases agrees with the expected anyonic statistics of the quasiparticles , while the common phase is given by the electron interference phase @xmath18 with an additional phase shift @xmath19 for even @xmath2 . the equivalence of the two distributions in the large - voltage limit proves that the bethe ansatz construction we implemented indeed describes the statistical transmutation of the effective flux through the mzi . in section 5 , we use the obtained generating function to find the first three cumulants of the transferred charge distribution . we study , in particular , the voltage dependence of the first and the second cumulants proportional to the average current and the shot noise , respectively . the ratio of the two , the fano factor , is an experimentally observable @xcite characteristics of the electron - quasiparticles decomposition . finally , we consider the third cumulant , which determines the asymmetry of the transferred charge distribution around average . in section 6 , we look at the special case @xmath20 which allows to obtain these cumulants in terms of elementary functions for arbitrary voltages . the results of this work are summarized in the conclusion . we start our discussion with the electronic model of mzi ( fig . 1 ) formed by two single - mode edges with filling factors @xmath21 , @xmath22 . electron operator @xmath23 of the edge @xmath17 is expressed using the standard bosonization approach @xcite as @xmath24 } .\ ] ] here @xmath25 are the two chiral bosonic modes propagating in the same direction ( to the right in fig . 1 ) , which satisfy the usual equal - time commutation relations @xmath26=i \pi \mbox{sgn}(x)\ , \delta_{lp}$ ] . the majorana fermions @xmath27 account for the mutual statistics of electrons in different edges , and @xmath28 is a common energy cut - off of the edge modes . the fermi momenta @xmath29 correspond to the average electron density in the edges , while the operators of the density fluctuations are : @xmath30 . in the symmetric case of equal times of excitation propagation between the contacts along the two edges , the two combinations of the bosonic operators @xmath25 that enter the electron tunneling terms of the two contacts can be expressed as the values at points @xmath31 of the same right - propagating chiral bosonic field : @xmath32 the langrangian describing electron tunneling in the two contacts can then be written as : @xmath33 \ , , \label{e2}\ ] ] where @xmath34 and @xmath35 are the absolute values and the phases of the dimensionless tunneling amplitudes . the products of the majorana fermions @xmath36 were omitted from the lagrangian ( [ e2 ] ) , since they cancel out in each perturbative order due to charge conservation . the phases @xmath35 include contributions from the external magnetic flux @xmath9 and from the average electron numbers @xmath37 on the two sides of the interferometer : @xmath38 + \mbox{const } \equiv -\kappa .\ ] ] the factor @xmath39 in the lagrangian ( [ e2 ] ) follows from the normalization of the bosonic field @xmath40 , which in the absence of tunneling is a free right - propagating chiral field . this field undergoes successive scattering at the two contacts by the tunneling terms of the lagrangian . the scattering breaks the charge conservation and therefore creates tunneling current . the applied voltage can be introduced at first as a shift of the incoming field of one of the edges : @xmath41 . as one can see from eqs . ( [ phi- ] ) , such a shift translates into the shift of the tunneling field @xmath42 . the thermodynamic bethe ansatz solution of the tunneling model was developed @xcite for a single - point tunneling contact with @xmath43 by application of one - particle boundary @xmath44-matrices @xcite to a distribution of the bosonic field excitations ( kinks , antikinks , and breathers ) introduced through the massless limit of the `` bulk '' sine - gordon model . this solution was generalized @xcite to the two tunneling contacts relevant for the mzi problem by successive application of two boundary @xmath44-matrices to the same distribution of the excitations of the model . for the charge transport , only the kink - antikink ( and vice verse ) transitions are important , and their boundary @xmath44-matrices are written as @xmath45 here the standard energy scales @xmath46 , @xmath47 , are used to characterize the tunneling strength at the individual contacts , and @xmath48 ) \over \nu \gamma(\nu/[2(1-\nu ) ] ) } .\ ] ] the explicit relation between the energy scales @xmath46 and electron tunneling amplitudes is given below . at zero temperature , dynamics of the liquid should be described with only one type of quasiparticles , e.g. , kinks , which fill out all available states with the `` bulk '' distribution @xmath49 , with @xmath50 being the quasiparticle momentum , up to some limiting momentum @xmath51 defined by the applied voltage . each momentum-@xmath50 quasiparticle undergoes successive scattering at the two tunneling contacts independently of other quasiparticles . the overall scattering process is described by the product of the two boundary @xmath44-matrices given by eq . ( [ e27 ] ) . our goal is to find the cumulant - generating function @xmath52 of the charge transfer between the two branches of the interferometer , which is defined , as usual , as the logarithm of the fourier transform of the probability distribution function of the transferred charge . we measure the charge in units of the elementary electron charge by setting @xmath53 . the independence of the scattering events of quasiparticles with different momenta implies then that @xmath52 can be found as a sum of logarithm of the generating functions of individual momentum states , and its long - time asymptotics is @xmath54 the generating function of one state with momentum @xmath50 @xmath55 is defined by the total transition probability @xmath56 of the momentum-@xmath50 kink into antikink . taking the product of the scattering matrices of the two contacts to find the total scattering matrix @xmath57 , and using the parametrization of the contact tunneling strengths as @xmath58 , we can write the probability @xmath56 from the corresponding matrix element of @xmath57 as @xmath59 here @xmath60 are the transition probabilities in the individual point contacts : @xmath61^{-1 } , \label{tau1}\ ] ] and the factor @xmath62 characterizes interference between the two contacts : @xmath63 without a loss of generality , we assume below a specific relation between the tunneling strength parameters of the two contacts , @xmath64 , and write them as @xmath65 , with @xmath66 . the aim of our subsequent derivation is to find the cumulant - generating function @xmath52 in terms of the two generating functions @xmath67 for charge transfer in an individual point contact that was found from the bethe ansatz solution by saleur and weiss @xcite . this derivation does not need the explicit expressions for @xmath49 and @xmath51 which can be found in @xcite . following the approach for one contact , we first relate @xmath52 in eq . ( [ lnp1 ] ) to the effective tunneling current . to do this , we introduce the generalized tunneling probability @xmath68 : @xmath69^{-1 } \ , . \label{tauc2}\ ] ] which is the solution of the following differential equation in the new parameter @xmath70 : @xmath71 that satisfies the initial condition @xmath72 . one can extend eq . ( [ p1 ] ) for @xmath73 to include the parameter @xmath70 through the substitution @xmath74 . equation ( [ tauc2 ] ) shows then that the logarithm of @xmath73 extended this way can be expressed as @xmath75\big{|}_{z=0}^{z = i\xi } \ , . \label{der1}\ ] ] calculating the derivatives of eq . ( [ der1 ] ) with respect to @xmath70 and @xmath76 , and using eqs . ( [ tauc2 ] ) , one can see that @xmath77 combining eq . ( [ der2 ] ) with ( [ lnp1 ] ) in which @xmath73 is extended to include the parameter @xmath70 , one sees that the cumulant - generating function satisfies the following relation @xmath78 which expresses it through the total tunneling current @xmath79 in the two contacts that is defined by the generalized tunneling probability @xmath80 . as the next step , one substitutes eq . ( [ tauc1 ] ) into ( [ tauc2 ] ) and casts the total tunneling probability @xmath80 into the following form @xmath81 where @xmath82 is defined by the conditions that @xmath83 and @xmath84 . differentiation of eq . ( [ theta1 ] ) shows that the coefficient in eq . ( [ tauc4 ] ) in front of the sum can be written as the derivative of @xmath85 : @xmath86 using eqs . ( [ tauc4 ] ) and ( [ theta2 ] ) in the definition of the tunneling current @xmath79 ( [ lnp2 ] ) we obtain an important relation @xmath87 which expresses the derivative of the cumulant - generating function ( [ lnp2 ] ) for the charge transfer in a symmetric interferometer in terms of the tunneling current @xmath88 in one point contact between the two edges with the effective filling factor @xmath89 . the current in one contact has been calculated @xcite from the bethe ansatz solution , and its tunneling conductance @xmath90=g_{1/m}[(v / a ) e^{-\theta/[2(m-1)]}]=i_{1/m}(\theta , v ) /v\ ] ] is given at zero temperature by a universal scaling function expressed in the form of the low- and high - voltage expansion series . integrating eq . ( [ ic ] ) over @xmath70 , and using the result in eq . ( [ lnp2 ] ) , we express the generating function @xmath91 in the following form : @xmath92 } ] \ , .\label{lnp3 } \end{aligned}\ ] ] the explicit expansion series for @xmath93 in eq . ( [ lnp3 ] ) allow integration in each order . the integration transforms the generating function ( [ lnp3 ] ) into the sum of the two generating functions @xmath67 for charge transfer in individual contacts and gives our key result : @xmath94 the single - contact generating function @xmath67 is known in terms of the low- and high - voltage expansion series @xcite : @xmath95 where @xmath96 and @xmath97 is the conductance quantum . equation ( [ lnps1 ] ) representing the total generating function @xmath98 as the sum of the two single - contact generating functions @xmath67 seems to suggest that the charge transfer in the mzi is divided into two independent processes associated with the two point contacts of the mzi . indeed , one manifestation of this division is that dependence of each of these processes on the bias voltage @xmath99 ( through the function @xmath100 ) is determined , as in eq . ( [ lnps2 ] ) , only by the characteristics energy scale @xmath46 of the corresponding junction . such a division , however , is not complete . the total generating function @xmath98 depends also on the charge dynamics in the interferometer as a whole , since each individual charge transfer process in one contact triggers multiple charge transfers involving interference between both contacts of the interferometer . information about this charge dynamics enters eq . ( [ lnps1 ] ) through the function @xmath101 determined by eq . ( [ theta1 ] ) and sensitive to both interferometer contacts . nevertheless , the interference can become irrelevant if the two contacts are strongly asymmetric . for @xmath102 , the generating function @xmath98 is well approximated at low voltages by the single - contact @xmath103 defined by @xmath104 , i.e. by the strongest electron tunneling amplitude @xmath105 see eq . ( [ w1 ] ) below . at large voltages , the generating function @xmath98 is approximated by the single - contact @xmath67 defined by @xmath106 , i.e. , by the weakest electron tunneling amplitude @xmath107 . at small voltages , when @xmath108 for both @xmath47 , the low - voltage expansions of the single - contact generating function @xmath67 ( [ lnps2 ] ) for both terms in eq . ( [ lnps1 ] ) can be combined as follows : @xmath109 \ , .\label{lnp4}\end{aligned}\ ] ] since @xmath110 is a polynomial of @xmath111 , and the latter , according to eq . ( [ theta1 ] ) , is a linear function of @xmath112 , this expansion of the mzi generating function shows that at low voltages , the charge transfer between the edges of the mzi is quantized in units of electron charge . more explicitly , using the standard expansion of @xmath113 ( see eq . 1.331.4 in ref . ) and the relations that follow from eqs . ( [ b1 ] ) and ( [ theta1 ] ) : @xmath114 \ , , \nonumber\\ r\equiv b\tanh\delta\theta_0= { |t_{1b}^{m-1}+t_{2b}^{m-1}e^{i\kappa}|^2 \over t_{1b}^{2(m-1)}+t_{2b}^{2(m-1 ) } } \ , , \label{r } \\ \cosh\delta\theta_0= { t_{1b}^{2(m-1)}+t_{2b}^{2(m-1 ) } \over 2(t_{1b}t_{2b})^{(m-1 ) } } , \nonumber\end{aligned}\ ] ] we bring eq . ( [ lnp4 ] ) into the following form : @xmath115^n \nonumber \\ \cdot \ , \big\ { [ 1+r(z-1)]^n + n \sum_{l=1}^{[n/2 ] } \frac{(-1)^lc^{n - l-1}_{l-1}}{l(2\cosh\delta\theta_0)^{2l } } \nonumber \\ \cdot \ , \big [ 1+r(z-1 ) \big]^{n-2l } \big\ } \big|^{z = e^{i\xi}}_{z=1 } . \label{lnp5 } \end{aligned}\ ] ] equation ( [ lnp5 ] ) quantifies our previous conclusions about properties of the mzi charge transfer statistics ( [ lnps1 ] ) in the low - voltage regime . in the limit of strongly different contacts , @xmath116 , one finds that @xmath117 and @xmath118 , so that the charge transfer statistics ( [ lnp5 ] ) approaches that of one point contact @xcite characterized by @xmath104 ( i.e. , the strongest electron tunneling amplitude @xmath105 ) with corrections in @xmath119 also quantized in the electron charge units . the @xmath120th term in the expansion of this statistics in powers of bias voltage @xmath99 corresponds in this case to tunneling of exactly @xmath120 electrons . by contrast , the @xmath120th order term of the general mzi transfer statistics ( [ lnp5 ] ) involves transfer of all numbers of electrons up to @xmath120 . in the lowest order in @xmath99 , the mzi statistics reduces to the poisson distribution , with the coefficient in front of @xmath121 in eq . ( [ lnp5 ] ) is equal to the average electron tunneling current . one can check this starting from eq . ( [ e2 ] ) by direct perturbative calculation @xcite , if the energy scales @xmath46 are expressed through the electron tunneling amplitudes @xmath34 : @xmath122 ^{{1/(m- 1 ) } } . \label{w1}\ ] ] at large voltages , @xmath123 , the combination large - voltage expansions of both single - contact generating functions @xmath67 in eq . ( [ lnps1 ] ) brings the total mzi generating function into the following form : @xmath124 \ , . \;\;\ ; \label{lnp6}\end{aligned}\ ] ] in terms of the parameters introduced in eq . ( [ r ] ) , this equation can be rewritten as : @xmath125^{{n\over m } } \nonumber \\ \cdot\ , \sum_\pm \big [ 1+r ( z-1 ) \pm \big ( [ 1+r(z-1)]^{2 } \nonumber \\ - \cosh^{-2}\delta\theta_0 \big)^{{1 \over 2 } } \big]^{{n\over m } } \big|^{z = e^{i\xi}}_{z=1 } . \;\;\ ; \label{lnp7}\end{aligned}\ ] ] we can see again that in the asymmetric limit , @xmath126 , when @xmath117 and @xmath118 , eq . ( [ lnp7 ] ) for the charge transfer statistics reduces to that of a single point contact @xcite . the dominant contact is now characterized by the larger quasiparticle tunneling amplitude @xmath127 ( i.e. , smaller electron tunneling amplitude @xmath107 ) and corresponding energy scale @xmath106 , related to @xmath127 as : @xmath128^{m/(m-1)}. \label{w2}\ ] ] the @xmath120th order term in the expansion ( [ lnp7 ] ) of the generating function corresponds to the transfer of the fractional charge @xmath129 by @xmath120 quasiparticles one can see , however , that the mzi transfer statistics ( [ lnp7 ] ) does not contain in general the terms @xmath130 that would correspond directly to transfer of individual quasiparticles of charge @xmath131 . in particular , the @xmath132 term of the expansion that gives the leading large - voltage contribution to the statistics , can not be interpreted as a poisson process of tunneling of independent quasiparticles , in contrast to the leading low - voltage term that did represent poisson process of individual electron tunneling events . the reason for this is the @xmath2-state dynamics of the effective flux through the interferometer associated with the quasiparticle tunneling , which introduces correlations in the tunneling process . these correlations can be most easily understood in the description of the quasiparticle tunneling based on kinetic equation . such an equation is discussed in the next section . to derive kinetic equation that reproduces the large - voltage asymptotics of the generating function ( [ lnp6 ] ) , and therefore provides a simple physical picture of the dynamics of quasiparticle tunneling , we start by rewriting this asymptotics in terms of the quasiparticle tunneling amplitudes @xmath133 . from the last relation in eq . ( [ r ] ) , we have @xmath134 . \label{der5}\ ] ] using this equation to transform the leading , @xmath132 , term in eq . ( [ lnp6 ] ) , and replacing the energy scales @xmath46 by the quasiparticle amplitudes @xmath133 with the help of eq . ( [ w2 ] ) , we have : @xmath135 \ , , \label{lnp8}\ ] ] where @xmath136 . kinetic equation describing the quasiparticle tunneling can be written down based on the following considerations . the general picture of the quasiparticle dynamics in the mzi discussed in the introduction implies that the quasiparticles create statistical contribution to the effective flux through the interferometer . because of this statistical contribution , the mzi can be found in @xmath2 separate states which differ by the effective flux , with each tunneling quasiparticle changing successively the state @xmath17 into @xmath137 modulo @xmath2 . since the total rates of the quasiparticle tunneling in the mzi depend on the effective flux , such dynamics of flux introduces correlations into quasiparticle transitions , separating naturally the processes of successive quasiparticle tunneling events into the groups of @xmath2 transitions . as usual , to cast the kinetic equation governing this flux dynamics into the form appropriate for the calculation of the cumulant - generating function , we multiply transition probabilities by a factor @xmath138 that keeps track of the transferred charge . then , we introduce the probabilities @xmath139 that at time @xmath140 the mzi is in the state @xmath17 and @xmath120 quasiparticles have been transferred through it . combining these probabilities into an @xmath2-dimensional vector @xmath141 , one can write the kinetic equation in the following matrix form : @xmath142 according to the qualitative picture of quasiparticle tunneling discussed above , the transition matrix has a simple form , with the only non - vanishing elements are those on the main diagonal , @xmath143 , and those with @xmath144 : @xmath145 here the kronecker symbol @xmath146 is defined modulo @xmath2 . the leading large - time asymtotics of the generating function for the probability distribution evolving according to the kinetic equation ( [ kineq ] ) is ( see , e.g. , @xcite ) : @xmath147 , where @xmath148 is the maximum eigenvalue of the transition matrix ( [ m ] ) . the structure of this matrix shows directly that the characteristic equation @xmath149 has the form : @xmath150 the maximum eigenvalue @xmath148 is the solution of this equation which goes to zero when @xmath151 , since all other eigenvalues of the matrix @xmath152 are negative . before trying to establish the general relation between the generating function obtained from this equation and the generating function ( [ lnp8 ] ) , we consider the simple case @xmath20 . in this case , eq . ( [ lnp8 ] ) for the large - voltage asymptotics of the generating function ( [ lnp6 ] ) , can be simplified further . first , we have from eq . ( [ r ] ) : @xmath153^{1/2 } .\ ] ] using this relation , eq . ( [ r ] ) , and eq . ( [ der5 ] ) with @xmath20 , we transform eq . ( [ lnp8 ] ) into : @xmath154^{1/2}\!\!\ ! -\sum_j w^2_j\big\ } . \label{lnp9}\end{aligned}\ ] ] on the other hand , in the kinetic - equation approach , eq . ( [ lambda1 ] ) be solved readily for @xmath20 giving the following expression for @xmath148 : @xmath155^{1/2 } -\sum_j \gamma_j \big)/2 \ , . \label{lambda2}\ ] ] this equation describes the generating function for the statistics of any transfer process consisting of the two steps with the rates @xmath156 , e.g. , incoherent charge transfer through a resonant level @xcite . comparison of eqs . ( [ lambda2 ] ) and ( [ lnp9 ] ) shows that the generating function ( [ lambda2 ] ) obtained from the kinetic equation reproduces the large - voltage asymptotics ( [ lnp9 ] ) of the generating function ( [ lnp6 ] ) , if the tunneling rates are taken as @xmath157 equation ( [ emp1 ] ) for the tunneling rates agrees with the physical picture of quasiparticle tunneling discussed above . statistical contribution to the effective flux though the mzi means that tunneling of each quasiparticle changes the phase between the interferometer contacts by @xmath158 in agreement with the quasiparticle anyonic exchange statistics . equation ( [ emp1 ] ) also shows that the quasiparticles see a phase shift @xmath159 in addition to the phase @xmath160 induced by the external magnetic flux . the origin of this phase shift is discussed below . following this logic , we look for solution of eq . ( [ lambda1 ] ) in the case of arbitrary @xmath2 taking the tunneling rates @xmath161 as : @xmath162 with some unknown @xmath163 . with these tunneling rates , eq . ( [ lambda1 ] ) reads @xmath164 where @xmath165 . to further transform eq . ( [ lambda3 ] ) , we use the basic identity : @xmath166 and the two identities that follow directly from it : @xmath167 = \cosh\delta\theta \label{iden } \\ - \ , ( -)^m \cos(m\phi ) \ , .\nonumber\end{aligned}\ ] ] the first one is obtained essentially by taking @xmath168 in eq . ( [ identity ] ) , while the second one follows from the first if the sum of the cosines is transformed into their product . by direct comparison of the second identity in ( [ iden ] ) with eq . ( [ lambda3 ] ) we see that @xmath148 as defined by @xmath169 in eq . ( [ lnp8 ] ) indeed solves eq . ( [ lambda1 ] ) if @xmath170 making use of eq . ( [ identity ] ) one more time , we calculate the product on the right - hand - side of eq . ( [ rr ] ) : @xmath171 one can see that with this expression for the product , eq . ( [ rr ] ) precisely coincides with the definition of @xmath172 by eq . ( [ r ] ) . indeed , combining all three relations in eq . ( [ r ] ) one can express @xmath172 as @xmath173 replacing @xmath46 in eq . ( [ theta3 ] ) with the amplitudes @xmath133 through eq . ( [ w2 ] ) , we see that eq . ( [ theta3 ] ) precisely coincides with eqs . ( [ rr ] ) and ( [ productgammas ] ) , if @xmath174 , i.e. , if the phase @xmath163 is chosen to satisfy the condition @xmath175 the deviation of the phase @xmath163 from @xmath176 in this equation is important only for even @xmath2 . in this case , it produces the shift @xmath177 of the interference phase in the quasiparticle tunneling rates from the value induced by the external magnetic field . this shift coincides with the phase acquired by one of the two klein factors of the mzi quasiparticle tunneling action of the mzi @xcite in the flux - diagonal representation due to the @xmath2-power periodicity condition for the klein factors , which corresponds physically to the requirement of the proper exchange statistics between electrons and quasiparticles . this phase shift ensures , actually , that there is no shift in the interference pattern of the tunnel current in the interferometer ( see the final results below ) between the regimes of electron and quasiparticle tunneling . so far , we have established the interpretation of the low- and high - voltage asymptotic behavior of the charge transfer statistics in terms of , respectively , tunneling of individual electrons and quasiparticles . in this section , we calculate the charge transfer cumulants in these two limits , with the emphasis on the quasiparticle limit which exhibits the non - trivial behavior of the cumulants . we also will use the generating function found in sec . [ lnp ] for arbitrary voltages to calculate the full voltage dependence of the cumulants , and to study the crossover between the two asymptotic regimes of electron and quasiparticle tunneling . the cumulants of the charge @xmath178 transferred through the interferometer during a large time interval @xmath140 can be found from the cumulant - generating function ( [ lnp2 ] ) by the standard relation : @xmath179 where the current @xmath79 is given by eq . ( [ ic ] ) . the first cumulant gives the average tunneling current : @xmath180 . at arbitrary voltages , the average current was calculated before in @xcite . its large-@xmath99 asymptotics can also be obtained directly from eqs . ( [ lnp7 ] ) and ( [ w2 ] ) : @xmath181 where the interference factor @xmath62 ( [ b1 ] ) can be expressed in term of the quasiparticle amplitudes @xmath182 as @xmath183 the second equality in eq . ( [ i1 ] ) provides direct interpretation of the asymptotics of the average current in terms of the quasiparticles transitions @xcite . it can be proven formally by making use of the identity @xmath184 \ , , \label{sum1}\ ] ] which can be obtained by directly differentiating individual rates @xmath161 in this equation . on the other hand , differentiating the product of all @xmath161s together , as given by eq . ( [ productgammas ] ) , and using eq . ( [ fi ] ) , we obtained the second equality in ( [ i1 ] ) . this equality agrees naturally with the simple solution of the quasiparticle kinetic equation , which gives the average tunneling rate as inverse of the average tunneling times in the different states of the interferometer . the second cumulant @xmath185 defines the spectral density of the current fluctuations at zero frequency @xmath186 , which at zero temperature reflects the shot noise associated with the charge transfer processes . from eq . ( [ ic ] ) , one can write the first derivative of the current as @xmath187 \nonumber \\ -(\partial_u\delta\theta(u))^2 \sum_\pm \partial_{\bar \theta } i_{1/m}(\bar \theta\mp \delta\theta(u),v)\ , . \label{dic } \end{aligned}\ ] ] substituting this formula into eq . ( [ cum ] ) and calculating the derivatives of @xmath188 ( [ theta2 ] ) for @xmath189 we obtain expression for the spectral density of current : @xmath190 it is convenient to characterize the short noise represented by this spectral density through the fano factor @xmath191 defined as @xmath192 . in the case of mzi , the fano factor reflects both the charge and statistics of the tunneling excitations and illustrates the transition between the electron and quasiparticle regimes . in the low - voltage limit , @xmath193 as a result of the regular poisson process of electron tunneling . to find the fano factor in the quasiparticle , large - voltage , limit , we start with eq . ( [ i2 ] ) which gives the following general expression for @xmath191 : @xmath194^{-1 } \big\ } \ , . \;\;\ ; \label{ff } \end{aligned}\ ] ] using eq . ( [ r ] ) , the fact that in the large - voltage limit only one quasiparticle tunneling term @xmath195 can be kept in the current @xmath88 , cf . ( [ i1 ] ) , and that with the parametrization of the energy scales @xmath46 with @xmath196 introduced above , @xmath197 , we get from eq . ( [ ff ] ) : @xmath198 the fano factor ( [ f ] ) corresponds to the dynamics of quasiparticle tunneling as described by the kinetic equation ( [ kineq ] ) . this can be seen by following the steps similar to that taken above for the average current . applying the differential operator from eq . ( [ sum1 ] ) to individual terms in the sum of the inverse tunneling rates @xmath161 , one obtains directly the following identity : @xmath199 on the other hand , replacing the sum of inverse @xmath161s in this equation with the corresponding expression from eq . ( [ i1 ] ) : @xmath200 \ , , \ ] ] and performing differentiation , we see that the large - voltage asymptotics ( [ f ] ) of the fano factor can be written in terms of the tunneling rates @xmath161 as @xmath201 this result agrees with the calculation @xcite based directly on the kinetic equation . because of the complex nature of the quasiparticle tunneling dynamics characterized by @xmath2 different tunneling rates @xmath161 , @xmath191 is not equal simply to the quasiparticle charge @xmath131 but varies as a function of parameters , e.g. the interference phase @xmath18 , between @xmath131 and @xmath202 . ( 3.,2.6 ) ( -0.2,-0.15)=3.4 in at arbitrary bias voltage @xmath99 , the fano factor @xmath191 should be plotted numerically . figure 2 shows @xmath191 in the case @xmath203 which corresponds , e.g. , to tunneling between the two @xmath204 edges . the curves are shown for different degrees of asymmetry between the two contacts of the interferometer and two values of interference phase , maximum constructive interference , @xmath205 , and complete destructive interference , @xmath206 . the range of variation of @xmath191 with the interference phase @xmath18 decreases with increasing junction asymmetry . in general , the curves show the transition between electron tunneling with @xmath193 at small voltages @xmath99 to quasiparticle tunneling at large voltages . in the quasiparticle regime , @xmath191 is still can be significantly different from @xmath207 because of the non - trivial three - state flux dynamics of the mzi . in particular , for identical junctions , the three total quasiparticle tunneling rates ( [ emp2 ] ) satisfy the relation : @xmath208 . taking into account eq . ( [ fi ] ) for the non - statistical contribution to the interference phase , one can see that under the condition of destructive interference , @xmath209 , the tunneling rate in one of the flux states of the interferometer , @xmath210 , is much smaller than the rates in the two other states . this mean that on the relevant large time scale set by the smallest rate , the three quasiparticles transition that transfer interferometer from state @xmath210 back to itself happen almost simultaneously , so that the three quasiparticle charges @xmath207 are effectively transferred together , restoring @xmath191 back to 1 . finally , we study the third charge transfer cumulant that characterizes the asymmetry around average of the transferred charge distribution , and has been measured experimentally for electron tunneling in metallic tunnel junctions see , e.g. , @xcite , and in quantum point contacts @xcite . as for the other cumulants , the large - time asympotic of the third cumulant is linear in time , and it can be characterized by the coefficient @xmath211 . to calculate this coefficient , we first find the second derivative of the tunnel current from eq . ( [ dic ] ) : @xmath212 ^ 2 \big)i - \nonumber \\ \sum_\pm \big [ 3\partial_u\delta\theta \partial^2_u\delta\theta \partial_{\bar \theta } \mp ( \partial_u\delta\theta)^3 \partial^2_{\bar \theta } \big ] i_{1/m}(\bar \theta\mp \delta\theta , v ) . \;\;\ ; \label{ddic } \end{aligned}\ ] ] derivatives of @xmath213 here can be found from eq . ( [ theta2 ] ) . in particular , the coefficient in front of @xmath214 in eq . ( [ ddic ] ) can be expressed as : @xmath215 ^ 2=1- 3\coth \delta \theta(u)\nonumber \\ \times \partial_u \delta\theta(u ) + [ 3\coth^2 \delta \theta(u)-1 ] ( \partial_u\delta\theta(u))^2\ , . \label{coeff } \end{aligned}\ ] ] substitution of eqs . ( [ ddic ] ) , ( [ coeff ] ) , and ( [ theta2 ] ) into eq . ( [ cum ] ) gives us the coefficient @xmath216 of the third cumulant : @xmath217 i \nonumber \\ - \sum_{j=1,2 } \big[3 b^2 ( 1-b\coth \delta \theta_0 ) \partial_\theta i_{1/m } ( \theta_j , v ) \;\;\ ; \label{c3 - 2 } \\ + ( -)^jb^3\partial^2_\theta i_{1/m}(\theta_j , v ) \big ] . \nonumber \end{aligned}\ ] ] the ratio @xmath218 has also been suggested @xcite as a possible alternative to the fano factor to characterize the charge of the tunneling particles . indeed , in a poisson process , @xmath219 is equal to the fano factor multiplied by the tunneling charge . therefore , in the case of mzi , @xmath219 ( [ c3 - 2 ] ) reduces to 1 in the low - voltage limit , as a result of the regular poisson electron tunneling . in the quasiparticle , large - voltage , limit , repeating the calculation similar to that leading to eq . ( [ ff ] ) , we can relate both factors as follows : @xmath220 for general @xmath2 , the results for the cumulants of the charge transfer statistics in the mzi discussed above can not be presented in a finite analytical form for arbitrary bias voltages . the situation is simpler for @xmath20 , when the kink - quasiparticles of the `` bulk '' sine - gordon model that provide the basis for the bethe - ansatz solution of the mzi transport are the regular fermions ( though carrying charge @xmath221 ) , and their distribution @xmath49 in eq . ( [ lnp1 ] ) is the fermi - dirac step - function . in practice , the @xmath20 regime should take place in the mzi formed by the two edges of different quantum hall liquids , with filling factors @xmath204 and @xmath222 . individual tunneling contacts of this type have been realized experimentally @xcite . using the fermi - dirac property of the distribution function @xmath49 for @xmath20 , one can obtain the cumulant generating function @xmath223 either by integration in eq . ( [ lnp1 ] ) , or equivalently , by direct substitution into eq . ( [ lnp3 ] ) of the known single - contact tunnel conductance @xmath224 , which can be expressed simply as @xmath225/2 $ ] . the resulting generating function is : @xmath226 \nonumber \\ + ( 1/2 ) \ln[1+y^2_j(u ) ] \big ) \big|^{u = i\xi}_{u=0 } \ , , \label{m2 } \end{aligned}\ ] ] where @xmath227 , and @xmath228 . this generating function can be combined with eq . ( [ cum ] ) to calculate the cumulants of the transferred change distribution for @xmath20 using the same steps as in the previous section . alternatively , one can substitute eq . ( [ lnp1 ] ) into ( [ cum ] ) and perform the integration directly . below we briefly discuss the first three cumulants obtained in this way . the first cumulant gives the average tunneling current in the mzi as @xcite : @xmath229 where @xmath230 , and @xmath231 are the characteristic quasiparticle tunneling rates in separate contacts . using the fact that @xmath232 for @xmath20 , one can see explicitly that the current agrees at large voltages with eq . ( [ i1 ] ) that follows from the quasiparticle kinetic equation with the tunneling rates ( [ emp1 ] ) : @xmath233 it is interesting to note that this agreement relies strongly on the shift of the interference phase ( [ fi ] ) from the externally - induced phase @xmath18 . ( 3.,2.6 ) ( -0.2,-0.15)=3.4 in the second cumulant gives the following expression for the fano factor at arbitrary voltages , including the transition region between electron and quasiparticle tunneling : @xmath234 \big\ } \ , . \label{fm2}\end{aligned}\ ] ] this equation is plotted in fig . ( [ fonehalf ] ) and describes the transition from @xmath193 for electron tunneling at small voltages to @xmath235 for quasiparticle tunneling at large voltages . one can see that the quasiparticle charge @xmath236 manifests itself most clearly for @xmath205 , when the total quasiparticle tunneling rates ( [ emp1 ] ) coincide , @xmath237 , regardless of the relation between the individual rates @xmath238 . similarly to the case @xmath203 illustrated in fig . 2 , the fano factor reduces to electron value 1 even in the quasiparticle regime , if @xmath239 and @xmath206 . in this case , one of the total quasipartical tunneling rates @xmath240 is much smaller than the other , so on the relevant large time scale set by the smaller rate the quasiparticles effectively tunnel together , restoring @xmath191 to 1 . ( 3.,2.6 ) ( -0.2,-0.15)=3.4 in calculating the third cumulant , we find the following expression for the `` alternative '' fano factor : @xmath241 ^ 2 } \big\ } \nonumber \ , , \end{aligned}\ ] ] where @xmath242 is defined by @xmath243 , @xmath244 . equation ( [ last ] ) is plotted in fig . [ f3onehalf ] , which shows @xmath219 as a function of the interference phase @xmath18 at several voltages . voltage dependence of @xmath219 is qualitatively very similar to that of the fano factor shown in fig . ( [ fonehalf ] ) : it approaches 1 at small voltages , in agreement with the underlying poisson tunneling process of electrons . at large voltages , eq . ( [ last ] ) reduces to the following form @xmath245 this expression can be understood in terms of the same two - state tunneling dynamics of the quasiparticles that was discussed above . also similarly to the fano factor , the voltage dependence of @xmath219 is non - monotonic , with a minimum between the regimes of electron and quasiparticle tunneling . the main qualitative difference between the noise - related fano factor and its third - cumulant alternative is that the minimum of @xmath219 can become negative for some values of parameters ( regime not shown in fig . [ f3onehalf ] ) . starting from the exact solution of the tunneling model of symmetric mach - zender interferometer in the fqhe regime , we have calculated the statistics of the charge transfer between interferometer edges . the obtained statistics shows the transition from electron tunneling at low voltages to tunneling of anyonic quasiparticles of the fractional charge @xmath5 and statistical angle @xmath19 at large voltages . deep in the electron tunneling regime , the dynamics of charge transfer is represented by the standard poisson process . dynamics of quasiparticle tunneling is more complicated and reflects the existence of @xmath2 effective flux states of the interferometer . the interference phase between the quasiparticle tunneling amplitudes in two contacts of the interferometer contains a contribution from the quasiparticle exchange statistics , making the quasiparticle tunneling rates in different interferometer states different . in general , the transition from electron to quasiparticle tunneling is reflected in the fano factor @xmath191 or its third - cumulant alternative @xmath219 , which both reach minima in the transition region . however , in the regime close to complete destructing interference ( interferometer phase @xmath206 and equal tunneling strength in the two contacts ) , both @xmath191 and @xmath219 have electron value 1 for all voltages .

the atlas and the cms collaborations have reported observations of a new particle with a mass of about 126 gev , which is considered to be the standard model ( sm ) higgs boson @xcite . if the particle is indeed the higgs boson , the sm inevitably involves the hierarchy problem . this unnaturalness indicates that there lies physics beyond the sm . there is another indication for physics beyond the sm . the precise measurement of the muon anomalous magnetic moment ( the muon @xmath1 ) @xcite has shown discrepancy from the sm prediction @xcite . with dedicated efforts to determine hadronic contributions precisely , the latest result is @xmath2 where the hadronic vacuum polarization is quoted from ref . @xcite , and the hadronic light - by - light contribution is from ref . the muon @xmath1 anomaly indicates physics beyond the sm at more than @xmath0 level . moreover , the difference is as large as the sm electroweak contribution , @xmath3 @xcite . if new physics is responsible for the discrepancy , its contribution is naively estimated as @xmath4 , where @xmath5 is a coupling constant of new particles to the muon , and @xmath6 is a typical scale of their masses . thus , new physics around the tev scale is required to involve strong couplings with the muon in order to solve the muon @xmath1 anomaly . the supersymmetry ( susy ) is a promising candidate for the tev - scale new physics . the minimal susy standard model ( mssm ) not only solves the hierarchy problem but also yields sizable contributions to the muon @xmath1 @xcite . the susy contributions to the muon @xmath1 are naturally enhanced by @xmath7 . as we will see in sec . [ sec : muon_g-2 ] , the muon @xmath1 anomaly is solved if the superparticles ( smuons , neutralinos , and charginos ) are around @xmath8 for @xmath9 . superparticles are searched for at the lhc . since none of them has been discovered , colored superparticles are considered to be heavier than @xmath10 @xcite . moreover , the higgs boson mass of @xmath11 indicates the scalar tops to be as heavy as @xmath12 @xcite and/or to have a large trilinear coupling to the up - type higgs boson @xcite . these two results naively contradict with the indication of the muon @xmath1 anomaly that superparticles have a mass of @xmath8 . in fact , some of the representative susy - breaking mediation mechanisms such as the minimal supergravity and the gauge mediation can not satisfy the above three phenomenological requirements simultaneously , ( i ) providing the higgs boson mass of @xmath11 , ( ii ) avoiding the constraints from the direct searches at the lhc , and ( iii ) solving the muon @xmath1 anomaly ( see ref . @xcite for a study on the minimal supergravity ) . from the viewpoint of model building , the muon @xmath1 is reconciled with the higgs boson mass of @xmath11 by extending the mssm so that extra contributions to the higgs potential appear @xcite . the above inconsistency implies a split mass spectrum of the superparticles as @xmath13 where the squarks are favored to be relatively heavy to explain the higgs boson mass and to satisfy the lhc bounds , while the sleptons , neutralinos and charginos are to be light to explain the muon @xmath1 discrepancy . this scenario forces us to change strategy of the susy search . the searches based on productions of colored superparticles , which have been the standard , are not always promising , because the particles can be away from the lhc reach without diminishing a virtue of the susy as a solution to the muon @xmath1 anomaly . in this letter , we investigate the lhc signatures assuming the mass hierarchy in eq . ( [ eq : rough_spectrum ] ) , with particular attention to the muon @xmath1 . we examine the direct productions of the superparticles which are relevant for the muon @xmath1 . the atlas and cms collaborations respectively reported results of their searches for this production channel @xcite . we will apply the result by the atlas to the above scenario , and will see that these searches are particularly important in the parameter regions where the muon @xmath1 is explained . in sec . [ sec : muon_g-2 ] , the susy contributions to the muon @xmath1 is reviewed , and the parameter regions in which the discrepancy is explained are clarified . the detailed mass spectra and parameter spaces are introduced in sec . [ sec : mass ] , and relevant susy searches at the lhc are summarized in sec . [ sec : lhc ] . in sec . [ sec : result ] , current lhc bounds on the muon @xmath1 parameter spaces are shown , and future prospects are discussed in sec . [ sec : future ] . the last section is devoted to the conclusion . the susy contributions to the muon @xmath1 are dominated by the chargino sneutrino and the neutralino smuon loop diagrams . at the leading order of @xmath14 and @xmath15 , where @xmath16 represents susy - breaking masses and the higgsino mass @xmath17 , they are evaluated as @xcite @xmath18 where @xmath19 is the muon mass , while @xmath20 and @xmath21 are the fine structure constants of the sm u(1)@xmath22 and the su(2)@xmath23 gauge symmetries , respectively . the loop functions are defined as and @xmath24 , are reduced from the functions , @xmath25 and @xmath26 , in ref . @xcite . ] @xmath27\ , , \\ & f_n(x , y)= xy \left [ \frac{-3+x+y+xy}{(x-1)^2(y-1)^2 } + \frac{2x\log x}{(x - y)(x-1)^3 } -\frac{2y\log y}{(x - y)(y-1)^3 } \right]\,.\end{aligned}\ ] ] they satisfy @xmath28 and are monochromatically increasing for @xmath29 and @xmath30 . in the limit of degenerate masses , they satisfy @xmath31 and @xmath32 . the arguments in the left - hand side of eqs . ([eq : bmulr ] ) show the superparticles which propagate in each loop diagram . if one of them decouples , the corresponding susy contribution is suppressed . ( [ eq : whsnu ] ) comes from the chargino sneutrino diagrams , and eqs . ( [ eq : whmul])([eq : bmulr ] ) are the neutralino smuon contributions . we can easily see that the parameters @xmath33 are relevant , where @xmath34 are the bino and wino masses , and @xmath35 are masses of the left- and right - handed smuons . note that the sneutrino mass @xmath36 is related to @xmath37 by the su(2)@xmath23 symmetry as @xmath38 . numerically , the susy contributions are evaluated as @xmath39 the susy contributions to the muon @xmath1 are enhanced when @xmath15 is large and @xmath16 is small . for @xmath40 and @xmath41 , they become @xmath42 , which can explain eq . . since the experimental value is larger than the sm prediction , @xmath43 is favored in most of the parameter region . throughout this letter , the gaugino masses are taken positive . let us investigate the susy contributions in detail to specify the parameter regions which should be searched for at the lhc . we will study two representative cases when the muon @xmath1 discrepancy is explained . other possibilities will be discussed in sec . [ sec : future ] . the first case is the one when the chargino sneutrino contribution ( [ eq : whlnu_n ] ) dominates the susy contributions . as can be seen in eqs . ( [ eq : whlnu_n])([eq : blr_n ] ) , this is valid when the relevant mass parameters of eq . are nearly degenerate . in fact , in wide susy models which explain the anomaly with non - decoupling higgsinos , the chargino sneutrino contribution ( [ eq : whlnu_n ] ) dominates . as the higgsino mass @xmath17 increases , the chargino sneutrino contribution decreases , while the neutralino smuon contribution becomes relevant . this is because the pure - bino contribution eq . is enhanced under the presence of a large @xmath17 resulting in a large left right mixing in the smuon mass matrix . the other contributions in eqs . ( [ eq : whsnu])([eq : bhmur ] ) are suppressed , where the higgsino propagates in the diagrams . in this situation , which we will take as the second case , the muon @xmath1 discrepancy is explained when both @xmath44 and @xmath45 are @xmath8 as will be seen in the next section . in this letter , we will study the susy searches at the lhc for the parameter regions of these two cases . the soft susy - breaking parameters are set at the susy scale , and we will not specify the models to realize them . it is interesting to mention that such parameter regions are realized in wide susy models which solve the muon @xmath1 anomaly ( see e.g. @xcite ) , in which much ingenuity is exercised in enhancing the higgs boson mass . as we take model independent approach , the following lhc analysis will be independent of the mechanisms which realize the higgs boson mass of @xmath11 . let us specify the mass spectrum to study lhc phenomena . we consider two representative cases in light of the muon @xmath1 : ( i ) the case when the chargino sneutrino contribution to the muon @xmath1 dominates , and ( ii ) that with a large @xmath17 parameter where the neutralino smuon diagram is relevant . the former contribution is controlled by @xmath46 , and the latter is by @xmath47 , according to eqs . and , respectively . in particular , the @xmath17 parameter is favored to be relatively small in the former case , while it is fairly large in the latter . the leading susy contributions to the muon @xmath1 are proportional to @xmath15 , while the lhc sensitivities in the following sections do not depend much on it . mainly through the left - right mixing of the stau mass matrix . @xcite [ footnote : stau ] ] it is fixed as @xmath48 throughout the lhc analysis . as @xmath15 increases , although the susy contributions to the muon @xmath1 are enhanced , the bottom and tau yukawa coupling constants are likely to blow up below the gut scale . if @xmath15 is lowered , the superparticle masses must be reduced to solve the muon @xmath1 anomaly . then , the lhc searches become easier . the soft masses of the left- and right - handed selectrons are respectively set to be the same as those of the smuons , which are denoted by @xmath49 and @xmath50 . the results in the following sections are almost independent of the selectron masses . the third - generation sleptons ( staus and tau - sneutrino ) are irrelevant for the susy contributions to the muon @xmath1 . they are taken to be decoupled in the following analysis to simplify the lhc studies . scenarios with light staus will be discussed in sec . [ sec : future ] . for the gaugino masses , the following two scenarios are analyzed : ( a ) they satisfy an approximate gut relation , @xmath51 , which is realized in a wide variety of susy models , and ( b ) the gluino is decoupled , while the bino and wino masses satisfy the same relation as ( a ) , @xmath52 . the susy contributions to the muon @xmath1 are sensitive to @xmath53 and @xmath54 , while @xmath55 barely affects the muon @xmath1 and the higgs boson mass , but the lhc study depends on it . in the case ( a ) , superparticles can be searched for by analyzing events with hard - jets plus missing transverse energy , which come from the gluino pair - production , while electroweak productions of superparticles are relevant in the both cases ( a ) and ( b ) . we will study both of ( a ) and ( b ) separately for the cases , ( i ) and ( ii ) . the bino and wino masses are set to satisfy @xmath56 throughout the analysis , and general cases will be discussed in sec . [ sec : future ] . it is assumed that the squarks are decoupled from the lhc sensitivities in the analysis . we set their soft masses as @xmath57 . they satisfy the current bounds on the superparticle masses obtained at the lhc @xcite , and also , heavy stops are preferred to realize the higgs boson mass of @xmath11 @xcite . the following results do not change as long as the masses are large enough . note that lighter squarks just tighten the lhc constraints , because their productions contribute to the susy events . the remaining susy parameters are the cp - odd higgs mass , @xmath58 , and scalar trilinear couplings ( @xmath59-terms ) . they are relevant neither for the muon @xmath1 nor the susy searches at the lhc . in the analysis , they are fixed as @xmath60 and @xmath61 , as a reference . in the above parameter space , the lightest superparticle ( lsp ) is either the bino - like neutralino , the neutral higgsino , or the sneutrino . the lsp is a candidate of the dark matter if the r - parity is conserved . in the last set of , @xmath66 is set to be slightly larger than @xmath67 in order to avoid a charged lsp in the whole parameter region , since the charged dark matter is excluded @xcite . the sneutrino lsp is also strongly disfavored by the direct searches @xcite . in the analysis , we will focus on the cosmologically favored regions . note that other superparticles such as the gravitino can be lighter than them , and such cases will be discussed in sec . [ sec : future ] . in order to solve the muon @xmath1 anomaly , the bino and wino masses are as small as @xmath8 . then , if we respect the approximate gut relation , the gluino has a mass less than a few tev , which is within the reach of the lhc . the gluinos are produced at the @xmath68 collisions , and decay into neutralinos or charginos with hard jets , @xmath69 similarly to the minimal supergravity models , searches for hard jets plus missing transverse energy are one of the most promising channels . currently , the atlas collaboration yields the most stringent bound in this category @xcite with analyzing their data corresponding to @xmath70 obtained at @xmath71 . we interpret this result in this letter , which will be referred to as `` j - search . '' the electroweak gaugino productions ( e.g. , @xmath72 ) are also searched for at the lhc . when sleptons are relatively light , the gauginos can produce hard leptons as @xmath73 as the muon @xmath1 prefers such lighter sleptons , the multi - lepton signatures are another promising channel . the atlas collaboration recently reported searches for events with three leptons plus missing transverse energy in the data of @xmath74 at @xmath71 @xcite . we interpret the result , which will be referred to as `` l - search . '' the multi - lepton signatures are particularly important , because they are yielded by the superparticles relevant for the muon @xmath1 , i.e. , sleptons , charginos , and neutralinos , and appear even in the decoupled gluino scenario ( b ) . those two atlas results are interpreted in the analysis with monte carlo simulation . on each model point , the mass spectra generated with softsusy3.4 @xcite are passed to susy - hit1.3 @xcite to calculate decay tables of the superparticles . here , the soft susy - breaking parameters are set at the susy scale , @xmath75 . the kinematical distributions of susy events are simulated by pythia6.4 @xcite with the atlas mc09 tune @xcite . the parton distribution functions ( pdfs ) are obtained from the cteq6l1 set @xcite . as the squarks are assumed to be decoupled , superparticles are produced via the gluino pair - production ( @xmath76 ) and the electroweak productions of neutralinos , charginos and sleptons ( @xmath77 ) . for the gluino productions , the cross sections are normalized by nlo @xmath78-factors obtained with prospino2 @xcite , where the cteq6l1 and cteq6.6 m pdfs @xcite are used . for the electroweak channels , the normalization factor is set as @xmath79 , which is a typical value in the parameter space studied in the next section . in addition , the susy contribution to the muon @xmath1 is evaluated by feynhiggs @xcite , which includes two - loop contributions . detector simulation is employed with delphes2.0 @xcite . we assume trigger efficiencies to be @xmath80 for simplicity . as for calorimeter configuration and resolutions , delphes original parameter card for the atlas detector is used . following the atlas analysis , jet reconstruction is done with the anti-@xmath81 jet clustering algorithm @xcite with the distance parameter @xmath82 . fastjet @xcite is utilized . we also simulate @xmath83-tagging , used as @xmath83-veto in the l - search , with simplified efficiency and fake rates estimated from refs . the missing transverse energy is read from delphes output without modification . we take lepton detection efficiencies into account , which are important particularly for the l - search . for electrons , the atlas collaboration utilizes three identification criteria . we consult ref . @xcite , and set the efficiency of the `` medium '' criterion as @xmath84 for @xmath85 and @xmath86 for @xmath87 , and that of the `` tight '' criterion as @xmath88 for @xmath85 and @xmath89 for @xmath87 . @xmath90 dependence of the efficiency is ignored . for muons , we interpret the results in ref . @xcite , and set the efficiency ( for @xmath91 ) as @xmath92 for @xmath93 and @xmath94 for @xmath95 . note that we further impose the lepton isolation criteria in the l - search , and the overlap removal in the j- and l - searches , in the same way as the atlas analysis . the signal regions ( srs ) are defined to be the same as those in the original atlas analyses . the details are summarized in the following subsections . throughout the analysis , we ignore criteria on the primary vertex , origins of lepton tracks , and jet qualities , which are designed to reduce noise backgrounds . the @xmath96 method is used to derive exclusions for each model point . the numbers of the signal events in srs are compared to the corresponding upper bounds obtained in the atlas reports . the analysis procedures are validated by comparing the simulations with the atlas results . we checked that calculated effective mass distributions of the signal events agree with those of the atlas in the j - search , and those of the missing transverse energy are reproduced for the l - search . exclusion plots are also simulated by using the upper bounds on the signal events obtained in the atlas analysis , and become consistent with those in the atlas papers . .definition of srs of the j - search . among the original 12 srs @xcite , the srs which are relevant for our exclusions are shown . the effective mass @xmath97 ( @xmath98 ) is defined as a scalar sum of @xmath99 and @xmath100 s of the leading @xmath101-jets ( all jets with @xmath102 ) . in the last two rows of c and e , the three values mean different srs , called as ` tight ' , ` medium ' , and ` loose ' , respectively . [ cols="^,^,^,^ " , ] the l - search is designed to search for electroweak productions of charginos and neutralinos via the s - channel exchange of a virtual sm gauge boson ( e.g. , @xmath103 ) . the event selections are summarized in table [ tab:2012 - 154 ] . events are required to have exactly three hard leptons . trigger efficiencies are not taken into account for simplicity . note that @xmath83-jet veto and @xmath104-cut are necessary to suppress events of sm electroweak backgrounds . the orange ( yellow ) band shows the region where the susy contributions explain the muon @xmath1 discrepancy at the @xmath105 ( @xmath106 ) level . the dark gray regions in ( a ) and ( c ) are excluded by lep searches for the neutralinos and charginos . the regions left to the blue dotted lines are excluded by the l - search . assuming the approximate gut relation for the gaugino masses , the regions left to the red lines are excluded by the j - search . these exclusions are at @xmath107 cl , and the theoretical uncertainty of @xmath108 is included in the hatched regions . the lsp is the lightest neutralino in the regions above the black thick lines , while the sneutrino is lightest below them . sleptons become lighter than neutralinos below the black dashed or dotted lines ( see the text for details ) . , title="fig:",scaledwidth=99.0% ] + ( a ) @xmath109 , @xmath110 explanations . the orange ( yellow ) band shows the region where the susy contributions explain the muon @xmath1 discrepancy at the @xmath105 ( @xmath106 ) level . the dark gray regions in ( a ) and ( c ) are excluded by lep searches for the neutralinos and charginos . the regions left to the blue dotted lines are excluded by the l - search . assuming the approximate gut relation for the gaugino masses , the regions left to the red lines are excluded by the j - search . these exclusions are at @xmath107 cl , and the theoretical uncertainty of @xmath108 is included in the hatched regions . the lsp is the lightest neutralino in the regions above the black thick lines , while the sneutrino is lightest below them . sleptons become lighter than neutralinos below the black dashed or dotted lines ( see the text for details ) . , title="fig:",scaledwidth=99.0% ] + ( b ) @xmath111 , @xmath110 + explanations . the orange ( yellow ) band shows the region where the susy contributions explain the muon @xmath1 discrepancy at the @xmath105 ( @xmath106 ) level . the dark gray regions in ( a ) and ( c ) are excluded by lep searches for the neutralinos and charginos . the regions left to the blue dotted lines are excluded by the l - search . assuming the approximate gut relation for the gaugino masses , the regions left to the red lines are excluded by the j - search . these exclusions are at @xmath107 cl , and the theoretical uncertainty of @xmath108 is included in the hatched regions . the lsp is the lightest neutralino in the regions above the black thick lines , while the sneutrino is lightest below them . sleptons become lighter than neutralinos below the black dashed or dotted lines ( see the text for details ) . , title="fig:",scaledwidth=99.0% ] + ( c ) @xmath112 , @xmath110 explanations . the orange ( yellow ) band shows the region where the susy contributions explain the muon @xmath1 discrepancy at the @xmath105 ( @xmath106 ) level . the dark gray regions in ( a ) and ( c ) are excluded by lep searches for the neutralinos and charginos . the regions left to the blue dotted lines are excluded by the l - search . assuming the approximate gut relation for the gaugino masses , the regions left to the red lines are excluded by the j - search . these exclusions are at @xmath107 cl , and the theoretical uncertainty of @xmath108 is included in the hatched regions . the lsp is the lightest neutralino in the regions above the black thick lines , while the sneutrino is lightest below them . sleptons become lighter than neutralinos below the black dashed or dotted lines ( see the text for details ) . , title="fig:",scaledwidth=99.0% ] + ( d ) @xmath113 , @xmath114 the lhc constraints and the muon @xmath1 are shown in fig . [ fig : result ] . in the orange ( yellow ) regions , the muon @xmath1 discrepancy is explained by the susy contributions at the @xmath105 ( @xmath106 ) level . the chargino sneutrino contribution dominates in the figures ( a ) , ( b ) and ( c ) , where the parameters are set as @xmath115 and @xmath116 , respectively . it is found that the contribution is suppressed as the wino mass increases . on the other hand , the neutralino smuon contribution dominates in the figure ( d ) with @xmath117 . the muon @xmath1 is explained even with large @xmath54 , while the susy contribution decreases as the left- and right - handed slepton masses increase . the lhc exclusion limits are displayed in fig . [ fig : result ] with the red - solid and blue - dotted lines at the 95% confidence level ( cl ) . including 30% errors of pdf and scale uncertainties , the limits distribute in the red / blue hatched regions . assuming the approximate gut relation , the regions left to the red lines are excluded by the j - search , and those to the blue lines are constrained by the l - search . if the gluino is decoupled , the red lines disappear and only the l - search provides a constraint , because the j - search targets the gluino - pair productions . the bounds from the l - search with the approximate gut relation are almost identical to those in the gluino decoupling limit . it is found that the susy searches at the lhc start to constrain the muon @xmath1 ( orange or yellow ) regions . the muon @xmath1 regions are also constrained by cosmology . noting that the bino mass is set to be a half of the wino mass , the sneutrino becomes the lsp below the black solid lines in fig . [ fig : result ] . due to radiative corrections , which are sizable since the soft susy - breaking parameters are set at @xmath118 . ] such parameter regions are disfavored by the dark matter searches @xcite . on the other hand , the lsp in the regions above the lines is the lightest neutralino . they are constrained by the j- and l - searches at the lhc . the lhc constraints are understood as follows . the j - search targets the gluino - pair productions . among the 12 srs defined in ref . @xcite , those with higher multiplicity of jets are relevant for the exclusion ( see table . [ tab:2012 - 109 ] ) . some of the jets originate in the gluino decay into charginos and neutralinos . others come from decays of the sm bosons that are decay products of the charginos and neutralinos . also , the processes are often associated with initial / final state radiations . assuming the approximate gut relation , the gluino mass is related to the wino mass , and the red lines are almost independent of the other model parameters in fig . [ fig : result ] . the j - search limit @xmath119 means @xmath120 , leading to @xmath121 , where @xmath122 is the gluino pole mass . this is almost the same as the constraint in the minimal supergravity models with heavy squarks @xcite . if the ratio @xmath123 is varied , the exclusion lines shift correspondingly . the l - search targets the electroweak superparticle productions . in particular , the wino productions are significant in the parameter regions in fig . [ fig : result ] . the search works only when hard leptons are produced at the wino decay . in the regions below the black dashed lines , where sleptons are lighter than the wino - like chargino / neutralino , the wino can decay into an on - shell slepton with a lepton , and the slepton produces another lepton when it decays . then , the sr1b provides a tight constraint . above the lines , the wino mainly decays into the sm bosons as long as the channels are open . in such regions , no limit is obtained by the l - search . the search becomes effective again if the mass splitting between the bino and the wino is small , where the wino decays into three - body final states , including leptons . since @xmath53 is set to be @xmath124 , the limit is determined by the wino mass , as can be seen from the vertical blue lines in fig . [ fig : result ] . here , the sr1a gives a limit . on the other hand , the search becomes less effective if sleptons are degenerate with the lightest neutralino , because the lepton at the slepton decay becomes soft . the charged slepton is degenerate with the lightest neutralino on the black dotted lines . note that the regions below the black solid lines are disfavored by cosmology as discussed above . it is emphasized that the l - search constraints are determined by the superparticles which are relevant for the muon @xmath1 . in figs . [ fig : result ] ( a)(c ) , the muon @xmath1 regions are partly excluded by the j- and/or l - searches . below the black dashed lines , tight constraints are obtained by the latter . above them , the limits are determined by the wino mass , i.e. , the gluino mass for the j - search and the splitting between @xmath53 and @xmath54 for the l - search . comparing the figures to each other , it is found that the muon @xmath1 regions with larger @xmath17 are easily detected at the lhc . as @xmath17 increases , @xmath54 is required to be smaller to explain the muon @xmath1 . if the sleptons are lighter than the wino , a wide region of the @xmath105 parameter space of the muon @xmath1 is already excluded in ( a ) and ( b ) . in contrast , almost all the muon @xmath1 regions are allowed by the lhc in ( c ) . in particular , the lepton productions are relatively suppressed for @xmath125 even below the black dashed line , because the winos can decay into the higgsinos , and the higgsinos rarely produce leptons due to tiny yukawa couplings . the sensitivity is slightly improved below the black dash - dotted line in ( c ) , where the sleptons are lighter than the second lightest neutralino . since @xmath126 , the second lightest neutralino does not decay into the sm bosons , but can produce the sleptons . here , the sr1a provides a limit from the higgsino direct productions . in figs . [ fig : result ] ( a)(c ) , @xmath50 is set to be @xmath127 . since the chargino sneutrino contribution is insensitive to @xmath50 , the muon @xmath1 regions do not change so much even if @xmath66 is varied . the lhc constraints do not depend much on @xmath50 either . even when it is as low as @xmath49 , the right - handed selectron and smuon are rarely produced in the cascade decays of the wino and higgsino , because they are singlet under the sm su(2)@xmath23 symmetry and have tiny yukawa interactions . if the mass is too small , the right - handed slepton becomes the lsp , and thus , excluded by the cosmology . the lhc constraints in fig . [ fig : result ] ( d ) are similar to those in ( b ) , where the wino composes the second lightest neutralino . above the black dashed line , since the sleptons are limited to be light by the muon @xmath1 , a large part of the @xmath105 region is already excluded by the j - search , while the l - search bound is determined similarly to ( a)(c ) . on the other hand , the muon @xmath1 is explained even with large @xmath54 . such a parameter region is challenging to be searched for at the lhc . ( see also the discussion in the next section . ) in fig . [ fig : result ] ( d ) , @xmath17 is set to be @xmath128 . as @xmath17 increases , the muon @xmath1 is enhanced , and thus , larger @xmath53 is allowed . however , the upper limit on the slepton masses is not relaxed so much , as noticed from eq . . moreover , if @xmath17 becomes too large , the electroweak vacuum can be destabilized on the smuon higgs plane . by naively scaling a bound on @xmath17 from the stau - higgs stability condition @xcite , the limit is estimated as @xmath129 for @xmath130 and @xmath131 . in contrast , the lhc constraints do not change , because the higgsinos are decoupled . on the other hand , if @xmath17 is set to be smaller , the upper bound on the wino mass gets tightened , and the lhc detection becomes easier . the lhc will be upgraded and run at @xmath132 or @xmath133 with the target luminosity of @xmath134 . further upgrades are proposed to realize @xmath135 or @xmath136 . there are several studies on future sensitivities of the susy searches . in this section , we discuss future searches for the susy parameter regions of the muon @xmath1 . multi - jet signature : : multi - jet searches are important for the approximate gut relation on the gaugino masses . according to the future sensitivity in ref . @xcite based on the minimal supergravity models , the gluino lighter than @xmath137 and @xmath138 can be discovered at the @xmath133 lhc with @xmath139 and @xmath140 , respectively . these gluino masses correspond to @xmath141 and @xmath142 under the approximate gut relation , which means a large parameter regions in fig . [ fig : result ] can be covered . since the sleptons are expected to be light particularly in a large wino mass region , signatures of multi - jets plus leptons may provide a better sensitivity than the expectations based on the minimal supergravity . multi - lepton signature : : multi - lepton searches works very well for the two cases : when sleptons are lighter than charginos and neutralinos , and when the wino mainly decays into three bodies . the former case is typically realized in the regions below the black dashed lines with @xmath143 as we saw in figs . [ fig : result ] ( a ) , ( b ) , and ( d ) , where the multi - leptons are provided from the cascade decays of winos . the latter case corresponds to the left - most regions in ( a)(d ) , where the three - body decay provides the leptons . it is important to study future sensitivities of this channel . @xmath144 signature from gaugino / higgsino decays : : if charginos and neutralinos are lighter than sleptons and can decay into the sm bosons , multi - lepton searches are not promising , but rather , we should explore searches for the sm bosons from the gauginos and/or higgsinos . for @xmath144 signature from gaugino pair productions , a study within the minimal supergravity framework in ref . @xcite shows that the wino masses of @xmath145 and @xmath146 can be covered at the @xmath133 lhc with @xmath139 and @xmath140 , respectively , in one lepton plus multi - bottoms signature . with this kind of searches , most of the muon @xmath1 regions above the black dashed lines in fig . [ fig : result ] are expected to be examined at the lhc , where the branching ratios of the sm boson channels are similar to those in the minimal supergravity models . + the @xmath144 signature is also expected from higgsino decays as long as higgsino is heavier than , but not degenerate to , the lsp . they are produced by the @xmath147-channel exchange of a virtual electroweak gauge boson , and their production cross sections are comparable to those of the winos . thus , the above sensitivity on the wino mass can be naively applied to the higgsino mass . @xmath148 or @xmath149 signature : : it may be possible to discover charginos and neutralinos via the @xmath150 or @xmath151 channels with a large missing transverse energy . since the bosons can be reconstructed by visible decay products or have sizable branching ratios of the bottom quark productions , the events could be discriminated from the sm backgrounds . we need further studies to explore such possibilities . same - sign leptons : : very recently , the authors in ref . @xcite studied the events with a same - sign lepton pair and no other leptons , obtained from same - sign @xmath152-bosons . they found that the signature is useful to search for the electroweak superparticle events . in our setup , this signature appears more frequently because the sleptons are lighter , and thus , may cover the muon @xmath1 regions . di - lepton channel search : : even with combining the above techniques , the bottom right corners of figs . [ fig : result ] ( c ) and ( d ) are still challenging . future multi - lepton searches at the lhc may not be sensitive enough to cover the regions completely since @xmath54 is very large . here , since sleptons are required to be light , di - lepton channels via the pair productions of sleptons might be promising . the search at the lhc is challenging because it suffers from huge sm backgrounds and their production cross sections are much smaller than those of the charginos and the neutralinos in this parameter region . the ilc is suitable to search for the di - lepton signatures . so far , we considered two representative scenarios to explain the muon @xmath1 discrepancy : with dominance of the chargino sneutrino contribution , and of the pure - bino smuon diagram . we have also set that the stau is heavy , and the lsp is within the mssm . let us touch on other possibilities . we have imposed @xmath153 , but this is not mandatory . as long as @xmath53 is smaller than , but not degenerate to , @xmath54 , and the winos are produced at the lhc , the above conclusions and discussions hold valid . on the other hand , if @xmath54 is too large with @xmath53 fixed , the muon @xmath1 regions disappear in figs . [ fig : result ] ( a)(c ) , whereas those in fig . [ fig : result ] ( d ) remain . searches for di - lepton signatures are necessary as discussed above . for the case when @xmath154 , the muon @xmath1 regions in figs . [ fig : result ] ( a)(c ) does not depend much on @xmath53 , while the lhc signatures are changed . signatures from the cascade decays of binos are less promising than those of winos due to a smaller cross section of the bino production . rather , signatures from the productions of higgsinos and sleptons are hopeful . moreover , a disappearing track at the lhc is expected to be observed . the track comes from the charged wino , which is almost degenerate with the neutral wino and decays into a neutral wino with a soft pion @xcite . so far , we investigated the parameter regions where the susy contribution to the muon @xmath1 is dominated either by the chargino sneutrino contribution or the pure - bino smuon contribution . let us mention other possibilities . first of all , the contribution of eq . can not take dominance , since it is always buried in that of eq . . next , if the left - handed smuon is decoupled , only eq . contributes to the muon @xmath1 , where the sign of @xmath155 is favored to be negative . lastly , eq . becomes dominant if the wino and the right - handed smuon are decoupled . however , the contributions of eqs . and are quantitatively small as shown in eqs . and , respectively . thus , the superparticles must be very light , and the scenarios are expected to be examined by searching for the sm boson or di - lepton channels . the staus and the tau sneutrino were supposed to be decoupled from the lhc phenomenology . if their masses are comparable to those of the selectron and smuon , physics becomes much involved . attention must be paid to the vacuum stability conditions on the stau - higgs plane especially when @xmath17 is large , i.e. , in fig . [ fig : result ] ( d ) @xcite . the decay branching ratio of the higgs boson into the di - photon may be affected in such parameter spaces ( see the footnote [ footnote : stau ] ) . as for the direct searches for the superparticles , since a stau tends to be light due to a large left - right mixing in the stau mass matrix , taus are likely to be produced in the decay chains of the superparticles . then the tau reconstruction becomes a central subject , which is one of the challenging topics of the lhc . because of the rich phenomenology , these scenarios are worth investigated for future . finally let us relax our restriction of the lsp . we focused on the neutralino lsp to avoid the cosmological constraints . however , they can be circumvented if the lsp is a superparticle which is not in the mssm , such as the gravitino and the axino . in those cases , charged particles or sneutrino can be the lsp among the mssm superparticles ( mssm - lsp ) , where the lhc signatures change correspondingly . if a sneutrino is the mssm - lsp , hard leptons are less produced because the left - handed charged sleptons are degenerate to the sneutrinos . although sneutrino decays into the lsp are not visible , the pair - production of the sleptons could be found by the mono - jet or mono - photon searches . in addition , this case might be checked by reconstructing soft particles that are produced by the quasi - degenerate charged slepton when it decays into the sneutrino . this search could be done at the ilc . on the other hand , when the mssm - lsp is a charged slepton , it leaves a track in the detectors if it is long - lived , or generates a hard lepton when it decays . finally , if the lightest neutralino is the mssm - lsp and decays into the lsp in the detectors , the signature can be distinguished from the backgrounds by searching for events with multiple ( displaced ) photons . all of these possibilities and future sensitivities on them will be studied elsewhere . the susy not only solves the hierarchy problem between the planck and electroweak scales , but also explain the 34@xmath156 deviation of the muon @xmath1 . the latter suggests that superparticles have masses of @xmath8 , whereas the higgs boson mass of @xmath11 and the lhc results on the direct susy searches indicate that they exist in @xmath12 . in this letter , we examined the parameter regions where the muon @xmath1 is explained with squarks decoupled . the muon @xmath1 anomaly is solved when electroweak superparticles are light and @xmath9 . they are searched for by the lhc j- and l - searches . we found that the wino is constrained to be heavier than @xmath157 by the j - search under the approximate gut relation on the gaugino masses , and than @xmath158 by the l - search irrespectively of the gluino mass . in addition , if sleptons are produced by decays of charginos and neutralinos , the l - search provides a severe constraint , and the bound can be as tight as @xmath159 . consequently , the muon @xmath1 parameter space starts to be constrained by the lhc independently of details of the susy models . we also discussed future prospects on the collider searches for the superparticles which are relevant for the muon @xmath1 . a wide parameter region is expected to be examined at the 13@xmath133 lhc . in particular , the sm boson channels of the electroweak superparticle decays such as the @xmath160 productions are promising for the study . however , the searches are challenging when the higgsinos are degenerate with the lightest neutralino , or if both of the wino and the higgsino are heavy . we need further studies to explore the whole parameter space where the muon @xmath1 anomaly is solved . the study in this letter is motivated by the muon @xmath1 anomaly . the @xmath161 deviation may originate in the uncertainties . currently they are dominated by those from the measurement and from theoretical calculation of the hadronic contributions . the former will be improved by the experiments at fermilab @xcite and j - parc @xcite . the reduction of the uncertainty in the hadronic vacuum polarization contribution requires more experimental data for the cross section of @xmath162 , e.g. , at vepp-2000 @xcite . the evaluations of the hadronic light - by - light contribution have been improving ( see ref . @xcite for a recent review ) , and lattice calculations are in development @xcite . if the anomaly is confirmed in future , the susy is one of the most attractive candidates for the solution . this letter is a first step towards the lhc searches for the scenario , and we hope it is useful for further studies . this work was supported by jsps kakenhi grant no . 23740172 ( m.e . ) , no . 21740164 ( k.h . ) , no . 22244021 ( k.h . ) and no . 228132 ( s.i . ) . the work of t.y . was supported by an advanced leading graduate course for photon science grant . collaboration , http://dx.doi.org/10.1016/j.physletb.2012.08.020[phys . ( 2012 ) 129 ] . collaboration , http://dx.doi.org/10.1016/j.physletb.2012.08.021[phys . ( 2012 ) 3061 ] . * muon g-2 * collaboration , http://dx.doi.org/10.1103/physrevd.73.072003[phys . rev . * d73 * ( 2006 ) 072003 ] ; b. l. roberts , http://dx.doi.org/10.1088/1674-1137/34/6/021[chin . * c34 * ( 2010 ) 741744 ] . k. hagiwara , a. d. martin , d. nomura , and t. teubner , http://dx.doi.org/10.1016/j.physletb.2007.04.012[phys . lett . ( 2007 ) 173179 ] ; t. teubner , k. hagiwara , r. liao , a. d. martin , and d. nomura , http://dx.doi.org/10.1088/1674-1137/34/6/019[chin . * c34 * ( 2010 ) 728734 ] ; k. hagiwara , r. liao , a. d. martin , d. nomura , and t. teubner , http://dx.doi.org/10.1088/0954-3899/38/8/085003[j . g * g38 * ( 2011 ) 085003 ] . m. davier , _ et al . _ , http://dx.doi.org/10.1140/epjc/s10052-009-1219-4[eur . j. ( 2010 ) 127136 ] ; m. davier , a. hoecker , b. malaescu , c. z. yuan , and z. zhang , http://dx.doi.org/10.1140/epjc/s10052-010-1246-1[eur . j. ( 2010 ) 19 ] ; m. davier , a. hoecker , b. malaescu , and z. zhang , http://dx.doi.org/10.1140/epjc/s10052-010-1515-z[eur . j. ( 2011 ) 1515 ] [ http://arxiv.org/abs/1010.4180[arxiv:1010.4180 ] ] . erratum http://dx.doi.org/10.1140/epjc/s10052-012-1874-8[ibid . * c72 * ( 2012 ) 1874 ] . j. prades , e. de rafael , and a. vainshtein , http://arxiv.org/abs/0901.0306[arxiv:0901.0306 ] ( 2009 ) . a. czarnecki , w. j. marciano , and a. vainshtein , http://dx.doi.org/10.1103/physrevd.67.073006[phys . * d67 * ( 2003 ) 073006 ] [ http://arxiv.org/abs/hep-ph/0212229[hep-ph/0212229 ] ] . erratum http://dx.doi.org/10.1103/physrevd.73.119901[ibid . * d73 * ( 2006 ) 119901 ] . j. l. lopez , d. v. nanopoulos , and x. wang , http://dx.doi.org/10.1103/physrevd.49.366[phys . * d49 * ( 1994 ) 366372 ] . u. chattopadhyay and p. nath , http://dx.doi.org/10.1103/physrevd.53.1648[phys . * d53 * ( 1996 ) 16481657 ] . t. moroi , http://dx.doi.org/10.1103/physrevd.53.6565 , 10.1103/physrevd.56.4424[phys . * d53 * ( 1996 ) 65656575 ] . collaboration , http://cdsweb.cern.ch/record/1472710[atlasconf2012109 ] ( 2012 ) . * cms * collaboration , http://dx.doi.org/10.1007/jhep10(2012)018[jhep * 1210 * ( 2012 ) 018 ] ; * cms * collaboration , http://dx.doi.org/10.1103/physrevlett.109.171803[phys . ( 2012 ) 171803 ] . y. okada , m. yamaguchi , and t. yanagida , http://dx.doi.org/10.1143/ptp.85.1[prog . * 85 * ( 1991 ) 16 ] ; j. r. ellis , g. ridolfi , and f. zwirner , http://dx.doi.org/10.1016/0370-2693(91)90863-l[phys . * b257 * ( 1991 ) 8391 ] ; h. e. haber and r. hempfling , http://dx.doi.org/10.1103/physrevlett.66.1815[phys . rev . ( 1991 ) 18151818 ] ; j. r. ellis , g. ridolfi , and f. zwirner , http://dx.doi.org/10.1016/0370-2693(91)90626-2[phys . * b262 * ( 1991 ) 477484 ] . y. okada , m. yamaguchi , and t. yanagida , http://dx.doi.org/10.1016/0370-2693(91)90642-4[phys . lett . ( 1991 ) 5458 ] . m. endo , k. hamaguchi , s. iwamoto , k. nakayama , and n. yokozaki , http://dx.doi.org/10.1103/physrevd.85.095006[phys . * d85 * ( 2012 ) 095006 ] . m. endo , k. hamaguchi , s. iwamoto , and n. yokozaki , http://dx.doi.org/10.1103/physrevd.84.075017[phys . * d84 * ( 2011 ) 075017 ] ; m. endo , k. hamaguchi , s. iwamoto , and n. yokozaki , http://dx.doi.org/10.1103/physrevd.85.095012[phys . rev . * d85 * ( 2012 ) 095012 ] ; m. endo , k. hamaguchi , s. iwamoto , and n. yokozaki , http://dx.doi.org/10.1007/jhep06(2012)060[jhep * 1206 * ( 2012 ) 060 ] ; m. endo , k. hamaguchi , k. ishikawa , s. iwamoto , and n. yokozaki , http://dx.doi.org/10.1007/jhep01(2013)181[jhep * 1301 * ( 2013 ) 181 ] . j. l. evans , m. ibe , and t. t. yanagida , http://dx.doi.org/10.1016/j.physletb.2011.10.031[phys . lett . ( 2011 ) 342348 ] ; j. l. evans , m. ibe , s. shirai , and t. t. yanagida , http://dx.doi.org/10.1103/physrevd.85.095004[phys . * d85 * ( 2012 ) 095004 ] . cho , k. hagiwara , y. matsumoto , and d. nomura , http://dx.doi.org/10.1007/jhep11(2011)068[jhep * 1111 * ( 2011 ) 068 ] . collaboration , http://cdsweb.cern.ch/record/1493493[atlasconf2012154 ] ( 2012 ) . m. carena , i. low , and c. e. m. wagner , http://dx.doi.org/10.1007/jhep08(2012)060[jhep * 1208 * ( 2012 ) 060 ] . t. kitahara , http://dx.doi.org/10.1007/jhep11(2012)021[jhep * 1211 * ( 2012 ) 021 ] . m. carena , s. gori , i. low , n. r. shah , and c. e. m. wagner , http://arxiv.org/abs/1211.6136[arxiv:1211.6136 ] ( 2012 ) . t. kitahara and t. yoshinaga , http://arxiv.org/abs/1303.0461[arxiv:1303.0461 ] ( 2013 ) . a. kudo and m. yamaguchi , http://dx.doi.org/10.1016/s0370-2693(01)00938-8[phys . ( 2001 ) 151155 ] . t. falk , k. a. olive , and m. srednicki , http://dx.doi.org/10.1016/0370-2693(94)90639-4[phys . ( 1994 ) 248251 ] . b. c. allanach , http://dx.doi.org/10.1016/s0010-4655(01)00460-x[comput . commun . * 143 * ( 2002 ) 305331 ] . a. djouadi , m. m. muhlleitner , and m. spira , acta phys . * b38 * ( 2007 ) 635644 . t. sjostrand , s. mrenna , and p. z. skands , http://dx.doi.org/10.1088/1126-6708/2006/05/026[jhep * 05 * ( 2006 ) 026 ] . collaboration , http://cds.cern.ch/record/1247375/[atlphyspub2010002 ] ( 2010 ) . w. beenakker , r. hopker , m. spira , and p. m. zerwas , http://dx.doi.org/10.1016/s0550-3213(97)00084-9[nucl . ( 1997 ) 51103 ] . s. heinemeyer , w. hollik , and g. weiglein , http://dx.doi.org/10.1016/s0010-4655(99)00364-1[comput . commun . * 124 * ( 2000 ) 7689 ] ; s. heinemeyer , w. hollik , and g. weiglein , http://dx.doi.org/10.1007/s100529900006[eur . . j. * c9 * ( 1999 ) 343366 ] ; g. degrassi , s. heinemeyer , w. hollik , p. slavich , and g. weiglein , http://dx.doi.org/10.1140/epjc/s2003-01152-2[eur . phys . j. ( 2003 ) 133143 ] ; m. frank , _ et al . _ , http://dx.doi.org/10.1088/1126-6708/2007/02/047[jhep * 0702 * ( 2007 ) 047 ] . s. ovyn , x. rouby , and v. lemaitre , http://arxiv.org/abs/0903.2225[arxiv:0903.2225 ] ( 2009 ) . m. cacciari , g. p. salam , and g. soyez , http://dx.doi.org/10.1088/1126-6708/2008/04/063[jhep * 0804 * ( 2008 ) 063 ] . m. cacciari , g. p. salam , and g. soyez , http://dx.doi.org/10.1140/epjc/s10052-012-1896-2[eur . j. ( 2012 ) 1896 ] ; m. cacciari and g. p. salam , http://dx.doi.org/10.1016/j.physletb.2006.08.037[phys . ( 2006 ) 5761 ] . collaboration , https://atlas.web.cern.ch/atlas/groups/physics/confnotes/atlas-conf-2012-040/[atlasconf2012040 ] ( 2012 ) . collaboration , http://dx.doi.org/10.1140/epjc/s10052-012-1909-1[eur . j. ( 2012 ) 1909 ] . collaboration , https://atlas.web.cern.ch/atlas/groups/physics/confnotes/atlas-conf-2011-063/[atlasconf2011063 ] ( 2011 ) . j. hisano and s. sugiyama , http://dx.doi.org/10.1016/j.physletb.2010.12.013[phys . ( 2011 ) 9296 ] . h. baer , v. barger , a. lessa , and x. tata , http://dx.doi.org/10.1103/physrevd.86.117701[phys . * d86 * ( 2012 ) 117701 ] . h. baer , _ et al . _ , http://arxiv.org/abs/1302.5816[arxiv:1302.5816 ] ( 2013 ) . collaboration , http://dx.doi.org/10.1007/jhep01(2013)131[jhep * 1301 * ( 2013 ) 131 ] . collaboration , http://dx.doi.org/10.1016/j.nuclphysbps.2011.06.038[nucl . * 218 * ( 2011 ) 237241 ] . collaboration , http://dx.doi.org/10.1088/1742-6596/295/1/012032[j . ( 2011 ) 012032 ] . y. m. shatunov , _ et al . _ in _ proceedings of the 7th european particle accelerator conference _ , j. l. laclare , _ et al . _ , eds . , conf . * c0006262 * ( 2000 ) 439441 . e. de rafael , http://arxiv.org/abs/1210.4705[arxiv:1210.4705 ] ( 2012 ) . t. blum , http://dx.doi.org/10.1103/physrevlett.91.052001[phys . rev . ( 2003 ) 052001 ] ; * qcdsf * collaboration , http://dx.doi.org/10.1016/j.nuclphysb.2004.03.026[nucl . phys . ( 2004 ) 135164 ] ; c. aubin and t. blum , http://dx.doi.org/10.1103/physrevd.75.114502[phys . * d75 * ( 2007 ) 114502 ] ; x. feng , k. jansen , m. petschlies , and d. b. renner , http://dx.doi.org/10.1103/physrevlett.107.081802[phys . ( 2011 ) 081802 ] ; p. boyle , l. del debbio , e. kerrane , and j. zanotti , http://dx.doi.org/10.1103/physrevd.85.074504[phys . * d85 * ( 2012 ) 074504 ] ; m. della morte , b. jager , a. juttner , and h. wittig , http://dx.doi.org/10.1007/jhep03(2012)055[jhep * 1203 * ( 2012 ) 055 ] .

a goal of monte carlo ( mc ) simulations in statistical mechanics @xcite is the calculation of ensemble mean values of thermodynamic quantities . ensemble mean values are multidimensional integrals over configuration space @xmath0 where @xmath1 is the probability of finding a system in the state defined by @xmath2 , and the functional form of @xmath1 depends on the ensemble investigated . mc simulations usually generate a sampling of configuration space @xmath3 by the use of a stochastic process with stationary probability @xmath4 . the quantity @xmath5 evaluated at @xmath6 is the output of the simulation @xmath7 , and its arithmetic mean value @xmath8 , in principle , must approach the ensemble mean value.@xcite in this paper we refer to the set of configurations generated in a monte carlo simulation as a time sequence , and we study the behavior of these temporal sequences @xmath9 and their arithmetic mean , to understand better how mc simulations approach ergodic behavior . it is important to emphasize that there are two time variables to consider . the time variable @xmath10 labels the separate configurations generated in a monte carlo walk . variations of properties with @xmath10 provide information about the short - time behavior of a mc simulation . the time variable @xmath11 labels the total length of the mc walk , and variations of computed properties with @xmath11 provide information about the convergence of the simulation on a long time scale . given an infinite time , the stochastic walker in a mc simulation visits every allowed point in configuration space.@xcite ergodic behavior is reached when the length of the walk is sufficiently long to sample configuration space appropriately.@xcite in practice , this does not mean that the space has been densely covered but that every region with non - negligible probability has been reached . in such a case we can say that the simulation is effectively ergodic or that it has reached the ergodic limit . for a finite walk , in the event of broken ergodicity @xcite , phase space is effectively disconnected . the different disconnected regions ( called components ) are separated by barriers of zero effective probability . if a stochastic walker starts its walk in one of these regions , it may not cross the barriers within the time of the simulation . if the simulation length is increased , some barriers may become accessible for the walker and phase space is better sampled . we can conclude that a time @xmath12 exists such that , for simulation lengths shorter than @xmath12 , the walker becomes trapped in one of the phase space components . for simulation lengths much larger than @xmath12 , phase space is effectively covered by the walker . in this study we imagine a system having more than one time scale @xmath13 . in a monte carlo simulation each scale comes from stochastic processes with different correlation times.@xcite a precise definition of the correlation times for monte carlo processes is given in section iii , but for the moment we can think of these correlation times as identical to physical time scales of the system under study . to understand these time scales more fully , it is useful to focus on an example . prototypical of systems having such disparate time scales are atomic and molecular clusters . typical cluster potential surfaces have many local minima separated by significant energy barriers . @xcite the local minima can be grouped into basins of similar energies , with each basin separated from other basins again by energy barriers . at short monte carlo times a cluster system executes small amplitude oscillations about one of its potential minima . we can think of these vibrational time scales as the shortest time scales that define a cluster system . as the simulation time is extended the system eventually hops between different local minima within the same basin . the time scale for the first hops between local minima can be considered the next shortest time scale for the simulation . at still longer monte carlo times , the system hops between different energy basins defining yet another time scale for the simulation . this grouping of time scales continues until the longest time scale for a given system is reached . at monte carlo times that are long compared to this longest time scale , the simulation is ergodic . consider a system with several time scales as mentioned above . if the length of the simulation is smaller than the smallest correlation time , the walker may become trapped in an effectively disconnected region and the sampling of phase space is incomplete . by increasing the time , the memory of the initial condition in the sampling decreases as the walker crosses to other previously unreachable regions . these oscillations and hoppings can be modeled by a superposition of stochastic processes with different correlation times . these processes with non - zero correlation times are known as _ colored noise _ processes ( as opposed to zero correlation time _ white noise _ processes ) . @xcite from the study of the autocorrelation functions of a stochastic model defined using these colored noise processes , we can verify that , at a fixed run length @xmath11 , there exist two different groups of processes ; those that contribute to the autocorrelation function with terms that decay like @xmath14 ( called _ diffusive _ processes ) , and those that contribute to the autocorrelation function with terms that decay slower than @xmath14 ( called _ non - diffusive _ processes ) . when the time of the simulation is increased , some non - diffusive processes at shorter run lengths , start to contribute to the autocorrelation function like diffusive processes . after the walk length reaches the largest correlation time @xmath15 , all processes contribute to the autocorrelation function with terms that decay like @xmath14 . at this point , the simulation is at the diffusive regime and effective ergodicity has been reached . a principal goal of this work is to investigate the way in which the mc output @xmath9 reaches the diffusive limit ( i.e. the ergodic limit ) by studying the properties of autocorrelation functions under changes of scale in time , @xmath16 with @xmath17 . by time scaling it is possible to infer the decay law of the non - diffusive contributions with respect to the total simulation time @xmath11 . the functional dependence of the non - diffusive contributions on the parameter @xmath18 that is used to scale @xmath11 is determined empirically . we have found the decay law so determined to be a particularly valuable method of concluding when a simulation can be considered ergodic . unlike previous studies @xcite that only have investigated the behavior of certain autocorrelation functions in the ergodic regime , by focusing on the approach to ergodic behavior we have a more careful monitor of the onset of ergodicity . once the non - diffusive contributions have decayed to a point where they are too small to be distinguished from zero to within the fluctuations of the calculation , we can say that the ergodic limit has been reached . the autocorrelation functions we use to measure the approach to the ergodic limit are based on one of the probes of ergodicity developed by thirumalai and co - workers @xcite , and is often called the _ energy metric_. the energy metric has been proposed as an alternative to other techniques @xcite ( like the study of the lyapunov exponents @xcite ) for the study of ergodic properties in molecular dynamics ( md ) simulations . the metric has been used to study the relative efficiency of mc simulation methods as well . @xcite the mc metric as used in the current work can easily be extended from the energy to other scalar observables of the system . we present two key issues in this paper . first , from the knowledge of the decay law of the non - diffusive contributions to the mc metric , we infer how long a simulation must be to be considered effectively ergodic . second , once the ergodic limit is reached , we can compare the results from different numerical algorithms to measure relative efficiencies . because the outcomes of mc simulations are noisy , we have found it useful to separate diffusive and non - diffusive terms in the mc metric with a fourier analysis so that we can neglect the high frequency components of the noise . this technique has given reproducible results . to test the match between the stochastic model and actual monte carlo simulations , we examine the approach to ergodic behavior in simulations of lennard - jones clusters . recently @xcite we have studied the thermodynamic properties of lennard - jones clusters as a function of temperature using both j - walking @xcite and parallel tempering methods . @xcite both simulation techniques require an initial high temperature that must be ergodic when metropolis monte carlo methods @xcite are used . if the metropolis method does not give ergodic results at the initial high temperature , systematic errors propagate to the lower temperatures in j - walking and parallel tempering simulations , and the results can be flawed or meaningless . in most monte carlo simulations of clusters at finite temperatures , @xcite the clusters are defined by enclosing the atoms within a constraining potential about the center of mass of the system . the constraining potential is necessary because clusters at finite temperatures have finite vapor pressures , and the association of any one atom with the cluster can be ill - defined . from experience @xcite we have found that if the radius of the constraining potential and the initial high temperature are not both carefully chosen , it can be difficult to attain ergodicity with metropolis methods . a key concern then is the choice of constraining radius and the choice of initial temperature . we verify the stochastic model by investigating monte carlo simulation results as a function of the temperature and the size of the constraining potential . the contents of the remainder of this paper are as follows . in section ii we motivate the studies that follow by examining numerally the behavior of a set of monte carlo simulations of a 13-particle lennard - jones cluster . this cluster system is used to illustrate the results throughout this paper . in section iii we introduce the stochastic model based on a continuous time sequence . in section iv we extend the model to discrete time sequences characteristic of actual monte carlo simulations . in section v we test the discrete stochastic model with applications to lennard - jones clusters and in section vi we summarize our conclusions . many of the key derivations needed for the developments are found in two appendices . before discussing the major developments of this work , it is useful to understand the nature of the problem we are attempting to solve by examining some numerical results on a prototypical system . we take the 13-particle lennard - jones cluster defined by the potential function @xmath19 + \sum_{i=1}^nv_c ( \vec{x}_i , r_c ) \label{pot}\ ] ] where @xmath20 and @xmath21 are the standard lennard - jones energy and length parameters , @xmath22 is the number of particles in the cluster ( 13 in the present case ) , @xmath23 is the distance between particles @xmath24 and @xmath25 @xmath26 and @xmath27 is the constraining potential discussed in sec . i @xmath28 where @xmath29 is the coordinate of the center of mass of the cluster and @xmath30 is the radius of the constraining sphere . the 13-particle lennard - jones cluster has a complex potential surface with many minima separated by significant energy barriers , @xcite and ergodicity problems associated with the simulation of properties of this system are well - known . @xcite we now consider a metropolis mc simulation of the average potential energy of the system in the canonical ensemble at temperature @xmath31(@xmath32 is the boltzmann constant ) . this average potential energy @xmath33 is defined by @xmath34 and is displayed in the upper panel of fig . 1 as a function of the walk length @xmath10 for 20 independent simulations each initialized from a random configuration . over the maximum time scale @xmath11 of the walks , it apparent that the potential energy averaged over each independent walk has not converged to the same result . such unreproducible behavior is symptomatic of a simulation not yet at the ergodic limit . = .48 the upper panel shows the time evolution " of @xmath33 ( in units of @xmath20 ) for @xmath35 independent experiments . the lower panel shows @xmath36 ( in units of @xmath37 ) vs. @xmath10 for the experiments of the upper panel . @xmath30 has been set to @xmath38 and @xmath31 . at least two basins with different energies are present . clearly , @xmath36 goes to a constant when @xmath10 is increased within the total time scale of the simulation . at the ergodic limit ( i.e. for the maximum walk length @xmath11 greater than that included in fig . 1 ) the averages displayed in the upper panel of fig . 1 must approach the same value for each walker . using related ideas developed elsewhere , @xcite the extent to which the walks approach the same limit can be measured in terms of a metric @xmath36 defined by @xmath39 ^ 2\,\,,\ ] ] in eq . ( [ omega1 ] ) @xmath40 represents the number of independent walks , and @xmath41 is the average potential energy computed in walk @xmath24 at mc time @xmath10 . the metric measures the energy fluctuations in the walk as a function of the walk length . for an ergodic simulation , the metric must decay to zero . for the 20 simulations of the 13-particle lennard - jones cluster , the metric as a function of @xmath10 is plotted in the lower panel of fig . 1 . rather than asymptotically approaching zero , over the short length of the walk displayed here , @xmath36 has decayed to a constant , and as discussed later in this paper , over the time scale of this simulation , @xmath36 can be qualitatively represented by the function @xmath42 where @xmath43 and @xmath44 are coefficients that are dependent on the total walk length @xmath11 . as @xmath11 is increased to a time where the walk is ergodic , @xmath44 must decay to zero . major goals of this work are to understand how @xmath44 decays and to use the discovered decay law to determine the onset of ergodic behavior . our approach is to introduce first a continuous stochastic model of a simulation followed by a discrete model more clearly linked to actual mc studies . we have discussed in the introduction how the output of mc simulations can be considered to be a combination of stochastic processes with different time scales , and how the contributions to autocorrelation function from these processes can vary when the length of the simulation is enlarged . here we present a continuous time model for the stochastic processes that occur in a simulation . even though a mc simulation occurs in a discrete time ( each mc point represents a time unit ) , we find that the continuous model helps to understand better the ideas used in the modeling of the mc output . in this section the ensemble mean value is used to find the expression for the autocorrelation functions of the model . although in actual numerical calculations the ensemble mean is replaced by a mean over a finite number of independent experiments , the results obtained here give information about the limit of an infinite sample . the stationary process used to sample space is a stochastic process . we assume the output of the mc simulation can be modeled by a linear superposition of stochastic processes with different correlation times @xmath45 , @xmath46 where @xmath47 is a constant , the random variable @xmath48 represents white noise processes with zero correlation time ( @xmath49 ) , and the @xmath50 are stochastic processes with correlation times @xmath51 . @xmath48 and @xmath52 have units of the square root of time , and @xmath53 and @xmath54 are constants with units of @xmath55 . if @xmath5 is chosen to be the the @xmath56-coordinate of a particle , @xmath53 and @xmath54 have units of a diffusion constant . consequently we refer to these constants as generalized diffusion coefficients . the white noise process has the following properties @xcite @xmath57 and the remaining colored noise processes are assumed to satisfy @xmath58 so that they represent processes with a memory @xmath59 . even though correlations between processes with different correlation times may be non - zero , we assume the processes to be independent , i.e. @xmath60 the memory function is assumed to be a continuous function that depends only on the distance between @xmath61 and @xmath62 disregarding the time origin ( stationary condition ) . the memory function represents the correlation between two times of the process @xmath63 . in our model we impose the condition @xmath64 the scope and implications of the leftmost inequality are explored in appendix a. in appendix a we also examine the conditions @xmath59 must satisfy in order to yield contributions to autocorrelation function that decay more weakly than @xmath65 . we now assume that this inequality can be taken as a bound to possible maxima of @xmath59 appearing at @xmath66 . the rightmost inequality enables us to assume @xmath59 is normalized @xmath67 we have identified here the time scale @xmath68 with the correlation time of the stochastic process @xmath63 . this identification is valid if @xmath69 which implies that the behavior of @xmath59 at large @xmath61 must be @xmath70 , or smaller . in addition , by the properties of the ensemble mean value , we have that for all real @xmath71 @xmath72 ^ 2\right\rangle \nonumber\\ & \leq & \left\langle g_{\ell}(t/\tau_{\ell})^2\right\rangle + 2\lambda\,\left\langle g_{\ell}(t/\tau_{\ell})\ , g_{\ell}(t'/\tau_{\ell } ) \right\rangle + \lambda^2\left\langle g_{\ell}(t'/\tau_{\ell})^2\right\rangle \nonumber\\ & \leq & \frac1{\tau_{\ell}}\left\ { f_{\ell}(0 ) + 2\lambda\ , f_{\ell}\left(\frac{|t - t'|}{\tau_{\ell}}\right ) + \lambda^2\ , f_{\ell}(0 ) \right\ } \label{a3}\,\,.\end{aligned}\ ] ] equation ( [ a3 ] ) must be true for all @xmath71 . therefore , the discriminant of the polynomial in @xmath71 must be non - positive @xmath73 \left[f_{\ell}\left(\frac{|t - t'|}{\tau_{\ell}}\right ) + \ , f_{\ell}(0)\right]\leq0\,\,.\ ] ] consequently , @xmath74 . other properties of @xmath59 are studied in appendix a. the ensemble mean value @xmath75 is time independent . the ensemble mean value of the noise processes is zero . therefore , @xmath47 must be equal to @xmath75 . processes defined by eq . ( [ eq:2 ] ) have two different components , uncorrelated white noise and correlated processes with correlation time @xmath68 . because the goal of the simulation is the calculation of the ensemble mean @xmath76 by the analysis of the time series , we study the behavior of the temporal mean @xmath77 @xmath78 where @xmath79 is a wiener process , @xcite @xmath80 with @xmath81 , and @xmath82 . equation ( [ eq:3 ] ) implies that the evolution of the temporal mean @xmath77 has the same structure as @xmath5 , with an uncorrelated term and terms with tailed correlation functions . the autocorrelation function of the process @xmath8 at times @xmath61 and @xmath62 is defined by @xmath83 where we have used eqs . ( [ a1 ] ) and ( [ a2 ] ) to neglect terms involving processes with different correlation times . because we have assumed @xmath59 is a continuous function , @xmath59 reaches its maximum and minimum value within any closed interval considered . the @xmath84th non - diffusive contribution to @xmath85 @xmath86 is bounded @xmath87 2 where @xmath88 , and @xmath89 is the time at which @xmath59 reaches its minimum value in the closed interval @xmath90 $ ] . there exists a @xmath91 $ ] @xcite such that , @xmath92 using eqs . ( [ wiener2 ] ) and ( [ tl ] ) in ( [ c1 ] ) , we find that @xmath93 for all times shorter than @xmath94 the autocorrelation function is the sum of _ diffusive contributions _ ( proportional to @xmath65 ) plus _ non - diffusive contributions_. these contributions implicitly depend on @xmath95 through @xmath96 . we assume that @xmath59 satisfies the conditions stated in appendix a , so that the dependence of @xmath59 on @xmath61 is _ weaker than _ @xmath65 ( for total time scales shorter than @xmath68 ; see appendix a ) . we next consider the behavior of eq . ( [ correl ] ) for time scales greater than @xmath94 . under the scale change @xmath97 such that @xmath98 , the contributions to the correlation function from the process with correlation time @xmath94 can be considered diffusive [ in other words , by virtue of eqs . ( [ noise2 ] ) and ( [ ggg ] ) , @xmath99 has become a delta function ] . with @xmath100 , the other processes preserve their old properties . then , the autocorrelation function can be expressed @xmath101 the complete derivation of eq . ( [ urenor ] ) can be found in appendix b. for a times larger than the correlation time @xmath15 , all contributions to the autocorrelation function are diffusive , the simulation can be considered ergodic , the sampling complete , and the temporal mean is equal to the ensemble mean within @xmath102 mean square fluctuations . monte carlo simulations generate discrete sequences @xmath103 of values of the quantity under study . additionally , in actual calculations the ensemble of sequences is represented by a finite rather than an infinite set . in this section , the model developed in the previous section is extended to finite sets of discrete sequences . we express the @xmath40 sequences @xmath104 , where the label @xmath105 ranges from 1 to @xmath40 . the exact ensemble mean value @xmath75 can be obtained in the limit that @xmath40 becomes infinite . in analogy with the model developed in section iii , each output is assumed to have the form @xmath106 where @xmath107 the true ensemble average @xmath75 does not depend on the index @xmath108 . in the discrete case we define a metric @xmath109 ^ 2\,\,,\ ] ] where the bars represent the temporal mean value @xmath110 with @xmath111 observe that in the present case , our finite sample of the infinite ensemble is the set of outcomes from @xmath40 independent numerical experiments . the metric we have defined in eq . ( [ omega ] ) can be contrasted with alternative metrics @xcite previously defined for molecular dynamics simulations . these alternative metrics examine the fluctuations of two simulations initialized from different components of configuration space averaged with respect to all the particles in the system . the metric we use in this work is determined using an average with respect to @xmath40 independent simulations that represent a subset of the full ensemble . using the model presented in eq . ( [ seq ] ) , we now develop a way to predict the behavior of the mc simulation in the non - ergodic and the ergodic regimes . we first consider the case that the total simulation time @xmath11 is larger than the first correlation time @xmath94 but shorter than @xmath112 , i.e. @xmath113 . the expression for @xmath36 is given by @xmath114 ^ 2\nonumber\\ & = & \frac2 m \sum_{i=1}^m\left(\overline{u}^{(i)}_k-\langle u\rangle\right)^2 - \frac4{m(m-1 ) } \sum_{i=2}^m \sum_{j=1}^{i-1 } \left(\overline{u}^{(i)}_k-\langle u\rangle\right ) \ , \left(\overline{u}^{(j)}_k-\langle u\rangle\right)\nonumber\\ & = & 2\,\gamma_0 \,\frac1 m \sum_{i=1}^m \left(\frac{w_k^{(i)}}k\right)^2 + 2\,\gamma_1 \,\frac1 m \sum_{i=1}^m \left(\frac{g_{1;\,k/\tau_1}^{(i)}}k\right)^2 + 2\,\sum_{\ell=2}^{\lambda}\gamma_{\ell } \,\frac1 m \sum_{i=1}^m \left(\frac{g_{\ell;\,k/\tau_{\ell}}^{(i)}}k\right)^2 \nonumber\\ & & + \,4\,\sum_{\ell=1}^{\lambda}\sqrt{\gamma_0 \gamma_{\ell } } \,\,\frac1 m \sum_{i=1}^m\frac{w_k^{(i ) } \,g_{\ell;\,k/\tau_{\ell } } ^{(i)}}{k^2 } + \,4\,\sum_{\ell=2}^{\lambda}\sum_{\ell'=1}^{\ell-1}\sqrt{\gamma_{\ell } \gamma_{\ell ' } } \,\,\frac1 m \sum_{i=1}^m\frac{g_{\ell;\,k/\tau_{\ell}}^{(i ) } \,g_{\ell';\,k/\tau_{\ell ' } } ^{(i)}}{k^2 } \nonumber\\ & & - \frac4{m(m-1 ) } \sum_{i=2}^m \sum_{j=1}^{i-1 } \left(\overline{u}^{(i)}_k-\langle u\rangle\right ) \ , \left(\overline{u}^{(j)}_k-\langle u\rangle\right)\,\ , . \end{aligned}\ ] ] if the number of experiments @xmath40 is sufficiently large , we can neglect terms involving processes with different correlation times , and products of sequences belonging to different experiments . under these assumptions we obtain @xmath115 2 equation ( [ om ] ) preserves the form of eq . ( [ correl ] ) . to make this statement explicit , let us rewrite eq . ( [ om ] ) as @xmath116 where @xmath117 in appendix b we present a study of the way non - diffusive contribution become diffusive under time scale changes . if @xmath40 is sufficiently large and @xmath118 , by virtue of appendix b , @xmath119 must roughly be a constant . by _ roughly a constant _ we mean a constant @xmath120 plus some rapidly fluctuating function @xmath121 , with the following properties : a ) @xmath122 and b ) @xmath123 . then @xmath124 if @xmath11 is enlarged , we expect to have a larger value of @xmath125 . @xmath126 is a quantity related to the memory functions @xmath59 with correlation times @xmath127 . in the continuous time model , the colored noise processes contribute to the autocorrelation function with terms proportional to @xmath128 , which are weakly dependent on @xmath61 ( see appendix a ) . we can expect @xmath126 to be weakly dependent on @xmath10 , and for sequences of length @xmath11 and for @xmath40 sufficiently large , we consider this quantity roughly to be a constant @xmath129 where @xmath130 represents additional random noise . then , for a given length @xmath131 , the mc metric @xmath36 can be approximated by @xmath132 where @xmath133 represents remaining stochastic noise from both contributions . in this approximation , @xmath125 and @xmath134 are the quantities that carry the long time dependence . short time features appear in the @xmath14 dependence and in the remaining noise @xmath135 . if the sequences considered are increased in size by a factor of @xmath18 , such that @xmath136 for a given @xmath137 , @xmath125 ( @xmath134 ) is increased ( decreased ) ( see appendix b ) . then , @xmath138 where @xmath139 must go to zero and @xmath140 must approach a constant when @xmath18 is increased . by virtue of the expected behavior of the non - diffusive contributions ( see appendix a ) , we propose the following expression for @xmath139 @xmath141 where @xmath142 is a decreasing function of @xmath18 . moreover , @xmath139 is a sum of non - diffusive contributions . as presented in appendix a , each non - diffusive contribution to the autocorrelation function has a relative variation smaller than the relative variation of the diffusive contribution , namely @xmath143 . if this inequality is applicable to the sum of non - diffusive contributions , we have that @xmath144 for all @xmath17 . then , @xmath145 must be either @xmath146 or @xmath147 with @xmath148 and @xmath149 . equation ( [ upa2 ] ) can be thought as the limit of eq . ( [ upa1 ] ) when the exponent goes to zero . we know of no _ a priori _ argument to justify eq . ( [ upa ] ) . however , as is discussed in section v , our numerical experience has shown eq . ( [ upa ] ) to be obeyed in all cases we have examined . our goal is to develop a criterion to decide when the simulation can be considered ergodic . from the previous considerations it is clear that the ergodic limit is reached when @xmath134 is indistinguishable from zero . the output from a mc simulation is usually noisy . therefore , @xmath135 can not be neglected . a useful way to separate diffusive and non - diffusive contributions and to eliminate the stochastic noise from eq . ( [ d ] ) , is to perform a fourier analysis of the function @xmath150 . let us define the frequencies @xmath151 , with @xmath152 . the discrete fourier transform of the function @xmath150 is the signal @xmath153 @xmath154 2 in general , @xmath155 is negligible except at high frequencies . for small positive values of the frequency we can make the approximation @xmath156 . from this approximation we have @xmath157 the real part of eq . ( [ fs ] ) for positive frequencies is @xmath158 even though simpler than eq . ( [ eq:10 ] ) , we have found eq . ( [ ns ] ) is more sensitive to the deviations of @xmath36 from the approximation eq . ( [ d ] ) . therefore , the data obtained from the real part is of poorer quality than the data obtained from the imaginary part . equation ( [ eq:10 ] ) implies that for a given simulation length @xmath11 , the contributions to the mc metric from the non - diffusive process can be determined from a simple relationship involving the fourier transform of the function @xmath150 at low frequencies . by increasing the length of the run @xmath11 by a factor of @xmath18 , it is possible to observe the dependence of @xmath139 on @xmath159 . the concepts developed in the previous sections are sufficiently general to be applied to any kind of mc simulation . we devote the present section to the application of the developments of this paper to the study of the lennard - jones 13-particle cluster in the canonical ensemble . this system has been introduced previously in sec . ii . some thermodynamic properties of clusters as a function of temperature exhibit rapid changes that are reminiscent of similar changes that occur for the same properties in bulk systems at phase transitions . in a bulk system a phase transition occurs at a single temperature . for clusters the rapid changes in thermodynamic properties occur over a finite temperature interval . to distinguish the temperature range where thermodynamic properties change rapidly in clusters from a true phase transition , we follow berry _ et al . _ @xcite and refer to such changes in physical properties as associated with a phase change . a common property that has been found to be useful in monitoring these phase change intervals of temperature is the heat capacity at constant volume @xcite @xmath160 where @xmath161 represents the classical canonical mean value . in this work we consider the bare metropolis ( met ) , @xcite j - walking ( jw ) , @xcite and parallel tempering ( pt ) @xcite approaches to monte carlo simulations . the free variable of all these methods is the reduced temperature @xmath162 . in pt and jw simulations , the highest temperature used ( @xmath163 ) must be sufficiently large to ensure that met is ergodic.@xcite from experience simulating a variety of systems , we have found that @xmath163 must also be lower than a temperature @xmath164 where cluster evaporation events become frequent . it is useful to think of @xmath164 as the cluster analogue of a boiling temperature . we have found that met is unable to sample the boiling phase change region for clusters ergodically , using total time scales accessible to current simulations . for the results that follow , @xmath165 is chosen to be represented by the potential energy of the system . in general @xmath165 can be any scalar property of the system . we define a pass to represent a set of single particle mc moves taken sequentially over the 13 particles in the cluster . we take @xmath165 to be the potential energy at the @xmath166th pass , in the @xmath167th experiment . using eq . ( [ fs ] ) we can write @xmath168 in the non - ergodic regime , @xmath169 grows with @xmath11 , while in the ergodic regime , the signal @xmath169 approaches a constant . we begin by displaying results obtained for a calculation that has not attained ergodicity over the time scale of the simulation . we examine the 13-particle lennard - jones cluster with the met algorithm setting @xmath170 at a temperature of @xmath171 . the temperature is chosen to be that typically used as the initial high temperature in jw and pt studies of lj@xmath172 . by choosing a large constraining radius , the evaporation events are so frequent at the chosen temperature that attaining ergodicity proves to be quite difficult . we demonstrate the effect of reducing the constraining radius shortly . = .48 the upper panel is the signal @xmath169 ( in units of @xmath37 ) vs. @xmath11 for @xmath170 at @xmath171 . from @xmath173 independent experiments , of lj@xmath172 . the length of the simulation is @xmath174 mc passes . the lower panel shows the time evolution " of @xmath175 ( in units of @xmath20 ) for 15 independent experiments . at least three basins with different energies are present . clearly , the simulation at this scale of time , is not ergodic . the number of replicas used in the calculation is @xmath173 , and @xmath176 . the upper panel of fig . 2 shows the signal @xmath169 [ evaluated using eq . ( [ eq : ft ] ) ] , which grows along the entire simulation . this is the behavior expected in the non - ergodic regime . in the lower panel we can see the time evolution " of the temporal mean values of 15 experiments . = .48 upper panel : @xmath139 ( in units of @xmath37 ) as a function of @xmath177 for @xmath170 , @xmath178 , @xmath179 , and @xmath180 . for the two larger radii the full line is the best fit to the data points , according to eq . ( [ upa ] ) with @xmath145 defined in eq . ( [ upa2 ] ) . the lower panel shows the linear behavior of @xmath181 vs. @xmath177 , for @xmath170 and @xmath178 . @xmath11 has been set to @xmath174 . there are three sets of curves , each of which is indicative of sampling of at least three different energy basins . at low values of @xmath11 the curves in the lower panel differ significantly . at @xmath182 the high energy basin curves begin to decrease in energy . for a value of @xmath11 larger than the data displayed in fig . 2 , the curves can be expected to coalesce with the low energy basin curves . it is clear that for @xmath183 , the simulation is not ergodic . in pt and jw studies it is essential that the initial high temperature walk be ergodic . ergodicity can be attained for lj@xmath172 by reducing the radius of the constraining potential so that evaporation events are rare . we now present a study of @xmath134 as a function of @xmath11 for several values of @xmath30 . to determine @xmath134 , we have calculated the fourier transform function @xmath153 using eq . ( [ eq : ft ] ) at a series of frequencies @xmath184 where @xmath185 has ranged from 1 to @xmath186 . this range of frequencies ensures the linear approximation used in eq . ( [ fs ] ) is valid while including sufficient numbers of points for accuracy.@xcite using eq . ( [ eq:10 ] ) , we have calculated the slope of the imaginary part of @xmath187 as a function of @xmath188 , for these frequencies . the data points appearing in fig . 3 are the mean value over twenty independent calculations of the slope of @xmath187 . = .48 @xmath139 ( in units of @xmath37 ) and its error vs. @xmath177 for @xmath189 and @xmath180 , with @xmath176 . when @xmath139 is on the order of its own error , the simulation can be considered ergodic . for @xmath190 the simulation becomes ergodic at @xmath191 ( @xmath192 ) . for @xmath189 a longer simulation is needed to reach ergodicity . starting from random configurations , we have performed @xmath193 met passes at @xmath171 . after this warmup process , we have created sequences of sizes @xmath194 , @xmath195 , @xmath196 , @xmath197 , @xmath198 . the results are presented in fig . 3 for @xmath170 , @xmath178 , @xmath179 , and @xmath180 . the upper panel shows @xmath139 as a function of @xmath177 , for fixed @xmath176 . we have chosen to present the data using base 2 logarithms for clarity ( each increase by 1 unit of @xmath177 represents a factor of 2 scale increase ) . all the data decrease with increasing @xmath18 , but only @xmath199 and @xmath200 appear to vanish to within the error bars over the time scale of the current simulation . in the lower panel we present @xmath181 as a function of @xmath177 for @xmath201 and @xmath178 . the decay law suggested in eq . ( [ upa ] ) with @xmath145 given by eq . ( [ upa2 ] ) is satisfied for both radii . = .48 the upper panel shows the decay behavior of @xmath139 ( in units of @xmath37 ) as a function of @xmath177 for pt and met , at the temperature of the melting peak of the heat capacity , @xmath202 . from eq . ( [ upa1 ] ) , we plot @xmath203 vs. @xmath204 , to extract the value of the exponent @xmath205 ( the slope of the linear fit ) . we have found @xmath206 for pt , and @xmath207 for met . the straight lines are the best linear fits of the data points . we have stated that the simulation can be considered effectively ergodic when @xmath134 is indistinguishable from zero . in fig . 4 we have plotted @xmath139 and its statistical error as a function of @xmath177 for @xmath208 and @xmath180 . for @xmath209 the crossing point of @xmath139 and its error is at @xmath210 . for @xmath189 the crossing point is at a @xmath211 . we can conclude that for @xmath171 and @xmath209 the simulation can be considered effectively ergodic after @xmath212 met passes . once a constraining radius is chosen , pt and jw simulations require the highest temperature @xmath163 be chosen so that met is ergodic . for a given @xmath30 , the extent of ergodicity can be tested using the same metric that has been used for determining the optimum value of @xmath30 , but by varying the temperature . for the parameters @xmath171 and @xmath209 the simulation is ergodic even at very short sequence lengths . we have found that for @xmath213 the simulations are not ergodic . to be sure that the parameters are appropriate , we have performed a short pt simulation ( @xmath174 passes , ten pt passes consists of nine met passes plus an exchange attempt ) with 40 equally spaced temperatures in the range @xmath214[0.028,0.393 ] in order to obtain a first estimate of the position of the melting and boiling temperature regions . the boiling peak in the specific heat appears to be located at a higher temperature than @xmath214 0.393 . moreover , the value of @xmath215 at @xmath31 is about one - half the value of @xmath215 at the temperature of the melting peak @xmath202 . from these results we feel it is safe to choose @xmath209 and @xmath171 for the calculations that follow . we now illustrate the convergence characteristics of @xmath134 when we increase the total time scale of the calculation by a factor @xmath18 . we illustrate this behavior using a pt simulation of lj@xmath172 , and we focus on results at the temperature of the melting peak in the heat capacity ( @xmath202 ) . we choose this temperature , because from experience @xcite we know the statistical fluctuations are large at the melting heat capacity maximum . the large statistical fluctuations make it possible to emphasize the behavior of @xmath134 . we have run the pt simulation at 40 equally spaced temperatures in the range @xmath214 [ 0.028,0.393 ] . the initial warmup time has been set to @xmath174 met passes , followed by @xmath195 pt passes . following the warm - up period , we perform simulations of @xmath216 , @xmath217 , @xmath218 , @xmath219 , @xmath220 , and @xmath221 pt passes . in each case the initial configuration has been taken to be the last configuration of the previous run . the output of the simulation are sequences of the potential energy . @xmath134 has been determined in the same way as in the calculation of the high temperature parameters ( presented in fig . 3 and fig . 4 ) . the data points appearing in the upper panel of fig . 5 are the mean value over twenty independent calculations of the slope of @xmath187 . in the lower panel of fig . 5 we have plotted @xmath203 as a function of @xmath204 , where @xmath176 and @xmath222 . the slope of the linear fit is the exponent @xmath205 , according to eq . ( [ upa1 ] ) . at the temperature of the melting peak , @xmath206 . it is of interest to perform a similar study of the behavior of @xmath134 as a function of the time scaling for an met calculation . we have taken the final configuration of the pt simulation at @xmath202 as an initial configuration , and we have performed a simple met simulation at that melting temperature . a graph of @xmath139 and @xmath203 as a function of @xmath204 for met is also presented in fig . 5 . from the upper panel of fig . 5 , it is evident that met results are not ergodic within the same scaled time as the pt results . it is also evident that the power law exponent for both met and pt are not distinguishable . similar studies of the power law using the jw method also give the same exponent . neither an increase in the number of temperatures nor changing the distribution of temperatures in both jw and pt simulations has any effect on the calculated exponent . = .48 comparison of the met and jw diffusion coefficients with the pt diffusion coefficient as a function of the reduced temperature . the dashed line represents equivalence between methods . by using the results to compare the relative efficiencies of met , jw and pt simulations for the lj@xmath172 system we have found that pt and jw simulations can be considered ergodic if the run length is on the order of @xmath217 passes , while met simulations that are initialized from configurations generated from an ergodic pt study are ergodic when the total run length consists of @xmath223 passes or more . in order to compare approaches , we have calculated @xmath224 as a function of the reduced temperature , for the three methods . the comparison of diffusion coefficients from different algorithms has also been used by andricioaei and straub @xcite . the comparison of jw and met with pt is presented in fig . the jw and pt simulations are found to have comparable efficiencies using @xmath224 as a measure for all calculated temperatures . at intermediate temperatures , met is significantly less efficient . we have chosen to truncate the jw study at @xmath225 . for temperatures below @xmath225 , jw simulations require significant effort , because a large set of external distributions must be generated . because at temperatures below @xmath225 lj@xmath172 is dominated by structures close to the lowest energy icosahedral isomer , we expect the jw and pt methods to have similar efficiencies ( as measured by @xmath224 ) for all temperatures . in this paper we have presented a study of the approach to the ergodic limit in mc simulations . in all the cases examined , the behavior of the mc metric @xmath36 can be approximated by eq . ( [ d ] ) , and the behavior of @xmath139 satisfies eq . ( [ upa ] ) . because the exponent @xmath205 is smaller than one for all the cases studied , the dependence of the non - diffusive contributions on @xmath36 is weaker ( in the sense of appendix a ) than the diffusive contributions . the assumption on which we have built the stochastic model have been verified numerically for a system having a sufficiently complex potential surface to be viewed as prototypical of a large set of many - particle systems . the mc metric used in this work appears to be a valuable tool to study the ergodicity properties of mc simulations . the non - ergodic components of the mc metric enable the prediction of the minimum length a mc simulation must have in order to be considered ergodic . the comparison of @xmath224 from different algorithms gives a reasonable estimate of their relative efficiencies . from the study of the melting region of 13 particle clusters , we have found that the exponent @xmath205 depends both on the method used and the nature of the potential energy function . we have performed calculations , not discussed in this work , where the functional form of the potential energy is modified . these studies have shown @xmath205 to be dependent on the details of the potential . we have not found the exponent @xmath205 to be a strong function of method . although pt and met have significantly different efficiencies as measured by their relative diffusion coefficients , @xmath205 is nearly the same in the two methods . the difference in the decay of @xmath134 appears to be dominated by the coefficient in eqs.([upa ] ) and ( [ upa1 ] ) rather than the exponent . as discussed in the text , parallel tempering and j - walking studies of many - particle systems must have an initial high temperature component that is chosen so that a met simulation is known to be ergodic . for cluster simulations that require an external constraining potential to define the cluster , the radius of the constraining potential must be carefully chosen in order to achieve ergodic results . we have found the metric and associated decay laws developed in this work to be a particularly valuable method of choosing these initial parameters in both parallel tempering and j - walking simulations . we also remark that the metric introduced here may be a more sensitive probe of ergodicity than may be required in some applications . for example in previous j - walking studies@xcite of the 13-particle lennard - jones cluster , the heat capacity curve determined with a constraining radius of 4@xmath21 is nearly indistinguishable from the curve obtained with a constraining radius of 2@xmath21 . from the results of this work , we know the initial high temperature walk is not ergodic when a constraining radius of 4@xmath21 is used . it is striking that the non - ergodicity as measured by the energy metric is not apparent in the heat capacity curve . we have constructed a metric based on an ensemble of mc trajectories . by using an ensemble we attempt to cover sufficient portions of space so that all components are accessible . in practice only a finite subset of a full ensemble can be included , and it is always possible that components of space are missed . in such a case @xmath134 may decay to zero numerically within the subspace , and the behavior may give misleading evidence that the simulation is ergodic . because components of space may be missed in any finite simulation , it is impossible to guarantee ergodicity . it is hoped by using a sufficiently large ensemble of trajectories to define the metric , the possibility of missing components is minimized . we would like to thank dr . o. osenda for helpful comments . this work has been supported in part by the national science foundation under grant numbers che-9714970 and cda-9724347 . this research has been supported in part by the phillips laboratory , air force material command , usaf , through the use of the mhpcc under cooperative agreement number f29601 - 93 - 0001 . the views and conclusions contained in this document are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements , either expressed or implied , of phillips laboratory or the u.s . government . we have considered two overall time scales for a mc simulation . properties calculated at short times ( labeled @xmath10 in the discrete case ) provide information about each step of the mc process , and properties averaged over the total simulation time ( labeled @xmath11 in the discrete case ) give information about the approach to ergodic behavior . when @xmath11 is sufficiently short we have both diffusive and non - diffusive contributions as a function of @xmath10 . in this appendix we explain the relative time dependence of the diffusive and non - diffusive contributions to the autocorrelation function . it has been assumed that the autocorrelation function eq . ( [ correl ] ) can be expressed as the sum of diffusive terms plus non - diffusive terms , i.e. @xmath226 where @xmath227 increasing the time variables by a factor @xmath17 , such that @xmath228 , with @xmath229 , we can study the relative variations of each contribution to the correlation function , diffusive and non - diffusive ( labeled by @xmath230 ) . in this appendix we only consider values of @xmath18 such that the transformation @xmath231 does not increase the time scale beyond the local correlation time . in appendix b values of @xmath18 are considered that do cross such time scales . by relative variations we mean @xmath232 the relative variation of each non - diffusive contribution is @xmath233 whereas the relative variation of the diffusive contribution is @xmath234 if @xmath235 for all pair of times @xmath61 and @xmath62 and for all @xmath17 such that @xmath236 , we say that the non - diffusive contributions are weaker than the diffusive contribution in their dependence on @xmath61 . we explore , in the remainder of this appendix , the properties @xmath59 must have in order that the inequality @xmath235 is satisfied . if the function @xmath237 @xmath238 satisfies the inequality @xmath239 then , @xmath237 is an increasing function of @xmath12 . for @xmath84 and @xmath61 fixed , the function @xmath240 evaluated in @xmath241 is @xmath242 then , for @xmath243 @xmath244 2 where @xmath245 $ ] . in the limit @xmath246 , and by virtue of the continuity of @xmath59 , the derivative takes the form @xmath247 then , @xmath248 , and @xmath237 is an increasing function of @xmath249 here we have presented the two first conditions @xmath59 must have , namely eqs . ( [ lemma1 ] ) and ( [ lemma2 ] ) . from eq . ( [ schwartz ] ) @xmath250 is a global maximum , and the memory functions must have a positive peak at zero . the area below that peak must be sufficiently large to satisfy eq . ( [ lemma1 ] ) . moreover , @xmath250 must be sufficiently large to satisfy eq . ( [ lemma2 ] ) , even at points where @xmath251 is a local maximum . then , to satisfy this lemma , we need a memory function with a sufficiently large global maximum at @xmath252 . suppose @xmath253 . if @xmath17 , then @xmath254 for all pair of times @xmath61 and @xmath62 . * demonstration : * under the change of scale in time @xmath231 , @xmath255 can be written @xmath256 then , the quotient @xmath257 is @xmath258 2 by eq . ( [ lemma1 ] ) , @xmath259 @xmath61 and @xmath12 . by the * lemma * the numerator is smaller than the denominator . then @xmath260 and then , @xmath261 suppose that @xmath17 is such that @xmath262 , @xmath253 , and all @xmath59 satisfy the lipschitz condition @xcite ( for all closed interval @xmath263 exists a real positive number @xmath264 such that @xmath265 for all @xmath56 and @xmath266 in @xmath263 ) . then @xmath267 if and only if @xmath59 is non - negative in the interval @xmath268 . if @xmath267 , then @xmath269 where the operations to reach eq . ( [ ineq ] ) are valid by using * corollary*. then @xmath270 + \int^{t_>}_{t_1}\!\ ! dt_2 \left[b\,f_{\ell}\left(\frac{b\,(t_2-t_1 ) } { \tau_{\ell}}\right)- \,f_{\ell}\left(\frac{t_2-t_1}{\tau_{\ell}}\right)\right ] \right\}\\ 0 & < & \int_0^{t_<}\!\ ! dt_1\left\ { \int_0^{t_1}\!\ ! dt \left[b\,f_{\ell}\left(\frac{bt}{\tau_{\ell}}\right)- \,f_{\ell}\left(\frac{t}{\tau_{\ell}}\right)\right ] + \int_0^{t_>-t_1}\!\ ! dt \left[b\,f_{\ell}\left(\frac{bt}{\tau_{\ell}}\right)- \,f_{\ell}\left(\frac{t}{\tau_{\ell}}\right)\right ] \right\}\\ 0 & < & \int_0^{t_<}\!\ ! dt_1\left\ { \int_0^{bt_1}\!\ ! dt \,\,f_{\ell}\left(\frac{t}{\tau_{\ell}}\right)- \int_0^{t_1}\!\ ! dt \,\,f_{\ell}\left(\frac{t}{\tau_{\ell}}\right ) + \int_0^{b(t_>-t_1)}\!\ ! dt \,\,f_{\ell}\left(\frac{t}{\tau_{\ell}}\right)- \int_0^{t_>-t_1}\!\ ! dt \,\,f_{\ell}\left(\frac{t}{\tau_{\ell}}\right ) \right\}\\ 0 & < & \int_0^{t_<}\!\ ! dt_1\left\ { \int_{t_1}^{bt_1}\!\ ! dt \,\,f_{\ell}\left(\frac{t}{\tau_{\ell}}\right ) + \int_{t_>-t_1}^{b(t_>-t_1)}\!\ ! dt \,\,f_{\ell}\left(\frac{t}{\tau_{\ell } } \right)\right\}\,\,.\label{k1}\end{aligned}\ ] ] using the intermediate value theorem , @xcite we have @xmath271\label{propx}\,\,,\end{aligned}\ ] ] where @xmath272 $ ] . let be @xmath273 and @xmath274 the values at which the intermediate value theorem is satisfied , in the intervals @xmath275 $ ] and @xmath276 $ ] respectively @xmath277 then , the remainder can be written as @xmath278+(t_>-t)\,\left[f_{\ell } \left(\frac{t^*_{\beta}(t)}{\tau_{\ell}}\right ) - f_{\ell}\left(\frac{t_>-t}{\tau_{\ell}}\right)\right]\right\}\,\,.\ ] ] by the lipschitz condition , we have that @xmath279\\ & < & \frac{c_{\ell}}{\tau_{\ell}}\,(b-1)\,\left(\frac23\,t_<^3+t_<t_>\ , ( t_>-t_<)\right)\\ & < & \frac23\,t_>^3\,\frac{c_{\ell}}{\tau_{\ell}}\,(b-1)\label{cota}\,\,,\end{aligned}\ ] ] where @xmath264 is a suitable positive real constant . using eqs . ( [ propx ] ) and ( [ rem ] ) in eq . ( [ k1 ] ) we have @xmath280 where @xmath281 is a continuous and differentiable function of @xmath61 . the inequality ( [ xx ] ) holds for any @xmath17 . suppose that @xmath282 . then , if @xmath18 is such that @xmath283 where @xmath284 , we have that @xmath285\end{aligned}\ ] ] in contradiction with the hypothesis that @xmath286 is negative . then @xmath287 let us define the function @xmath288 where @xmath289 . the right derivative at @xmath252 of @xmath290 is @xmath291 2 where @xmath292 $ ] and @xmath293 $ ] . thus @xmath294 if the right derivative at 0 of @xmath290 is negative , @xmath290 approaches @xmath295 from below , when @xmath296 . there exists a time @xmath297 , such that @xmath298 , in contradiction with eq . ( [ xxx ] ) . then , @xmath59 must be non - negative for @xmath289 . by the property eq . ( [ schwartz ] ) @xmath250 must be positive . this proves that @xmath299 for @xmath300 . to demonstrate that if @xmath59 is positive yields @xmath301 ( i.e. the converse ) , follow the argument backwards , from eq . ( [ k1 ] ) . @xmath302 in conclusion , if the memory functions are positive , satisfy the lipschitz condition , and satisfy the condition eqs . ( [ lemma1 ] ) and ( [ lemma2 ] ) , the non - diffusive contributions are more weakly dependent on time than @xmath65 . the results of the present appendix are valid in the limit of a complete ensemble . in our numerical experiments only partial samples of the ensemble can be considered . the memory functions that appear in our numerical calculations come from partial mean values of the product of discontinuous functions ( every noise process is a discontinuous function ) . these memory functions are discontinuous . the behavior of the non - diffusive contributions observed in our numerical experiments is in agreement with these analytic ( infinite ensemble limit ) results . we can infer that there might be a version of the theorem applied to discontinuous memory functions , but we have been unable to develop such a theorem . in this appendix we show the behavior of the function @xmath303 when its correlation time is changed according to @xmath304 , with @xmath305 ; i.e. when the total simulation time is scaled to exceed the correlation time of the first colored noise process . where @xmath309 and @xmath310 . we want to compute this contribution both within the neighborhood @xmath311 as well as outside such a region . to do so , we can split the integral in eq . ( [ eq:6 ] ) in three parts where @xmath322 ( @xmath89 ) is the time in the interval @xmath323 $ ] at which the function @xmath303 reaches its maximum ( minimum ) value . because @xmath303 is continuous , there exists @xmath324 $ ] at which we can observe that the correlation time @xmath351 goes to zero when @xmath18 is increased . the function @xmath352 becomes negligible outside a neighborhood of @xmath252 [ observe eqs . ( [ i11 ] ) and ( [ i33 ] ) ] . equation ( [ norm ] ) holds , then , if @xmath18 is sufficiently large , @xmath353 can be considered a delta function . the integrals @xmath316 and @xmath326 become zero , and the integrals involving @xmath252 in the expression of @xmath315 converge to one . @xmath315 becomes the same argument can be used when @xmath18 is such that @xmath356 . after such changes in the time scale , the diffusion coefficient @xmath357 is enlarged , and the non - diffusive contributions are reduced . there is an ultimate scale change , such that @xmath358 . beyond this maximum time scale the process can be considered diffusive . j. p. valleau and s. g. whittington , a guide to monte carlo for statistical mechanics : 1 highways , in _ statistical mechanics , part a : equilibrium techniques _ , modern theoretical chemistry series , vol . 5 , chap . 4 , b. berne ed . ( plenum , new york , 1976 ) .

many years of effort have been spent on trying to understand the structure and evolution of the super star cluster ngc2070 , at the centre of the 30doradus nebula in the large magellanic cloud ( otherwise known as the tarantula nebula ) since its discovery as such by lacaille in 1751 . in this case , the prize is clearly worth the effort as this object may hold the key to answers to many fundamental questions regarding the physics of the formation of stars , stellar clusters and galaxies and their role in the cosmos ( e.g. brandl 2005 ) . thanks also to this effort , we now know much about the brightest stars more massive than @xmath5m@xmath6that can be more easily detected and resolved . in spite of the rather large distances ( @xmath7kpc ) involved ( hunter et al . 1995a ) , these objects can be studied even in the very compact core of ngc2070 , referred to as r136 , which was first resolved unambiguously by the hubble space telescope ( hst ) using the faint object camera ( weigelt et al . 1991 ; de marchi et al . but we still know very little about the other part of this population consisting of the fainter , low mass main sequence ( ms ) and pre - ms ( pms ) stars and brown dwarfs . these might be present in even larger numbers if the tapered power law imf proposed recently by de marchi , paresce & portegies zwaart ( 2010 ) for young and very old clusters continues to the hydrogen burning limit and beyond . their number and spatial distribution are critical for understanding the low mass star formation process in a cluster of very massive stars and their role in determining the ultimate fate of the cluster itself . the current observational situation regarding low mass stars in ngc2070 is contradictory . although there is good evidence for a significant population of these stars , the precise nature of their spatial and mass distributions is not well understood . sirianni et al ( 2000 ) , using @xmath8- and @xmath9-band hst observations claim that the mass function ( mf ) flattens below @xmath10m@xmath6 , while zinnecker et al . ( 2002 ) and more recently andersen et al . ( 2010 ) , using hst observations in the h band , find no such feature in their data down to their completeness limit around 1m@xmath6 . so far , little can be said of what happens below this limit , mainly because these stars in ngc2070 and its compact core r136 are very difficult to detect and characterize accurately . first , because of their inherent faintness especially with respect to the high background due to the numerous much brighter stars in the field . second , because they scatter widely across the colour magnitude diagram ( cmd ) making accurate subtraction of field stars and older cluster pms stars and subdwarfs near the ms unreliable without useful separation criteria . finally , the patchy extinction across the face of the nebula ( e.g. indebetouw et al . 2009 ) needs to be properly accounted for in order to accurately determine their true brightness . for these reasons , use of the wide field camera 3 ( wfc3 ) , the new panchromatic camera on hst , is , in principle , ideal for the task of establishing the main characteristics of the low mass population in ngc2070 . with its high spatial resolution over a wide field encompassing most of the central part of the cluster and its unprecedented sensitivity in the near uv and ir , the wfc3 allows us to tackle for the first time most of the problems summarised above with some hope of resolving them . in this paper , we present the first results of a deep survey of ngc2070 obtained in 2009 october as part of the early release science programme with the wfc3 camera . this project has as its ultimate objective that of determining as accurately as possible the properties of the low - mass initial mf in this cluster , which could be a prototype of the much larger and distant unresolved super star clusters in starburst galaxies . in order for this type of study to become possible , we must first carefully identify and distinguish from one another stars of different ages , since several generations of objects are known to populate the 30dor region ( e.g. walborn & blades 1997 ) . so far , studies of the star formation history of 30dor have been limited to the most massive stars , typically above 20m@xmath6 , and hence to the past 10myr or so , because these works rely on spectroscopy of individual objects that can not easily be obtained for less massive stars ( e.g. selman et al . 1999 ) . clearly , access to a wider mass range is crucial , so the goal of this paper is to understand how star formation has proceeded for much smaller objects , i.e. stars with masses as low as @xmath11m@xmath6 . therefore , in this work we concentrate on a systematic search for low - mass stars in the pms phase , which we can identify in a reliable way and which allow us to study how star formation has proceeded in this area over the past @xmath3myr . only then will it be possible to address the properties of the mf , which will be the subject of future work . the structure of this paper is as follows . we describe the observations in section2 , the reduction of the data in section3 and the photometric analysis in section4 . the correction for extinction and the resultant corrected cmds are presented in section5 . the hertzsprung russell ( hr ) diagram and its analysis with recent pms models are the topic of section6 , where we also discuss the preliminary physical implications of our results on the star formation history in these regions . a summary is provided in section7 . the observations discussed in this paper were obtained in 2009 october ( 20 27 ) with the wfc3 on board the hst . we refer the reader to dressel et al . ( 2010 ) for information about the wfc3 instrument and its performances . six broad - band and one narrow - band filters were used in these observations , as shown in table[tab1 ] , where the number of exposures and the total exposure times in each filter are also given . note that an extensive dithering pattern was employed , combining both long and short exposures , in order to improve the sampling of the telescope s point spread function ( psf ) , to minimise the effects of detector blemishes , to cover the gap between the two detectors in the field of view , and to recover the magnitudes of the brightest objects that would otherwise saturate in long exposures . lcr[bh ] f336w & 24 & @xmath12 + f438w & 16 & @xmath13 + f555w & 20 & @xmath14 + f656n & 8 & @xmath15 + f814w & 20 & @xmath16 + f110w & 9 & @xmath17 + f160w & 12 & @xmath18 [ tab1 ] for illustration purposes , figure[fig0 ] presents a four - colour composite image of the observed field . the colours result from assigning different hues to each monochromatic image , as indicated in the figure caption . the bright compact star cluster in the upper right - hand side of the frame is r136 , the core of the massive ngc2070 association . a cursory glance at figure[fig0 ] already shows that the centre of the cluster is dominated by bright blue stars , although a handful of bright red objects are seen to the east of it , suggesting the presence of an older generation of stars . this is consistent with the conclusions of selman et al . ( 1999 ) , who found at least three generations of massive stars , with the youngest and bluest members more centrally concentrated . we will return in more detail to the spatial distribution of these objects in section4 . throughout this paper we will concentrate on the observations in the optical wavelength range , and in particular on the f555w , f656n and f814w bands ( hereafter referred to as @xmath8 , @xmath19 and @xmath9 , respectively ) , whereas the entire wfc3 dataset including the observations at ir wavelengths will be covered in more detail in a forthcoming paper ( beccari et al . , in preparation ) . as regards the data reduction , the long and short exposures were treated in a different way . the deep datasets were combined using the iraf _ multidrizzle _ task , with the appropriate calibration files and geometric distortion coefficients for the uvis channel of the wfc3 . given the large number of deep frames available in the uvis filters and the careful dithering strategy adopted for these observations , we were able to derive one single `` drizzled '' image per band , free of cosmic rays and blemishes and with a rather sharp and well sampled psf , ideal for the detection of faint stars . producing a deep , high - quality combined image per filter is particularly useful in a field like 30dor where stars are embedded in extensive nebulosity associated with the gas that makes the photometry and the determination of the background more difficult on individual images . star detection and identification was done by running the automated iraf _ daofind _ routine separately on the @xmath8- and @xmath9-band drizzled frames , with the detection threshold set at @xmath20 above the local background level . in order for objects to be classified as real stars , we required that they be detected in both filters . the presence of numerous gas filaments and irregularities in the background can cause spurious detections , particularly in the h@xmath1 band , where considerable emission is present ( see figure[fig0 ] . for this reason , we visually inspected the detected stars and excluded from the bona - fide list all the detections not clearly associated with point - like sources . more specifically , we applied an unsharp - masking filter ( e.g. malin 1977 ) , with a smoothing radius of 15 pixel meant to subtract the low - frequency signal . unsharp masking makes structures and filaments in the nebular gas easier to see than in the direct image . in this way we compared the structures and filaments with the positions of the detected stars , in order to avoid spurious detections ( see beccari et al . 2010 for an application of this method to the case of the massive galactic cluster ngc3603 observed with the same camera ) . once the master list was defined in this way , we proceeded to measuring the magnitudes of each star with the iraf _ phot _ aperture photometry task . since the central regions are rather crowded , we opted for a photometric aperture of 3 pixel radius and a background annulus enclosing the area between 4 and 7 pixel radius around the stars . following de marchi et al . ( 1993 ) , we took as background value the mode of the counts within the annulus , in such a way to minimise the contamination due to neighbouring objects . this procedure was not applied directly on the combined drizzled frames , but rather on a subset of those . more precisely , having at our disposal 16 deep exposures both in the @xmath8 and @xmath9 bands , we used multidrizzle to build for each filter four drizzled images , each comprising four individual exposures . we then ran the iraf phot task on each of them , using as input coordinates the master list derived before . in this way , we could derive four independent measurements of the magnitude for each object in both the @xmath8 and @xmath9 band and the standard deviation of the four measurements immediately provides the photometric uncertainty . we followed the same procedure in all other visible bands , including h@xmath1 for which the four combined images comprise each two individual exposures . the approach followed for the short exposures ( ranging from a @xmath21s to 30s in duration ) is different , owing to the limited number of exposures available in each filter . in that case we performed the photometry on each individual flat - fielded image using psf fitting , having first built a proper psf model on each frame using isolated and well sampled stars , which was then fitted to each object in the images by means of the standard daophot / allstar routine . the reason why we did not use psf - fitting photometry also for the deep exposures is that image combination with multidrizzle does not necessarily preserve the uniformity of the psf . the two photometric catalogues obtained in this way for the long and short exposures were independently calibrated following the specific recipes developed by kalirai et al . ( 2009 ) for the uvis channel of the wfc3 . briefly , all magnitudes were first rescaled to an aperture of @xmath22 and then transformed into the vegamag system using the synthetic zero points given by kalirai et al . ( 2009 ) . the accuracy of the photometry in both catalogues is very good . the typical photometric uncertainty over the entire sample ranges from @xmath23mag at @xmath24 or @xmath25 to @xmath26mag at @xmath27 or @xmath28 ( note that in the following sections we will limit our study to objects with the smallest uncertainties , as indicated ) . finally , the two photometric catalogues were combined and merged so as to avoid duplications . we used stars in the magnitude range @xmath29 , where errors are small , to verify the integrity of the photometry by comparing the magnitudes derived for the same star in the long and short exposures . we found the match to be excellent , with any differences in magnitude being comparable to the photometric uncertainties in the short exposures ( @xmath30mag ) . the final combined catalogue contains 22291 stars with well defined fluxes in the @xmath8 and @xmath9 bands . of these , 18142 objects are also detected in h@xmath1 . the cmd of all objects with combined photometric uncertainty in the @xmath8 and @xmath9 bands not exceeding @xmath31mag is shown in figure[fig1 ] . the cmd shown in figure[fig1 ] is the deepest so far obtained for this cluster . at @xmath32 it reaches about 3mag deeper than the cmds previously obtained with the wfpc2 camera on board the _ hst _ in these bands ( sirianni et al . 2000 ; hunter et al . note that this is not only due to the improved sensitivity of the wfc3 camera , but also to the considerably shorter exposure times in the i band and the smaller field of view of the wfpc2 camera , which limited the observations to the most crowded regions of the cluster . the striking difference between the cmd of figure[fig1 ] and that of sirianni et al . ( 2000 ) is the prominent lower ms occupying the region fainter than @xmath33 and bluer than @xmath32 and revealing the presence in this field of a considerably older stellar population than the @xmath34myr old stars associated with r136 ( e.g. de marchi et al . 1993 ; hunter et al . 1995a ; massey & hunter 1998 ) . we will discuss the age distribution of the stars in this field in sections5 and 6 . the other interesting feature seen in figure[fig1 ] is the large number of objects in the range @xmath35 that appear brighter than the lower ms . their location is consistent with that of the pms population studied by sirianni et al . ( 2000 ; see also andersen et al . 2009 ) , but their number is much larger than those detected before . we discuss the properties of these stars in section5 , but we first need to address an important issue that affects the cmd , namely interstellar extinction . the width of the upper ms seen in figure[fig1 ] , at @xmath36 , is much larger than the photometric uncertainty at those magnitudes ( indicated by the horizontal bars on the left - hand side of the figure ) , revealing that there is a wide spread of extinction in this field . actually , there is general consensus that differential reddening is present in these regions . from the analysis of high - resolution narrow - band images , hunter et al . ( 1995b ) concluded that the nebulosity in this area is not uniform even on scales of order @xmath37 . however , in a subsequent study of the stellar populations in this region hunter et al . ( 1996 ) opted for a uniform reddening correction across the whole field , since they concluded that uncertainties ( e.g. red leak ) in the response of the wfpc2 filters did not justify a star - by - star correction . also sirianni et al . ( 2000 ) used the same @xmath38 value for all stars , although this may have lead to larger uncertainties in their mass determination of individual pms stars , as selman et al . ( 1999 ) and more recently andersen et al . ( 2009 ) have pointed out . in order to overcome this problem , in section4.1 we will use a robust method based on the determination of the individual reddening corrections for a large group of bright stars ( romaniello 1998 ) , which are then used to derive a correction also for fainter neighbouring objects . the young population in this field , responsible for the upper ms in figure[fig1 ] , has an age of @xmath34myr ( e.g. de marchi et al . 1993 ; hunter et al . 1995a ) or perhaps younger ( @xmath10myr ) if wolf rayet stars are still burning hydrogen ( massey & hunter 1998 ; de koter , heap & hubeny 1998 ; walborn et al . one can therefore determine the reddening towards each of those stars by measuring their colour and magnitude displacements with respect to a theoretical isochrone for the appropriate age , metallicity and distance . as for the metallicity , throughout this paper we will assume @xmath39 or about one third @xmath40 , since this is a typical value for the lmc ( e.g. hill , andrievsky & spite 1995 ; geha et al . 1998 ) . as regards the distance to 30dor , we will use @xmath41kpc since this is the geometrical distance to the neighbouring sn1987 object ( panagia et al . 1991 , later updated in panagia 1999).the resulting distance modulus of @xmath42 is fully consistent with that adopted by ( walborn & blades 1997 ) . we show in figure[fig2 ] as a solid line ( in blue in the online version ) an isochrone for an age of 3myr and a metallicity of @xmath39 , from marigo et al . ( 2008 ) , and the adopted distance modulus of @xmath42 . having assumed a younger age , i.e. 1 or 2myr , would have only affected the shape of the isochrone in the top @xmath43mag range , which we will not consider in this analysis . the isochrone is also already reddened by the amount corresponding to the intervening absorption along the line of sight due to the milky way , which fitzpatrick & savage ( 1984 ) quantified in @xmath44 , in turn corresponding to @xmath45 . the other curves on the left - hand side of the figure ( in blue in the online version ) represent the same isochrone reddened , for purpose of illustration , by an additional amount of extinction , with the long - dashed , short - dashed and dot - dashed lines corresponding respectively to @xmath46 , @xmath47 and @xmath48 . as for the extinction law , for the portion in addition to the @xmath44 due to the milky way we have used the law determined spectroscopically by scuderi et al . ( 1996 ) for star2 in the field of sn1987a , which we take as representative of the 30dor region as well . the arrow in the figure shows the corresponding reddening vector for @xmath49 . the long - dashed line , corresponding to @xmath46 , forms a natural left envelope to the upper ms at @xmath50 and leaves very few objects in the range @xmath51 at those magnitudes . this is expected because of the considerable amount of foreground extinction known to be present in the lmc towards 30dor : although the @xmath52 value that we find is somewhat larger than the @xmath53 reported by fitzpatrick ( 1985 ) , the latter value has been determined for the brightest stars at the very centre of 30dor that generally lie in evacuated cavities of hii gas of their own making and should therefore have a lower extinction . note that , at magnitudes fainter than @xmath33 , where the photometric uncertainty is still rather small ( see bars in figure[fig1 ] ) , there is a large number of objects between the @xmath46 and @xmath44 solid curves . in fact , the latter offers a much better fit to the lower ms , suggesting that these stars belong to an older foreground population . at least some of them could be members of the much older lmc field , which is known to have ages in excess of @xmath43gyr ( e.g. panagia et al . 2000 ; harris & zaritsky 2004 ) . interestingly , there are also a few objects that appear bluer than the @xmath44 curve . although they are in general consistent with our photometric uncertainties , some of them could actually be very young stars with a circumstellar disc seen at high inclination ( @xmath54 ) . objects of this type would appear somewhat bluer than their photospheric colour due to light scattering on the circumstellar disc , as well as several magnitudes fainter than their photospheric brightness due to extinction along the line of sight caused by an almost edge - on disc . however , according to the models of robitaille et al . ( 2006 ) for the spectral energy distribution of young stars seen at various viewing angles , objects of this type can only account for a few percent of the total young population , so the vast majority of stars near the @xmath44 solid curve must belong to the field in the lmc foreground . the presence of a foreground population of older objects becomes even more obvious when one looks at the other set of curves on the right - hand side of the figure ( shown in orange in the online version ) , corresponding to isochrones for an age of 2gyr and the same metallicity , distance , and reddening values used for the 3myr isochrones , also from the models of marigo et al . the 2gyr isochrones are shown up to a mass of @xmath55m@xmath6 , corresponding to the approximate location of the red giant clump ( rc ) , a characteristic feature of the cmd of low - mass , metal - rich stars experiencing their he - burning phase . it represents the counterpart at high metallicity of the horizontal branch seen in cmds of metal- poor globular clusters . the rc usually appears as a very tight , well defined feature in the cmd of stellar populations that could in principle be used as a distance indicator ( e.g. stanek & garnavich 1998 ; paczynski & stanek 1998 ; cole 1998 ; girardi et al . 1998 ; girardi & salaris 2001 ) . when there is patchy absorption , like in our case , and the amount of extinction varies from place to place , the rc becomes in practice an elongated ellipse running parallel to the direction of the reddening vector . this is precisely the nature of the apparent sequence of points running diagonally between @xmath56 and @xmath57 . this feature can also be used to derive the extinction law ( de marchi , panagia & romaniello 2007 ; panagia , de marchi & romaniello 2008 ) . the implications of this finding for the properties of the absorbing material in this and other lmc fields will be addressed in detail elsewhere ( de marchi & panagia 2011 , in preparation ) , but it is immediately clear from figure[fig2 ] that the young and old stars present in this field have rather different extinction ranges . while stars in the rc span the entire range @xmath58 , there are in practice no objects in the upper ms with @xmath59 . therefore , the reddening correction that we can derive using the objects in the upper ms will only be valid for those stars and , in a statistical sense , for low - mass stars of similar age if they have a similar spatial distribution , but it will not be appropriate for older objects , for which we will have to resort to an average correction . the procedure that we followed for reddening correction is explained in detail in the following subsection . we have used all the upper ms objects inside the box shown in figure[fig2 ] as reference stars for our determination of the reddening distribution across the field , with a total of about 800 stars selected in this way . assuming that they are all ms stars , we have measured their displacement from the solid isochrone along the reddening vector , whose direction is given by the arrow and the tilted sides of the box ( as mentioned above , we adopted the extinction law of scuderi et al . we have explicitly excluded the brightest objects where an age difference of 1 or 2myr could have an effect , albeit small , on the shape of the isochrone . the reddening distribution ranges from @xmath60 or @xmath61 to @xmath62 or @xmath63 , corresponding respectively to the 17% and 83% distribution limits , with a median value of @xmath64 or @xmath49 . the difference between the 17% and 83% limits indicates a @xmath65 spread of @xmath66mag in @xmath67 , indicating that reddening varies considerably in this field . the reddening values towards each of the selected ms stars were then used to derive a reddening correction for the objects in their vicinities . this procedure implicitly assumes that stars projected within a relatively small region of the field have similar extinction values , even though no information is available on the depth of their distribution nor on the uniformity of the absorbing material . in order to verify the validity of this assumption , we first applied the reddening correction to the upper ms stars themselves . each star was corrected using the average of the extinction values of the nearest upper ms stars , excluding the star itself ( no distance weighing was used , since we have no information on the depth of the distribution ) . after experimenting with a different number of neighbours ( namely 1 , 5 , 10 , and 20 ) , we concluded that using the 5 closest neighbours produces the tightest upper ms . this choice corresponds to using reference stars within a radius of less than @xmath68 , with a typical effective radius of @xmath69 . after reddening correction , the @xmath70 spread in the upper ms drops to @xmath71mag and the upper ms itself appears considerably tighter in the cmd than in the raw data ( see figure[fig6 ] ) . this result convinced us that the method , albeit statistical in nature , can be applied to other objects in the field , provided that they have an age and spatial distribution consistent with that of the upper ms stars . the cmd and isochrones in figure[fig2 ] help us to identify stars of similar age . this is the case of ms stars roughly brighter than @xmath72 and bluer than @xmath73 and of young pms objects roughly redder and fainter than the dot - dashed 2gyr isochrone ( these objects are shown as dark dots in the figure ) . however , the cmd alone does not allow us to assign an age to objects falling in the region between the solid and dot - dashed 2gyr isochrones ( shown as light dots ) , since they could be pms stars of various ages , low - mass ms stars , or red giants . therefore , in order to identify bona - fide young objects in this region we have to resort to a more reliable youth indicator , namely a strong h@xmath1 excess emission . this feature is characteristic of low - mass star formation and is attributed to the accretion process , whereby gravitational energy , released by infalling matter , ionises and excites the surrounding gas ( e.g. knigl 1991 ; shu et al . 1994 ) . a detailed account on how to reliably detect these objects using multi - colour photometry is given in de marchi et al . ( 2010 ) and de marchi et al . the method , already employed also by beccari et al . ( 2010 ) to search for objects with h@xmath1 excess in ngc3603 , will be used in section5 for an accurate selection of bona - fide pms stars after reddening correction . briefly , the method uses the median @xmath74 dereddened colour of stars with small ( @xmath75mag ) photometric uncertainties in the three bands ( @xmath76 and @xmath19 ) to define a reference template that is used to identify objects with excess h@xmath1 emission . however , as de marchi et al . ( 2010 ) have shown , even before reddening correction the colour colour diagram @xmath77 , @xmath78 provides a robust identification of stars with h@xmath1 excess , since in these bands the reddening vector runs almost parallel to the median photospheric colours of normal stars . therefore , following that procedure , we looked initially for stars whose @xmath77 colour exceeds by at least three times the photometric uncertainty the @xmath77 colour of normal stars ( i.e. those with no excess ) with the same @xmath78 colour . a more stringent selection will be applied after reddening correction ( see section5 ) . the pms candidates selected in this way , a total of 2374 , are shown as thick dots in figure[fig3 ] , where the dashed line represents the median photospheric @xmath77 colour as a function of @xmath78 . for comparison , we also show as a thin dotted line the corresponding colours derived from the model atmospheres of bessell , castelli & pletz ( 1998 ) for stars with effective temperatures in the range 3500 k @xmath79 40000k , surface gravity @xmath80 and a metallicity index @xmath81=-0.5 $ ] , as appropriate for the lmc ( dufour 1984 ) . the theoretical colours were calculated for the specific filters of the wfc3 camera and were reddened by @xmath82 , following the extinction law of scuderi et al . ( 1996 ) mentioned above for the purpose of comparison with the data . note that , due to the still preliminary calibration of the wfc3 photometric system in the h@xmath1 band , we had to apply a small correction to the @xmath77 colour of @xmath83 mag to force the @xmath77 colour of vega to be 0 at @xmath84 , as is appropriate for the vegamag photometric system . the objects with h@xmath1 excess emission are also shown as thick symbols in the cmd of figure[fig4 ] . besides the 3myr and 2gyr isochrones , we show there also the isochrones for pms stars of age 5myr ( upper thin line ) and 10myr ( lower thin line ) from the models of deglinnocenti et al . ( 2008 ) for the assumed lmc metallicity ( @xmath39 ) and having taken @xmath46 for the reddening . these isochrones allow us to distinguish in an approximate way between younger and older pms stars ( a more accurate age determination requires the reddening correction and will be possible only at the end of this process ) and to study their spatial distribution . using the 5myr pms isochrone as a reference , we have marked as thick dots and crosses respectively the objects above and below that line , and hence approximately younger and older than 5myr . in figure[fig5 ] we show the radial distribution of these low - mass stars with h@xmath1 excess , drawn from the nominal centre of r136 . the short - dashed line corresponds to stars younger than @xmath85myr , the long - dashed line to stars older than 10myr , while the distribution of upper ms stars is shown by the solid line , rescaled by a factor of @xmath86 to make the comparison easier . younger and older pms stars have markedly different distributions : older stars have a rather uniform distribution and are almost absent inside a radius of @xmath87 , while younger objects show a density gradient that follows remarkably well that of upper ms stars down to @xmath37 , where photometric incompleteness starts to dominate as indicated by the rapidly increasing photometric uncertainties in these regions . the choice of which approximate age ( 5myr or 10myr ) to adopt for separating `` younger '' and `` older '' stars is somewhat arbitrary , but the dotted line in figure[fig5 ] , corresponding to stars with age in the range @xmath88myr , suggests that objects with h@xmath1 excess older than @xmath89myr have a radial distribution remarkably different from that of upper ms stars . this is perhaps not surprising , considering that 5myr is about twice the age of the massive members of r136 ( @xmath90myr ; de marchi et al . 1993 ; hunter et al . 1995a ; massey & hunter 1998 ; de koter et al . 1998 ; walborn et al . 1999 ) . we will discuss these differences in the radial distribution in more detail in section6 . for now , figure[fig5 ] confirms that it is appropriate to use the reddening values derived from upper ms stars also for similarly young low - mass objects , but it suggests that this type of correction would not be suitable for older stars outside of the inner @xmath91 . as figure[fig5 ] implies , these objects are uniformly distributed over the field and are thus consistent with being in the foreground of r136 . in order to obtain at least an estimate of their reddening , we can use the stars with h@xmath1 excess that appear bluer than the isochrone for @xmath46 , i.e. the long dashed line in figures[fig2 ] and [ fig4 ] . as mentioned earlier , @xmath46 is the lower limit to the reddening towards the upper ms stars of r136 , so objects bluer than the long - dashed line in those figures are most likely in the foreground . a total of 144 stars satisfy this condition and , assuming that they are in the late stages of pms evolution , i.e. close to the ms , we derive for them a median reddening value of @xmath82 . if these objects were in a considerably earlier stage of their pms evolution , this procedure would give us an upper limit to their reddening . on the other hand , the fact that they are uniformly distributed over the field indicates that they can not be as young as the other much more concentrated objects with h@xmath1 excess ( see figure[fig5 ] ) . we therefore take @xmath82 as a representative value for the reddening towards all these stars and in general for all objects with h@xmath1 excess shown as crosses in figure[fig4 ] . we can now correct for extinction the magnitudes of all objects shown in figure[fig4 ] in the following way . all stars located within @xmath92 of the centre of r136 are corrected individually using the average extinction value of the nearest five upper ms stars . as regards objects outside @xmath92 , those with an age lower than @xmath85myr ( i.e. thick and thin dots in figure[fig4 ] ) are corrected in the same way , whereas older stars with excess ( crosses in figure[fig4 ] , red in the online version ) are corrected using the average @xmath82 value . no correction is attempted for other objects , since we can not determine their ages on the basis of the cmd alone , and they are not considered further in this work . note that when the magnitude of an object is corrected for extinction using its nearest neighbours , it is assumed that the placement of the object with respect to the absorbing material is similar to that of the reference stars in its vicinities . this process is accompanied by some statistical fluctuations , since as we have seen there is an intrinsic spread of @xmath71mag in the @xmath67 distribution of our upper ms reference stars . however , the uncertainties that this assumption can introduce in the final magnitudes of the stars are relatively small . the typical standard deviation of the @xmath67 values of the five nearest neighbours is @xmath93mag and , more generally , over the entire field these statistical uncertainties are not expected to exceed the reddening spread measured for the upper ms reference stars before reddening correction , i.e. @xmath66mag in @xmath67 or @xmath94mag in @xmath38 . the cmd corrected for extinction in this way is shown in figure[fig6 ] . it must be understood that it does not cover all objects that were observed , but only those for which we could determine a reliable correction for reddening . the symbol types are the same as used in figure[fig4 ] , the solid line represents the theoretical isochrone from marigo et al . ( 2008 ) for a ms age of 3myr , a metallicity of @xmath39 and a distance modulus of @xmath42 , whereas the dashed line corresponds to the isochrone of deglinnocenti et al . ( 2008 ) for a pms age of 5myr and the same metallicity and distance modulus . the pms isochrone of deglinnocenti et al . ( 2008 ) has been translated from the theoretical plane ( @xmath95 , @xmath96 ) to the plane of the observations using the model atmospheres of bessell et al . ( 1998 ) folded through the specific wfc3 filters employed here . the pms isochrone meets the ms at @xmath97 and it is reassuring that the two curves are in excellent agreement at brighter magnitudes , since they are based on independent models and different input physics . the cmd corrected for extinction in figure[fig6 ] reveals that at dereddened @xmath98 magnitudes fainter than @xmath99 there is a conspicuous number of objects that are above ( i.e. brighter than ) the ms and , as such , they are consistent with a population of pms stars . these objects extend down to @xmath100 , or more than three magnitudes deeper than the limit reached by sirianni et al . ( 2000 ) and andersen et al . ( 2009 ) , who detected and studied the brightest members of this population . as already mentioned in section4 , having corrected the photometry for extinction allows us to better clarify and confirm the pms nature of these stars by looking at the specific signatures of the mass accretion process that is supposed to accompany this evolutionary phase . the accretion process is believed to be the cause of the strong excess emission observed in these objects ( e.g. knigl 1991 ; shu et al . for example , the presence of a strong h@xmath1 emission line ( with equivalent width in excess of @xmath101 ) in young stellar objects is normally interpreted as the signature of ongoing accretion ( e.g. feigelson & montmerle 1999 ; white & basri 2003 ) . note that , although it is customary to use negative equivalent widths for emission lines , in this work we will use @xmath102 to refer to the equivalent width of the h@xmath1 emission line , defined as a positive quantity . in section4 we used this method for a preliminary coarse identification of stars with h@xmath1 excess emission , but here we employ very conservative criteria to refine the selection of bona - fide pms stars . in particular , we now limit the search only to stars with a combined photometric uncertainty in the @xmath76 and @xmath19 bands of less than @xmath103mag and fainter than @xmath104 ( corresponding to @xmath105l@xmath106 ) and we look for objects with dereddened @xmath107 colour excess at least four times larger than the photometric uncertainty in those bands ( equivalent to a @xmath108 detection limit ) . in section4 we used as a reference template the median photospheric colours of stars with small photometric uncertainty , the majority of which are older ms stars and have no h@xmath1 excess . this approach can not be followed here since many of the objects for which we could compute a reddening correction have h@xmath1 excess . as a reference template we use instead the theoretical photospheric colours derived from model atmospheres . as figure[fig3 ] shows , the models of bessell et al . ( 1998 ) for surface gravity @xmath109 and metallicity @xmath81 = -0.5 $ ] are in excellent agreement with the observed colours . this was also shown to be the case for stars in the field of sn1987a ( de marchi et al . 2010 ) and in ngc346 ( de marchi et al . 2011 ) . besides selecting stars with excess emission above the @xmath108 level , we also imposed constraints on the equivalent width of the h@xmath1 emission , @xmath102 . following the procedure outlined in de marchi et al . ( 2010 ) to derive @xmath102 from the observed colours , we further restricted the selection to objects with @xmath110 in order to avoid contamination by stars with significant chromospheric activity . furthermore , since at temperatures @xmath111k or colours @xmath112 the sample could be contaminated by be stars that are evolving off the ms , for those objects we set a more stringent condition , namely @xmath113 . in a survey of about 100 stars of type be in the galaxy ( cot & waters 1987 ) , only one star is found with @xmath113 and the largest majority have values in the range from @xmath114 to @xmath115 . a total of 1159 objects satisfy these conservative detection conditions : we will take them hereafter as bona - fide pms stars . obviously , less stringent limits would result in a larger number of pms stars , albeit with lower significance , but it must be understood that , even without changing the selection criteria , the actual number of pms stars in this field must be considerably larger . this is because our detection and classification relies on the presence of h@xmath1 excess emission at the time of the observations , but very young populations are known to show large variations in their h@xmath1 emission over hours or days ( e.g. fernandez et al . 1995 ; smith et al . 1999 ; alencar et al . 2001 ) , with only about one third of them at any given time being active h@xmath1 emitters with @xmath116 ( panagia et al . 2000 ; de marchi et al . 2010 ; de marchi et al . 2011 ) . therefore , the 1159 stars identified as bona - fide pms objects are those that were undergoing active mass accretion and therefore showed h@xmath1 excess emission at the time of these observations . on the other hand , since the duty cycles of any two stars are completely independent of one another , the sample of bona - fide pms stars identified in this way is representative of the entire pms population in this field . hereafter , unless otherwise indicated , when talking about pms stars we will refer exclusively to the 1159 bona - fide pms objects identified through their h@xmath1 excess emission . we show in figure[fig7 ] their locations in the cmd ( thick dots ) . the majority occupy the region above the ms where young pms objects are expected , but there are many located at or near the ms , suggesting an older age . a similar situation has already been noticed in other star forming regions , namely the sn1987a field in the lmc ( de marchi et al . 2010 ) , ngc3603 in the milky way ( beccari et al . 2010 ) , and ngc346 in the small magellanic cloud ( de marchi et al . 2011 ) . the skeptical reader might be concerned that the h@xmath1 excess emission of many of the objects near the ms could in reality be due to nebular contamination of the gas present in the field . however , as mentioned in section3 , we specifically excluded from our photometry all objects for which the photometric aperture would be contaminated by gas filaments . if the filament completely fills both the core aperture used to measure the flux of the star and the surrounding background annulus , it will not alter the photometry of the star , since it will be equivalent to a uniform background . obviously , if the background annulus does not properly sample the contamination , for instance because the filament only covers part of it , the emission present in the gas may be erroneously associated to the object ( for an illustration of this effect see beccari et al . precisely for this reason , as discussed in section3 , we have removed from the initial star lists all objects in the vicinity of a filament . therefore , we are confident that the h@xmath1 excess emission of our bona - fide pms stars is an intrinsic feature of those objects and we must interpret their distribution across the cmd as a sign of an age spread . more quantitative information on the age of these objects can be obtained from the comparison with pms isochrones , shown in figure[fig8 ] , where the effective temperatures ( @xmath117 ) and luminosities ( @xmath118 ) of these stars are arranged in the hr diagram . the value of @xmath117 was derived for each star directly from the dereddened @xmath78 colour , since this is an excellent index for temperature determinations in the range @xmath119k ( e.g. von braun et al . 1998 ; bessell et al . 1998 ) . for the conversion we made use of the models of bessell et al . ( 1998 ) already discussed in section4 . as for the bolometric luminosity , it was obtained from the dereddened @xmath8 magnitude and the same models , having assumed a distance modulus of @xmath42 . the solid lines in the figure corresponds to the pms evolutionary track from the pisa group ( deglinnocenti et al . 2008 ; tognelli et al . 2011 ) for metallicity @xmath39 . the tracks correspond to stellar masses from @xmath120m@xmath6to @xmath121m@xmath6and reveal the presence of bona - fide pms stars over the entire mass range . the isochrones ( dashed lines ) correspond , from right to left , to ages ranging from @xmath122myr through to 32myr with a constant age step of a factor of 2 , selected in such a way that our photometric uncertainties are slightly smaller than the typical separation between the isochrones . pms objects are seen across the entire age range . the comparison of the hr diagram with pms evolutionary tracks and isochrones allows us to quantify the age spread of the pms stars that was already evident in the cmd . individual stellar masses and ages were derived via interpolation over a finer grid than the one shown in the figure . we followed the interpolation procedure developed by romaniello ( 1998 ) , which does not make assumptions on the properties of the population , such as the functional form of the imf . on the basis of the measurement errors , this procedure provides the probability distribution for each individual star to have a given value of the mass and age ( the method is conceptually identical to the one presented recently by da rio et al . 2010 ) . an important aspect to consider is the uncertainty associated with the age determination of our pms stars from their location in the hr diagram . besides photometric errors in the @xmath8 and @xmath9-band data ( which are always less than @xmath31mag ) , the major sources of uncertainty are the effects of reddening and the comparison with stellar models . as regards the latter , de marchi et al . ( 2010 ) show that , excluding possible inaccuracies in the models input physics and interpolation errors , the largest source of uncertainty comes from the use of models that do not properly describe the stellar population under study , e.g. because of the wrong metallicity , and from differences between models of various authors . they conclude that systematic uncertainties of order 20% are to be expected for the mass and possibly higher for the age . as for reddening , dust extinction systematically displaces stars in the hr diagram towards lower luminosities and effective temperatures . therefore , if the photometry of a pms star is not corrected for reddening , its age will typically be underestimated . a study currently being conducted ( de marchi & panagia , in prep . ) shows that for reddening @xmath123 there is a rather linear relationship of the type @xmath124 , where @xmath125 and @xmath126 are the ages before and after reddening correction . for example , as discussed in section4 , the spread of the upper ms in figure[fig1 ] corresponds to a reddening spread of @xmath127mag in @xmath67 . therefore , if we had used the average reddening value of @xmath49 to correct the magnitudes of all young stars , instead of using the values of the nearest neighbours , we would have introduced an age uncertainty of a factor of @xmath55 . similarly , if reddening correction for older stars with excess ( i.e. those indicated by crosses in figure[fig4 ] ) had been done using the average @xmath38 value of the nearest upper ms neighbours instead of @xmath82 , the derived ages would be typically twice as old . as a further example , the double arrow in figure[fig8 ] indicates the amount and direction of the displacement in the hr diagram caused by the typical uncertainty of @xmath128mag in @xmath67 associated with our reddening correction ( see section4.1 ) . in the specific case shown , the corresponding uncertainty on the age is a factor of @xmath55 while the uncertainty on the mass is less than 15% . this analysis suggests that the typical uncertainty on the ages that we derive is less than a factor of 2 . for this reason , the age distribution of pms stars is shown in figure[fig9 ] using as bin size a constant logarithmic step of a factor of 2 , which does not suffer from the relative age uncertainties . the solid line gives the number of stars inside each age bin as a function of time , whereas the dashed line provides an apparent value of the star formation rate in units of stars per myr ( i.e. the number of stars in each bin divided by the size of the bin ) . the dashed line seems to suggest a rather uniform star formation rate over the past 4myr , of order @xmath129m@xmath6myr@xmath130 ( the mass of our pms stars younger than 4myr ranges from @xmath131m@xmath6to @xmath132m@xmath6 , respectively the 17 and 83 percentiles , with an average of 2m@xmath6 ) , and a progressive drop at earlier times . obviously , this is necessarily a lower limit to the star formation rate , since our measurements only account for pms stars in the range @xmath133m@xmath6that had h@xmath1 excess emission at the time of the observations . furthermore , since the fraction of pms stars in an active state of accretion is expected to decrease with age ( see de marchi , panagia & sabbi 2011 ) , the apparent drop in the star formation beyond 2myr can at least in part be explained . keeping in mind that we can only provide a lower limit to the star formation rate , near the peak of the distribution , at @xmath43myr , this limit corresponds to @xmath134m@xmath6myr@xmath130 ( the average mass of those objects is @xmath135m@xmath6 ) , while at @xmath136myr it drops by over an order of magnitude to @xmath101m@xmath6myr@xmath130 ( average mass @xmath137m@xmath6 ) . a more detailed study of the star formation rate of 30 dor will be the topic of a future paper , but the histograms in figure[fig9 ] already show quite convincingly that star formation has been present in this region for an extended period of time , exceeding 30myr . from a spectroscopic study of massive stars ( @xmath138m@xmath6 ) within a radius of @xmath139 of r136 , selman et al . ( 1999 ) concluded that recent star formation in this region shows at least three distinct bursts , respectively @xmath140 and @xmath55myr ago , with increasing strength . they also found that the younger generation is considerably more concentrated than the others and is responsible for most of the star - formation within 6pc or @xmath141 of the cluster centre . it is therefore useful to compare this picture with the spatial distribution of our pms stars . although we do not have the age resolution needed to establish unambiguously whether the formation has been continuous during the past 30myr or whether repeated bursts have occurred , we can look at the current age distribution for clues as to the relationships between stars of different ages . the age distribution of these pms stars is characterised by a median of @xmath142myr , with about @xmath143 of the stars being younger than @xmath2myr and an equal number older than @xmath4myr . hereafter we will refer to these two groups , comprising 415 objects each , as `` younger '' and `` older '' pms stars , respectively . note that , with a typical age uncertainty not exceeding a factor of 2 , these two groups are well separated in age , since the former has a median age of @xmath10myr and the latter @xmath144myr . interestingly , also their projected spatial density distributions are very different . they are shown in figure[fig10 ] by contour lines of different colour , with logarithmic scaling , plotted over an image of the field in the h@xmath1 band . the contour plots have been derived after smoothing the distribution with a gaussian beam with size @xmath145 or @xmath43pc , as indicated by the circle in the lower left corner of the figure . the lowest contour level corresponds to a local density of pms stars twice as high as the average pms stars density over the entire field . the step between levels corresponds to a factor of @xmath146 . there is a pronounced difference in the spatial distribution of younger and older pms stars . the former are concentrated near the centre of the r136 cluster , where they form a density plateau , while older pms stars occupy preferentially the region to the east of r136 and are remarkably absent from the cluster centre . since these stars have a rather uniform distribution across the field , we should expect to find them also near the location of the young r136 cluster , contrary to what we observe . furthermore , with a median age of @xmath144myr , these stars are considerably older than the @xmath10myr old members of r136 , so we should not expect any spatial relationship ( either a correlation or anti - correlation ) between the distributions of the two types of objects , unless something else related to the younger population prevents us from detecting older pms stars near the cluster centre . photometric incompleteness due to the young massive stars could seem an obvious culprit , since the younger pms stars are on average brighter than the older objects . however , our analysis shows that the difference in the distribution of younger and older pms stars remains the same as shown in figure[fig10 ] even when we restrict the choice of objects to the same magnitude range ( e.g. @xmath147 ) . therefore , the apparent paucity of older pms stars near the centre is not caused by selection effects in their detection . a more likely explanation is the erosion of circumstellar discs via photo - evaporation caused by the far - ultraviolet ( fuv ) radiation of the young massive members of r136 . if the discs of pms stars are disrupted , the accretion process dwindles and these objects would no longer be detectable through their h@xmath1 excess emission . to test this hypothesis , we have looked at how older pms stars of various h@xmath1 luminosity @xmath148 are distributed in the field . in figure[fig11 ] we show , as an example , all pms stars with ages in the range @xmath149myr ( 115 objects in total ) using different symbols according to their h@xmath1 luminosity . dots correspond to stars with @xmath150l@xmath151 , squares to @xmath152l@xmath151 , and triangles to intermediate values . these values have been selected in such a way to have an equal number of objects in each group . in the background we show as a reference a negative image of the field in the h@xmath1 band . the distribution of the three classes of objects is not random : stars with higher @xmath148 are typically closer to the centre of r136 and located in areas where the background has a higher emission , whereas stars with lower @xmath148 are further away and preferentially in regions where the background emission is low . although systematic differences in the mass of the three classes of objects could produce the observed distribution of @xmath148 , this can be excluded since the median mass of the three groups of stars is the same at @xmath153m@xmath6 . we can also exclude that this is an artefact of our photometry since the background is subtracted locally from each star ( see section3 ) and therefore the diffuse h@xmath1 nebular emission does not affect the intrinsic h@xmath1 brightness of individual objects . however , a fraction of the h@xmath1 luminosity of the stars closest to the centre of r136 could in principle be due to fluorescence of their circumstellar discs caused by the very strong ionising radiation of the massive stars . it is , therefore , important to quantify this effect . according to massey & hunter ( 1998 ) , adopting the temperature calibration of vacca , garmany & shull ( 1996 ) , the total bolometric luminosity of the @xmath3 brightest and most massive stars in r136 is of order @xmath154l@xmath106 and most of them are of spectral type o4 or earlier . based on this information and following panagia ( 1973 ) , we have determined at which distance from the centre of r136 the effects of the ionising radiation impinging face - on on a circumstellar disc of radius 100au would contribute significantly to the h@xmath1 luminosity of its star . the median @xmath148 value of the pms stars shown in figure[fig11 ] is @xmath155l@xmath106 , so we have considered three cases of contamination , at approximately the 10% , 50% and 100% level . the dotted , dot - dashed and solid circles shown in figure[fig11 ] correspond respectively to the distances at which the contribution to the h@xmath1 luminosity due to fluorescence is @xmath156 and @xmath157l@xmath106 ( note that these distances are necessarily upper limits since they are projected on the plane of the sky and we have no information on the distribution of these objects along the line of sight ) . it is immediately clear that the effects of fluorescence can only be important for a handful of stars , leaving the largest majority of old pms objects in this field unaffected . the origin of the apparent correlation between their positions and h@xmath1 luminosities has , therefore , to be found elsewhere . a natural explanation would be that stars in regions of low emission , where the density of the gas is lower , such as those to the east of r136 , are much less shielded from the radiation of massive stars than objects in denser regions , such as those to the north and west of r136 . therefore , their discs have been exposed to photo - evaporation for a longer time and their h@xmath1 luminosity is today further reduced . in support of this hypothesis , the photometry of all stars enclosed in the low - background region resembling a `` christmas tree '' and centered at ( @xmath158 ) , which is better seen in figure[fig0 ] , is consistent with a very low extinction value , close to the @xmath44 indicated by fitzpatrick & savage ( 1984 ) as the contribution of the milky way to the intervening absorption along the line of sight . furthermore , this region is characterised by very low emission at ir ( e.g. indebetouw et al . 2009 ) and radio wavelengths ( e.g. johansson et al . 1998 ; rubio et al . 1998 ) , confirming that it is devoid of both dust and gas . the region also corresponds with a high - velocity structure seen by chu & kennicutt ( 1994 ) in their detailed h@xmath1 echelle study of the kinematics of 30dor . these authors find a bright redshifted knot @xmath159 east of r136 , with a velocity offset of @xmath160 km s@xmath130 and very low emission . the knot is part of a fast expanding shell to the east of the cluster that correlates well with a region of diffuse x - ray emission seen in images obtained with the _ einstein _ satellite . these authors conclude that both the high - velocity features and the x - ray emission are related to high - velocity shocks from fast stellar winds and supernovae . this picture is supported by the later study of 30dor by townsley et al . ( 2006 ) with the _ chandra x - ray observatory_. their observations reveal rather strong diffuse emission in the area of the `` christmas tree '' ( their region 6 , coincident with the east fast shell of chu & kennicutt 1994 ) with an absorption - corrected x - ray surface brightness of @xmath161erg s@xmath130 pc@xmath162 in the @xmath163kev band . a spectral fit to this emission shows a soft thermal plasma with @xmath164kev ( townsley et al . therefore , in such an evacuated region , the discs of pms stars must be much more exposed to photo - evaporation . conversely , as discussed in section4 , extinction towards the young massive stars at the centre of r136 is both higher and highly variable , indicating a much denser environment where gas and dust can intercept a significant fraction of the uv radiation from the massive central stars , thereby reducing the efficiency of the photo - evaporation process . indeed , most of the pms stars with @xmath165l@xmath106 in figure[fig11 ] are in regions of significant @xmath166co@xmath167 emission , suggesting a high density of molecular gas , and the column density in hi in front of r136 is known to be very high ( @xmath168 cm@xmath162 ; de boer , koornneef & savage 1980 ) . thus , although this picture is still necessarily qualitative , it provides a plausible explanation of the observed distribution of h@xmath1 luminosities of older pms stars across the field via the effects of photo - evaporation . these effects have been discussed by de marchi et al . ( 2010 ) in the field of sn1987a . in that environment , located @xmath170 or 300 pc sw of 30dor , the pms stars around the hottest massive young stars are both less numerous and fainter in h@xmath1 emission , suggesting that their circumstellar discs have been eroded more efficiently by enhanced photo - evaporation . unlike the relatively simple case of the sn1987a field where little gas is present , the interstellar medium in 30 dor is known to be much denser and highly fragmented , with considerable density variations across the field ( e.g. poglitsch et al . 1995 ) . from the h@xmath1 echelle study mentioned above , chu & kennicutt ( 1994 ) conclude that the kinematics of 30dor is dominated by a large number of expanding structures , over scales of 2 20pc and with expansion velocities of 100 300 km s@xmath130 and kinetic energies of @xmath171erg . these structures are probably associated with supernova remnants embedded in supershells produced by stellar winds and supernovae explosions . these processes can release a large amount of kinetic energy into the surrounding environment , and as such they could in principle contribute to the dissipation of circumstellar discs , adding to the effects of photo - evaporation . to better understand this , it is useful to calculate the amount of kinetic energy possibly deposited on such a disc by a nearby supernova explosion . we will assume hereafter a total energy release of @xmath172foe @xmath173erg per supernova explosion , a circumstellar radius @xmath174au around a pms star located at a distance of @xmath175pc from the supernova . the maximum fraction of energy deposited on such a disc hit face - on by the blast can be written as @xmath176 and amounts to @xmath177erg for the specific parameters assumed here . the motion that this energy release can impart onto the material of the disc depends on the total mass and size of the disc and hence on its density , which in turn could depend e.g. on the radial location . in the simplified case of a homogeneous disc of mass @xmath178 that we express as a fraction @xmath179 of the total stellar mass @xmath180 , i.e. @xmath181 , the velocity @xmath182 imparted by the ejecta to the disc material can be expressed as : @xmath183^{-1 } \ , \bigg [ \frac{r}{100 \ , { \rm au } } \bigg ] \ , \left [ \phi\ , \frac{m}{{\rm m}_\odot } \right]^{-1/2 } \ , { \rm km~s}^{-1}.\ ] ] de marchi et al . ( 2011 ) have shown that stars in the magellanic clouds accrete a substantial fraction of their total mass during the pms phase , particularly at low masses , so for an order - of - magnitude calculation it is not unreasonable to assume @xmath184@xmath185 . we can see in this way that for a star with @xmath186m@xmath6at a distance of 10pc , the supernova ejecta hitting the disc face - on would impart a motion to its material with a velocity of @xmath43 km s@xmath130 . this value is about a factor of 5 smaller than the escape velocity at the edge of the disc for the typical pms stars that we detect in the `` christmas tree '' ( median mass @xmath187m@xmath6 ) . for a disc seen under a smaller solid angle , the effect would be proportionally smaller . on the other hand , less massive discs around less massive stars located closer to the supernova would obviously experience a much stronger effect . for instance for a star of mass @xmath188m@xmath6at a distance of 1pc from the supernova , the impulse received by the disc material would be sufficient to exceed the escape velocity at @xmath189au . this could significanly affect or possibly even stop the accretion process , eventually limiting the final mass of the star . on the other hand , with an average surface density of @xmath190pms stars per pc@xmath191 , only a very small fraction of the discs around objects in this field would be severely affected by any one supernova explosion . in summary , it is clear that the interplay between shocks and photo - evaporation effects due to the massive members of r136 can have important consequences on disc erosion , potentially affecting the formation of low - mass stars in these regions . however , without more specific kinematical information for individual objects , it remains difficult to quantify the expected effects on the circumstellar discs of pms stars using photometry alone . for a better understanding of this complex region and a more quantitative characterisation of the differences between the regions preferentially occupied by older pms stars with high and low @xmath148 , we may have to await high spectral and spatial - resolution spectroscopic observations to obtain a map of the amount and kinematics of the gas in the interstellar medium along the lines of sight towards these objects . finally , we briefly address the implications of our findings for the current understanding of whether and how different generations of stars in 30 dor are causally connected . walborn & blades ( 1987 ) and walborn ( 1991 ) reported three early o stars embedded in dense nebular knots to the north and west of r136 and suggested that these objects could belong to a younger generation , triggered by the energy released by r136 . using ground - based spectral classification of a large number of ob stars in the 30 dor nebula , walborn & blades ( 1997 ) studied the ages and spatial distribution of these objects , discovering an additional eight early o stars embedded in dense nebular knots . from a morphological analysis of these regions , they concluded that 30 dor has a two - stage starburst structure in which the energetic activity of the @xmath90myr old r136 has triggered the formation of a younger ( @xmath192myr ) generation of stars at a typical projected distance of @xmath193pc , inside the dense molecular clouds that surround r136 to the northeast and west . later ir observations with nicmos ( walborn et al . 1999 ; brandner et al . 2001 ) resolved many of these knots into many new ir sources , including multiple systems , clusters , and nebular structures , indicating active star formation . from mainly morphological considerations in this and other galactic star forming regions , walborn et al . ( 1999 ; see also walborn 2002 ) suggested that this mode of two - stage starburst , with generations separated by @xmath10myr , is ubiquitous . they concluded that in 30 dor the first generation of stars , still present , is now triggering the birth of the second , peripheral generation of very massive stars , which in a few million years will be the only one left , after the most massive stars in the core of r136 have disappeared . these authors predict 30 dor to become a giant shell h ii region like n11 and ngc 604 , with its most massive stars around the periphery . our discovery of a considerable spread in the ages of pms stars in this region , with the younger generation confined to the cluster s centre and older objects uniformly distributed across the field ( see figure[fig10 ] ) , raises some questions as to the validity of this picture at least as regards the formation of low - mass stars triggered by older ( massive ) objects in their vicinities . for instance , if we compare the projected distribution of pms stars with ages of less than 2myr with that of objects in the age - range @xmath194myr we see that they have the same centre , with the former slightly more concentrated , rather than distributed in the periphery of the latter as suggested in walborn s ( 2002 ) triggered massive star formation scenario . as already shown in figure[fig10 ] , the stars around the periphery of r136 are systematically older and not younger than the members of r136 itself . the same conclusion had been reached by selman et al . ( 1999 ) from a spectroscopic study of the massive stars ( @xmath195m@xmath6 ) in these regions . our findings suggest that if causal effects are present between different generations of stars , their interpretation can not rely only on the relative ages and projected distributions of the objects , but must consider proper kinematical and possibly dynamical information . while it is undeniable that we are seeing a case of sequential star formation , in the sense that there are separate generations of stars over time , no clear signs of causal effects are seen . therefore , although the concept of triggered star formation may well be valid , its relevance in the case of 30 dor has yet to be demonstrated . we have studied the properties of the stellar populations in the central regions of the 30dor cluster in the lmc using observations obtained with the wfc3 camera on board the hst in the @xmath8 , @xmath9 and @xmath19 bands . the main results of this work can be summarised as follows . 1 . our photometry delivers the deepest cmd so far obtained for this cluster , reaching down to @xmath196 or about 3 mag deeper than previous hst observations , and reveals for the first time a prominent lower ms fainter than @xmath33 . we attribute this feature to a considerably older stellar population than the @xmath34myr old stars associated with r136 . the cmd shows also a large number of objects in the range @xmath35 that appear brighter than the lower ms and that are consistent with , but extend much deeper than , the population of pms stars detected and studied by sirianni et al . ( 2000 ) and andersen et al . the observations also reveal the presence of considerable differential extinction across the field . we quantify the total extinction towards massive ms stars younger than @xmath34myr to be in the range @xmath197 ( respectively the 17% and 83% levels ) with a median value of @xmath49 . we show that these stars can be used to derive a statistical reddening correction also for stars in their vicinities , if they have a similar age and spatial distribution . however , this is not the case for older stars , which show a broader reddening distribution starting already at @xmath44 ( corresponding to the contribution of the milky way along the line of sight towards 30dor ) . a proper correction for extinction , therefore , requires at least an approximate knowledge of the age of the stars , which is not possible for all objects in the cmd . 3 . in order to distinguish relatively young objects from older stars , we looked for stars with h@xmath1 excess emission , since this feature identifies candidate pms stars , and we studied their distribution both across the field and in the cmd . younger pms stars follow the spatial distribution of massive stars and this allows us to correct them for reddening individually , using the extinction information derived for the massive objects . on the other hand , older pms objects are uniformly distributed over the field and appear to be systematically less reddened than younger pms stars . we derive for them an average extinction correction of @xmath82 . 4 . reddening correction allows us to proceed to a more accurate identification of bona - fide pm stars . we searched for stars with @xmath77 colour exceeding that of normal stars at the @xmath108 level or more and with @xmath198 ( or @xmath199 for stars with @xmath200k ) , in order to avoid contamination by stars with chromospheric activity or by the winds of be stars . a total of 1159 objects satisfy these stringent conditions and we take them as bona - fide pms stars . we derived the age of pms stars through the comparison with theoretical pms isochrones for metallicity @xmath39 , as appropriate for the lmc . about @xmath201 of these objects are younger than 4myr and an equal number are older than 12myr , indicating that star formation in this field has proceeded over a long time , although our age resolution does not allow us to discriminate between an extended episode and a series of short and frequent bursts . as regards the past 4myr , we obtain a lower limit to the star formation rate of @xmath129m@xmath6myr@xmath130 for objects in the range @xmath202m@xmath6 . we find a pronounced difference in the spatial distribution of younger ( @xmath203myr ) and older ( @xmath204myr ) pms stars . the former are concentrated near the centre of r136 , whereas older pms objects are mostly located to the east of it and are remarkably absent near the cluster centre . this effect , which is not due to photometric incompleteness , is in general consistent with the photo - evaporation of the circumstellar discs of older stars caused by the uv radiation of the massive members of r136 . however , not all older pms stars are affected in the same way , since some are in regions where the high density of the molecular gas can reduce the efficiency of the photo - evaporation process . our results show that several generations of stars are present in this field , since there are objects with ages ranging from less than 1 to over 30myr , although no clear signs of causal effects or of triggered star formation are seen . it appears , however , that the presence of the massive r136 cluster is sculpting the region and affecting low - mass neighbouring objects through the effects of photo - evaporation . we thank an anonymous referee , whose suggestions have helped us to improve the presentation of this work . this paper is based on early release science observations made by the wfc3 scientific oversight committee . we are grateful to the director of the space telescope science institute for awarding director s discretionary time for this programme . fp is grateful to the space science department of esa for their hospitality via the visitor programme . np acknowledges partial support by hst nasa grants go11547.06a and go11653.12a , and stsci ddrf grant d0001.82435 . zinnecker , h. , andersen , m. , brandl , b. , brandner , w. , hunter , d. , larson , r. , mccaughrean , m. , meylan , g. , moneti , a. 2002 , in `` extragalactic star clusters , '' iau symp . 207 , eds . d. geisler , e. grebel , d. minniti ( san francisco : asp ) , 531

vertex operators are one of the central objects in string theory . they represent cohomology classes of the brst operator . the brst cohomology depends on the chosen background , and in fact describes the tangent space to the moduli space of backgrounds at the chosen point . in particular , let us look at the pure spinor superstring theory in expansion around flat space . the structure of massless brst cohomology in flat space is more or less clear , but it appears that it has never been explicitly spelled out in the literature . the present paper is aimed at filling this gap . for the closed _ bosonic _ string the cohomology was computed in @xcite . we will here do a similar computation for the pure spinor superstring , but with the following difference . it is well known that the physically relevant cohomology problem is the so - called semirelative cohomology @xcite , which is @xmath1 acting on the vertex operators @xmath2 satisfying the following condition : @xmath3 this condition was built - in into the computations of @xcite . in the pure spinor superstring , the construction of the @xmath4-ghost is very subtle . in our paper we will compute the `` naive '' cohomology of @xmath1 , without taking into account ( [ semirelativecondition ] ) . failure to take into account ( [ semirelativecondition ] ) leads to some strange results : 1 . nonphysical vertex operators , _ i.e. _ elements of the brst cohomology which do not correspond to any linearized sugra solutions 2 . absence of the dilaton zero mode 3 . nontrivial cohomology at the ghost number three problems 1 and 2 are removed if we require the existence of the dilaton superfield @xmath5 ( see @xcite and the discussion in section [ sec : commentonnonphysicalstates ] ) . to defeat the ghost number three cohomology is more difficult . it is dangerous as a potential obstacle for continuing an infinitesimal solution to a finite solution ( _ i.e. _ obstructed deformations of the flat spacetime ) . such obstructions would render the theory physically inconsistent . in bosonic string , all linearized deformations are unobstructed . one explanation is that the _ semi - relative _ cohomology at the ghost number three is zero , and therefore there is no obstacle . more precisely , the higher order correction to @xmath2 are controlled by the string field equation @xcite : @xmath6 since the ghost number four cohomology is zero , @xmath7 is in the image of @xmath8 . in fact , the pre - image could be chosen to be annihilated by @xmath9 , and this shows that eq . ( [ stringfieldequation ] ) can be resolved order by order in the deformation parameter . unfortunately , we do not have such a proof in the pure spinor formalism . it follows from the consistency of @xcite that there is actually no obstacle in extending the infinitesimal deformation to higher orders . even though the ghost number three cohomology is nonzero , the actual obstruction vanishes for physical states . it would be good to have a transparent proof of this fact using the language of brst cohomology and vertex operators . this would probably require the use of the composite @xmath4-ghost . in the rest of this introductory section we will review general facts about the brst cohomology and its relation to the deformations of the worldsheet sigma - model . then in section [ sec : typeiibvsmaxwell ] we will review the cohomology of the classical electrodynamics , and explain how to reduce the cohomology of the type iib brst operator in flat space to the cohomology of electrodynamics . the relation will involve the computation of the cohomology of the algebra of translations with coefficients in the space of solutions of susy maxwell equations ( section [ sec : cohomologyofclassicalelectrodynamics ] ) and the tensor produce of two copies of such spaces ( section [ sec : cohomologyoftensorproduct ] ) . the results on brst cohomology are summarized in sections [ sec : brstcohomology ] and [ sec : actionofsusy ] . it was shown in @xcite that classical solutions of the type iib supergravity are in one - to - one correspondence with two - dimensional sigma - models satisfying certain axioms . most importantly , there should be two nilpotent odd symmetries @xmath10 and @xmath11 : @xmath12 also , there should be conserved charge known as the `` ghost number '' , with both @xmath10 and @xmath11 having ghost number @xmath13 . suppose that we are given such a sigma - model . a natural question is , how can it be deformed ? deformations of the sigma - model are the deformations of the action : @xmath14 where @xmath15 is some operator . if @xmath15 vanishes on - shell , then such deformation is trivial , as it can be undone by a field redefinition . suppose that the deformation is non - trivial . the condition that the deformed action still has a pair of nilpotent symmetries is equivalent to requiring the existence of @xmath16 and @xmath17 such that on - shell : @xmath18 here @xmath19 means `` equivalent on - shell '' , _ i.e. _ `` equivalent modulo the equations of motion '' . explicitly , ( [ separateqlandqr ] ) implies the existence of infinitesimal transformations @xmath20 and @xmath21 ( vector fields on the field space ) such that : @xmath22 were @xmath23 is the sigma - model lagrangian . ( the @xmath24 of ( [ totalderivativeonshell ] ) may be different from the @xmath25 of ( [ separateqlandqr ] ) because the variation of the lagrangian is proportional to the equations of motion only modulo a total derivative ) . then @xmath26 and @xmath27 are both symmetries of the deformed action ( [ deformationoftheaction ] ) . actually they are nilpotent : @xmath28 this is automatically true because all those anticommutators would be conserved charges of the ghost number two . in this paper we study vertices which are homogeneous polynomials of @xmath29 and @xmath30 . the conserved charges of the ghost number two are polynomials of low degree . therefore if @xmath15 is of large enough degree in @xmath29 and @xmath30 , then the nilpotence condition ( [ anticommutatorautomaticallyzero ] ) is satisfied . it is enough to verify ( [ separateqlandqr ] ) for @xmath31 : @xmath32 conditions ( [ separateqlandqr ] ) and ( [ q ] ) are equivalent , because @xmath33 and @xmath34 are independent , as both left and right ghost number are conserved . in fact , any linear combination @xmath35 with nonzero constant @xmath36 and @xmath37 can be choosen as a brst operator ; all such complexes are quasi - isomorphic to each other . operators @xmath15 satisfying the condition ( [ q ] ) are called `` integrated vertices '' . notice that @xmath38 is a one - form of the ghost number one , and @xmath39 ; this typically and @xmath30 ; some of our results are only valid under the assumption that the degree of the polynomial is large enough ; exceptions may happen for vertices which do not depend on @xmath29 ] implies @xmath41 , because there are no conserved charges of the ghost number two . this @xmath2 is called the _ unintegrated vertex _ corresponding to the integrated vertex @xmath15 : @xmath42 it is also possible to revert this procedure and go from @xmath2 back to @xmath15 . this involves the assumption about the vanishing of the cohomology at the nonzero conformal weight requires the absence of local conserved charges with nonzero ghost number ; going back ( from @xmath2 to @xmath15 ) requires the vanishing of the cohomology in the sector with positive conformal dimension ] . although ( to the best of our knowledge ) the proof of this vanishing theorem has never been given , we feel that the statement is true . notice that the construction of @xcite establishes the correspondence between integrated and unintegrated vertices independently of this assumption . although ( in its current form ) it only works in flat space and in @xmath43 , it also teaches us something about the generic curved background . for example , it tells us that the map @xmath44 is injective . indeed , suppose that existed an integrated vertex @xmath15 such that @xmath45 and @xmath46 ( _ i.e. _ nonzero @xmath15 gives @xmath2 ) . let us expand such @xmath15 in taylor series around a fixed point in the curved space - time , and take the leading term . this should give us the flat space vertex . since the map @xmath44 is injective in flat space , the leading term in @xmath2 should also be nonzero . this means that , if @xmath2 gets killed , then @xmath15 can not survive either . in any case , our * working hypothesis * is : * at the linearized level the deformations of the action are in one - to - one correspondence with the brst cohomology of @xmath31 at the ghost number two if @xmath15 is an integrated vertex operator , then ( [ deformationoftheaction ] ) defines a deformation of the sigma - model action to the first order in @xmath47 . it is natural to ask , if the deformation can be continued to higher orders of @xmath47 . an obstacle can , in principle , arise already at the order @xmath48 . once we deform the action as in ( [ deformationoftheaction ] ) , the brst operator gets deformed : @xmath49 here @xmath50 is such that : @xmath51 where @xmath23 is the sigma - model lagrangian ( the existence of such @xmath50 follows from the fact that @xmath52 is a total derivative on - shell , this is in the definition of an integrated vertex operator ) . let us consider the following expression : @xmath53 where @xmath54 is the hamiltonian generating @xmath55 : @xmath56 it was proven in @xcite that exists a ghost - number - three operator @xmath57 such that : @xmath58 moreover , the cohomology class of @xmath57 is the obstacle for extending the deformation to the order @xmath48 . the same analysis can be extended to higher orders in @xmath47 . [ [ conclusion ] ] conclusion : + + + + + + + + + + + if the brst cohomology at the ghost number three is zero then any infinitesimal deformation can be continued to a finite deformation , at least as a power series in @xmath47 . however , if the brst cohomology at the ghost number three is nonzero , then there is a potential obstacle . [ [ comment - on - the - derivation - in ] ] comment on the derivation in @xcite + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in @xcite we concentrated on the perturbation theory around @xmath43 , while in the present paper we work in flat space . some of the assumptions leading to eq . ( [ definitionofobstacle ] ) do not work literally in flat space . for example , conserved charges with nonzero ghost number ( besides the brst charge ) do exist in flat space @xcite . however , these charges do not depend on @xmath29 . if we restrict ourselves to the polynomial expressions with large enough degree , then the arguments of @xcite do apply . [ [ another - way - of - looking - at - the - obstacle ] ] another way of looking at the obstacle + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + suppose that we have an unintegrated vertex operator @xmath2 of the ghost number two . suppose that we deform the action as in ( [ deformationoftheaction ] ) by _ some _ integrated operator @xmath59 ( which is related by the descent procedure to _ some other _ integrated vertex @xmath60 ) . the brst operator gets deformed : @xmath61 . the question is , does @xmath2 survive such a deformation ? in other words it is possible to correct @xmath62 in such a way that @xmath63 ? if the cohomology at the ghost number three is trivial , then this is always possible . otherwise , further analysis is needed : one has to prove that the ghost number three vertex @xmath64 is @xmath8-exact . [ [ a - simpler - related - phenomenon ] ] a simpler related phenomenon + + + + + + + + + + + + + + + + + + + + + + + + + + + + similar thing happens at the ghost number one . in flat space , there is a nontrivial cohomology at the ghost number one , corresponding to the global symmetries . however , a generic perturbation of the flat space will kill all this ghost number one cohomology . this is obvious , as generic linearized sugra solution does not have any global symmetries . what we want to stress , is the cohomological interpretation of why the ghost number one cohomology gets killed : the existence of the ghost number two cohomology . in this paper we will show that the ghost number three cohomology is nozero . the more or less general example of a cohomologically nontrivial ghost number three vertex can be obtained as follows . let us consider a ghost number two vertex for an exponential linearized solution , for example a ramond - ramond excitation : @xmath65 \right)_{\alpha } p^{\alpha\hat{\beta } } \left ( ( \theta_r\gamma^m\lambda_r)(\theta_r\gamma_m ) + [ \lambda_r\theta_r^{\geq 4 } ] \right)_{\hat{\beta}}\end{aligned}\ ] ] where @xmath66 is a constant polarization tensor , @xmath67 . suppose that @xmath68 is a constant vector such that @xmath69 . let us consider : @xmath70 notice that @xmath71 is brst closed . we will prove in section [ sec : diracdiracsector ] that it is not brst exact , but the relative sign in ( [ v3 ] ) is minus . with the plus sign it would be @xmath72 . also notice that @xmath73 is the ghost number one unintegrated vertex corresponding to the global conserved charge of translations ( the momentum of the string ) . vertices of the ghost number three transform in the same representation of the super - poincare algebra as the linearized sugra solutions . ( in particular , the obstacle for @xmath71 to be brst - exact is in fact the scalar @xmath74 , so all the polarization is in @xmath66 . ) the integrated vertex corresponding to ( [ v3 ] ) can be constructed as follows . let @xmath75 be the integrated vertex corresponging to @xmath76 . let @xmath77 be the conserved current corresponding to @xmath73 : @xmath78 since @xmath75 is an integrated vertex , exists a 1-form @xmath38 such that @xmath79 . let us denote : @xmath80 we have : @xmath81 the next step is : @xmath82 we conclude that @xmath83 is the integrated vertex operator corresponding to @xmath71 . it is a two - form of the ghost number one . in this paper we will study _ polynomial _ vertices , _ i.e. _ vertices depending on @xmath29 polynomially . the exponential vertices ( [ exponentialv2 ] ) and ( [ v3 ] ) are sums of infinitely many polynomial vertices . indeed , @xmath84 can be decomposed in the taylor series , and the brst operator preserves the degree of a polynomial ( we assign degree @xmath85 to @xmath29 and degree @xmath86 to @xmath30 and @xmath87 ) . polynomial vertices are , essentially , harmonic polynomials of @xmath29 dressed with some appropriate @xmath30-dependence . a low degree example of a polynomial vertex of the ghost number three has been previously constructed in the revised version of @xcite . it is equivalent to the linear term in the expansion of @xmath71 in powers of @xmath29 . we will prove in section [ sec : ghostnumberfour ] that the pure spinor cohomology is zero at the ghost number four . we have proven in @xcite that the pure spinor cohomology is zero at the ghost number greater than four . this implies that the ghost number three cohomology survives the deformation from flat space - time to generic curved space - time . ( however , in the case of a generic curved space - time , there are no ghost number one vertices ; therefore the construction of section [ sec : introghostnumberthree ] does not work . ) consider an unintegrated vertex operator @xmath2 and the corresponding deformation of the sigma - model . can we extend it to the second order in the deformation parameter ? the potential obstacle is the ghost number 3 cohomology class @xmath57 defined in eq . ( [ definitionofobstacle ] ) . it is bilinear in @xmath2 : @xmath88\!]}\ ] ] we will show that @xmath57 transforms in the linearized supergravity multiplet ( _ i.e. _ in the same representation as @xmath2 , modulo some discrete states ) . the map @xmath89 given by ( [ wisgb ] ) defined by ( [ wisgb ] ) should commute with the action of the supersymmetry , in particular with the translations . moreover , one can see that : @xmath90 ( for example , for the linear dilaton background analized in @xcite , @xmath91 $ ] and @xmath92 $ ] . ) this implies that @xmath93\!]}$ ] can only be nonzero if either @xmath94 or @xmath76 is a low degree polynomial . it should be possible to complete this argument , which would provide a proof of the vanishing of the obstructions to most of the deformations of the flat space at the second order ( but this proof will not work at higher orders ) . in section [ sec : typeiibvsmaxwell ] we explain how to compute the massless brst cohomology of the type ii sugra by relating it to the brst cohomology of the maxwell theory using the spectral sequence of a bicomplex . in sections [ sec : cohomologyofclassicalelectrodynamics],[sec : zerothcohomology],[sec : cohomologyoftensorproduct ] and [ sec : secondcohomologytensorproduct ] we compute the second page of that spectral sequence . in section [ sec : brstcohomology ] we finally compute the spectrum of massless states , and in section [ sec : actionofsusy ] we study the action of supersymmetry on the ghost number three vertices . for the first reading , we recommend the following sequence : section [ sec : introduction ] @xmath95 section [ sec : typeiibvsmaxwell ] @xmath95 section [ sec : brstcohomology ] . then sections [ sec : cohomologyofclassicalelectrodynamics],[sec : zerothcohomology],[sec : cohomologyoftensorproduct ] and [ sec : secondcohomologytensorproduct ] could be read at the second pass . we will compute the cohomology of the type iib brst complex by relating it to the super - maxwell brst complex . the cohomology of the super - maxwell brst complex : @xmath96 is only nontrivial at the ghost numbers 0 and 1 . at the ghost number 0 the cohomology is constants : @xmath97 . at the ghost number 1 , the cohomology is in one - to - one correspondence with the solutions of the free maxwell equation and the free dirac equation . the vanishing of the cohomology at the ghost numbers two and three is equivalent to the following statements : 1 . for any current @xmath98 such that @xmath99 always exists the gauge field @xmath100 satisfying @xmath101}=0 $ ] and @xmath102 2 . for any antichiral spinor @xmath103 exists a chiral spinor @xmath104 such that @xmath105 3 . for any @xmath106 exists @xmath98 such that @xmath107 [ [ example ] ] example : + + + + + + + + let us look at the ghost number two cohomology . the leading term in the @xmath30-expansion is either @xmath108 or @xmath109 . let us for example investigate the first possibility . the following expression is in the cohomology of @xmath110 : @xmath111 now let us study the effect of the @xmath112-term in ( [ qsupermaxwell ] ) . for ( [ zeromodeghostnumbertwo ] ) to survive the action of @xmath113 we need : @xmath114 the `` something '' on the right hand side always exists , because any expression of the form @xmath115 $ ] annihilated by @xmath116 is automatically in the image of @xmath116 . it remains to investigate the possibility of ( [ zeromodeghostnumbertwo ] ) being @xmath8-exact : @xmath117 this is possible iff @xmath118 . but for any @xmath119 we can find @xmath120 such that @xmath118 . this implies that any expression of the type ( [ zeromodeghostnumbertwo ] ) is always brst - trivial . the class with the leading term @xmath109 is analyzed similarly . [ [ conclusion-1 ] ] conclusion : + + + + + + + + + + + @xmath121 here `` @xmath122 '' stands for the direct sum of the space of solutions of the maxwell equations and the space of solutions of the dirac equation . we now want to relate the super - maxwell complex to the type iib sugra complex . [ [ comment - in - the - revised - version ] ] comment in the revised version + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + it is possible to modify the definition of the brst complex by imposing the constraint that the cochains are annihilated by @xmath123 . in this case @xmath124 is nonzero and in fact isomorphic ( perhaps modulo some zero modes ) to @xmath125 see the recent work @xcite and references there . we do not impose any such constraints . therefore our brst complex has @xmath126 for open strings . but for closed strings , we still get the massless @xmath127 nonzero ( and isomorphic to @xmath128 up to zero modes ) . let us consider the tensor product of two smaxwell complexes : @xmath129 the operator @xmath130 acts on the space of functions @xmath131 . we will denote @xmath10 and @xmath11 the two terms on the right hand side of ( [ brstsmaxwtimessmaxw ] ) . this is the `` doubled '' brst complex . the difference with the type iib sugra brst complex is the splitting @xmath132 . in the type iib brst complex there is no separation of @xmath29 into @xmath133 and @xmath134 : @xmath135 the difference with ( [ brstsmaxwtimessmaxw ] ) is that the left and the right parts have a common @xmath29 instead of separate @xmath133 and @xmath134 ; the operator @xmath136 acts on the space of functions @xmath137 . the computation of the cohomology of ( [ brstsmaxwtimessmaxw ] ) is straightforward , because it is just the tensor product of two maxwell complexes ( [ qsupermaxwell ] ) ; therefore the cohomology is : @xmath138 where the spaces @xmath139 are given by eqs . ( [ smaxw0 ] ) , ( [ smaxw1 ] ) and ( [ smaxwhigher ] ) . to compute the cohomology of ( [ qsugra ] ) , we relate it to the cohomology of ( [ brstsmaxwtimessmaxw ] ) by the following trick . let us introduce a formal fermionic variable @xmath141 and the operator : @xmath142 ( we call it @xmath143 because it can be thought of as the cohomology of the abelian lie algebra of translations . ) let us consider the bicomplex : @xmath144 consider two ways of computing the cohomology of @xmath145 . we can either compute first the cohomology of @xmath143 , and then consider @xmath146 as a perturbation . or , first compute @xmath147 and then act on it by @xmath143 . this means that there are two different spectral sequences , both converging to @xmath148 . [ [ first - q_rm - lie - then - q_l - q_r ] ] first @xmath143 , then @xmath146 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + because of the poincare lemma , the cohomology of @xmath143 is only nontrivial in the ghost number 0 , and is represented by the functions @xmath149 . therefore the type iib brst complex is equivalent to the cohomology of @xmath146 acting on the cohomology of @xmath143 : @xmath150 [ [ first - q_l - q_r - then - q_rm - lie ] ] first @xmath146 , then @xmath143 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + now let us first compute the cohomology of @xmath146 , and then consider @xmath143 as a perturbation . the resulting spectral sequence will be denoted @xmath140 . it computes the cohomology of the sugra brst complex : @xmath151 therefore , the only nontrivial components are : @xmath152 all other components are zero . the only potentially nonzero differentials are : @xmath153 therefore , in order to compute the brst cohomology of sugra , we have to : * first compute the cohomology of @xmath143 with coefficients in spaces of solutions of the classical electrodynamics and their tensor products * then compute the differentials @xmath154 the first step will be elaborated in sections [ sec : cohomologyofclassicalelectrodynamics ] , [ sec : zerothcohomology ] and [ sec : cohomologyoftensorproduct ] , and the second in section [ sec : brstcohomology ] . in the previous section we related the cohomology of the sugra complex to the lie algebra cohomology of the algebra of translations @xmath155 with coefficients in the tensor product of solutions of maxwell and dirac equations . in order to compute it , we will first compute the cohomology with coefficients in the single space of solutions of maxwell and dirac equations . then , in the next section , we will proceed to compute the cohomology with coefficients in the tensor product of two such spaces . consider the space of solutions of the vacuum maxwell equations : @xmath156 } = 0\ ] ] depending on a parameter @xmath141 , a free grassmann variable . we need to calculate the cohomology of the operator @xmath157 acting on this space . we will start by computing the cohomology of divergenceless currents . consider the space @xmath158 of one - forms @xmath159 satisfying @xmath160 . this is a subspace of the space of all 1-forms @xmath161 : @xmath162 this gives the long exact sequence of cohomology : @xmath163 we conclude : @xmath164 now we proceed to the cohomology of the maxwell complex . a solution of the maxwell equation is completely characterized by its curvature . the space of solutions is therefore the same as the space of closed 2-forms @xmath165 satisfying @xmath166 . it is included in the following short exact sequence : @xmath167 where @xmath168 is the space of all closed 2-forms . the corresponding long exact sequence reads : @xmath169 to calculate the cohomology of @xmath168 we use : @xmath170 and @xmath171 this implies that for @xmath172 : @xmath173 . therefore , we obtain from ( [ longexactsequenceformaxwell ] ) : @xmath174}dx^m\wedge dx^n \\ h^1({\rm maxwell } ) = \ ; & { \bf c}^{d } \oplus \lambda^3{\bf c}^{d}\;:\ ; c_kf_l dx^k\wedge dx^l\;\mbox { \tt\small and } \ ; f_{[klm]}c^kdx^l\wedge dx^m \label{h1maxwell}\\ h^2({\rm maxwell } ) = \ ; & { \bf c } \oplus \lambda^4{\bf c}^{d}\;:\ ; c_kc_ldx^k\wedge dx^l\;\mbox { \tt\small and } \ ; f_{[ijkl]}c^ic^jdx^k\wedge dx^l \label{h2maxwell}\\ h^{n>2}({\rm maxwell } ) = \ ; & \lambda^{n+2}{\bf c}^d\;:\ ; f_{[j_1\ldots j_{n+2 } ] } c^{j_1}c^{j_2}\cdots c^{j_n } dx^{j_{n+1}}\wedge dx^{j_{n+2 } } \end{aligned}\ ] ] notice that all these cohomology classes are represented by the _ constant _ field strength . in other words , the _ dilatation symmetry _ @xmath175 acts as zero in cohomology . let @xmath176 denote the space of solutions of the dirac equations , and @xmath177 the space of chiral - spinor - valued functions , and @xmath178 the antichiral - spinor - valued functions . there is a short exact sequence : @xmath179 this leads to the long exact sequence of the cohomologies : @xmath180 therefore : @xmath181 this is the direct sum : @xmath182 the space @xmath183 can be thought of as the space of functions : @xmath184\;;\;[pq]}(x ) \end{aligned}\ ] ] satisfying : @xmath185\;;\;[pq ] } \;&= 0 \label{leftjacobionfx } \\ f_{[mn]\;;\;[pq}\stackrel{\leftarrow}{\partial}_{r ] } \;&= 0 \label{rightjacobionfx } \\ \partial^mf_{[mn]\;;\;[pq ] } \ ; & = 0 \label{lefteqmonfx } \\ f_{[mn]\;;\;[pq]}\stackrel{\leftarrow}{\partial}^q \ ; & = 0 \label{righteqmonfx}\end{aligned}\ ] ] eqs . ( [ leftjacobionfx ] ) and ( [ rightjacobionfx ] ) together imply that : @xmath186 } = & \ ; \mbox{const } \label{constantfantisymm } \\ g^{mp}g^{nq}f_{[mn]\;;\;[pq ] } = & \ ; \mbox{const } \label{constantdoubletrace}\end{aligned}\ ] ] we can write : @xmath187\;;\;[pq ] } \;= \partial_{[m}a^l_{n]\;;\;[pq ] } = a^r_{[mn]\;;\;[p}\stackrel{\leftarrow}{\partial}_{q]}\ ] ] a consequence of eqs . ( [ leftjacobionfx ] ) , ( [ rightjacobionfx ] ) , ( [ lefteqmonfx ] ) , ( [ righteqmonfx ] ) is the existence of @xmath188 and @xmath189 such that : @xmath190\;;\;p } \;&=\partial_p\phi^r_n \label{phirdef } \\ a^l_{m\;;\;[pq]}\stackrel{\leftarrow}{\partial}^q \;&=\partial_m\phi^l_p \label{phildef}\end{aligned}\ ] ] this implies : @xmath191\;;\;[pq]}(x ) = { 1\over 2 } \partial_q\left ( g^{np}a^r_{[mn]\;;\;p } + \phi^r_m \right ) = { 1\over 2}\partial_m\left ( g^{np}a^l_{n\;;\;[pq ] } + \phi^l_q \right)\end{aligned}\ ] ] let us denote : @xmath192\;;\;p } + \phi^r_m\;,\;\ ; \\ b^l_q \;&= g^{np}a^l_{n\;;\;[pq ] } + \phi^l_q\end{aligned}\ ] ] in particular : @xmath193 although @xmath194;p}$ ] and @xmath195 are only defined by ( [ phirdef ] ) up to : @xmath196;p } & \mapsto a^r_{[mn];p } + \partial_p \chi^r_{mn } \\ \phi_n^r & \mapsto \phi_n^r + \partial^m \chi_{mn}\;,\end{aligned}\ ] ] this ambiguity does not affect the definition of @xmath197 ( and similarly @xmath198 ) . notice that : @xmath199 } = \;- \partial_{[p}b^l_{q ] } = \ ; & \mbox{const } \\ \partial^p b^l_p = \;\partial^p b^r_p = \ ; & \mbox{const}\end{aligned}\ ] ] let us denote : @xmath200 then : @xmath201 } \ ; & = 0 \\ \partial_{(q}a^{-}_{m ) } \ ; & = 0 \label{eqaminus}\end{aligned}\ ] ] the physical meaning of @xmath202 will be explained in section [ sec : cale02 ] . having computed the cohomology of @xmath143 with values in maxwell and dirac solutions , we will now use it to compute the cohomology with values in the tensor product @xmath203 . again , we will use some spectral sequence . in order to distinguish it from the spectral sequence of section [ sec : typeiibvsmaxwell ] , we will use the notation ] @xmath204 ( that other one was denoted @xmath140 ) . the following group is part of the ghost number 3 cohomology : @xmath205 in this section we will calculate this cohomology group . the differential @xmath143 is realized on the space of bispinors @xmath206 satisfying : @xmath207 the differential @xmath143 acts as follows : @xmath208 let us introduce the filtration by the degree @xmath209 : @xmath210 then @xmath211 is the leading ( of degree zero ) term in @xmath143 and @xmath212 is subleading ( of degree one ) . let us calculate the cohomology of @xmath143 using the spectral sequence of this filtration . the first page @xmath213 is : @xmath214 where @xmath215 consists of polynomials with @xmath216 . schematically , @xmath213 consists of expressions of the form @xmath217\ ] ] satisfying both left and right dirac equations , representing the cohomology of @xmath211 . just to remember : @xmath218 where @xmath219 means degree in @xmath29 . because of section [ sec : cohomologyinmaxwell ] , the cohomology of @xmath211 is localized on @xmath220 , and either @xmath221 or @xmath222 . this means that the only nontrivial components of @xmath213 are the ones represented by the following expressions : @xmath223 here , as usual , we denote @xmath224 . the only nontrivial differential is @xmath225 . the cohomology of this differential is @xmath226 . notice that @xmath227 . indeed , the construction of @xmath228 involves the inversion of @xmath211 and therefore any expression in the image of @xmath229 is necessarily in the image of @xmath230 . but @xmath230 acts as zero on @xmath231 and therefore also on @xmath232 . therefore our spectral sequence converges at the second page : @xmath233 . the condition that the cohomology class of an expression of the form ( [ cr ] ) is cancelled by the @xmath235 of an expression of the form ( [ p ] ) is : @xmath236 indeed , for any @xmath237 solving the right dirac equation ( [ diracequationonp ] ) we can tautologically write : @xmath238 then ( [ trivialityofr ] ) is the necessary and sufficient condition that @xmath239 satisfies both @xmath240 _ and _ @xmath241 . ( and , moreover , any presentation of @xmath242 as @xmath243 plus @xmath244 will necessarily be of the form ( [ presentationofr ] ) . ) [ [ comment ] ] comment + + + + + + + those @xmath245 which satisfy @xmath246 are in the kernel of @xmath235 , and therefore they form @xmath247 . they are in the ghost number two cohomology ( the ramond - ramond fields ) . we have previously explained that @xmath229 is zero ; if it were not zero , it would have killed the ghost number two cohomology . notice that any @xmath245 satisfying ( [ trivialityofr ] ) and ( [ diracequationonp ] ) is automatically harmonic : @xmath248 , therefore ( [ trivialityofr ] ) and ( [ diracequationonp ] ) imply that @xmath249 satisfies the left dirac equation : @xmath250 this means that : @xmath251 in the rest of this section we will prove that this is the only obstacle , _ i.e. _ any @xmath249 satisfying ( [ rsatisfiesleftdirac ] ) can be represented as ( [ trivialityofr ] ) , ( [ diracequationonp ] ) . in this section we will prove that if @xmath249 is a polynomial of nonzero degree ( _ i.e. _ not a constant ) , than ( [ diracoperatorisobstacle ] ) is the only obstacle to the triviality of @xmath249 . notice that it is always possible to solve for @xmath245 to satisfy ( [ trivialityofr ] ) , but @xmath245 will not necessarily satisfy ( [ diracequationonp ] ) . but if the dirac equation ( [ rsatisfiesleftdirac ] ) is satisfied , then we have : @xmath252 we will now prove that ( [ dldrpiszero ] ) and ( [ pisharmonic ] ) imply that exist @xmath253 and @xmath254 such that : @xmath255 this implies that @xmath245 can be chosen to satisfy the right dirac equation , and therefore @xmath249 is in the image of @xmath235 . [ [ proof . ] ] proof . + + + + + + let us switch from the bispinor notations to the forms notations . the left dirac operator corresponds to @xmath256 while the right dirac operator is @xmath257 . ( [ pisharmonic ] ) implies that @xmath258 while eq . ( [ dldrpiszero ] ) implies that @xmath259 . therefore we have : @xmath260 we will now prove that under the condition ( [ ddeltapiszero ] ) exist @xmath253 and @xmath254 such that : @xmath261 it is useful to keep in mind the cohomology of the de rham @xmath262 on harmonic forms is : @xmath263 ( the @xmath264 is generated by @xmath265 ) . [ [ case - when - p - is - a-5-form ] ] case when @xmath245 is a 5-form + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in this case we will write @xmath266 instead of @xmath245 . since @xmath267 , exists a harmonic 3-form @xmath268 such that : @xmath269 similarly , as @xmath270 , exists a harmonic 7-form @xmath271 such that : @xmath272 furthermore , there exist harmonic @xmath273 and @xmath274 such that : @xmath275 this implies that @xmath276 and therefore exists a harmonic form @xmath277 such that @xmath278 . similarly , @xmath279 . therefore the following @xmath253 and @xmath254 satisfy ( [ plandpr ] ) : @xmath280 [ [ case - when - p - is - a-3-form - plus-7-form ] ] case when @xmath245 is a 3-form plus 7-form + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the 7-form part of @xmath245 is related to the 3-form part by the condition that @xmath245 is self - dual . in this case we will write @xmath281 instead of @xmath245 . ( this @xmath268 has nothing to do with the @xmath268 of the previous paragraph . ) since @xmath282 and @xmath283 , exists harmonic @xmath273 such that : @xmath284 this implies that @xmath276 . similarly , exists a harmonic @xmath266 such that : @xmath285 this automatically implies : @xmath272 also exists a harmonic @xmath274 such that : @xmath286 we take : @xmath287 [ [ case - when - p - is - a-1-form - plus - a-9-form ] ] case when @xmath245 is a 1-form plus a 9-form + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + now suppose that @xmath288 . let us first assume that the degree of @xmath245 is more than 1 . we have : @xmath289 similarly @xmath290 . now we have : @xmath291 now consider the case when the degree of @xmath245 is one , _ i.e. _ @xmath245 is linear in @xmath29 . in this case we can have @xmath292 . this corresponds to the @xmath249 of ( [ trivialityofr ] ) a constant proportional to unit matrix . the corresonding element of @xmath293 is : @xmath294 it corresponds to the following ghost number three vertex : @xmath295 [ [ conclusion-2 ] ] conclusion + + + + + + + + + + we conclude that the main obstacle for ( [ cr ] ) to be trivial is @xmath296 . ( and besides that , there is also a case when @xmath249 is a constant times a unit matrix , which results in a nontrivial vertex ( [ vertexghostnumberthreedimfour ] ) . ) if @xmath296 , then there is a nontrivial cohomology class of the form : @xmath297 + r_2[c x_l^2 x_r^{(2m-3 ) } ] + \ldots + r_{2m-1}[c x_l^{(2m-1 ) } ] \label{seriesforrepresentativeh1diracdirac}\end{aligned}\ ] ] indeed , acting on the leading term @xmath298 with @xmath212 we get an expression of the form @xmath299 $ ] , which does not depend on @xmath133 and therefore is annihilated by @xmath211 . but since @xmath300 , this expression is automatically of the form @xmath301 $ ] . continuing this process we get ( [ seriesforrepresentativeh1diracdirac ] ) . let us consider the ghost number three vertex @xmath71 given by eq . ( [ v3 ] ) , and expand it in the taylor series in @xmath29 and @xmath30 . we assign to @xmath29 degree @xmath85 and to @xmath87 and @xmath30 degree @xmath302 . the brst operator preserves this degree . in particular , every term in the expansion is a brst - closed _ polynomial _ of @xmath303 . it is enough to prove the nontriviality term by term . let us consider the extended space @xmath304 . in this extended space , we get : @xmath305 the corresponding element of @xmath306 is given by : @xmath307 consider the expansion in powers of @xmath133 . the leading term is @xmath308 . we observe : @xmath309 and @xmath310 satisfies the left dirac equation . therefore ( [ adotcpexp ] ) is equivalent to @xmath311 . comparing this with ( [ seriesforrepresentativeh1diracdirac ] ) , we get : @xmath312 then ( [ diracoperatorisobstacle ] ) implies that @xmath71 represents a nontrivial cohomology class . [ [ ghost - number - three - vertex - of ] ] ghost number three vertex of @xcite + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + can be obtained as the first order of expansion of ( [ adotcpexp ] ) in powers of @xmath29 . indeed , at the first order of the @xmath29-expansion @xmath313 . notice that the expression : @xmath314 satisfies the left dirac equation ( we use @xmath315 ) . therefore @xmath313 is equivalent to @xmath316 . therefore the leading term of the @xmath29-linear part of ( [ adotcpexp ] ) is equivalent to @xmath317 which is the leading term of the vertex constructed in @xcite . in this section we will compute the cohomology of @xmath318 on the solutions of bi - maxwell equations . solutions of bi - maxwell equations are defined as expressions of the form : @xmath319\;;\;[m } { \stackrel{\leftarrow}{\partial}\over\partial x_r^{n ] } } \right)\ ; dx_r^m\wedge dx_r^n\;\ ] ] satisfying the left and right maxwell equations : @xmath320\;;\;[m}{\stackrel{\leftarrow}{\partial}\over\partial x_r^{n ] } } = 0 \\ { \partial\over\partial x_l^{[p } } { \cal a}(x_l , x_r)_{q]\;;\;[m } { \stackrel{\leftarrow}{\partial}\over\partial x_r^{n ] } } { \stackrel{\leftarrow}{\partial}\over\partial x_r^{n } } = 0\end{aligned}\ ] ] notice that we have left and right indices , separated with the semicolon . we use the notations @xmath321 . the expression @xmath322 means the same as @xmath323 . the sole purpose of such notations is to improve the readability of the formulas , as they allow us to naturally separate left and right indices . [ [ definition ] ] definition + + + + + + + + + + as in section [ sec : diracdiracsector ] , we will use the filtration by the powers of @xmath133 , _ i.e. _ treat @xmath133 as being small . the elements of @xmath204 are of the type : @xmath324\;dx_r\wedge dx_r + \ldots\ ] ] where @xmath325 stands for terms of the type @xmath326\;dx_r\wedge dx_r$ ] with @xmath327 , which are factored out when we consider @xmath328 modulo @xmath329 . for a polynomial element @xmath330 , of the total order @xmath331 in @xmath133 and @xmath134 , there is an expansion in powers of @xmath133 : @xmath332 where @xmath333 does not depend on @xmath133 , @xmath334 is linear in @xmath133 , _ etc._. [ [ the - structure - of - e_2pq ] ] the structure of @xmath226 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the following is the most general ( up to the @xmath335-exact terms ) ansatz for the leading term @xmath333 : @xmath336}\;dx_r^m\wedge dx_r^n\ ; + \label{ansatzmaxwellmaxwell } \\ \;&+ c^p dx_l^q \wedge dx_l^r\ ; b(x_r)_{pqr\;;\;[m}\stackrel{\leftarrow}{\partial}_{n]}\;dx_r^m\wedge dx_r^n\ ; \nonumber \\ \mbox{\small\tt with } \ ; & a(x_r)_{q\;;\;[m}\stackrel{\leftarrow}{\partial}_{n ] } \stackrel{\leftarrow}{\partial}_m = 0 \label{ansatzmaxwellmaxwelleqm}\\ & b(x_r)_{pqr\;;\;[m}\stackrel{\leftarrow}{\partial}_{n ] } \stackrel{\leftarrow}{\partial}_m = 0 \label{ansatzmaxwellmaxwelleqmforb}\\ & \partial^qa(x_r)_{q\;;\;[m}\stackrel{\leftarrow}{\partial}_{n ] } = 0 \label{ansatzmaxwellmaxwellunobstructeda}\\ & \partial_{[p}b(x_r)_{qrs]\;;\;[m}\;\stackrel{\leftarrow}{\partial}_{n ] } = 0 \label{ansatzmaxwellmaxwellunobstructedb}\\ { \cal a}^{(0)}_{q\;;\;m}\ ; & \mbox { \tt\small represents } \mbox { \tt\small an element of } e_2^{-{m-1\over 2},{m+1\over 2 } } \mbox { \tt\small , see ( \ref{e1pqmaxwellmaxwell})}\end{aligned}\ ] ] here @xmath337 and @xmath338\;;\;m}$ ] are polynomials in @xmath134 of the order @xmath331 . they correspond to the two terms in ( [ h1maxwell ] ) . ( [ ansatzmaxwellmaxwelleqm ] ) and ( [ ansatzmaxwellmaxwelleqmforb ] ) enforce the right maxwell equation . ( the left maxwell equation is automatically satisfied because @xmath339 does not depend on @xmath133 . ( [ ansatzmaxwellmaxwellunobstructeda ] ) and ( [ ansatzmaxwellmaxwellunobstructedb ] ) are the conditions for being in the kernel of @xmath235 . in other words , those are the conditions for the existence of @xmath340 linear in @xmath133 and @xmath341 such that : @xmath342}\;dx_r^m\wedge dx_r^n\ ; + \nonumber \\ & + \ ; c^p dx_l^q \wedge dx_l^r\ ; b(x_r)_{pqr\;;\;[m}\stackrel{\leftarrow}{\partial}_{n]}\;dx_r^m\wedge dx_r^n\ ; \big ) \ ; = \nonumber \\ = \ ; & c^r{\partial\over\partial x^r_l } \left ( dx_l^p\wedge dx_l^q\;\partial_{[p } a^{(1)}(c , x_l , x_r)_{q]\;;\;[m } \stackrel{\leftarrow}{\partial}_{n]}\;dx_r^m\wedge dx_r^n\ ; \right ) \\ \mbox{\tt\small and } & a^{(1)}(x_l , x_r)_{q\;;\;[m}\stackrel{\leftarrow}{\partial\over \partial x_r^{n ] } } \stackrel{\leftarrow}{\partial\over\partial x_r^m } \ ; = \ ; { \partial\over\partial x_l^p}{\partial\over\partial x_l^{[p } } a^{(1)}(x_l , x_r)_{q]\;;\;m } \;= \;0\end{aligned}\ ] ] eq . ( [ ansatzmaxwellmaxwellunobstructeda ] ) is the vanishing of the obstacle proportional to the first term in ( [ h2maxwell ] ) , and eq . ( [ ansatzmaxwellmaxwellunobstructedb ] ) is to avoid hitting the second term in ( [ h2maxwell ] ) . remember that we are working in the polynomial sector , _ i.e. _ @xmath343}$ ] is a homogeneous polynomial in @xmath134 . let us first assume that the degree of the polynomial is nonzero : @xmath344 } \;\neq\ ; \mbox{const}\ ] ] then ( [ ansatzmaxwellmaxwellunobstructedb ] ) implies that we can remove the term @xmath345}\;dx_r^m\wedge dx_r^n$ ] , by adding to @xmath346 an element in @xmath347 . indeed , this is equivalent to the existence of the following two objects : * @xmath348 satisfying @xmath349}\stackrel{\leftarrow}{\partial}_{m } = 0 $ ] and * @xmath350 linear in @xmath133 satisfying left and right maxwell equations : @xmath351\;;\;[m}\stackrel{\leftarrow}{\partial\over\partial x_r^{n ] } } \;&= 0 \nonumber \\ { \partial\over\partial x_l^p } g(x_l , x_r)_{pq\;;\;[m}\stackrel{\leftarrow}{\partial\over\partial x_r^{n ] } } \;&= 0 \nonumber \\ g(x_l , x_r)_{pq\;;\;[m}\stackrel{\leftarrow}{\partial\over\partial x_r^{n]}}{\stackrel{\leftarrow}{\partial}\over\partial x_r^{m } } \;&= 0 \nonumber \end{aligned}\ ] ] such that : @xmath352}\;dx_r^m\wedge dx_r^n \ ; + o(x_l ) = \nonumber \\ = \;\;\;\ ; & c^k\left({\partial\over\partial x^k_l } - { \partial\over\partial x^k_r}\right)\ ; \big(\;\;\ ; dx_l^p\wedge dx_l^q\ ; c(x_r)_{pq\;;\;[m}\stackrel{\leftarrow}{\partial}_{n]}\;dx_r^m\wedge dx_r^n \ ; + \nonumber \\ & \phantom{c^k\left({\partial\over\partial x^k_r } - { \partial\over\partial x^k_r}\right ) } \;\;+\ ; dx_l^p\wedge dx_l^q\ ; g(x_l , x_r)_{pq\;;\;[m } { \stackrel{\leftarrow}{\partial}\over\partial x_r^{n ] } } \;dx_r^m\wedge dx_r^n\;\ ; \big ) \;+ \nonumber \\ \ ; & + c_{p } dx_l^p\wedge dx_l^q\ ; \widetilde{a}(x_r)_{q\;;\;[m}\stackrel{\leftarrow}{\partial}_{n]}\ ; dx_r^m\wedge dx_r^n\ ; + \ ; o(x_l ) \ ; \label{removingbterm } \end{aligned}\ ] ] here @xmath353}$ ] is some correction to the @xmath354}$ ] of ( [ ansatzmaxwellmaxwell ] ) . ( in other words , when we gauge away the @xmath355-term , this leads to some change in the @xmath339 term : @xmath356 . ) the existence of such @xmath348 and @xmath350 follows from ( [ ansatzmaxwellmaxwellunobstructedb ] ) and the fact that @xmath357 is zero in polinomials of the degree @xmath358 , in the following way in two different ways . first , we use @xmath359 to argue that the leading term can be reduced to the form ( [ ansatzmaxwellmaxwell ] ) . then we use the _ vanishing _ of @xmath360 in polynomials of high enough degree to remove the b - term by adding @xmath361 . ] . ( [ ansatzmaxwellmaxwellunobstructedb ] ) implies that exists @xmath348 satisfying the right maxwell equation , such that : @xmath362 } = - \partial_{[p } c_{qr]\;;\;[m } \stackrel{\leftarrow}{\partial}_{n]}\ ] ] therefore , in computing the first line on the rhs of ( [ removingbterm ] ) , the @xmath363 gives zero as @xmath364 does not depend on @xmath133 , and when acting with @xmath365 , we get : @xmath366}\;dx_r^m\wedge dx_r^n \;\ ] ] this has to be understood as an element of @xmath367 , _ i.e. _ modulo the image of @xmath363 . this ambiguity is described by the second line on the rhs of ( [ removingbterm ] ) , the term containing @xmath350 . this term can be used to remove the components other than those listed in eq . ( [ h1maxwell ] ) ; the component @xmath368 corresponds to @xmath369 , and the component @xmath370 kills the b - term . we conclude that we can get rid of the @xmath355-term in ( [ ansatzmaxwellmaxwell ] ) by adding to @xmath371 an element in @xmath347 . now let us consider the case when @xmath343}$ ] is constant : @xmath344 } = \mbox{const}\ ] ] consider the total antisymmetrization : @xmath372 } = b_{[pqr\;;\;m}(x_r)\stackrel{\leftarrow}{\partial}_{n]}\ ] ] in this case the @xmath355-term in ( [ ansatzmaxwellmaxwell ] ) can not be gauged away , as @xmath373}$ ] represents a nontrivial cohmology class of @xmath374 . however , we will show in section [ sec : ghostnumberfour ] that this is cancelled by the @xmath375 . in other words , for our ansatz to survive on @xmath376 we need to put @xmath373}$ ] to zero : @xmath377 } = 0\ ] ] let us therefore assume that @xmath378 . can the remaining @xmath339-term also be in the image of @xmath235 ? let us define the double field strength @xmath379;[mn]}$ ] as follows : @xmath380;[mn ] } = \partial_{[p}a_{q]\;;\;[m}\stackrel{\leftarrow}{\partial}_{n]}\ ] ] this double field strength has the following properties : @xmath381 } \ ; & = 0 \mbox { \tt\small ( total antisymmetrization ) } \\ \partial_{[p}f_{qr]\;;\;mn } \;&= 0 \\ f_{qr\;;\;[mn}\stackrel{\leftarrow}{\partial}_{k ] } \ ; & = 0 \\ \partial^p f_{[pq];[mn ] } \;&= 0 \\ f_{[pq];[mn ] } \stackrel{\leftarrow}{\partial}^m \;&= 0 \\ \delta f_{[pq];[mn ] } = 0\end{aligned}\ ] ] we will now show that @xmath382 is trivial iff @xmath383 . we have to understand under which conditions the class with the leading term ( [ ansatzmaxwellmaxwell ] ) is trivial , _ i.e. _ can be obtained by acting with @xmath318 on something : @xmath384\;;\;[m}\stackrel{\leftarrow}{\partial}_{n]}\;dx_r^m\wedge dx_r^n\ ; + \ldots = \\ = \ ; & c^j\left ( { \partial\over\partial x_l^j } - { \partial\over\partial x_r^j}\right)\ ; \left ( dx_l^p\wedge dx_l^q\ ; w(x_r)_{pq\;;\;[m}{\stackrel{\leftarrow}{\partial}\over \partial x_r^{n]}}\ ; dx_r^m\wedge dx_r^n\ ; + \ldots \right ) \nonumber\end{aligned}\ ] ] this property is equivalent to the existence of @xmath385 satisfying : @xmath386 } \ ; = \ ; & \partial^{p } w_{pq \;;\ ; [ m } \stackrel{\leftarrow}{\partial}_{n ] } \label{afromw } \\ \mbox{\tt\small with } \ ; & \partial_{[p}w_{qr]\;;\;[m}\;\stackrel{\leftarrow}{\partial}_{n ] } = 0 \label{wisleftclosed}\\ \mbox{\tt\small and } \ ; & w_{pq\;;\;[m } \stackrel{\leftarrow}{\partial}_{n ] } \stackrel{\leftarrow}{\partial}_{n } = 0 \label{rightmaxwellonw}\end{aligned}\ ] ] notice that the ghost number two vertices correspond to @xmath387 satisfying ( [ wisleftclosed ] ) , ( [ rightmaxwellonw ] ) and @xmath388 } = 0 $ ] instead of ( [ afromw ] ) . if @xmath339 can be expressed through @xmath57 as in ( [ afromw ] ) , then we have : @xmath389\;;\;[m}\stackrel{\leftarrow}{\partial}_{n ] } = -\;\partial_{[p}\partial^rw_{q]r\;;\;[m}\;\stackrel{\leftarrow}{\partial}_{n ] } = { 1\over 2}\partial_r\partial^rw_{pq\;;\;[m}\;\stackrel{\leftarrow}{\partial}_{n ] } \;= 0\end{aligned}\ ] ] this means that : @xmath390 we will now prove that this is the only obstacle . in other words , if @xmath383 , then ( [ maxwellmaxwellexpansion ] ) is cohomologically trivial . let @xmath391 be the space of polynomial expressions of the form : @xmath392 } \;\ ; \mbox{\tt\small satisfying } \ ; \phi_{[mn } \stackrel{\leftarrow}{\partial}_{k ] } = 0 \;\ ; \mbox{\tt\small and } \ ; \phi_{mn}\stackrel{\leftarrow}{\partial}_n = 0\ ] ] let @xmath393 be the subspace of @xmath391 consisting of polynomials of the order @xmath209 in @xmath29 , _ i.e. _ @xmath394 . notice that such @xmath395 are automatically harmonic . the operator @xmath396 acts on such expressions , and is nilpotent : @xmath397 [ [ lemma ] ] lemma + + + + + @xmath398 indeed , @xmath399 is acyclic , as @xmath400 is the same as already computed in section [ sec : cohomologyinmaxwell ] cohomology of the translations algebra on the solutions of the maxwell equations , and it is zero for @xmath401 . this means that it is always possible to gauge away the term with the highest number of @xmath402 , and therefore the cohomology of @xmath396 is zero . the proof of @xmath403 is identical to the proof of @xmath404 after applying the hodge dual operation on the @xmath341 ghosts . ( [ ansatzmaxwellmaxwelleqm ] ) implies that the expression @xmath405}$ ] belongs to @xmath406 . ( [ ansatzmaxwellmaxwellunobstructeda ] ) implies that it is annihilated by @xmath407 . since @xmath403 , exists @xmath408 such that : @xmath409 } = \delta_l \left ( dx_l^p\wedge dx_l^q\;\phi^{(2)}_{pq\;;\;mn } \right)\ ] ] now suppose that @xmath383 . this implies that we can find @xmath410 , @xmath411 , @xmath412 and @xmath413 ( all elements of @xmath391 ) satisfying : @xmath414 } = ( \delta_l + d_l ) \left ( \phi^{(2)}_{mn } + \phi^{(4)}_{mn } + \phi^{(6)}_{mn } + \phi^{(8)}_{mn } + \phi^{(10)}_{mn } \right)\ ] ] ( here each @xmath415 is a polynomial of the degree @xmath416 in @xmath402 . ) indeed , as elements of @xmath391 are harmonic functions , @xmath417 implies @xmath418 and therefore the existence of @xmath410 such that @xmath419 . and so on until @xmath420 . since @xmath421 is a top form , exists @xmath422 such that @xmath423 . furthermore , @xmath424 , therefore exists @xmath425 such that @xmath426 . continuing , we get @xmath427 , @xmath428 and finally @xmath429 . let us denote : @xmath430 then we get : @xmath431 } = ( \delta_l + d_l ) \big ( \phi - ( \delta_l + d_l)\psi\big)\ ] ] notice that @xmath432 is of the form : @xmath433 this concludes the proof that the ansatz ( [ ansatzmaxwellmaxwell ] ) is trivial iff @xmath379\;;\;[mn ] } = 0 $ ] . [ [ case - n0 ] ] case @xmath434 + + + + + + + + + + + + + + + + + + + + + + the vanishing lemma ( [ vanishingofcohomologydplusdelta ] ) does not work in the case @xmath434 , in this case the cohomology of @xmath399 is given by the formulas of section [ sec : cohomologyinmaxwell ] with the replacement @xmath435 , @xmath436 . similarly , the cohomology of @xmath407 is obtained via the hodge duality . therefore , it is necessary to repeat the analysis taking into account this nontrivial cohomology . there is no obstacle to satisfy ( [ fromatophi2 ] ) , even if @xmath437 } = \mbox{const}$ ] , because there are no 11-forms and therefore the cohomology of @xmath407 vanishes on expressions which are monomials of the first order in @xmath402 . there are potential obstacles in completing the chain ( [ chainofphis ] ) . we will not do the analysis here , but just point out that by rotational symmetry , the potential obstacles are proportional to the following constant tensors : the total antisymmetrization and the contraction : @xmath438 } \label{totalantisymmc } \\ { \cal c}_n \;&= g^{pm}a_{p\;;\;[m}\stackrel{\leftarrow}{\partial}_{n ] } \label{contractionc}\end{aligned}\ ] ] consider the following ansatz for the leading term of the expasion in powers of @xmath133 : @xmath439 } dx_r^m\wedge dx_r^n\ ] ] where @xmath440 satisfies the maxwell equation @xmath441 . this is in the image of @xmath235 when exists @xmath5 such that : @xmath442 } \ ; = \ ; & \gamma^k\partial_k \phi_{[m}\stackrel{\leftarrow}{\partial}_{n ] } \label{diracmaxwelltrivialization}\\ \mbox{\small\tt and } \ ; & \phi \stackrel{\leftarrow}{\bf m } = 0 \nonumber\end{aligned}\ ] ] then it follows that @xmath443 } = 0 $ ] and therefore : @xmath444 } = 0\end{aligned}\ ] ] if @xmath440 does not satisfy this equation , then the trivialization ( [ diracmaxwelltrivialization ] ) is impossible . notice that @xmath441 , therefore @xmath445}$ ] is automatically annihilated by @xmath446 . but it is not necessarily annihilated by the left dirac operator . we conclude that @xmath447}$ ] is an obstacle for ( [ diracmaxwellansatz ] ) to be trivial . consider the following ansatz for the leading term : @xmath448 where @xmath440 satisfies the right dirac equation @xmath449 and also @xmath450 . this is trivial if exists @xmath451 such that : @xmath452 this implies that @xmath453 and therefore @xmath454 } = 0 $ ] . therefore @xmath454}$ ] is an obstacle for ( [ maxwelldiracansatz ] ) to be trivial . for the polynomials of nonzero degree this is the only obstacle . indeed , suppose that @xmath455 } = 0 $ ] . as the cohomology of dirac solutions at the nonzero degree is zero , this implies that : @xmath456 where @xmath457 satisfies the right dirac equation . the cohomology of @xmath458 on the solutions of the dirac equation is zero , therefore exists @xmath459 such that @xmath460 where @xmath461 satisfies the dirac equation . this implies ( [ trivialityofpsimaxwelldirac ] ) . the term @xmath462 is generated by two types of terms : @xmath463}\ ; dx_r^m\wedge dx_r^n \;+ \nonumber \\ + & c^pc^qdx_l^r\wedge dx_l^s \;b_{[pqrs]\,;\,[m}(x_r)\stackrel{\leftarrow}{\partial}_{n]}\ ; dx_r^m\wedge dx_r^n\end{aligned}\ ] ] where @xmath464}$ ] is a polynomial of degree @xmath331 in @xmath134 . under @xmath465 the first term cancels with the right hand side of eq . ( [ ansatzmaxwellmaxwellunobstructeda ] ) , because @xmath466 . the second term for non - constant @xmath467\,;\,[m}\stackrel{\leftarrow}{\partial}_{n]}$ ] cancels with the right hand side of eq . ( [ ansatzmaxwellmaxwellunobstructedb ] ) . the constant @xmath467\,;\,[m}\stackrel{\leftarrow}{\partial}_{n ] } = \mbox{const}$ ] ( _ i.e. _ @xmath468 ) generates @xmath469 ( because @xmath470 ; the @xmath229 acts as @xmath143 on maxwell solutions ) : @xmath471 we are now ready to compute the cohomology of @xmath472 . the corresponding part of @xmath473 consists of two parts : @xmath474 however , there is a nontrivial @xmath475 , which cancels the @xmath476 antisymmetric part of @xmath477 with @xmath478 . we are left with : @xmath479 these vertices are in one - to - one correspondence with the generators of the super - poincare algebra . the corresponding part of @xmath473 consists of three parts : @xmath480 we have already seen that @xmath478 gets killed by the @xmath229 : @xmath481 let us look at @xmath482 . we have : @xmath483 the interpretation is as follows : * @xmath484 corresponds to the linear dilaton and the `` asymmetric linear dilaton '' ( the nonphysical vertex of @xcite with constant @xmath485 ) * one copy of @xmath486 cancels under @xmath229 with @xmath487 * another copy of @xmath488 is the nsns @xmath355-field strength @xmath489 * two copies of @xmath490 are both unphysical this was computed in section [ sec : zerothcohomology ] . we identify @xmath492 as @xmath493 ( the gradient of the dilaton ) and @xmath485 is the unphysical state of @xcite . notice that eq . ( [ eqaminus ] ) implies that @xmath494 , _ i.e. _ @xmath485 is a linear function of @xmath29 . notice that eqs . ( [ phildef ] ) and ( [ phirdef ] ) define @xmath495 only up to a constant , and therefore @xmath202 is defined only up to a constant . this is because linear dilaton and linear asymmetric dilaton have already been counted in @xmath496 . [ [ conclusion-3 ] ] conclusion + + + + + + + + + + as expected , @xmath497 has the quantum numbers of the nsns sector of the linearized type iib sugra , modulo some zero mode subtleties . the symmetric part @xmath498 corresponds to the riemann curvature tensor @xmath499[mn]}$ ] , and the antisymmetric part @xmath500 to @xmath501\;[m } \stackrel{\leftarrow}{\partial}_{n]}$ ] . there are the following nonphysical states : [ cols="^ , < " , ] they have the quantum numbers of the adjoint representation of the super - poincare algebra . in the bosonic string theory , the nonphysical states were removed by imposing the constraint @xmath502 @xcite . this is probably possible also in the pure spinor approach , as the pure spinor @xmath4-ghost was constructed in the nonminimal formalism @xcite . but there is also another way of removing the nonphysical states , which we will now describe . as we discussed in the introduction , the brst closedness of the vertex operator is a necessary and sufficient condition for the corresponding deformation of the classical worldsheet action to have the classical brst invariance . however , at the one - loop level there is an anomaly which is cancelled by the fradkin - tseytlin term @xcite : @xmath503 here @xmath5 is the _ dilaton superfield_. the only place where @xmath5 enters is the fradkin - tseytlin term ( [ fradkintseytlinterm ] ) , which does not matter at the _ classical _ level . it is , in this sense , `` invisible '' in the classical theory . the condition of the one - loop brst invariance implies that @xmath5 is related to the `` visible '' superfields ( those which enter in the main part of the worldsheet action ) by some equations : @xmath504 where @xmath505 and @xmath506 on the right hand side are some function of the `` visible '' superfields . in this sense , @xmath5 is determined , unambiguously up to a constant , from the `` visible '' superfields . however , it turns out that for some classical backgrounds the equations ( [ dilatonleft ] ) and ( [ dilatonright ] ) are incompatible @xcite . such backgrounds , in our terminology , are _ nonphysical_. being perfectly consistent from the point of view of the classical worldsheet sigma - model , they however fail at the one - loop level . this is somewhat unusual , as the typical situation is that differential equations are `` generally speaking incompatible , but sometimes become compatible '' . here we have the opposite situation . equations ( [ dilatonleft ] ) and ( [ dilatonright ] ) for @xmath5 are compatible for the vast majority of backgrounds , but become incompatible on a finite - dimensional nonphysical component . in other words , * physical and non - physical deformations are `` mutually obstructed''*. + roughly speaking , this can be understood as follows . the compatibility conditions for equations ( [ dilatonleft ] ) and ( [ dilatonright ] ) include the equation : @xmath507 both @xmath505 and @xmath508 are defined in terms of other sugra fields , which already satisfy the sugra constraints . these constraints translate into some constraint on @xmath509 ( which is therefore automatically satisfied ) . surprizingly , that automatic constraint seems to be not @xmath510 but rather @xmath511 , _ i.e. _ the _ derivatives _ of @xmath509 being zero ( @xcite , _ cp . ( [ constantfantisymm ] ) and ( [ constantdoubletrace ] ) ) . in order to kill the nonphysical component , we just have to require that this constant is zero ; this is why the nonphysical component is finite - dimensional . we observe that the nonphysical operators seem to be in correspondence with the global symmetries . this should have a natural interpretation in terms of the action of the @xmath4-ghost : @xmath512 but , as we explained : * instead of imposing the condition @xmath513 , one can request the existence of the dilaton superfield @xmath5 notice that including @xmath5 also solves the following problem . our analysis , based on the naive brst cohomology , failed to identify the dilaton zero mode . but once we include @xmath5 , the dilaton is identified as the lowest component of @xmath5 , and in particular the zero mode of the dilaton is recovered . most of the ghost number three vertex operators transform in the same representation as ghost number two vertex operators . this is in line with the picture : @xmath514 notice that the map ( [ bfromthreetotwo ] ) lowers the polynomial degree of the vertex by 2 , as the @xmath4-ghost should . for example , in the dirac - dirac sector , the ghost number 3 vertex is of the form @xmath242 ; to produce the bispinor field we remove @xmath515 and then act with the left dirac operator : @xmath516 removing @xmath515 lowers the degree by one , and then @xmath517 again lowers the degree by one . let us look more carefully at the subtleties which arise when we consider polynomial vertices of low degree . this is @xmath488 . it cancels with part of @xmath482 see section [ sec : cale11 ] . this is @xmath519 . first of all , we have restrict to the kernel of @xmath520 . this kills one copy of @xmath521 . but also , we have to take a factorspace over the image of @xmath522 . this cancels another copy of @xmath523 against the ( [ constantfantisymm ] ) and one copy of @xmath524 against the ( [ constantdoubletrace ] ) . for example , @xmath525 cancels the diagonal @xmath523 as : @xmath526 and a similar computation shows that @xmath527 cancels a diagonal copy of @xmath524 . another copy of @xmath528 does not seem to cancel with anything : @xmath529 [ [ dirac - dirac - sector ] ] dirac - dirac sector + + + + + + + + + + + + + + + + + + there are the following obstacles to triviality : 1 . the bispinor @xmath531 , which satisfies both left and right dirac equations 2 . there is also a discrete state ( [ vertexghostnumberthreedimfour ] ) which corresponds to @xmath249 being a constant times the unit matrix [ [ maxwell - maxwell - sector-1 ] ] maxwell - maxwell sector + + + + + + + + + + + + + + + + + + + + + + there are the following obstacles to triviality : 1 . the double field strength @xmath532\;;\;[pq]}$ ] of section [ sec : doublefieldstrength ] 2 . if the double field strength is zero , then there are constant tensors @xmath533 and @xmath534 defined in ( [ totalantisymmc ] ) and ( [ contractionc ] ) first let us look at the double field strength . notice that the formulas of section [ sec : doublefieldstrength ] are almost identical to section ( [ sec : cale02 ] @xmath535 [ sec : zerothcohomology ] ) . the only difference is that the @xmath532\;;\;[pq]}$ ] of section [ sec : zerothcohomology ] is not required to be of the form ( [ definedoublefieldstrength ] ) with @xmath536 satisfying eq . ( [ ansatzmaxwellmaxwellunobstructeda ] ) . as the @xmath532\;;\;[pq]}$ ] of section [ sec : doublefieldstrength ] _ is _ required to be of such a form , it automatically satisfies : @xmath191\;;\;[pq]}(x ) \;&= { 1\over 2 } \partial_{[m } g^{np } a_{n]\;;\;[p}\stackrel{\leftarrow}{\partial}_{q]}\end{aligned}\ ] ] eq . ( [ ansatzmaxwellmaxwellunobstructeda ] ) implies the existence of @xmath5 such that @xmath537 . taking also into account eq . ( [ ansatzmaxwellmaxwelleqm ] ) , we get : @xmath538\;;\;[pq]}(x ) = \partial_m\partial_q \left[{1\over 8}\left(g^{pn}a_{p\;;\;n } - \phi\right)\right]\ ] ] this is the same equation as we got in section ( [ sec : zerothcohomology ] ) , except there is no unphysical @xmath485 . on the other hand , there are @xmath533 and @xmath534 defined in ( [ totalantisymmc ] ) and ( [ contractionc ] ) , which should be mapped by @xmath539 to @xmath540 and the dilaton gradient . also , there is the discrete state ( [ vertexghostnumberthreedimfour ] ) . notice that in our computation we missed the dilaton zero mode , as the corresponding vertex is probably a brst variation of something that is not annihilated by @xmath541 @xcite . it is possible that the discrete state ( [ vertexghostnumberthreedimfour ] ) mapped by @xmath542 to the dilaton zero mode . however , there is also another discrete state at the ghost number three : eq . ( [ discretestatefrome21 ] ) . therefore our computations seem to confirm eq . ( [ bfromthreetotwo ] ) , except that we see _ two _ ghost scalar ghost number three discrete states : eq . ( [ vertexghostnumberthreedimfour ] ) and eq . ( [ discretestatefrome21 ] ) . the term @xmath544 was computed in section [ sec : secondcohomologytensorproduct ] : @xmath471 it cancels with half of : @xmath545 ( and another half of @xmath546 then cancels with @xmath547 ) . this pattern persists for @xmath548 , giving the short exact sequences : @xmath549 \;\stackrel{d_2}{\longrightarrow}\ ; \nonumber \\ \;\stackrel{d_2}{\longrightarrow}\ ; & [ { \cal e}^{p+2,1 } = \lambda^{p+4}{\bf c}^{10}\oplus \lambda^{p+4}{\bf c}^{10 } ] \;\stackrel{d_2}{\longrightarrow}\ ; \nonumber \\ \;\stackrel{d_2}{\longrightarrow}\ ; & [ { \cal e}^{p+4,0 } = \lambda^{p+4}{\bf c}^{10}]\end{aligned}\ ] ] the term @xmath551 is nonzero , but it cancels with the @xmath229 of @xmath552 . the space @xmath554 is killed by the @xmath229 of @xmath555 . indeed , let us consider the following element of @xmath555 with constant @xmath556}$ ] : @xmath557}\ ; dx_r^m\wedge dx_r^n\ ] ] ( this is a particular case of ( [ ansatzmaxwellmaxwell ] ) with zero @xmath339 and constant @xmath355 . ) being an element of @xmath555 , this is a @xmath341-dependent element of the cohomology of @xmath146 , parametrized by a left times right field strength . we need to act on this by the @xmath375 . for that , we need to know the actual ( @xmath341-dependent ) vertex , which is built using the left and right _ vector potentials _ , _ i.e. _ @xmath558}\ ; x_r^m ( \theta_r\gamma^n\lambda_r)$ ] . the @xmath143 on the vertex is not zero : @xmath559 } x_r^m(\theta_r\gamma^n\lambda_r ) - c^p x_l^q ( \theta_l\gamma^r\lambda_l ) b_{pqr\;;\;[m}\stackrel{\leftarrow}{\partial}_{n ] } c^m(\theta_r\gamma^n\lambda_r)\ ] ] but is a pure gauge , namely it is @xmath146 of : @xmath560 } x_r^m ( \theta_r\gamma^n\lambda_r ) + c^p x_l^q ( \theta_l\gamma^r\lambda_r ) b_{pqr\;;\;[m}\stackrel{\leftarrow}{\partial}_{n ] } c^m x_r^n\ ] ] and the @xmath143 of this gives : @xmath561}x_r^m(\theta_r\gamma^n\lambda_r ) + c^p x_l^q ( \theta_l\gamma^r\lambda_r ) b_{pqr\;;\;[m}\stackrel{\leftarrow}{\partial}_{n ] } c^m c^n + \\ + c^pc^q x_l^r b_{pqr\;;\;[m}\stackrel{\leftarrow}{\partial}_{n ] } c^m ( \theta_r\gamma^n\lambda_r ) - c^p c^q ( \theta_l\gamma^r\lambda_l ) b_{pqr\;;\;[m}\stackrel{\leftarrow}{\partial}_{n ] } c^m x_r^n\end{aligned}\ ] ] the second row is @xmath562 . and the first row is equivalent , in the maxwell cohomology , to the expression : @xmath563 } ( dx_r^m\wedge dx_r^n - dx_l^m\wedge dx_l^n)\ ] ] where @xmath373}$ ] is defined in ( [ defbpqrmn ] ) this can be used to kill any class of the form : @xmath564}(dx_r^s\wedge dx_r^t - dx_l^s\wedge dx_l^t)\ ] ] in @xmath565 . the classes of the form : @xmath566}(dx_r^s\wedge dx_r^t + dx_l^s\wedge dx_l^t)\ ] ] are not in the image of @xmath229 . however , the @xmath229 of them is nonzero , giving an element of @xmath567 of the form @xmath568 . we conclude that @xmath569 . in this section we will study the action of the supersymmetry on the ghost number three vertices . we will first act by the left supersymmetry on the element of the maxwell - dirac sector , an see that the result is some element of the dirac - dirac sector . then we will act my the left supersymmetry on the dirac - dirac sector , which will bring us back to the maxwell - dirac sector . we will verify that the anticommutator of two supersymmetries is a translation . let us consider an element of the maxwell - dirac sector , of the following form : @xmath570 where @xmath325 stand for elements of the lower degree in @xmath134 ( which have dependence on @xmath133 ) . let us act on it by the _ left _ supersymmetry with the parameter @xmath571 , which we will call @xmath572 . to evaluate the action of this supersymmetry , we will use the formulas from section 6.1.3 of @xcite ( where @xmath572 was denoted @xmath573 , and @xmath571 was called @xmath574 ) . we get the following element of the dirac - dirac sector : @xmath575\epsilon\;\psi_m + \ldots\ ] ] we observe : @xmath576\epsilon\;\psi_m \right ) = \gamma^j{\partial\over\partial x_l^j}\left ( { 9\over 10 } \hat{x}_l\gamma^m\epsilon\psi_m \right)\ ] ] this implies that ( [ susyonpsigivesprer ] ) gives the same cohomology class as : @xmath577 in notations of section [ sec : definer ] we have : @xmath578 the obstacle to the triviality is : @xmath579}(x_r)\ ] ] ( this is a bispinor : @xmath580}(x_r))^{\dot{\beta}}$ ] ) we want to calculate the action of the supersymmetry with the parameter @xmath581 on the class : @xmath582 this is a bit tricky , becase the leading term @xmath583 does not contribute , and we have to analyze the subleading term proportional to @xmath133 : @xmath584 where @xmath325 stand for terms of the higher order in @xmath133 . again , we use the formulas from @xcite . when acting by the supersymmetry with the parameter @xmath581 , we are getting the following element of the maxwell - dirac sector : @xmath585}r(x_r ) - { 1\over 6}\gamma_{m]}c^k\partial_kr(x_r ) + { 5\over 6}c_{m]}\gamma^k\partial_kr(x_r ) \right)dx_l^n\wedge dx_l^m\end{aligned}\ ] ] this can be written as follows : @xmath586k } - { 5\over 6 } \gamma_{k[n}\delta^l_{m ] } - { 1\over 6 } \gamma_{nm}\delta^l_k\end{aligned}\ ] ] we can add @xmath587 then we get : @xmath588k } - { 5\over 6 } \gamma_{k[n}\delta^l_{m ] } + { 5\over 6 } \gamma_{nm}\delta^l_k\end{aligned}\ ] ] now @xmath589}=0 $ ] and @xmath590 . consider the following tensor : @xmath591k } - { 5\over 6 } \gamma_{k[n}\delta^l_{m ] } + { 5\over 6 } \gamma_{nm}\delta^l_k\ ] ] it satisfies @xmath592}=0 $ ] and @xmath593 . therefore the cohomology class does not change if we replace @xmath594 as follows : @xmath595k}\ ] ] indeed : @xmath596 we conclude that the supersymmetry with the parameter @xmath581 brings @xmath597 to @xmath598 . when @xmath249 is given by ( [ susyonpsigivesr ] ) , we get : @xmath599 this is in agreement with the fact that the anticommutator of two susy transformations is a translation . ghost number three vertices transform in the linearized type iib supergravity supermultiplet . we would like to thank nathan berkovits for useful discussions . this work was supported in part by the ministry of education and science of the russian federation under the project 14.740.11.0347 `` integrable and algebro - geometric structures in string theory and quantum field theory '' , and in part by the rfbr grant 15 - 01 - 99504 `` string theory and integrable systems '' . o. chandia , a. mikhailov , and b. c. vallilo , _ a construction of integrated vertex operator in the pure spinor sigma - model in @xmath600 _ , _ jhep _ * 1311 * ( 2013 ) 124 doi : * 10.1007/jhep11(2013)124 * [ arxiv/1306.0145 ] . o. a. bedoya , l. bevilaqua , a. mikhailov , and v. o. rivelles , _ notes on beta - deformations of the pure spinor superstring in ads(5 ) x s(5 ) _ , _ nucl.phys . _ * b848 * ( 2011 ) 155215 doi : * 10.1016/j.nuclphysb.2011.02.012 * [ arxiv/1005.0049 ] .

the influence of aperiodic ( dis-)order on the thermodynamic properties and the critical behaviour of spin systems has been an active research area , particularly since the discovery of quasicrystals in 1984 . in view of the many articles existing , we can not give proper credit to all contributors ; instead , we would like to refer the reader to a recent review @xcite which contains a rather complete bibliography on this subject . for ferromagnetic ising systems , heuristic scaling arguments , put forward by luck @xcite , lead to qualitative predictions on the relevance of disorder implemented through an aperiodic variation of the coupling constants which go beyond the original relevance - irrelevance criterion of harris @xcite . for the ising quantum chain ( iqc ) , or the equivalent 2@xmath2 classical ising model @xcite with layered disorder , this relevance criterion reads as follows : the critical behaviour is of onsager - type as long as the fluctuations in the sequence of coupling constants are bounded , whereas for unbounded fluctuations the critical singularities resemble those of the randomly disordered case . this behaviour had first been conjectured by tracy @xcite on ths basis of his results for two different examples of three - letter substitution rules . of particular interest is the marginal case with logarithmically ( in the system size ) diverging fluctuations . here , the critical exponents are predicted to depend continuously on the coupling constants @xcite . meanwhile , some exact results were obtained for substitution chains with marginal fluctuations of the couplings . for a number of substitution chains , the surface magnetisation critical exponent could be calculated @xcite ; and , recently , the correlation length critical exponent was derived for two special substitution rules using an exact renormalisation approach @xcite . all these results are in accordance with luck s predictions . the real space renormalisation method we use in this article is based on the decimation process introduced in ref . our setup is quite different , though ; and in particular we are going to show how _ arbitrary _ substitution rules can be analysed in terms of @xmath3-matrices and their star - product formalism . in this way , the dependence of the critical behaviour on the nature of the fluctuations can be derived explicitly . the analysis is carried out in detail for two - letter substitution chains , with the result that luck s relevance criterion emerges from the renormalisation relations in a natural way . an extended renormalisation scheme for the most general case of @xmath4-letter substitution rules is presented in an appendix and discussed for certain examples . for infinite classes of substitution chains with marginal fluctuations , including the two - letter chains , an analytic expression for the correlation length critical exponent is given . this article is organised as follows . in sec . 2 , we give a brief outline of the model , including some introductory remarks on substitution rules and on properties of the corresponding substitution sequences . then , in sec . 3 , we introduce the renormalisation transformation for a restricted class of @xmath4-letter substitution chains , comprising all two - letter substitution chains . the dependence of the renormalisation flow on the fluctuations of the couplings is shown and the consequences are discussed in detail . subsequently , in sec . 4 , we summarise the implications of our results for the critical properties of the iqc for the three different types of fluctuations in the sequence of the coupling constants . our conclusions and an outlook on future developments are presented in sec . the paper contains two appendices related to the content of sec . 3 : in appendix a , the solution of the eigenvalue problem for the case of a two - letter substitution sequence is derived ; and finally , generalising the discussion of sec . 3 , the renormalisation transformation for arbitrary @xmath4-letter substitution rules is explained in appendix b. the iqc in a transverse magnetic field is defined by the hamiltonian @xmath5 acting on the tensor - product space @xmath6 . here , the @xmath7 denote the transverse fields at the sites . the @xmath8 are site - dependent coupling constants , and the operators @xmath9 denote pauli s matrices acting on the @xmath10th site . in what follows , we consider closed chains with _ periodic _ boundary conditions which are defined by @xmath11 ; various other kinds of boundary conditions can be chosen without changing the key results . for a general set of coupling constants @xmath8 and fields @xmath7 , the iqc can be written as a free - fermion model via a jordan - wigner transformation and can then be diagonalised by a bogoljubov - valatin transformation , resulting in @xcite @xmath12 where @xmath13 , @xmath14 are @xmath15 fermionic creation and annihilation operators and @xmath16 is some constant . the dimensionless excitation energies @xmath17 ( which can be chosen to be positive and ordered , i.e. , @xmath18 , while properly adjusting the constant @xmath16 ) satisfy the linear difference equations @xmath19 normally , one proceeds by eliminating either @xmath20 or @xmath21 . in this way , the problem of diagonalising the @xmath22 matrix @xmath23 can be reduced to the diagonalisation of an @xmath24 matrix . from the set of @xmath15 eigenvalues ( fermion frequencies ) @xmath25 of the latter , the total spectrum of the complete hamiltonian @xmath23 is recovered by considering the sums of the elements of any of its @xmath26 subsets . for periodic boundary conditions , the situation is in fact a little more complicated ; in order to avoid the appearance of a non - local number operator in the fermionic hamiltonian one has to consider so - called _ mixed sector _ hamiltonians . these are obtained by combining the _ even _ sector ( with respect to the operator @xmath27 that commutes with @xmath28 ) of the hamiltonian ( [ eq : qc ] ) with _ periodic _ boundary conditions with the _ odd _ sector of the same hamiltonian with _ antiperiodic _ boundary conditions ( defined by @xmath29 ) , and vice versa @xcite . the iqc is equivalent to the 2@xmath2 classical ising model with a layered modulation of the interactions @xcite . the correspondence shows up through an extremely anisotropic limit ( also called the @xmath30-continuum limit @xcite ) of the coupling constants of the latter , which is believed not to alter the critical behaviour . explicitly , one establishes the following relations : the _ temperature _ @xmath31 of the classical model corresponds to the _ transverse field _ @xmath32 of the quantum chain , the _ free energy _ translates into the _ ground - state energy _ and the counterpart of the _ correlation length @xmath33 parallel _ to the layers is the _ inverse mass gap _ @xmath34 of the iqc . in our setting , the mass gap is just given by the modulus of the smallest fermionic excitation @xmath35 of ( [ ffh ] ) . according to finite - size scaling , we thus expect a scaling behaviour of @xmath35 at the critical point as @xmath36 provided the correlation length scales with a power law at all . hereby , the scaling exponent @xmath37 translates into the correlation length critical exponent @xmath38 parallel to the layers in terminology of the 2@xmath2 statistical system , while @xmath39 retains its unperturbed onsager value @xmath40 . the hamiltonian ( [ eq : qc ] ) is _ critical _ ( in the sense that the energy gap in the excitation spectrum vanishes , i.e. , @xmath41 ) in the thermodynamic limit if the _ geometric mean _ ( rather than the arithmetic mean ) of the couplings , normalised by the variables of the transverse field , is equal to one , i.e. , @xmath42 compare ref . the repetitive structure of the substitution sequences we are going to consider guarantees that the mean is well - defined @xcite . let us now assume that the coupling constants @xmath43 ( @xmath44 ) are chosen from a set of @xmath4 possible values @xmath45 ( @xmath46 ) , the value @xmath45 occuring with a frequency @xmath47 ( where @xmath48 ) . for simplicity , we take the field to be constant , i.e. , @xmath49 . then , condition ( [ eq : critcond ] ) can be fulfilled by choosing @xmath4 positive real numbers @xmath50 with @xmath51 , and setting the couplings to @xmath52 where @xmath53 . unless stated otherwise , this parametrisation of the critical surface will be used in the sequel . let us now consider coupling constants @xmath8 drawn from a set of values @xmath54 , where the label @xmath55 runs over the letters of a ( finite ) @xmath4-letter alphabet @xmath56 . we choose the couplings @xmath8 according to sequences generated by iterated application of a substitution rule @xmath57 on @xmath56 . for the introduction of the concept and some notation , consider a simple example on a two - letter alphabet @xmath58 : @xmath59 we start with an initial word , e.g. @xmath60 , and define @xmath61 . @xmath62 denotes the corresponding substitution matrix ( whose elements count the number of letters @xmath63 and @xmath64 in the words @xmath65 and @xmath66 ) , which has the eigenvalues @xmath67 . the entries of the , statistically normalised , ( right ) eigenvector @xmath68 , corresponding to the perron - frobenius eigenvalue @xmath69 , give the frequencies of the letters in the limit word @xmath70 . in order to ensure that , starting from any seed , the letter frequencies converge to unique , positive values in the limit chain , we restrict ourselves to _ primitive _ substitution rules @xmath71 throughout this paper . for the corresponding substitution matrices @xmath72 , this means that the entries of @xmath73 are _ strictly positive _ for any integer @xmath74 ( and a suitably chosen finite @xmath75 ) . this is obviously true for the examples ( [ eq : sub ] ) . associated to a word @xmath76 of length ( i.e. , number of letters ) @xmath77 , one obtains finite ( closed ) quantum chains with @xmath15 sites by setting @xmath78 whenever the @xmath10th letter of the word @xmath76 is an @xmath63 , and @xmath79 otherwise , the final letter of @xmath76 determining the coupling between the last and the first spin in the chain . while @xmath80 , the largest eigenvalue of the substitution matrix , describes the asymptotic scaling of the chain length with the number of iterations @xmath81 , @xmath82 , the _ fluctuation _ behaviour of the exchange couplings is connected to the second - largest eigenvalue @xmath83 ( where it is the _ modulus _ of the eigenvalue that matters , hence by `` second - largest eigenvalue '' we always refer to the eigenvalue with the second - largest absolute value ) . in order to quantify the fluctuations , let us define @xmath84 where @xmath85 is the average coupling in the limit chain . fluctuations stay _ bounded _ as long as @xmath86 , but for @xmath87 they grow with iterated substitutions as a power of the chain length , @xmath88 , governed by the so - called _ wandering exponent _ @xmath89 that is given by @xmath90 in the limiting case @xmath91 , @xmath92 stays constant when we prolong the chain by complete iteration steps . generically , however , finite fluctuations within the substitutes @xmath93 of the letters @xmath94 will increase the total fluctuation in the vicinity of these points by a certain amount @xmath95 : @xmath96 for some word @xmath97 with @xmath98 and fixed @xmath99 . by repeated use of this argument , we thus expect fluctuations that _ diverge logarithmically _ with the length of the chain ( since the length grows exponentially with the number of iterations ) , see also ref . @xcite . on the other hand , the same argument also shows that , in this case , the fluctuation behaviour is not completely determined by the _ substitution matrix _ , but may also depend on the details of the actual _ substitution rule _ @xcite ; and for special choices the fluctuations may still stay bounded even though the second - largest eigenvalue is @xmath91 . coming back to our examples ( [ eq : sub ] ) , let us concentrate on the three cases @xmath100 ( which yields the famous _ fibonacci sequence _ ) , @xmath101 ( the _ period - doubling sequence _ ) , and @xmath102 ( the _ binary non - pisot sequence _ ) . as the second - largest eigenvalues of the substitution matrices @xmath62 fulfill @xmath103 , @xmath104 , and @xmath105 , this includes an example of bounded fluctuations ( @xmath100 ) , a marginal case with logarithmically divergent fluctuations ( @xmath101 ) and an example of strong fluctuations ( @xmath102 ) , hence all three interesting cases are illustrated @xcite . another substitution chain that is referred to several times in the text is the celebrated _ thue - morse sequence _ @xmath106 rather than reducing the set of linear equations ( [ fermeq ] ) to @xmath15 dimensions , we proceed ( as does ref . @xcite ) with a direct treatment as a @xmath107-dimensional eigenvalue problem of the real symmetric matrix @xmath108{$h_{\scriptscriptstyle n-1}^{}$ } & 0&\makebox[0pt]{$\varepsilon_{\scriptscriptstyle n-1}^{}$}&0\\ 0&0&\cdots&0&0&\makebox[0pt]{$\varepsilon_{\scriptscriptstyle n-1}^{}$ } & 0&h_n^{}\\ \varepsilon_n^{}&0&\cdots&0&0&0&h_n^{}&0 \end{array}\right)\ ; .\ ] ] since @xmath109 is symmetric and bipartite after an appropriate reordering of the basis . ] , the eigenvalues occur in real pairs of opposite sign . but changing the sign of an eigenvalue simply means interchanging the corresponding creation and annihilation operators in eq . ( [ ffh ] ) , thus we can restrict ourselves to the positive part of the spectrum . in what follows , we concentrate on the case of a constant field @xmath110 , where we may set @xmath111 by an appropriate choice of the energy scale . of course , one can not possibly expect analytical results for an arbitrary sequence of couplings . however , for _ substitution _ sequences , more can be said using an exact renormalisation scheme , and , for an interesting class of chains with marginal fluctuations , the scaling exponent @xmath37 ( [ eq : scal ] ) can be determined exactly and given in a closed form . consider now sequences of couplings chosen according to a substitution rule @xmath112 on an @xmath4-letter alphabet @xmath113 . here , the @xmath114 are @xmath4 arbitrary words of finite length in the alphabet @xmath56 . indeed , for the case of _ aperiodic two - letter _ substitution chains , this form imposes no restriction at all , since these can all be generated by substitution rules of the form ( [ subs ] ) . to see this , note that any two - letter substitution rule @xmath115 can be transformed appropriately by inner automorphisms , i.e. , by replacing @xmath116 [ where a natural choice of the word @xmath117 is the common beginning , if any , of the words @xmath118 and @xmath119 , and/or by considering @xmath120 instead . obviously , this does not alter the limit chain . note that this is not possible for certain _ periodic _ chains , namely those which are generated by a substitution rule that replaces all letters by the same word ( the period ) , or by a multiple of it . it is no problem , however , to treat this case separately . the eigenvalue problem for the matrix @xmath109 ( [ fh ] ) can be reformulated using transfer matrices in the usual way . our renormalisation transformation ( rt ) is , however , most conveniently described in terms of the star - product formalism , borrowed from scattering theory @xcite . we therefore introduce new @xmath3-transfer matrices , called @xmath3-matrices from now on , by @xmath121 with @xmath122 where the star - product ( ) of two @xmath123 matrices is defined as @xmath124 in our case , according to eq . ( [ fh ] ) , the elementary @xmath3-matrices are @xmath125 - \varepsilon_{j+1}^{-1 } h & \varepsilon_{j+1}^{-1 } \lambda \end{array } \right ) \ ; , \ ] ] where @xmath126 denotes an eigenvalue may be _ any _ eigenvalue of @xmath109 . only later , after expanding the renormalisation equations in powers of @xmath126 [ eq . ( [ hren ] ) and below ] , we shall restrict ourselves to eigenvalues that vanish in the limit of an infinite system size . ] of @xmath109 . in order to establish the renormalisation scheme , we introduce a number of additional parameters for these @xmath3-matrices . for each letter @xmath127 , we define two asymmetry - coefficients @xmath128 and a field variable @xmath129 that enter the @xmath130 different elementary @xmath3-matrices through ) and ( [ fh ] ) in that the field variables are now labeled in the same way as the coupling constants ( i.e. , as the bonds ) rather then by the sites of the chain . ] @xmath131 the rt now just _ reverses _ the substitution steps . in our framework , this is simply done by building -products of the @xmath3-transfer matrices that correspond to the words @xmath114 given by the substitution rule ( [ subs ] ) . schematically , this works in the following fashion @xmath132 \cdots & \mbox{\large { \mbox{\boldmath $ a_i^{}$ } } } & \mbox{\large { \mbox{\boldmath $ a_j^{}$ } } } & \mbox{\large { \mbox{\boldmath $ a_k^{}$ } } } & \cdots \\ & \multicolumn{3}{c } { \begin{picture}(50,30)(0,0 ) \thicklines \put ( 0,20){\oval(100,35)[br ] } \put ( 50,20){\vector(0,1){5 } } \end{picture } \hphantom{\mbox{\large $ a$ } } \begin{picture}(116,30)(0,0 ) \thicklines \put ( 0,20){\line(0,1){5 } } \put ( 58,20){\oval(116,35)[b ] } \put(116,20){\vector(0,1){5 } } \put ( 58,15){\makebox[0pt]{\large $ \tilde{s}_{a_i^{}|a_j^{}}$ } } \end{picture } \hphantom{\mbox{\large $ a$ } } \begin{picture}(116,30)(0,0 ) \thicklines \put ( 0,20){\line(0,1){5 } } \put ( 58,20){\oval(116,35)[b ] } \put(116,20){\vector(0,1){5 } } \put ( 58,15){\makebox[0pt]{\large $ \tilde{s}_{a_j^{}|a_k^{}}$ } } \end{picture } \hphantom{\mbox{\large $ a$ } } \begin{picture}(50,30)(0,0 ) \thicklines \put ( 0,20){\line(0,1){5 } } \put ( 50,20){\oval(100,35)[bl ] } \end{picture } } & \end{array}\ ] ] where @xmath133 denotes the @xmath134th letter of the word @xmath114 . in this way , we obtain renormalised @xmath3-matrices @xmath135 - \varepsilon_{a_j}^{-1 } \tilde{h}_{a_i } & \varepsilon_{a_j}^{-1 } \tilde{\kappa}_{a_i}^-\tilde{\lambda } \end{array } \right ) \ ; , \ ] ] compare eq . ( [ sj ] ) . as an _ additional condition _ on the rt , and in accordance with their initial values , we keep the product of the @xmath136 asymmetry - coefficients @xmath137 fixed : @xmath138 now , the rt for the different parameters can be read off directly from the -product relations . explicitly , we obtain the following recursion relations @xmath139[r]{$\displaystyle h^{(1)}_{a_i}$ } = \makebox[10ex][l]{$\displaystyle h_{a_i}^ { } \ ; , $ } \qquad \makebox [ 5ex][r]{$\displaystyle h^{(k+1)}_{a_i}$ } = \makebox[25ex][l]{$\displaystyle \frac{-\varepsilon_{w_i^k } h_{w_i^k } h^{(k)}_{a_i } } { \varepsilon_{w_i^k}^2 -\lambda_{w_i^k+}^ { } \lambda_{a_i-}^{(k)}}\ ; , $ } \qquad \makebox [ 5ex][r]{$\displaystyle\tilde{h}_{a_i}^{}$ } = \makebox[15ex][l]{$\displaystyle h^{(|w_i|+1)}_{a_i}\ ; ; $ } & \\[1 mm ] \label{rg2 } & \makebox [ 5ex][r]{$\displaystyle\lambda_{a_i-}^{(1)}$ } = \makebox[10ex][l]{$\displaystyle\lambda_{a_i-}^{}\ ; , $ } \qquad \makebox [ 5ex][r]{$\displaystyle\lambda_{a_i-}^{(k+1)}$ } = \makebox[25ex][l]{$\displaystyle\lambda_{w_i^k-}^ { } + \frac{h_{w_i^k}^2\lambda_{a_i-}^{(k ) } } { \varepsilon_{w_i^k}^2 - \lambda_{w_i^k+}^ { } \lambda_{a_i-}^{(k)}}\ ; , $ } \qquad \makebox [ 5ex][r]{$\displaystyle\tilde{\lambda}_{a_i-}^{}$ } = \makebox[15ex][l]{$\displaystyle\lambda_{a_i-}^{(|w_i|+1)}\ ; ; $ } & \\[1 mm ] \label{rg3 } & \makebox [ 5ex][r]{$\displaystyle\lambda_{a_i+}^{(1)}$ } = \makebox[10ex][l]{$\displaystyle\lambda_{a_i+}^{}\ ; , $ } \qquad \makebox [ 5ex][r]{$\displaystyle\lambda_{a_i+}^{(k+1)}$ } = \makebox[25ex][l]{$\displaystyle\lambda_{a_i+}^{(k ) } + \frac{\lambda_{w_i^k+}^ { } \left[h^{(k)}_{a_i}\right]^2 } { \varepsilon_{w_i^k}^2 - \lambda_{w_i^k+}^ { } \lambda_{a_i-}^{(k)}}\ ; , $ } \qquad \makebox [ 5ex][r]{$\displaystyle\tilde{\lambda}_{a_i+}^{}$ } = \makebox[15ex][l]{$\displaystyle\lambda_{a_i+}^{(|w_i|+1)}\ ; ; $ } & \end{aligned}\ ] ] for the field variables @xmath140 and for @xmath141 that are defined as @xmath142 using eq . ( [ kappacond ] ) , we obtain the renormalised fermion frequencies @xmath143 as @xmath144 note the following properties of our renormalisation approach : * the rt is _ exact _ , since no new parameters are introduced in a renormalisation step . altogether , there are @xmath145 parameters for @xmath4 different couplings : effectively @xmath146 asymmetry - parameters @xmath147 , the @xmath4 fields @xmath129 , and the eigenvalue @xmath126 . note that _ the coupling constants are not renormalised _ in our scheme . * the rescaling factor of the system size is just given by the leading eigenvalue @xmath80 of the substitution matrix corresponding to the substitution rule under consideration . * the critical point of the model , defined by the vanishing of the mass gap ( @xmath148 ) , corresponds to a fixed point of the rt for @xmath126 , as can easily be seen from eqs . ( [ rg2 ] ) and ( [ rg3 ] ) . * the scaling behaviour in the vicinity of the critical point is governed by the leading - order term of the rt as @xmath149 . we now proceed to establish the rt of the spectrum to leading order for @xmath150 explicitly . this is most easily done by noting that , in the product ( [ sren ] ) , the common prefactors @xmath151 of the -products ( [ star ] ) gives rise to higher - order corrections only ; and therefore , for the present purpose , the denominator can be neglected completely . consider first the renormalisation of the field variables . we obtain @xmath152 instead of working directly with the rt for the fields , it is advantageous to consider new variables @xmath153 to linear order in @xmath126 , the rt of @xmath154 then yields the simple matrix form @xmath155 where @xmath156 is just the _ transpose _ of the substitution matrix of @xmath71 . now , take a look at the initial conditions @xmath157}$ ] for @xmath158 which are @xmath159 } = - 2\log \varepsilon_{a_i}$ ] . as stated above , the entries of the ( statistically normalised ) perron - frobenius eigenvector @xmath68 are the asymptotic letter frequencies @xmath160 . we thus obtain @xmath161 = 0\ ] ] for an _ arbitrary _ set of couplings fulfilling the criticality condition ( [ eq : critcond ] ) . hence , @xmath157}$ ] is nothing but an arbitrary ( real ) linear combination of the ( right ) eigenvectors of @xmath156 ( hence of the _ left _ eigenvectors of @xmath162 ) _ excluding _ the perron - frobenius one . thus , for a generic choice of critical couplings , it scales with the second - largest eigenvalue , @xmath163}\| \sim |\lambda_2|^k$ ] , under the iteration @xmath164}={\mbox{\boldmath $ m$}}_{\varrho}^t { { \mbox{\boldmath $ y$}}}^{[k]}$ ] . translated into the terminology of the renormalisation group , we recognise @xmath158 as a _ scaling field _ transforming according to @xmath165 where the corresponding _ renormalisation group eigenvalue _ is just the wandering exponent @xmath89 defined in eq . ( [ beta ] ) . this demonstrates explicitly that power law , logarithmic , or bounded fluctuations lead to relevant , marginal , or irrelevant scaling fields , respectively . we are going to describe below how this affects the critical behaviour . we now proceed to the rt for @xmath166 . defining @xmath167 \prod_{\ell = 1}^{|w_i| } \frac{h_{w_i^\ell}^2}{\varepsilon_{w_i^\ell}^2 } \ ; , \end{aligned}\ ] ] we find , by induction , as the linear part of the rt of @xmath166 @xmath168 setting @xmath169 , this can also be written in a matrix form @xmath170 since all the components of the matrices @xmath171 and of the vectors @xmath172 are positive , the vectors converge to the perron - frobenius eigenvectors of @xmath171 under iteration of the rt . let @xmath173 denote the perron - frobenius eigenvalues of @xmath171 . from eq . ( [ kappacond2 ] ) , we conclude that @xmath174 for further convenience , we transform @xmath175 ( under conservation of the spectrum ) according to @xmath176 with a diagonal transformation matrix @xmath177 . as new entries , we find , dropping the index @xmath31 , @xmath178 this form shows explicitly that , to linear order in @xmath126 , only the _ reduced couplings _ @xmath179 enter in the rt . finally , since the rescaling factor of the chain length is the largest eigenvalue @xmath80 of the substitution matrix , we formally obtain the scaling exponent @xmath37 through @xmath180 let us now look at the different cases of fluctuation behaviour in more detail . because @xmath181 is perpendicular to the perron - frobenius eigenvector of the substitution matrix , and since all other eigenvalues are smaller than one in modulus , we find @xmath163}\|\to 0 $ ] for all substitution chains with bounded fluctuations , hence @xmath182 . evaluating the transformation matrices @xmath171 at this fixed point of the fields , we obtain from eqs . ( [ pminus ] ) and ( [ pplus ] ) @xmath183 and hence @xmath184 thus we have @xmath185 for substitution chains with bounded fluctuations , and the low - energy spectrum at criticality scales as @xmath186 hereby , the @xmath126-transformation induces a finite renormalisation of the _ fermion velocity _ @xmath187 in comparison to the uniform chain @xmath188}/\lambda}\ ] ] where @xmath189}$ ] is the image of @xmath190}=\lambda$ ] after a @xmath134-fold application of the rt . in general , all renormalisation steps contribute in eq . ( [ vren ] ) . however , if _ all _ eigenvalues of the substitution matrix but the perron - frobenius one are zero , the fixed point of the rt is already reached after the first renormalisation step and @xmath187 is obtained immediately . this is shown below for periodic chains , but also allows to calculate the fermion velocity of the thue - morse and related chains . using the equidistribution property of quasiperiodic _ cut - and - project _ chains , like our first example ( [ eq : sub ] ) with @xmath100 ( the fibonacci sequence ) , the limit in eq . ( [ vren ] ) can be carried through explicitly for these cases , too . since this has already been done in ref . @xcite , we do not repeat the arguments here , and merely cite the result @xmath191 for two couplings parametrised as in eq . ( [ eq : para ] ) . in this short part , we briefly show how the known results for the fermion velocity @xmath187 for _ periodic _ quantum chains can be recovered easily within our renormalisation scheme . by a substitution rule of the form @xmath192 _ any _ periodic chain with period @xmath193 can be generated . in this simple situation , we can take all the fields @xmath194 and asymmetry - parameters @xmath195 to be equal . by virtue of the criticality condition ( [ eq : critcond ] ) , the reduced couplings transform as @xmath196 [ see eq . ( [ hren ] ) ] , and we obtain the ( fermionic ) hamiltonian of a _ uniform _ chain ( @xmath197 ) after a single renormalisation step . note that the additional asymmetry - parameters result in an irrelevant similarity transformation of the matrix ( [ fh ] ) only . the renormalisation factor for @xmath126 can be deduced from eqs . ( [ pminus ] ) and ( [ pplus ] ) , explicitly it reads ( using eq . ( [ eq : critcond ] ) , and normalising to @xmath198 ) : @xmath199 \label{v } & = & \sum_{k=1}^{|w|}\sum_{l=1}^{|w| } \varepsilon_k^2 \cdots \varepsilon_{k+l-1({\rm mod}\,|w|)}^2 \end{aligned}\ ] ] with @xmath200 . the critical scaling of the fermion frequencies now results in @xmath201 where @xmath187 is given in eq . ( [ v ] ) , and where the well - known scaling behaviour of the uniform chain is recovered for @xmath202 . this result was first derived in the context of the 2@xmath2 classical ising model with layered periodicity in ref . @xcite . in general , there is no closed expression for the fermion velocity @xmath187 . for the special case of a uniform chain of @xmath203-couplings with periodic @xmath204-defects on every @xmath81th site , we derive @xmath205}{\sinh [ \log ( r)]/m } \ ; .\ ] ] note that @xmath206 for @xmath207 ( simple periodic chain ) , but also @xmath208 for @xmath209 since the parametrisation ( [ eq : para ] ) implies that @xmath210 in this limit . we now consider the case where the second - largest eigenvalue(s ) of the substitution matrix ( and of its transpose ) have @xmath211 . let @xmath212 denote the corresponding @xmath81-dimensional ( joined ) eigenspace of @xmath213 , where @xmath214 . by eqs . ( [ y1 ] ) and ( [ y2 ] ) , @xmath215}$ ] converges to the projection of @xmath216}$ ] on @xmath212 under iteration of the rt . the critical surface being @xmath217-dimensional , we can parametrise @xmath218}$ ] in the renormalisation limit as @xmath219 with @xmath81 vectors @xmath220 that span @xmath212 and coefficients @xmath221 , @xmath222 . this implies @xmath223 and , if the vector components @xmath224 are chosen as integers , the entries of the matrices @xmath171 are polynomials with positive coefficients of @xmath81 parameters @xmath225 . if all @xmath226 are equal to 1 , we obviously obtain @xmath227 furthermore , if @xmath215}$ ] converges to an eigenvector of @xmath213 to the eigenvalue @xmath228 , @xmath175 and @xmath229 are related by @xmath230 to obtain the scaling exponent @xmath37 from eq . ( [ zz ] ) , we need to know the perron - frobenius eigenvalues @xmath173 of @xmath171 . there is no explicit solution to that problem in the general case . in what follows , we present the complete answers for a number of more special , yet infinite , classes of substitution rules . let @xmath231 be any word from the alphabet , and @xmath232 its @xmath134th cyclic permutation . consider now a substitution rule @xmath71 of the form ( [ subs ] ) with @xmath233^{l_i } \ ; .\ ] ] one special case is to take equal words @xmath234 for all @xmath235 . the corresponding substitution matrices indeed have @xmath236 as their second - largest eigenvalue , with degeneracy @xmath237 , since the column vectors of @xmath238 differ by scalar factors only . thus the fields stay at their original values and are not renormalised at all . to derive @xmath173 as a function of the parameters @xmath239 , we simply note that the special form of @xmath71 transforms this property to @xmath171 , not just for the point @xmath240 , but _ independently _ of the parameters . we thus obtain @xmath241 and the scaling exponent is given by @xmath242 where the @xmath243 are given in eqs . ( [ pminus ] ) and ( [ pplus ] ) as functions of the couplings . since @xmath173 are polynomials of the parameters @xmath244 with only positive coefficients , and @xmath245 , we conclude @xmath246 . as a function of the parameters , @xmath37 is either purely _ convex _ , with unique minimum @xmath247 ( the limit of the uniform chain ) , or it is _ constant _ : it is interesting to take a closer look at the latter case . obviously , this means that @xmath249_{ii}$ ] , independently of the parameters . as can be shown , for substitution rules of the form ( [ subs ] ) with ( [ rhoform2 ] ) , this is possible if and only if @xmath231 is a word ( or a power of a word ) with all letters @xmath250 appearing precisely once and if the last letter of @xmath114 is @xmath250 . this means , however , that the resulting chain is _ periodic _ and has bounded rather than logarithmically diverging fluctuations . as stated above , each two - letter substitution chain can be generated by a substitution rule of the form ( [ subs ] ) . for any marginal substitution rule , we can assume that the second - largest eigenvalue of the substitution matrix is @xmath251 . if necessary , this can be achieved by going to @xmath252 . as a consequence , @xmath216}$ ] in eq . ( [ y1 ] ) is eigenvector to the eigenvalue @xmath228 of @xmath213 , and as above the fields are already at their fixed points . in this case , the entries of the transformation matrices @xmath171 are polynomials of a single parameter @xmath253 that determines the critical couplings . the crucial point now is to show that like in the case discussed above the matrices @xmath171 have an eigenvalue @xmath254 independently of @xmath253 and of the detailed form of the substitution rule . this assertion is proved in appendix a. then , the perron - frobenius eigenvalues of @xmath171 are given as above ( [ mu ] ) , and we obtain the critical scaling exponent @xmath255 where @xmath80 is the leading eigenvalue of the substitution matrix . as in the case discussed above , @xmath256 is constant ( @xmath257 ) only for the special case of a periodic chain ( with period @xmath258 ) . otherwise , @xmath37 is a convex function of @xmath253 , with a unique minimum @xmath259 where the ( reduced ) couplings are equal . on the other hand , it is easy to see that the periodic chain with period @xmath258 is the only two - letter substitution chain with bounded letter fluctuations that can be generated by a substitution rule of the form ( [ subs ] ) and that has @xmath254 as the second - largest eigenvalue of the substitution matrix . we thus conclude that @xmath257 _ if and only if _ the fluctuations are bounded , and @xmath260 otherwise . the _ period - doubling chain _ , given by the substitution rule ( [ eq : sub ] ) with @xmath101 , falls into both classes described above . we consider double substitution steps , eliminating blocks of the form @xmath261 and obtain the scaling exponent @xmath262 this result was reported before in ref . if the length of the chain is increased by simple substitution steps , the coefficients @xmath263 in eq . ( [ zz ] ) converge _ separately _ for an even and an odd number of steps . effectively , this can be seen as a renormalisation of the couplings according to @xmath264 in a single renormalisation step . hence we obtain @xmath265 . as an example that is not contained in the above classes , consider the three - letter substitution @xmath266 with eigenvalues of the substitution matrix @xmath267 and @xmath236 . the eigenvector to eigenvalue 1 of the transposed substitution matrix @xmath213 is @xmath268 , and with eq . ( [ y0 ] ) we conclude that @xmath269 and @xmath270 . evaluating the rt of @xmath172 at this fixed point , we obtain the critical scaling exponent @xmath271 note that , for @xmath272 , the fluctuations of the couplings are indeed bounded , and we obtain @xmath273 as it should be . in our last example , the couplings follow the thue - morse sequence on the _ sites _ of the chain , rather than on the bonds as in all cases discussed so far . the coupling constants for a site - problem are given as a function of the two adjacent bonds and thus can take in general @xmath130 different values ( @xmath274 in this case ) . it can , however , easily be shown that this can be reformulated as a bond - problem with a substitution chain on an @xmath130-letter alphabet @xcite . in this case , the resulting four - letter substitution rule is @xmath275 the asymptotic letter frequencies are @xmath276 and @xmath277 ; and so we have @xmath278 ( for @xmath279 ) as the criticality condition ( [ eq : critcond ] ) . a possible parametrisation is @xmath280 , @xmath281 , @xmath282 , and @xmath283 . the eigenvalues of the substitution matrix are @xmath284 , @xmath285 , @xmath285 , and @xmath286 , we thus expect marginal fluctuations , whereas the thue - morse bond - problem has bounded fluctuations . following the above steps , we obtain a scaling exponent @xmath287}{\log ( 2 ) } \ ; .\ ] ] note that for @xmath288 , @xmath272 we obtain another coding of the period - doubling chain , while the special case @xmath289 , @xmath290 leads us back to the thue - morse bond - problem with bounded fluctuations ( with @xmath291 , and thus @xmath273 ) . for @xmath87 , fluctuations diverge with a power law . as a consequence , the vector @xmath215}$ ] finally scales with @xmath292 under the rt . since the matrix elements of @xmath171 are proportional to @xmath293 , at least a part of them eventually diverge like @xmath294 [ note that the components of @xmath181 have different sign because of eq . ( [ y2 ] ) ] . estimating the product of the perron - frobenius eigenvalues @xmath173 by the product of the corresponding traces , we obtain @xmath295 and thus @xmath296 where the _ wandering exponent _ @xmath89 is defined in eq . ( [ beta ] ) . let us discuss this in more detail for the case of two - letter substitutions . we take @xmath297 with corresponding eigenvector @xmath298 of @xmath213 and parametrise the critical couplings as @xmath299 since @xmath300 and @xmath301 have different sign , either one of these two is bigger than one for a fixed value @xmath302 , and without loss of generality we assume @xmath303 . note that we have the following relation @xmath304 we now show that @xmath305 $ ] . from eq . ( [ pminus ] ) , this is immediately clear for @xmath306 , but it also holds for @xmath307 . to see this , consider the products of subsequent couplings of the word @xmath65 . the product of the entire word is given by eq . ( [ product ] ) . if the last letter of @xmath65 is an @xmath63 , this enters @xmath307 and we are done . on the other hand , if the last letter is a @xmath64 , consider the product of all couplings but the last @xmath308 , which is even larger . iterating this argument , we finally arrive at a letter @xmath63 which proves our estimate of @xmath307 . using the standard eigenvalue formula , we obtain the scaling of @xmath173 under the rt , and thereby @xmath309 here , @xmath310 is some constant that may depends on the sign of @xmath311 , but apart from that is independent of @xmath253 and @xmath15 . this scaling form was also predicted in @xcite . in this section , we give a short summary of the results of our renormalisation treatment and discuss their consequences for the critical properties of the iqc . here , the behaviour of the low - energy part of the the spectrum at criticality is essentially the same as for the uniform chain . the correlation length critical exponents stays at its onsager value @xmath312 . for periodic chains ( or periodic variations of the interactions along the layers of the 2@xmath2 statistical system ) , the results from renormalisation show that the smallest fermion frequencies are only altered by a common factor @xmath187 ( fermion velocity ) , thus also conformal properties persist and the whole model is described by the central charge @xmath313 conformal field theory of a massless free majorana fermion : @xmath314 with a fermion velocity @xmath187 given by ( [ v]).the scaling dimensions @xmath315 are odd integers for periodic boundary conditions ( mixed sector hamiltonian ) and odd half - integers for free boundaries @xcite . numerical results @xcite confirm this conformal behaviour even for non - periodic chains with bounded fluctuations . for the critical scaling , we can make the following _ ansatz _ : @xmath316 the index @xmath253 here denotes the dependence on the position on the critical surface . the predicted scaling behaviour @xcite is confirmed analytically for all marginal two - letter substitution chains and some infinite classes of @xmath4-letter substitution chains . @xmath317 converges to the scaling coefficients @xmath263 if we increase the length of the chain @xmath15 by ( multiple ) substitution steps , in other words @xmath318 is asymptotically periodic for @xmath319 with period @xmath320 , where @xmath80 is the scaling factor of the system size . ( [ z ] ) implies @xmath321 , and @xmath273 is only obtained in the limit of the uniform chain where the fluctuations actually vanish . we further note that , in the explicitly solved cases , @xmath37 is invariant under inversion of all couplings . thus this is a symmetry of the lower part of the critical spectrum , while it is easy to see that it is not a symmetry of the entire finite - size spectrum . as a consequence , @xmath37 quadratically approaches the isotropic value @xmath322 for small deviations of the couplings from the uniform strength @xmath323 , as was also predicted by luck @xcite . in the terminology of the 2@xmath2 classical ising model , we have @xmath324 for any marginal disorder within layers of couplings . since the correlation length critical exponent perpendicular to the layers remains @xmath325 , marginal fluctuations introduce a relevant anisotropy of the model , as was noticed in ref . @xcite . due to anisotropic hyperscaling , a _ negative _ specific heat exponent @xmath326 is expected . within the renormalisation approach , the length of the chain @xmath15 is increased by ( multiple ) substitution steps . thus , it is a priori not clear what happens to the scaling dimensions when we increase @xmath15 arbitrarily . indeed , numerical studies ( for free boundary conditions ) show that , if we increase @xmath15 by arbitrary amounts , while @xmath37 converges , @xmath327 stays finite but will not tend to a unique limit . this is in contrast to the bounded fluctuation case . in more detail , the following properties of @xmath327 have been observed in all examples ( of two - letter substitution chains ) studied : * given the couplings , @xmath327 appears to be bounded for all @xmath15 and @xmath10 . * for @xmath328 , the effect of @xmath10 on @xmath329 is a mere shift : @xmath330 . * as a consequence of the rt , @xmath331 , @xmath318 is asymptotically periodic . * the amplitude of @xmath329 vanishes in the limit of the uniform chain , @xmath332 , as @xmath329 approaches continuously the constant value @xmath333 . these points can immediately be translated into properties of the integrated density of states ( idos ) , the inverse of which is given by @xmath334 while the first two points imply the convergence of this quantity , the third point above implies the lower part of the idos to be self - similar with a discrete scaling symmetry @xmath335 where @xmath336 is the asymptotic period of @xmath329 . mind that in general not the spectrum ( the lowest fermion frequencies ) , but only the idos displays this scaling behaviour for @xmath337 . this scaling property has consequences for the possible gap structure of the critical spectrum , where the results of the renormalisation approach at least lead to necessary conditions for gap - labelling theorems . but since within this class more spectral properties can be found analytically using trace maps , we postpone the discussion to a forthcoming publication . in substitution sequences with strong fluctuations , as in our third example , the binary non - pisot sequence [ @xmath102 in eq . ( [ eq : sub ] ) ] , their divergence is described @xcite by the wandering exponent @xmath89 ( [ beta ] ) which determines the exponential closing of the spectral gaps as @xmath338 here , @xmath339 is the second moment of the distribution of coupling constants along the chain @xcite which for the binary non - pisot sequence is given by @xmath340 and compatibility with the uniform case requires @xmath341 and @xmath342 . from the results of the renormalisation approach , @xmath318 should be asymptotically periodic for @xmath331 . this is confirmed by numerical investigations . indeed , the latter suggest the very same properties as given above in the case of marginal fluctuations . in particular , we observe @xmath343 for @xmath344 . again , this means that the integrated spectral density converges in the thermodynamic limit and its lower part shows the following scaling property at the critical point : @xmath345 using a renormalisation approach , we obtained analytic results for the critical behaviour of a class of ising models intermediate between uniformly ( or periodically ) ordered and randomly disordered systems . the systems under consideration are ferromagnetic ising quantum chains whose interaction constants are modulated according to substitution sequences . luck s criterion @xcite for the relevance of aperiodic disorder on the critical behaviour is fully confirmed for these systems , bringing along increasing evidence that the _ fluctuation of the interactions _ is the basic concept that demarcates the onsager universality class from models with weaker critical singularities . for two - letter substitution rules , this analysis was carried out in detail , showing how the renormalisation flow is determined by the nature of the fluctuations . an exact renormalisation formalism for the most general case of @xmath4-letter substitution rules was also presented and applied to examples . a number of quantitative results have been displayed , mainly for the correlation length critical exponent ( resp . the scaling exponent @xmath37 of the mass gap ) , confirming and refining luck s predictions for bounded , marginal , and relevant fluctuations of the coupling constants . in particular , an exact formula has been obtained for the scaling exponent of arbitrary two - letter substitution chains , containing previous results as special cases . in this light , the case of two - letter substitution rules appears to be rather well understood by now . what remains to be done in the @xmath4-letter case is a thorough discrimination between quantum ( substitution ) chains with bounded and with marginal fluctuations of the couplings . this is connected with the problem of determining all substitution rules with second - largest eigenvalue of the substitution matrix on the unit circle , which nevertheless result in a chain with bounded rather than marginal fluctuations . also , the investigation of ordered cases that do not stem from a substitution rule might add some insight as they can possess a higher degree of complexity . unfortunately , renormalisation techniques will probably be unsuitable here . of primary interest are , of course , extensions to higher dimensions . a natural first step is to study two - dimensional statistical systems with two - dimensional rather than one - dimensional variations of the interactions . as can be seen from some special cases @xcite , analytic results are possible in principle , but they are not generic and solvable cases might actually be misleading . further analysis is necessary , and some results can be expected from numerical approaches , especially since structures generated by substitution rules are well - suited for ( numerical ) renormalisation . first results in this direction have been obtained recently @xcite and will be extended soon . part of this work was done while jh was visiting melbourne university . it is a great pleasure to thank paul pearce for his kind hospitality . financial support from the german science foundation ( dfg ) and the studienstiftung des deutschen volkes is gratefully acknowledged . we consider two - letter chains generated by substitution rules @xmath346 which have @xmath347 as the second - largest eigenvalue of the corresponding substitution matrix . as a consequence , one finds the constraints @xmath348 on the words @xmath65 and @xmath66 and the letter frequencies @xmath349 in the limit word , where @xmath350 denotes the number of letters @xmath94 contained in the word @xmath351 . we parametrise the criticality condition ( [ eq : critcond ] ) @xmath352 ( setting @xmath353 ) for the corresponding ising chain by @xmath354 with _ coprime _ integers @xmath134 and @xmath355 , such that @xmath356 . then @xmath357 divides @xmath358 and @xmath359 . the scaling exponent of the critical spectrum is a simple expression of the leading eigenvalues of two matrices @xmath360 where the entries @xmath361 are defined in eqs . ( [ pminus ] ) and ( [ pplus ] ) . they are polynomials in @xmath253 and @xmath362 . in what follows , we prove that @xmath171 have an eigenvalue @xmath254 , _ independently _ of the value of @xmath253 and of the detailed form of the substitution rule @xmath71 . since we have @xmath363 , it suffices to concentrate on @xmath175 . we define two polynomials @xmath364 corresponding to the words @xmath65 and @xmath66 by @xmath365 taking the frequencies @xmath349 of the letters as given , the possible words @xmath65 and @xmath66 are subject to the constraints ( [ eq : wawb ] ) . we define an _ elementary polynomial _ @xmath366 to be the polynomial @xmath364 corresponding to a word @xmath367 of minimal length fulfilling this condition . since @xmath134 and @xmath355 are coprime , each @xmath367 consists of @xmath134 @xmath63 s and @xmath355 @xmath64 s and leads to one of @xmath368 different elementary polynomials possible . let in the following @xmath231 be @xmath65 or @xmath66 . we now introduce a graphical representation of the word @xmath231 and the accompanying polynomials . for given @xmath231 , construct iteratively a step function @xmath369 \to \mathbb{z}$ ] as follows : @xmath370 note that the criticality conditions imply @xmath371 . an example for the case @xmath101 , @xmath372 is shown in figure [ fig1 ] . for @xmath373 ( @xmath374 ) , @xmath375 ( resp . @xmath376 ) is given by the `` sum over the _ upward _ steps '' of @xmath377 ( proceeding in positive @xmath378-direction ) , while @xmath379 ( resp . @xmath380 ) is given by the `` sum over the _ steps '' , each step contributing a term @xmath381 . in our example ( for @xmath373 ) , @xmath382 and @xmath383 . in a first step , we now show that @xmath364 can be expressed as sums over elementary polynomials @xmath384 . _ lemma:_for an arbitrary word @xmath231 fulfilling the criticality condition for given @xmath349 , there are integers @xmath250 and elementary polynomials @xmath385 , such that @xmath386 _ proof:_consider the substrings @xmath387 of @xmath231 of length @xmath357 starting with the letter @xmath388 . for each @xmath387 containing more than @xmath134 @xmath63 s , there has to be an @xmath389 with less than @xmath134 @xmath63 s , because of the criticality condition for @xmath231 . since the number of @xmath63 s contained in successive substrings @xmath387 and @xmath390 differs at most by one , there is at least one substring consisting of @xmath134 @xmath63 s and @xmath355 @xmath64 s . in the graphical representation , this appears as a restriction of @xmath377 to an interval of length @xmath357 with the endpoints of the graph taking the same value . the corresponding term within @xmath364 is obviously some elementary polynomial times a power of @xmath253 . we may now eliminate this substring from @xmath231 ( cut out the interval in the graphical picture ) and start the argument again . we proceed by showing that the elementary polynomials differ only in a ( polynomial ) factor that is independent of @xmath391 . _ proposition:_given @xmath134 and @xmath355 , there are polynomials @xmath392 and @xmath393 such that for every elementary polynomial @xmath394 @xmath395\ ] ] where @xmath396 is a polynomial that does not depend on @xmath391 . _ proof:_an elementary polynomial @xmath394 takes the following general form @xmath397 where @xmath398 are integers . defining @xmath399 the proof then follows by direct calculation . we remark that with this choice of @xmath400 and @xmath401 , @xmath402 can also be shown to be a polynomial in @xmath253 and @xmath362 . now , we are in the position to complete our argument . from eqs . ( [ eq : pab ] ) , ( [ lem ] ) and ( [ prop ] ) , we conclude that @xmath403 thus it follows that @xmath404 , and hence @xmath236 is an eigenvalue of @xmath175 for an _ arbitrary _ value of @xmath253 . the leading eigenvalues of @xmath405 are given by eq . ( [ mu ] ) . in the above discussion , the coupling constants were chosen according to a substitution rule of the special form ( [ subs ] ) . while any ( infinite ) substitution chain with two letters can be generated that way , this is no longer the case for three or more different letters ( resp . couplings ) . in this appendix , we present a generalised version of the renormalisation formalism to deal with the general case . consider a substitution rule of the form @xmath407 the main problem in the general case is to maintain the recursive structure of the renormalisation procedure within the elimination process . by fixing the first letter of the substitutes in eq . ( [ subs ] ) , the appropriate couplings of the renormalised chain were given . this is no longer possible here ; however , we can proceed as follows . * fix any letter @xmath250 within the word @xmath408 , thus defining two words @xmath409 and @xmath410 as shown in eq . ( [ subs2 ] ) . note that the letter @xmath250 appears in the word @xmath114 ( at least for a suitably chosen power @xmath411 of the substitution rule ) for primitive substitutions . * redefine the substitution rule as a function on _ pairs _ of letters in the following way @xmath412 obviously , ( [ subs3 ] ) leads to the same limit chain as ( [ subs2 ] ) . the substitution matrix of @xmath115 has a dimension of at most @xmath130 ( pairs @xmath413 that do not appear in the original chain need not be included in @xmath115 ) . actually , @xmath414 is just the substitution matrix of the corresponding _ site_-problem , where the type of site follows the sequence , and the bond ( resp . the interaction along it ) is a function of its two endpoints @xcite . note , however , that the couplings here are still attached to the bonds . * in the renormalisation formalism , now define @xmath3-matrices , fields and asymmetry - parameters corresponding to pairs of letters @xmath415 renormalisation again means inverting the substitution process by integrating out all degrees of freedom attached to the words @xmath409 and @xmath410 . we obtain renormalised @xmath3-matrices through @xmath416 * the rt ( and their linear orders in @xmath126 ) for the ( at most ) @xmath417 parameters are then obtained analogous to section 3 . for a classification of the critical behaviour based on the fluctuations we need to know about the spectrum @xmath418 of @xmath414 in dependence of the spectrum @xmath419 of @xmath72 . for an appropriate power of @xmath71 , it is indeed possible to show that @xmath418 contains @xmath419 and that all additional eigenvalues are either 1 or 0 . we only sketch the proof here . without restriction of the general case , we can assume that the first and the last letters of all words @xmath114 remain fixed under the substitution ; that is @xmath420 ^ 1 $ ] for the first letter , and equivalently for the last letter . this is always fulfilled by a finite power @xmath421 of the substitution rule . we assume the @xmath4 words @xmath114 to have @xmath134 different first letters and @xmath10 different last letters . taking a look at the associated pair substitution matrix @xmath414 , it is easy to show that @xmath422 since we always have @xmath423 and the numbers @xmath134 and @xmath10 remain fixed because of the above assumption , the traces of @xmath424 and @xmath425 differ only by the constant @xmath426 for arbitrary @xmath81 . we conclude that @xmath418 containes @xmath419 and the all additional eigenvalues are fixed points under any power , thus @xmath285 or @xmath286 . we will , however , not give a detailed discussion of the general case here , but rather illustrate the method by applying it to a special example . consider the substitution rule of the `` circle sequence '' . this example of a quasiperiodic chain ( in the sense that it has a pure point fourier spectrum ) has been studied numerically in ref . @xcite @xmath427 with @xmath252 being of the form ( [ subs3 ] ) . the eigenvalues of @xmath428 are @xmath429 , @xmath285 , and @xmath430 , where @xmath431 is the golden mean we thus expect marginal fluctuations . the asymptotic frequencies of the letters are @xmath432 , @xmath433 , and @xmath434 , respectively . we choose @xmath435 , @xmath436 , and @xmath437 as the parametrisation of the critical couplings @xmath438 . since only the five pairs @xmath258 , @xmath439 , @xmath440 , @xmath441 , and @xmath442 appear in the chain , the dimension of the pair transfer matrix @xmath443 is just five . the spectrum @xmath444 consists of the spectrum @xmath445 and of two additional eigenvalues 1 . identifying pairs of couplings with the first one , we clearly have @xmath446 , @xmath447 , and @xmath448 . thus the vector of the logarithms of the reduced couplings , @xmath449 , is sill perpendicular to the perron - frobenius eigenvector of @xmath414 and converges in the renormalisation limit to a linear combination of two eigenvectors corresponding to an eigenvalue 1 of @xmath450 : @xmath451 . while both parameters , @xmath253 and @xmath452 , enter the transformation matrices @xmath171 of @xmath172 , @xmath452 can actually be eliminated by a similarity transformation . such a behaviour is always to be expected if a parameter is connected to an eigenvector ( like in this case ) that corresponds to an _ additional _ eigenvalue 1 of @xmath453 , not contained in @xmath445 . the perron - frobenius eigenvalues of @xmath171 can be determined explicitly , and we obtain for the scaling exponent @xmath454 finally , we like to show an example of a three - letter substitution rule @xmath455 with bounded fluctuations , which nevertheless has @xmath236 as the second - largest eigenvalue of the substitution matrix . this is the bond substitution that corresponds to the site - problem given by @xmath456 . fluctuations are indeed bounded here , since we just obtain the fibonacci chain by identifying @xmath63 and @xmath64 , but also by identifying @xmath63 and @xmath310 . calculating the scaling exponent by the above method , we obtain @xmath273 , as was to be expected . 99 h. au - yang and b.m . mccoy , theory of layered ising models : thermodynamics , _ phys . rev . * b10 * _ ( 1974 ) 88691 . m. baake , p. chaselon and m. schlottmann , the ising quantum chain with defects ( ii ) . the @xmath457 kac moody spectra , _ nucl . phys . _ * b314 * ( 1989 ) 62545 . m. baake , u. grimm and r. baxter , a critical ising model on the labyrinth , _ int . j. mod . * b8 * ( 1994 ) 3579600 . m. baake , u. grimm and d. joseph , trace maps , invariants and some of their applications , _ int . j. mod . _ * b7 * ( 1993 ) 152750 . berche and b. berche , crossover between aperiodic and homogeneous semi - infinite critical behaviors in multilayered two - dimensional ising models , _ phys . * b * ( to appear ) . b. berche , p .- e . berche , m. henkel , f. igli , p. lajk , s. morgan and l. turban , anisotropic scaling in layered aperiodic ising systems , _ j. phys . _ * a28 * ( 1995 ) l16571 . b. berche , p .- e . berche and l. turban , marginal anisotropy in layered aperiodic ising systems , _ j. phys . i france _ * 6 * ( 1996 ) 62140 . cardy , effect of boundary conditions on the operator content of two - dimensional conformally invariant theories , _ nucl . phys . _ * b275 * ( 1986 ) 20018 ; and : conformal invariance , in : _ phase transitions and critical phenomena _ , eds . c. domb and j.l . lebowitz , academic press , london ( 1987 ) , pp . dumont , summation formulae for substitutions on a finite alphabet , in : _ number theory and physics _ , luck , p. moussa , and m. waldschmidt , springer , berlin ( 1990 ) , pp . e. fradkin , l. susskind , order and disorder in gauge systems and magnets , _ phys . * d17 * ( 1978 ) 263758 . u. grimm , the quantum ising chain with a generalized defect , _ nucl . _ * b340 * ( 1990 ) 63358 . u. grimm and m. baake , non - periodic ising quantum chains and conformal invariance , _ j. stat . phys . _ * 74 * ( 1994 ) 123345 . u. grimm and m. baake , aperiodic ising models , in : _ the mathematics of long - range aperiodic order _ , ed . . moody , kluwer , dordrecht ( 1997 ) , pp . harris , effect of random defects on the critical behaviour of ising models , _ j. phys . _ * c7 * ( 1974 ) 167192 . f. igli and l. turban , common trends in the critical behavior of the ising and directed walk models , _ phys . * 77 * ( 1996 ) 12069 . lieb , t. schulz and d. mattis , two soluble models of an antiferromagnetic chain , _ ann . phys . ( ny ) _ * 16 * ( 1961 ) 40766 . j. kogut , an introduction to lattice gauge theory and spin systems , _ rev . * 51 * ( 1997 ) 656713 . luck , critical behavior of the aperiodic quantum ising chain in a transverse magnetic field , _ j. stat . * 72 * ( 1993 ) 41758 ; and : a classification of critical phenomena on quasi - crystals and other aperiodic structures , _ europhys . * 24 * ( 1993 ) 35964 . f. lunnon and p. a. b. pleasants , quasicrystallographic tilings , _ j. math . pures et appl . _ * 66 * ( 1987 ) 21763 . pfeuty , an exact result for the 1d random ising model in a transverse field , _ phys . lett . _ * a72 * ( 1979 ) 24546 m. queffelec , _ substitution dynamical systems _ , lecture notes in mathematics , vol . 1294 , springer , berlin ( 1987 ) . r. redheffer , difference equations and functional equations in transmission - line theory , in : _ modern mathematics for the engineer _ e. f. beckenbach , mcgraw hill , new york ( 1961 ) , pp . . h. simon , _ ferromagnetische isingmodelle auf aperiodischen strukturen _ , phd thesis , universitt tbingen , dissertations druck darmstadt ( 1997 ) . tracy , universality classes of some aperiodic ising models , _ j. phys . _ * a21 * ( 1988 ) l6035 . l. turban , p .- e . berche and b. berche , surface magnetisation of aperiodic ising systems : a comparative study of bond and site problems , _ j. phys . _ * a27 * ( 1994 ) 634966 . l. turban , f. igli and b. berche , surface magnetization and critical behavior of aperiodic ising quantum chains , _ phys . rev . _ * 49 * ( 1994 ) 12695702 .

according to roger myerson @xcite , the 1950 publication of nash s paper on equilibria was a watershed event not just for game theory , but for economics in general . the new general equilibrium concept , and its established universality , was an impetus for understanding rationality in much more general economic contexts , and inspired the important price equilibrium results by arrow and debreu . myerson argues convincingly in @xcite that the concept of nash equilibrium lies at the foundations of modern economic thought . seen from an algorithmic perspective , however , the nash equilibrium suffers from two important problems : first , it is not clear how to find it efficiently ( the same is true for the arrow - debreu variety for markets and prices ) . this shortcoming had already been identified by economists since the 1950s , and much effort has been devoted to algorithms for finding nash equilibria , see @xcite for examples from a very extensive literature . none of these algorithms came with polynomial - time guarantees , however , and the recent result @xcite establishing that the problem is ppad - complete explains why . of the many algorithmic approaches proposed by economists over the past 50 years for finding nash equilibria , most have been shown by now to require exponential time in the worst case @xcite . one exception is an important algorithmic genre known as _ homotopy methods _ @xcite ; see @xcite for a recent survey . in topology , homotopy _ is a continuous transformation from one function to another ( as , for example , between two paths joining two points on a map ) . the homotopy method starts with a fixpoint problem that is easy to solve ( say , a rotation of a disc around its center ) , and continuously transforms it into the problem in hand , by `` pivoting '' to new fixpoints along the way . a theorem by browder @xcite establishes the validity of this method in the limit , by showing the existence of a continuous path of fixpoints that joins two fixpoints of the initial and the final problems . the second algorithmic obstacle for the nash equilibrium concept is _ multiplicity . _ games have multiple equilibria , and markets many price equilibria , and thus the corresponding equilibrium concepts are only _ nondeterministic predictions _ ( oxymoron intended ) . in price equilibria , this multiplicity has been blamed for _ economic crises : _ the path guaranteed by browder s theorem is non - monotonic , going back and forth in time . as a result , equilibria vanish at its folds , leaving the market in turmoil @xcite . in games , a proposed remedy for multiplicity is the so - called focal point theory see , e.g. , @xcite p. 414 , postulating that players implicitly coordinate their equilibrium choice by focusing on the most obvious , or mutually advantageous , equilibrium ; repeated play and learning ( see , e.g. , @xcite ) can also be considered a remedy for multiplicity . in 1975 , harsanyi proposed the _ tracing procedure _ @xcite for battling equilibrium multiplicity , a theory further explicated in his joint 1988 book `` a general theory of equilibrium selection in games '' with selten @xcite ( harsanyi and selten shared in 1994 the nobel prize with nash ) . the tracing procedure asserts that players engaged in a game @xmath0 play at first a simple game @xmath1 , in which their prior beliefs about the other players behavior result in a dominant strategy . as time @xmath2 progresses , and their priors are falsified by life , they play a more and more realistic game @xmath3 , until , at time @xmath4 , they end up playing the intended game @xmath0 . they show that , for almost all games , tracing the equilibrium path of this process results in a unique equilibrium . notice the parallel with the homotopy method ; apparently the two were discovered independently . [ [ our - results ] ] our results + + + + + + + + + + + _ this paper is a complexity - theoretic critique of the tracing procedure and the homotopy method : _ we show that finding the solutions they prescribe requires the power of . in particular , finding the brouwer fixpoint that would have been discovered by the homotopy method , for a simple starting function and an adversarial final one , is -complete . the same is true , via standard reductions , for price equilibria . we also construct examples where the homotopy method not only will undergo an exponential number of pivots ( this was expected since @xcite ) , but will suffer an exponential number of _ direction reversals . _ as for the tracing procedure , we show that it is -complete to find the nash equilibrium selected by it , even in two - player games , and even if the initial game has dominant strategies obtained from priors , exactly as prescribed by harsanyi and selten . we extend this result to homotopy - based algorithms where the starting game depends on the final game and show that it is -complete to implement the herings - van den elzen , herings - peeters , and van den elzen - talman algorithms for finding equilibria in games . finally , it is particularly noteworthy that -completeness prevails even for finding the solutions that would be returned by the classical lemke - howson algorithm , a simplex - like method that had long been considered an oasis of conceptual simplicity and ( until @xcite ) of algorithmic hope in this field . this reinforces the `` exponentially long paths '' result of @xcite with a new result which says that , subject only to the hardness of , no short cuts to lemke - howson solutions are possible ( for any of the different initial choices of the algorithm ) . since it is known that the lemke - howson algorithm can be expressed as a homotopy @xcite , this result can also be seen as a powerful specialization of our first result . the algorithms we consider solve problems in the complexity class ppad , which is contained in tfnp , the class of all total function problems in np . another prominent complexity class contained in tfnp is pls ( for polynomial local search ) . many common problems in pls ( e.g. , local max cut and finding pure equilibria of congestion games ) are complete under a so - called _ tight _ pls - reduction , implying that the corresponding standard local search algorithm is exponential ( for certain starting configurations and _ any _ choices of the local search algorithm ) . furthermore , one can conclude that the computational problem of finding a local optimum reachable from a given starting configuration by local search is -complete . no such concept of tight reductions is known for ppad , and our results can be seen as addressing this deficiency . specifically , we show the -completeness ( and exponential worst - case behaviour ) of a number of homotopy - based algorithms for finding equilibria . our reductions start with the problem other end of this line ( oeotl ) , which is related to the problem end of the line used in the definition is ppad , seeking not just any end of a path , but the other end of the particular path starting at the origin . oeotl was known to be -complete since @xcite , but this fact has so far remained unexploited for proving lower bounds for other problems . [ [ outline - of - the - paper ] ] outline of the paper + + + + + + + + + + + + + + + + + + + + in section [ sec : prelim_hom ] , we give an overview of the linear homotopy method as applied to brouwer functions and games . in section [ sec : com ] , we recall the -complete problem oeotl ( other end of this line ) , which serves as the starting point for all our main reductions . in section [ sec : hom ] , we show that the linear homotopy method to compute a brouwer fixpoint is -complete , which is proved in section [ sec : hom : red ] . in section [ sec : lt ] , we establish the -completeness of the linear tracing procedure for two - player strategic form games for a special starting game that is independent of the final game . these results are extended to starting games that depend on the final game in section [ sec : extvdet - hp ] , where we show that it is -complete to implement the herings - van den elzen , herings - peeters , and van den elzen - talman algorithms for computing equilibria of games . the techniques of @xcite are central to both section [ sec : hom ] and section [ sec : lt ] and are recalled and extended along the way . finally , in section [ sec : extlh ] , we show that it is -complete to find any solution of a two - player game by the lemke - howson algorithm . a _ brouwer function _ @xmath5 is a continuous function from a convex and compact domain @xmath6 to itself ; by brouwer s fixpoint theorem there exists @xmath7 such that @xmath8 . a _ homotopy _ between two functions @xmath9 and @xmath10 ( where @xmath11 and @xmath12 are topological spaces ) is a continuous function @xmath13\times x\longrightarrow y$ ] such that for all @xmath14 , @xmath15 and @xmath16 . in this paper , we are interested in the special case where @xmath17 , for @xmath6 a closed compact subset of euclidean space , such as a cube . thus , @xmath18 and @xmath19 are brouwer functions on @xmath6 . given two continuous functions @xmath20 , the _ linear homotopy _ is given by the expression @xmath21 , and ( if @xmath6 is convex ) results in a continuum of brouwer functions @xmath22 given by @xmath23 for @xmath24 $ ] . browder s fixpoint theorem @xcite ( not to be confused with brouwer s fixpoint theorem ) asserts that given a homotopy connecting @xmath18 and @xmath19 , there is a path in @xmath25\times d$ ] from some fixpoint of @xmath18 to some fixpoint of @xmath19 , such that for every point @xmath26 on that path , @xmath27 is a fixpoint of @xmath28 . the homotopy method @xcite for finding a fixpoint of @xmath19 selects @xmath18 to have a unique and easy to find fixpoint , and essentially follows such a path . as noted in @xcite , we do not expect the path to be monotonic in @xmath2 indeed , we show in the full paper section [ sec : exp - changes ] that an exponential number of direction reversals is possible . we are often interested in _ approximate _ fixpoints . if @xmath5 is a brouwer function , an @xmath29-approximate fixpoint is a point @xmath27 such that @xmath30 ( we shall use the @xmath31 metric throughout ) . it follows from browder s theorem that , for any @xmath32 , there is a finite sequence @xmath33 of @xmath29-approximate fixpoints of @xmath34 , for some @xmath35 and @xmath36 , such that any two consecutive fixpoints in the sequence are at most @xmath29 apart . we shall be interested in the following problem , which we call browder fixpoint : given two arithmetic circuits computing two functions @xmath18 and @xmath19 from @xmath25^d$ ] to itself with lipschitz constant @xmath37 , an @xmath38 , where @xmath18 has a unique fixpoint @xmath39 , find an @xmath29-approximate fixpoint @xmath40 of @xmath19 that is connected via a sequence of @xmath29-approximate fixpoints to @xmath39 . ( to make this definition precise , we of course have to identify classes of functions from which @xmath18 and @xmath19 may be drawn . ) notice that the homotopy method for computing brouwer fixpoints provides a solution to this problem . homotopies can be defined very similarly also for games . given two games @xmath41 of the same type ( number of players and strategies ) , we consider @xmath42 , where it is the players utilities that are interpolated . it is routine to extend this definition to more general classes of games , such as graphical games @xcite ( in which case , in addition to the players and strategies , the two graphs must be the same ) . browder s theorem , via nash s reduction , establishes that there is a path of approximate nash equilibria here as well . the problem linear tracing is the following : given two games @xmath1 and @xmath43 , an @xmath38 , and a nash equilibrium @xmath39 of @xmath1 , find an @xmath29-approximate nash equilibrium @xmath40 of @xmath43 that is connected via a sequence of @xmath29-approximate nash equilibria to @xmath39 . it is easy to see that linear tracing is in , and it can be checked that the algorithm of herings and van den elzen @xcite achieves this(see section [ sec : lt : membership ] ) . browder fixpoint is also in . we consider directed graphs on @xmath44 vertices represented as @xmath45-bit vectors . the arcs are represented by two polynomial - size circuits @xmath46 and @xmath47 , each having @xmath45 inputs and outputs , as follows . there is an arc from vertex @xmath48 to @xmath49 provided that @xmath50 and @xmath51 . notice that all vertices of the graph have both indegree and outdegree @xmath52 or @xmath53 , that is , the graph consists of paths , cycles , and isolated vertices . an @xmath54-graph with parameter @xmath45 is a graph on @xmath55 specified by circuits @xmath46 and @xmath47 , as described above , subject to the constraint that vertex @xmath56 has no incoming arc but does have an outgoing arc . the problem end of the line is the problem of finding a vertex of a given @xmath54-graph other than @xmath56 which has at most one incident arc . note that this problem is in the class tfnp of total search problems in np : there exists a solution that could be obtained by following the directed path that starts at @xmath56 , and any given solution may be efficiently checked for correctness . the class ppad @xcite is defined as all search problems polynomial - time reducible to end of the line . the problem other end of this line ( which we will subsequently abbreviate to oeotl ) is the problem of finding the end of _ the particular path _ that starts at @xmath56 . in contrast with end of the line , a given solution to an instance of oeotl has no obvious concise certificate that it is the correct endpoint , so while oeotl is a total search problem , it is apparently not an np total search problem . in fact , we have the following ( theorem 2 of @xcite ) , which is the starting - point of our reductions . [ thm : oeotl ] @xcite oeotl is -complete . we build on the ideas of @xcite ( showing that unrestricted nash equilibria can efficiently encode end of the line solutions ) to show how equilibria defined by homotopies can efficiently encode oeotl solutions . in @xcite , each instance @xmath57 of end of the line is reduced to a game @xmath58 in such a way that any nash equilibrium of @xmath58 efficiently encodes a solution to @xmath57 . here we reduce @xmath57 to a homotopy problem defined by two games , @xmath59 , where @xmath1 depends only on @xmath45 , while @xmath60 encodes the circuits in @xmath57 , and is an extension of @xmath58 of @xcite . we establish that the associated linear homotopy corresponds to the naive `` follow the line '' approach to solving oeotl ; technically , a suffix of the homotopy path corresponds to following the line , and the existence of the relevant suffix is established in a non - constructive way , using the intermediate value theorem . in extending the result to the lemke - howson algorithm , the main technical obstacle is the initial choice of which `` label to drop '' , leading to multiple disjoint paths in the mixed - strategy profile space . we have to ensure that all of the @xmath61 solutions ( one for each pure strategy ) efficiently encode the solution to @xmath57 , where @xmath57 is treated as an instance of oeotl . this is done by embedding two copies of the game @xmath60 inside a larger one in such a way that at least one copy does not contain the initially - dropped label , and arguing that any lemke - howson equilibrium restricted to this copy ends up encoding the unique solution to @xmath57 . in this section we give detailed definitions of classes of fixpoint and approximate fixpoint computation problems . in section [ sec : hom : def ] , we review the definition of s and related concepts from chen et al . @xcite , here applied to a three dimensional domain . in section [ sec : hom : impl ] , we review the techniques of @xcite for implementing s as arithmetic circuits . in section [ sec : hom : red ] , we prove theorem [ thm : pspace - brouwer ] , the main result of section [ sec : hom ] , in which we establish the -completeness of a linear homotopy for finding a fixpoint of a brouwer function . @xmath62 will denote a complexity parameter of problem instances . we define a sequence @xmath63 of `` basic brouwer functions '' having unique known fixpoints . for each @xmath45 we define a class of brouwer functions whose members encode @xmath54-graphs on @xmath55 . the homotopy of equation ( [ eq : homotopy ] ) defines a class of functions @xmath28 , @xmath24 $ ] , that interpolate between @xmath18 and @xmath19 and specifies a particular fixpoint of @xmath19 . we will show that from that fixpoint , we can efficiently recover a solution to oeotl for the graph encoded by @xmath19 . let @xmath64 be the unit 3-d cube @xmath25 ^ 3 $ ] . for @xmath62 let @xmath65 denote a partition of @xmath64 into @xmath66 `` cubelets '' , @xmath67 ; @xmath68 is an axis - aligned cube of length @xmath69 whose vertex closest to the origin has coordinates @xmath70 . we define a _ brouwer - mapping circuit _ in a similar way to the definition in @xcite , here specialized to the case of 3 dimensions . we also introduce some variations of the definition , as follows : [ def : bmf ] ( brouwer - mapping circuit / function ; basic ; dgp - style ; partial ) a _ brouwer - mapping circuit ( bmc ) _ is a directed boolean circuit with @xmath71 input nodes and @xmath72 output nodes . note that any bmc @xmath73 has an associated _ ( ) _ @xmath74 that maps any cubelet @xmath68 to one of the four colors @xmath75 . we require the colors of all exterior cubelets to be predetermined as follows . for @xmath76 , @xmath77 . for @xmath78 , @xmath79 , @xmath80 . for @xmath81 , @xmath82 , @xmath83 . all other exterior cubelets are mapped to @xmath52 . the _ basic _ @xmath84 has the additional property that all internal cubelets get mapped to 0 . notice that @xmath85 is computable by a bmc of size polynomial in @xmath45 . a _ dgp - style _ is one that is derived from an @xmath54-graph in the manner of @xcite , and so is computable with a bmc of size polynomial in the size of circuits @xmath46 and @xmath47 . ( proposition [ prop : graph ] notes the relevant property of dgp - style s . ) a _ partial _ @xmath86 is defined with respect to a set @xmath87 ; @xmath86 assigns a color to elements of @xmath88 but @xmath86 may be undefined on non - elements of @xmath88 . [ prop : graph ] the following problem is -complete . given a brouwer - mapping circuit @xmath73 , find a point in @xmath64 that is a vertex of 4 cubelets mapped to all 4 colors by the associated @xmath89 , and which is connected to the origin via cubelets having colors other than 0 . this is a total search problem : the topological intuition is that there is a line that is adjacent to the colors @xmath90 and has one end at @xmath91 . the other end must be inside @xmath64 and adjacent to color 0 , since no other exterior point is adjacent to the colors @xmath90 . we note in passing that if we did not make the `` connected to the origin '' requirement , the problem would be ppad - complete . the proof of proposition [ prop : graph ] in the full paper , applies the reduction of @xcite from end of the line to the version of the problem where a _ panchromatic vertex _ is sought that is adjacent to all colors . the resulting @xmath90-colored line has a structure that faithfully simulates the arcs of the @xmath54-graph from which is was derived . panchromatic vertices correspond to end of the line solutions , and are linked - up with the @xmath90-colored line , whose structure corresponds to the end of the line graph , and whose orientation arises from the clockwise order of @xmath90 around it . we reduce from oeotl ( theorem [ thm : oeotl ] ) : let @xmath92 be an @xmath54-graph . let @xmath89 be a dgp - style derived from @xmath92 , whose circuit @xmath73 efficiently encodes @xmath92 . given a @xmath89 , define a _ @xmath90-chromatic vertex _ to be one that is shared by 3 cubelets with colors 1 , 2 and 3 . by construction , the only exterior @xmath90-chromatic vertex for any is @xmath91 . form a digraph @xmath93 on @xmath90-chromatic vertices by adding an arc between any pair that share a cubelet , directed such that if it pointing away from a viewer , its adjacent colors @xmath53 , @xmath72 , @xmath94 will appear in clockwise order around it . the reduction of @xcite ensures @xmath93 has indegree / outdegree at most 1 . panchromatic vertex _ to be one that belongs to 4 cubelets of all 4 different colors . by construction , for all s there is a path of @xmath90-chromatic vertices starting at @xmath91 and ending at a unique panchromatic vertex @xmath95 . @xmath95 is a solution ; it can be found in polynomial space by following this path . let @xmath93 be the graph on @xmath89 s @xmath90-chromatic vertices as described above . the reduction of @xcite has the following properties , from which the result follows . 1 . each vertex @xmath48 of @xmath92 has an associated @xmath90-chromatic vertex @xmath96 of @xmath93 ; @xmath48 and @xmath96 may be computed in polynomial time from each other . for @xmath97 , @xmath98 . @xmath48 is a solution to end of the line if and only if @xmath96 is panchromatic . each arc @xmath99 of @xmath92 corresponds to a sequence of edges of @xmath93 that connect @xmath96 to @xmath100 . each connected component of cubelets colored with @xmath90 , corresponds to a connected component of @xmath92 . we review a class of functions used to establish ppad - completeness of graphical and strategic - form games . recall that @xmath64 denotes the 3-dimensional unit cube ; we consider continuous functions @xmath101 having the following structure . each function is an arithmetic circuit composed of nodes , with each node taking inputs from up to 2 other nodes , and producing an output , for example , the sum of its inputs . all values are constrained to @xmath25 $ ] , so a node that adds its inputs would output 1 if their sum is greater than 1 . identify 3 nodes as `` input nodes '' and another 3 as `` output nodes '' , so if @xmath5 is a continuous function from @xmath64 to @xmath64 , it has a brouwer fixpoint . [ def : circuit ] a _ linear arithmetic circuit _ is an arithmetic circuit that computes a function from @xmath64 to @xmath64 , represented by a directed graph whose nodes are `` gates '' that perform certain basic arithmetic operations on their inputs as follows . each gate takes as input 0 , 1 or 2 real values in @xmath25 $ ] and outputs a single real value in @xmath25 $ ] , where the output of a gate may be the sum / difference / max / min of two inputs , or a constant multiple of a single input , or no input and constant output . ( an output value is set to 1 if for example two inputs that sum to more than 1 are input to a `` sum '' gate . ) we also allow `` comparator gates '' in which the output of such a gate evaluates to 1 ( respectively , 0 ) if its first input is greater ( respectively , less ) than the second input , and may take any value if they are equal . let @xmath102 . let @xmath103 , @xmath104 , @xmath105 , @xmath106 . [ def : impl ] we shall say that a @xmath86 is _ implemented by an arithmetic circuit @xmath107 _ if whenever @xmath108 , then @xmath109 when @xmath27 is at the center of @xmath68 . for @xmath27 not at a center , @xmath110 should be a convex combination of values of @xmath111 for cubelet centers @xmath112 within @xmath113 distance @xmath69 of @xmath27 . given @xmath114 computed by such a @xmath107 , we shall similarly say that @xmath5 _ implements _ @xmath86 . [ obs : impl ] if @xmath5 implements @xmath86 , then any fixpoints of @xmath5 must lie within distance @xmath69 of panchromatic vertices of @xmath86 , and vice versa . [ thm : bmf2circ ] a brouwer - mapping function having complexity parameter @xmath45 can be implemented using a linear arithmetic circuit having @xmath115 gates , that computes a continuous function . the proof gives a new technique to implement any brouwer - mapping function @xmath86 as a _ continuous _ function @xmath5 that uses a linear arithmetic circuit . this is in contrast with the corresponding techniques of @xcite that used a sampling - based approach in order to smooth the transition between distinct cubelets . the sampling - based approach results in discontinuous functions , where browder s theorem would not be applicable ( although it could still be applied to a continuous approximation ) . the technique only works in constant dimension ; if can be extended to higher dimension using the `` snake - embeddings '' of @xcite . ( see section [ sec : polyerror ] . ) the general idea of the technique is to take a simplicial decomposition of the domain @xmath64 , give rules for obtaining the values of @xmath5 at the vertices of the decomposition , and linearly interpolate within each simplex . let @xmath116 be a brouwer - mapping function . we construct a continuous brouwer function @xmath114 computed by a linear arithmetic circuit @xmath107 as follows . for @xmath27 at the center of cubelet @xmath68 , set @xmath117 where @xmath118 . for @xmath27 a vertex of cubelets @xmath119 , set @xmath120 to be the average of @xmath121 for all points @xmath112 at the centers of members of @xmath122 . the relevant points @xmath112 can be obtained using a polynomial - sized piece of circuitry . let @xmath88 be a simplicial decomposition of the unit cube consisting of 12 simplices that share a vertex at the center of the cube , and all other vertices are vertices of the cube . let @xmath123 be the simplicial decomposition of cubelet @xmath68 obtained by scaling @xmath88 down to @xmath68 . applied to all cubelets in @xmath65 this results in a highly regular decomposition @xmath124 of @xmath64 into @xmath125 simplices . for any @xmath126 , @xmath127 is obtained by linearly interpolating between the vertices of the simplex in @xmath124 that contains @xmath27 . clearly @xmath5 is continuous . the result follows from the following claim : @xmath5 as defined above , may be computed by a linear arithmetic circuit of size polynomial in @xmath45 . if @xmath27 is not a vertex of @xmath124 , the circuit can determine the vertices of a simplex @xmath128 that contains @xmath27 . there may be more than one such simplex , in which case _ it does not matter which is chosen_. the circuit has 12 cases to consider , depending on the orientation of @xmath129 . each case can be handled in the same general manner , by subtracting some vertex @xmath48 of @xmath129 from @xmath27 , and multiplying @xmath130 by some constants ( the coefficients of the linear function that interpolated between the vertices of @xmath129 ) . note that we never need to multiply two computed quantities together , multiplication only ever takes place between a computed quantity and a constant , as required for a linear arithmetic circuit . in this subsection , we establish the -completeness of the problem browder fixpoint , mentioned in the introduction , which can now be made precise as follows . we use two s @xmath131 and @xmath132 , where @xmath131 is the _ basic _ of definition [ def : bmf ] , and @xmath132 shall be a dgp - style that encodes an instance of end of the line as constructed in @xcite . let @xmath18 and @xmath19 be implementations of @xmath131 and @xmath132 using linear arithmetic circuits as described in the proof of theorem [ thm : bmf2circ ] . for @xmath114 let @xmath133 denote the @xmath134-th component of @xmath5 . for @xmath135 let @xmath136 where in ( [ eq : homotopy ] ) , the outputs of operators @xmath137 and @xmath138 are restricted to lie in @xmath25 $ ] ( so , rounding to 0 or 1 if needed ) . @xmath28 interpolates continuously between @xmath18 and @xmath19 and is constructed from them using elements of the linear arithmetic circuits of definition [ def : circuit ] ( which is useful later ; the natural alternative @xmath139 does not have this property . ) [ obs : cts ] for all @xmath24 $ ] , @xmath140 is lipschitz continuous , with lipschitz value @xmath141 . @xmath18 has a unique fixpoint close to @xmath142 . @xmath18 is a `` basic brouwer function '' which forms the starting - point of homotopies we consider . hence observation [ obs : cts ] and browder s fixpoint theorem implicitly define a corresponding fixpoint of @xmath19 . define an _ approximate fixpoint _ of @xmath114 to be a point @xmath126 with @xmath143 ( recall @xmath102 ) . [ thm : pspace - brouwer ] it is -complete to find , within accuracy @xmath69 , the coordinates of the fixpoint of @xmath19 that corresponds to the homotopy of ( [ eq : homotopy ] ) . it is also -complete to find the coordinates of an approximate fixpoint of @xmath19 that would be obtained by following a sequence of approximate fixpoints of @xmath28 in which consecutive points are within distance @xmath144 of each other . we reduce from the problem defined in proposition [ prop : graph ] as follows . let @xmath73 be a brouwer - mapping circuit derived from oeotl - instance @xmath54 using proposition [ prop : graph ] and let @xmath74 be the function computed by @xmath73 . let @xmath145 be the function computed by a linear arithmetic circuit that implements @xmath89 , and @xmath18 be computed by a circuit that implements the basic @xmath131 ( where both implementations apply theorem [ thm : bmf2circ ] ) . @xmath28 is given by ( [ eq : homotopy ] ) . let @xmath47 be a connected subset of @xmath146 $ ] such that for any @xmath147 , @xmath27 is a fixpoint of @xmath28 , and @xmath47 contains @xmath148 and @xmath149 . browder s fixpoint theorem ( with observation [ obs : cts ] ) assures us that such a @xmath47 exists . we claim that @xmath40 is within distance @xmath69 of the unique solution to @xmath73 of the problem specified in proposition [ prop : graph ] ( and hence , given @xmath40 we can easily construct this solution ) . suppose otherwise . for @xmath40 to be a fixpoint ( even an approximate one ) of @xmath19 , by observation [ obs : impl ] it must be within distance @xmath69 of a panchromatic vertex @xmath48 of @xmath89 . but now , @xmath48 is not connected to the origin via non - zero cubelets of @xmath89 . by connectivity of @xmath47 , there must exist @xmath147 such that @xmath27 lies within a cubelet @xmath150 where @xmath151 . we may assume further that @xmath27 is at least @xmath69 distant from any non - zero cubelet of @xmath89 . this follows provided we assume that connected components of non - zero cubelets of @xmath89 are separated from each other by a layer of 0-colored cubelets of thickness at least 3 . this may be safely assumed by increasing @xmath45 by a factor of 3 and subdividing the cubelets . we note that 1 . each entry of vector @xmath152 is @xmath153 , and 2 . each entry of @xmath154 is @xmath153 . it follows that for @xmath24 $ ] , each entry of @xmath155 is less than @xmath156 , since coordinatewise , @xmath157 . that means that @xmath27 can not be an approximate fixpoint of any @xmath158 , contradicting the assumption as required . since @xmath27 is at least @xmath69 distant from any non - zero cubelet of @xmath89 , it is also at least @xmath69 distant from any non - zero cubelet of @xmath131 , since for any cubelet @xmath159 , @xmath160 . the implementation of any @xmath86 as a function @xmath5 computed by a linear arithmetic circuit , as referred to in theorem [ thm : bmf2circ ] , ensures that @xmath120 is a convex combination of vectors @xmath121 for cubelet centers @xmath112 in the vicinity of @xmath27 , and since all those cubelet centers are colored 0 , we have that the entries of @xmath161 are all less than @xmath156 , as required . we now turn to games and nash equilibrium . let @xmath0 denote an @xmath162 game that we wish to solve , assumed to be chosen by an adversary . @xmath1 is a game with a unique `` obvious '' solution . in @xmath1 each player receives payoff 1 for his first action , and payoff 0 for all others , regardless of what the other player does . @xmath163 in the problem linear tracing the solution consists of the nash equilibrium of @xmath0 that is connected to the unique equilibrium @xmath164 of @xmath1 via equilibria of convex combinations @xmath165 . we can also define an approximate version of this problem , where instances include an additional parameter @xmath29 , and we seek an @xmath29-nash equilibrium that is connected to the solution of @xmath166 via a sequence of @xmath29-approximate solutions of @xmath167 . for the two - player case we assume @xmath168 . for more than 2 players , we need a positive @xmath29 to ensure that solutions can be written down as rational numbers . [ thm : lt ] linear tracing is -complete for 2-person games . the same result then holds for strategic - form games with more than 2 players . it holds for a value of @xmath29 that is exponentially small ; we could again use the ideas of @xcite to obtain a version where @xmath29 is inverse polynomial . our reduction uses the result of the previous section , along with earlier reductions between strategic - form games and graphical games . @xmath1 has a similar role to the basic brouwer function @xmath18 , but the correspondence is indirect ; generally @xmath18 is associated with one of the `` intermediate games '' @xmath169 for @xmath170 . the following is a brief overview of the rest of this section [ sec : lt ] . membership of can be deduced from @xcite . the reduction from the -complete discrete brouwer fixpoint problem of the previous section , applies the idea from @xcite of going via _ graphical games _ to normal - form games . we derive a type of graphical game in which a specific player ( denoted @xmath171 ) acts as a switch , allowing the remaining players to simulate either the basic brouwer - mapping function , or one associated with an instance of the search for a discrete brouwer fixpoint . @xmath171 governs this behavior via his choice of either one of two alternative strategies , and we show that a _ continuous _ path of equilibria from one choice to the other , results in an equilibrium that ultimately represents a solution to oeotl . the graphical game is then encoded as a 2-player game such that the linear - tracing procedure corresponds to this continuous path of equilibria in the graphical game . herings and van den elzen @xcite show how to find approximate equilibria on multi - player games , implicitly constructing a degree-2 graph that has a vertex corresponding to @xmath172 , the nash equilibrium of @xmath1 . given a simplicial decomposition of @xmath173 $ ] ( where @xmath6 is the space of mixed strategies of @xmath0 ) vertices of the graph correspond to simplices and subsimplices , and edges are implicitly defined by a lexicographical pivoting rule that governs a choice of movement from simplex to adjacent simplex , at each step of the algorithm . it can be checked that this algorithm establishes membership of for multiplayer linear tracing . in a graphical game @xcite , each player is a vertex of a graph , and his payoffs depend on his own and his neighbors actions . for a low - degree graph , this is one way that games having many players may be represented concisely . a homotopy between two graphical games @xmath174 and @xmath175 would require that these games have the same underlying graph , so that they differ only in their numerical payoffs . in the graphical games considered here , each player has just 2 actions and 3 neighbors . the main result of this section is [ prop : gg ] consider graphical games that contain a special player @xmath171 whose payoffs are constant ( unaffected by his own actions or the other players ) . the following problem is -complete : find a nash equilibrium of the game where @xmath171 plays 1 , that is topologically connected to a nash equilibrium in which @xmath171 plays 0 , via a path of nash equilibria in which @xmath171 plays mixed strategies . let @xmath18 and @xmath19 be functions computed by linear arithmetic circuits that implement brouwer - mapping functions @xmath131 and @xmath132 , where @xmath131 is the `` basic '' of definition [ def : bmf ] , and @xmath132 is a dgp - style that encodes some instance of end of the line . in a graphical game in which all players have 2 pure strategies denoted 0 and 1 , given a mixed - strategy profile for the players we let @xmath176 $ ] denote the probability that player @xmath48 plays 1 . [ def : simulation ] ( linear graphical game ; simulation of bmf s and partial bmf s ) given a bmf @xmath86 , we construct an associated graphical game @xmath177 as follows . @xmath177 has 3 special players @xmath178 whose strategies @xmath179,\pp[v_y],\pp[v_z])$ ] represent a point in @xmath64 . if @xmath86 is implemented by @xmath114 we use gadgets of @xcite to simulate the nodes in the arithmetic circuit that computes @xmath5 ( each node of the circuit has an additional associated player in @xmath177 ) . the game can pay them to adjust @xmath179,\pp[v_y],\pp[v_z])$ ] in the direction @xmath180,\pp[v_y],\pp[v_z])$ ] @xmath181,\pp[v_y],\pp[v_z])$ ] . then the players @xmath178 are incentivized to play @xmath180,\pp[v_y],\pp[v_z])$ ] . consequently a nash equilibrium of @xmath177 corresponds to a fixpoint of @xmath5 . moreover , an @xmath29-nash equilibrium corresponds to a @xmath182-approximate fixpoint of @xmath5 . we call @xmath177 a _ linear graphical game _ since we only allow players whose payoffs cause them to simulate the gates of linear arithmetic circuits . a game of the above kind is said to _ simulate _ we say further that a game @xmath183 simulates a partial bmf on a subset @xmath46 of cubelets , if for any @xmath184 , when @xmath179,\pp[v_y],\pp[v_z])$ ] lie at the center of @xmath185 the players @xmath178 are incentivized to play @xmath179,\pp[v_y],\pp[v_z])+\delta_c$ ] , where @xmath186 . [ lem : switch ] given any linear graphical game @xmath175 that simulates a @xmath132 , we can efficiently construct a new game @xmath187 having a player @xmath188 whose behavior can either cause @xmath183 to simulate @xmath132 ( if @xmath188 plays 1 ) or cause @xmath183 to simulate @xmath131 if instead @xmath188 plays 0 . @xmath171 shall serve as a `` switch '' , in allowing the game to switch between simulating @xmath131 and @xmath132 ( using an additional 3 players @xmath189 whose strategies represent a point in @xmath64 ) according to whether @xmath171 plays 0 or 1 . of course , @xmath171 has a key role in the associated two - player game . for @xmath190 , let @xmath191 be a graphical game constructed from @xmath192 according to definitions [ def : bmf ] , [ def : impl ] , [ def : simulation ] . @xmath191 has 3 players / vertices whose mixed strategies , as represented by the probabilities that they play 1 , represent a point in @xmath64 . denote these players @xmath193 . construct a `` combined '' game @xmath187 as follows . @xmath187 contains all the players in @xmath174 and @xmath175 together with a new player @xmath171 , where @xmath171 has the same fixed payoff for playing either 0 or 1 . we add 3 players @xmath189 whose mixed strategies represent a point in @xmath64 , and players @xmath194 , whose behavior is governed by @xmath195 $ & = & $ ( \pp[v^0_x]-\pp[v_{\switch}])+(\pp[v^1_x]-(1-\pp[v_{\switch}]))$ \\ $ \pp[v^+_x]$ & = & $ \max(\pp[\bar{v}^+_x],\min(\pp[v^0_x],\pp[v^1_x]))$ \end{tabular}\ ] ] ( and similar expressions for @xmath196 and @xmath197 ) where the parentheses in the above expression are important since the outputs of the operators @xmath137 and @xmath138 are truncated to lie in @xmath25 $ ] . players from @xmath174 and @xmath175 that take input from nodes @xmath198 or @xmath199 respectively , are then modified to take that input from @xmath200 instead . this completes the construction . of proposition [ prop : gg ] : we reduce from the circuit homotopy of theorem [ thm : pspace - brouwer ] . let @xmath201\}$ ] be an instance of this circuit homotopy . construct @xmath43 from @xmath19 as per definition [ def : simulation ] . construct @xmath187 as in lemma [ lem : switch ] , and we make the following observation . [ intermediate ] suppose that in @xmath187 we have @xmath202=t\in(0,1)$ ] . the resulting game @xmath203 simulates a partial @xmath158 which is implemented by a brouwer function @xmath28 that is ( pointwise ) a convex combination of @xmath18 and @xmath19 and is defined on the subset of cubelets where @xmath204 . given a homotopy path of nash equilibria of @xmath187 that start at the unique equilibrium of @xmath187 that satisfies @xmath202=0 $ ] and ends at an equilibrium of @xmath187 in which @xmath202=1 $ ] , there is a corresponding homotopy path from the fixpoint of @xmath18 and a fixpoint of @xmath19 ( noting that ( [ eq : gg - homotopy ] ) is essentially the same as ( [ eq : homotopy ] ) ) . that concludes the proof of proposition [ prop : gg ] . the following version of lemma [ lem : switch ] is useful in the construction for lemke - howson solutions , later on . [ cor : switches ] given any linear graphical game @xmath175 that simulates a @xmath132 , we can efficiently construct a new game @xmath187 having 2 players @xmath171 and @xmath205 whose behavior can either cause @xmath187 to simulate @xmath132 ( if both @xmath171 , @xmath205 play 1 ) or cause @xmath187 to simulate @xmath131 if instead either or both play 0 . the proof of the previous lemma is modified as follows . we re - use equation ( [ eq : gg - homotopy ] ) for players @xmath189 . the 3 players @xmath206 whose mixed strategies represent a point in @xmath64 have behavior governed by @xmath207 $ & = & $ ( \pp[v^0_x]-\pp[v'_{\switch}])+(\pp[v^+_x]-(1-\pp[v'_{\switch}]))$ \\ $ \pp[v^{++}_x]$ & = & $ \max(\pp[\bar{v}^{++}_x],\min(\pp[v^0_x],\pp[v^+_x]))$ \end{tabular}\ ] ] again with similar expressions for @xmath208 and @xmath209 . in this subsection we prove the following theorem , from which theorem [ thm : lt ] follows since we have previously noted membership of . [ thm : ltdetails ] it is -hard to compute the nash equilibrium of a given 2-player normal - form game @xmath43 , that is obtained via the linear homotopy that starts from @xmath1 , a version of @xmath43 where the payoffs have been changed to give each player payoff 1 for his first strategy and 0 for the others . we reduce from the graphical game problem of proposition [ prop : gg ] . let @xmath187 be a linear graphical game that includes a player @xmath188 as per proposition [ prop : gg ] . first , modify @xmath187 to give @xmath188 a small payment ( say , @xmath210 ) to play 1 , and zero to play 0 . we define a homotopy between two - player strategic - form games @xmath1 and @xmath43 such that equilibria of @xmath43 efficiently encode equilibria of @xmath187 , and equilibria of @xmath169 encode equilibria of versions of @xmath187 where @xmath188 has a bias towards playing 0 . we use the reduction of @xcite ( section 6.1 ) from graphical games to 2-player games ( a similar reduction is used in @xcite ( section 7 ) to express _ generalized circuits _ ( similar to our linear arithmetic circuits ) as 2-player games ) . given in the context of a mixed - strategy profile , let @xmath211 $ ] denote the probability allocated to pure strategy @xmath212 by its player . [ def : circgame ] a _ circuit - encoding _ 2-player game @xmath0 has a corresponding graphical game @xmath183 where the graph of @xmath183 is bipartite ; denote it @xmath213 ; each player ( vertex ) in @xmath183 has 2 actions ( denote them 0 and 1 ) and payoffs that depend on the behavior of 2 other players in the opposite side of @xmath92 s bipartition . each vertex / action pair @xmath214 of @xmath183 has a corresponding strategy in @xmath0 ; for @xmath215 , @xmath214 belongs to the row player and for @xmath216 , @xmath214 belongs to the column player . the payoffs in @xmath0 are designed to ensure that in a nash equilibrium of @xmath0 * @xmath217+\pr[(v,1 ) ] \geq 1/2n$ ] where @xmath45 is the number of players in @xmath183 * if in @xmath183 , @xmath48 plays 1 with probability @xmath218/(\pr[(v,0)]+\pr[(v,1)])$ ] then we have a nash equilibrium of @xmath183 . let @xmath0 be a circuit - encoding game derived from @xmath187 according to definition [ def : circgame ] . associate @xmath171 with 2 strategies of of the column player of @xmath0 , and let @xmath219 and @xmath220 be these strategies . hence a nash equilibrium of @xmath0 corresponds to one of @xmath187 where the value @xmath202 $ ] is given by the value @xmath221/(\pr[s^c_k]+\pr[s^c_{k+1}])$ ] . if we take a circuit - encoding 2-player game , and award one of the players a small bonus to play @xmath214 , then this corresponds to incentivizing the player @xmath48 in @xmath183 to select strategy @xmath222 . the corresponding incentive for @xmath48 will be larger , but only polynomially larger . let @xmath1 be a @xmath223 game with strategies @xmath224 for the row player , and @xmath225 for the column player . payoffs are as follows : each player receives 1 for playing @xmath226 or @xmath227 , and 0 for @xmath228 or @xmath229 for @xmath230 . ( thus @xmath1 is a @xmath223 version of equation ( [ eqn1 ] ) . ) rescale the payoffs of @xmath0 to all lie in the range @xmath231 $ ] . let @xmath43 be a @xmath223 game with strategies @xmath232 for the row player , and @xmath233 for the column player . payoffs are as follows : * @xmath164 results in payoffs @xmath234 for the players . * @xmath235 for @xmath230 results in payoffs @xmath236 . * @xmath237 for @xmath230 results in payoffs @xmath238 for @xmath239 , and @xmath240 ( for @xmath241 inverse polynomial in @xmath45 ) for @xmath242 * the rest of @xmath43 is a copy of @xmath0 above . @xmath243 let @xmath244 . the above payoffs have been chosen so that nash equilibria satisfy : in @xmath43 , players do not use @xmath226 or @xmath227 ; in @xmath245 , players both have a proper mixture of @xmath226 and @xmath227 with their other strategies . since @xmath0 s payoffs were rescaled to lie in @xmath231 $ ] , @xmath246 $ ] and @xmath247 $ ] can be shown to lie in @xmath248 $ ] , which can be checked from the following payoff ranges for @xmath245 : @xmath249,[0\cdotp54,0\cdotp66])$ \\ \cline{2 - 3 } \end{tabular}\ ] ] _ thus a continuous path of equilibria should at some stage allocate gradually less and less probability to @xmath226 and @xmath227 as @xmath2 increases . _ [ subgamene ] in any nash equilibrium @xmath250 of @xmath43 , the players assign probability 0 to @xmath226 and @xmath227 , and consequently @xmath250 consists of a nash equilibrium of @xmath0 , restricting to strategies @xmath228 , @xmath251 for @xmath252 . since @xmath244 , we can write @xmath169 as @xmath253 @xmath254 the general idea is as follows . consider the browder path of equilibria that begins from the unique equilibrium of @xmath1 ( where initially both players play @xmath226 , @xmath227 ) . as @xmath2 increases , the players will start to use the other strategies . at that stage , consider the distribution of their mixed strategies restricted to @xmath255 and @xmath256 . these distributions will constitute a nash equilibrium of a version of @xmath0 in which the column player receives a small bonus for playing @xmath219 . as @xmath2 increases to 1 , the bonus decreases continuously to 0 , and we recover observation [ subgamene ] . now , recall from definition [ def : circgame ] that the way @xcite reduce graphical games to two - player games , is to associate each player @xmath48 in the graphical game with two strategies in the two - player game , both belonging to the same player . the division of probability between those two strategies represents the probability that @xmath48 plays 1 . consider @xmath171 now , corresponding to @xmath219 and @xmath220 . @xmath171 is , in the graphical game , mildly incentivized to play 1 , but for @xmath257 the @xmath241 in the two - player game @xmath169 pushes it the other way , towards 0 . as a result , a nash equilibrium of @xmath169 may simulate a nash equilibrium of @xmath203 where @xmath258\in(0,1)$ ] . as @xmath2 increases and the contribution from @xmath241 decreases , this process corresponds to raising @xmath202 $ ] continuously ( but not monotonically ) from 0 to 1 . [ lem : gt ] let @xmath250 be a nash equilibrium of @xmath169 in which @xmath246<1 $ ] and @xmath247<1 $ ] . let @xmath259 be the probability distributions over @xmath260 and @xmath261 obtained by taking each value @xmath262 $ ] ( for @xmath263 , @xmath264 ) and dividing it by @xmath265 $ ] . then @xmath259 is a nash equilibrium of a version of @xmath0 where the column player receives a bonus of @xmath266/(1-\pr[s^r_0])$ ] for @xmath219 . in nash equilibrium @xmath250 , @xmath267 s strategy @xmath227 contributes the same quantity to each one of @xmath268 s strategies @xmath255 . so the values @xmath269,\ldots,\pr[s^r_n]$ ] must form a best response to @xmath267 s mixed strategy from @xmath259 . the column player receives a bonus @xmath270 $ ] specific to @xmath219 , arising from the possibility that row player plays 0 . he also receives an additional @xmath271 for all strategies @xmath229 for @xmath230 , but that uniform bonus has no further effect on his preference amongst @xmath256 . so in @xmath250 , @xmath272,\ldots,\pr[s^c_n]$ ] is a best response to a mixture of @xmath0 weighted by @xmath273 $ ] and the probability @xmath246 $ ] of a bonus @xmath266 $ ] for playing @xmath219 . this is equivalent to a best response to a version of @xmath0 with a bonus of @xmath266/(1-\pr[s^r_0])$ ] for playing @xmath219 . consider the path of equilibria connecting equilibrium @xmath172 of @xmath1 to equilibrium @xmath274 of @xmath43 . by lemma [ lem : gt ] we can choose @xmath241 such that in any equilibrium of @xmath275 we have @xmath221=0 $ ] . we also have that in any equilibrium of @xmath43 , @xmath276=0 $ ] . consider the longest suffix of the path for which @xmath277 for all games @xmath169 that appear in that suffix . the corresponding equilibria assign weight strictly less than 1 to @xmath226 and @xmath227 , so lemma [ lem : gt ] may be used to recover corresponding equilibria of versions of @xmath0 which in turn correspond to versions of @xmath187 in which initially , @xmath188 is incentivized to play 0 , and finally , @xmath188 is incentivized to play 1 . in the previous section , we showed the -completeness of finding the nash equilibrium of a two - player game that is associated with a homotopy that uses a specific simple starting - game that is not derived from the game of interest . in the literature on homotopy methods , starting with harsanyi @xcite , the starting - game is usually derived from the game of interest by positing a prior distribution over the players pure strategies , and using a starting - game whose payoffs are the result of playing against this prior distribution . in this section , we extend the result of section [ sec : lt ] to handle these starting - games and thus obtain results for the herings - van den elzen @xcite and herings - peeters @xcite algorithms , which use the same underlying homotopy , and the van den elzen - talman @xcite algorithm , which uses a different homotopy . all three algorithms have been shown under certain conditions to mimic the harsayni - selten linear tracing procedure . for each algorithm , we use the uniform distribution as the prior distribution , which is a natural choice . the van den elzen - talman algorithm uses a homotopy based on a starting mixed - strategy profile @xmath48 . letting @xmath278 be the set of mixed - strategy profiles , let @xmath279 be the set of convex combinations @xmath280 . in the notation of @xcite , the van den elzen - talman algorithm restricted to the two - player case uses the homotopy @xmath281 where for @xmath282 , @xmath283 denotes the best responses of player @xmath134 to mixed strategy @xmath284 , restricted to @xmath279 . it is -complete to compute equilibria that result from the above van den elzen - talman homotopy . the proof works by giving each player a dummy strategy which does not get used in any equilibrium , but whose payoffs ensure that the uniform distribution ( at @xmath285 ) over all players strategies results in payoffs that looks like @xmath1 . a similar trick works for the homotopy of @xcite . ( sketch ) it can be checked that the algorithm uses polynomial space . for the hardness , we reduce from linear tracing ; consider a game @xmath286 for which we seek an equilibrium that results from starting with @xmath166 as in section [ sec : lt ] . of the form of ( [ eqn1 ] ) . suppose we take a game @xmath0 from section [ sec : lt ] and give each player an additional strategy as follows . let @xmath287 and @xmath288 be the new strategies , for the row and column player respectively . @xmath287 has a payoff of @xmath289 for the row player , regardless of how the column player plays ( thus , @xmath287 is dominated by all the other strategies ) . the payoffs to the column player are chosen in such a way that , if in fact the row player uses the uniform distribution over @xmath290 , then the column player s payoffs will be 1 for @xmath227 and 0 for @xmath229 , for @xmath230 . the new strategy @xmath288 has a similar definition . note that the new payoffs are at most @xmath45 in absolute value . these new strategies ensure that we have the desired @xmath1 of section [ sec : lt ] , when we restrict each player to his first @xmath45 strategies . strategies @xmath287 and @xmath288 are not used in any nash equilibrium of @xmath169 , since they are strictly dominated for all @xmath2 . we let @xmath48 be the uniform distribution . in @xmath167 , the row plays a mixture @xmath291 while the column player plays @xmath292 , where @xmath293 and @xmath294 are mixed strategies whose support do not include @xmath287 and @xmath288 , so they constitute a nash equilibrium of a version of @xmath286 in which there is a bonus to play @xmath226 and @xmath227 . this bonus drops continuously to zero , so it is equivalent to the linear - tracing homotopy . the algorithms of herings - van den elzen @xcite and herings - peeters @xcite are based on an identical homotopy and differ only in the numerical technique used to follow the homotopy path . we can show using essentially the same construction as above that is is -hard to compute the equilibria found by these homotopies . to do so we can again construct a starting game by giving the row / column players new strategies @xmath287 and @xmath288 chosen to have low payoffs to the row ( respectively , column ) players , but whose payoffs to the opponent are chosen such that if either player played the uniform distribution , the opponent would receive a higher payoff for his first strategy ( either @xmath226 or @xmath227 ) than the others , which would all receive the same ( lower ) payoffs . the lemke - howson ( l - h ) algorithm is an important and rich research subject in and by itself within game theory ; for the purposes of this reduction , it is helpful to take a point of view that considers the l - h algorithm as a homotopy @xcite , where an arbitrary strategy ( the one whose label is dropped initially ) is given a large `` bonus '' to be played , so that the unique equilibrium consists of that strategy together with its best response from the other player ; the homotopy arises from reducing that bonus continuously to zero . [ lhtheorem ] it is -complete to find _ any _ of the solutions of a 2-player game that are constructed by the lemke - howson algorithm . the remainder of this section proves theorem [ lhtheorem ] , the hardness being established by a reduction from the graphical game problem of proposition [ prop : gg ] , extending the ideas of the reduction for linear tracing ( theorems [ thm : lt ] , [ thm : ltdetails ] ) . a new technical challenge here is that the choice of initially dropped label results in @xmath61 alternative homotopy paths , and we must ensure that any of the ( up to ) @xmath61 solution can encode the single solution to some instance of linear tracing . suppose that some strategy has been given this `` l - h bonus '' , and a browder path of nash equilibria is obtained from reducing that bonus to zero . as before let @xmath24 $ ] be a parameter that denotes the distance from the starting game of the homotopy to the game of interest , so that @xmath295 is a multiplicative weight for the bonus in intermediate games . consider the browder path it is piecewise linear , a topologically well - behaved line . let @xmath296 $ ] parameterize points along the browder path an equilibrium @xmath297 is the one that is a fraction @xmath298 of the distance along the path ( starting at the version of the game with the l - h bonus ) . so , multiple values of @xmath298 can correspond to the same value of @xmath2 . here we mostly focus on @xmath298 rather than @xmath2 . the following construction addresses the issue that an arbitrary strategy may receive the l - h bonus . we embed two copies of a circuit - encoding game @xmath0 ( definition [ def : circgame ] ) into a game instance for the lemke - howson algorithm . at least one of those copies of @xmath0 will not contain the strategy that receives the l - h bonus . the l - h homotopy , restricted to that copy of @xmath0 , will simulate the homotopy of section [ sec : lt ] . in figure [ lhdiagram ] , @xmath286 denotes a circuit - encoding @xmath299 game ( note the two copies ) whose payoffs have been rescaled to lie in the interval @xmath300 $ ] . @xmath286 is assumed to have an associated graphical game with two `` switch '' players @xmath301 , @xmath302 that affect the equilibria of @xmath286 according to corollary [ cor : switches ] . they will correspond to the first pair of each of @xmath286 s players strategies @xmath303 and @xmath304 such that , * if both @xmath305=1 $ ] and @xmath306=1 $ ] , @xmath286 s equilibrium encodes a solution to an end of the line instance that is efficiently encoded by @xmath286 ; * if either @xmath305=0 $ ] or @xmath306=0 $ ] , @xmath286 encodes the `` basic '' brouwer - mapping function ; * if we add a bonus to the row player for his first strategy @xmath226 that is less than some threshold @xmath307 , it will result in @xmath246=0 $ ] and hence @xmath305=1 $ ] , and similarly for the column player with respect to @xmath227 and @xmath302 . ( we will see that such bonuses occur , and they decrease at @xmath308 . ) * notation . * @xmath309 and @xmath310 denote sets of the players strategies as shown in figure [ lhdiagram ] . in the context of a mixed - strategy profile , @xmath311 $ ] denotes the probability that the column player uses @xmath107 ; @xmath312 $ ] that he chooses an element of @xmath313 , and so on . let @xmath314+\pr[d]+\pr[c']+\pr[d']$ ] , a function of distance along the browder path . we note the following facts * @xmath315 ( if , say , a column player strategy receives the l - h bonus , then the row player will play some pure best response , either @xmath316 or @xmath317 ; so @xmath318=1 $ ] or @xmath319=1 $ ] . ) * @xmath320 ( shown in lemma [ lem : endpath ] ) together with the key observation that @xmath321 is a continuous function of @xmath298 , implying : for some @xmath322 $ ] , @xmath323 , and for @xmath324 , @xmath325 . let @xmath326 be the copy of @xmath286 that does not contain the strategy that receives the l - h bonus . ( if one of @xmath107 , @xmath6 , @xmath316 or @xmath317 receive the l - h bonus , then @xmath326 may be either copy of @xmath286 . ) for any @xmath11 , at least one player @xmath327 has an additional bonus at least @xmath328 to play @xmath329 in @xmath326 ( suppose for example @xmath311+\pr[d]\geq x/2 $ ] and @xmath327 is the row player ; figure [ lhdiagram ] awards additional @xmath330 to @xmath327 when @xmath107 or @xmath6 is played ) . but neither player s bonus exceeds @xmath331 . as @xmath298 increases from @xmath332 to 1 , @xmath321 goes down from @xmath333 to at most @xmath334 . we will establish that when @xmath335 , @xmath336 contains a solution to a `` biased '' version of @xmath326 where one of the players first strategies ( i.e. @xmath226 or @xmath227 ) has an additional bonus ( enough to ensure @xmath269=0 $ ] and @xmath305=0 $ ] , in the case of the row player ) . furthermore , when @xmath337 , we have that @xmath297 contains a solution to @xmath326 , only with smaller biases . these biases are associated with `` switch '' strategies in the graphical game associated with @xmath286 . let @xmath332 be the largest value of @xmath298 where @xmath321 is large enough that one of the bonuses sets @xmath338=1 $ ] in @xmath326 ( for @xmath339 ) . between @xmath332 and @xmath337 we pass through a continuum of equilibria where @xmath338 $ ] changes from 1 to 0 ; equivalently @xmath340 $ ] changes from 0 to 1 , and the resulting equilibrium at @xmath337 corresponds to a solution to oeotl . let @xmath297 be a solution of @xmath341 . if @xmath326 is the bottom right - hand copy of @xmath286 in figure [ lhdiagram ] , then if the distributions over @xmath73 and @xmath342 are normalised to 1 , we have a nash equilibrium of a game @xmath343 where the row player has an additional bonus of @xmath344+\pr[d])/\pr[b']$ ] to play his first strategy @xmath226 , and the column player has an additional bonus of @xmath345+\pr[d'])/\pr[b]$ ] to play his first strategy @xmath227 . by symmetry , a similar result also holds in the case that @xmath326 is the top right - hand copy of the @xmath286 . payoffs to the row player are unaffected by the column player s distribution over @xmath313 . meanwhile , @xmath107 and @xmath6 lead to an additional bonus of @xmath346 ( weighted by the probability that @xmath107 and @xmath6 are used by the column player ) for the row player to use the top row of @xmath342 . [ lem : endpath ] at @xmath4 ( equivalently , @xmath337 ) we have in any nash equilibrium , that @xmath311\leq\frac{1}{m}$ ] , @xmath347\leq\frac{1}{m}$ ] , @xmath318\leq\frac{1}{m}$ ] and @xmath319\leq\frac{1}{m}$ ] . since @xmath348 we have @xmath320 . game @xmath286 embedded in the top - left and bottom - right regions , with payoff rescaled to @xmath300 $ ] . in the top - right and bottom - left regions are copies of a @xmath299 game that give the column player a payoff of 0 and the row player a payoff of 1.each of @xmath313 , @xmath73 , @xmath349 , @xmath342 denotes a set of @xmath45 strategies . @xmath107 , @xmath6 , @xmath316 and @xmath317 are individual strategies.in the proofs we put @xmath350 , @xmath330.,title="fig:",width=240 ] game @xmath286 embedded in the top - left and bottom - right regions , with payoff rescaled to @xmath300 $ ] . in the top - right and bottom - left regions are copies of a @xmath299 game that give the column player a payoff of 0 and the row player a payoff of 1.each of @xmath313 , @xmath73 , @xmath349 , @xmath342 denotes a set of @xmath45 strategies . @xmath107 , @xmath6 , @xmath316 and @xmath317 are individual strategies.in the proofs we put @xmath350 , @xmath330.,title="fig:",width=576 ] a proof of lemma [ lem : endpath ] may be found in the full version . of lemma [ lem : endpath ] . we give the proof that @xmath311\leq\frac{1}{m}$ ] ; by symmetry the other claims are similar . assume for contradiction that @xmath311>\frac{1}{m}$ ] . we know that at least one strategy from @xmath351 gives positive payoff , since the places where they can obtain a negative payoff are `` equal and opposite '' . indeed , it can be checked that the payoff to at least one member of @xmath313 and @xmath73 is at least @xmath352 . the ( column player s ) payoffs from @xmath107 and @xmath6 must sum to zero , so if @xmath311>0 $ ] then @xmath347=0 $ ] ( @xmath6 gets negative payoff and is a strictly worse response @xmath313 or @xmath73 . ) a similar argument for the row player s payoffs establishes that one or both of @xmath316 and @xmath317 gets zero probability . given that @xmath311>\frac{1}{m}$ ] we can deduce that @xmath353=0 $ ] due to being a worse response than @xmath342 : @xmath107 contributes at least @xmath354 to @xmath355 ; @xmath313 contributes a positive amount ; @xmath6 has zero probability so contributes nothing ; @xmath73 contributes @xmath356 . we noted above that @xmath318=0 $ ] or @xmath319=0 $ ] ( or both ) . consider two cases : * case 1 : * @xmath319=0 $ ] . deleting strategies with probability zero , we are left with the following structure : r|c|c|c| & & & + @xmath342 & @xmath357 & @xmath358,-m)$ ] & @xmath359,[0\cdotp4,0\cdotp6])$ ] + @xmath316 & @xmath360)$ ] & @xmath361 & @xmath362)$ ] + comparing @xmath107 with @xmath313 , we need @xmath318=1 $ ] to avoid @xmath313 being a better response than @xmath107 ( which is supposed to have positive probability @xmath363 ) . if @xmath318=1 $ ] , @xmath73 is a worse response than the others , but when the column player uses only @xmath313 and @xmath107 , @xmath316 has much lower payoff than @xmath342 , contradicting assumption that @xmath318 $ ] is positive . this leaves us with case 2 : * case 2 : * @xmath318=0 $ ] . we get , after deleting zero - probability strategies , r|c|c|c| & & & + @xmath342 & @xmath357 & @xmath358,-m)$ ] & @xmath359,[0\cdotp4,0\cdotp6])$ ] + @xmath317 & @xmath362)$ ] & @xmath364 & @xmath360)$ ] + here the contradiction is immediate since @xmath73 is a strictly better response than @xmath107 , preventing @xmath311>0 $ ] . [ lem : lbweight ] assume that @xmath365 in figure [ lhdiagram ] and that @xmath348 . suppose that @xmath366 . then @xmath312\geq\frac{1}{10}$ ] , @xmath367\geq\frac{1}{10}$ ] , @xmath353\geq\frac{1}{10}$ ] , @xmath368\geq\frac{1}{10}$ ] . a proof of lemma [ lem : lbweight ] may be found in the full version . we need to consider two cases in detail : case 1 assumes that an element of @xmath349 received the l - h bonus and case 2 assumes that @xmath107 received the bonus . all other possibilities are essentially the same as these , by symmetry . suppose first that a strategy from @xmath349 has been given the l - h bonus , and we are at a nash equilibrium where @xmath369 . first we prove that the row player strategies satisfy @xmath353\geq\frac{1}{10}$ ] , @xmath368\geq\frac{1}{10}$ ] . since @xmath370 , we have @xmath353+\pr[b']\geq \frac{3}{4}$ ] . suppose for a contradiction that @xmath368<\frac{1}{10}$ ] , so that @xmath353\geq 0\cdotp65 $ ] . then ( for @xmath365 ) , @xmath107 is the unique best response for the column player and hence @xmath311=1 $ ] . this implies @xmath371 , contradicting the assumption that @xmath370 . similarly , if @xmath353<\frac{1}{10}$ ] then @xmath368\geq 0\cdotp65 $ ] , then @xmath6 is the column player s unique best response , hence @xmath347=1 $ ] , again contradicting @xmath370 . next we prove that the column player strategies satisfy @xmath312\geq\frac{1}{10}$ ] , @xmath367\geq\frac{1}{10}$ ] . suppose for a contradiction that @xmath367<\frac{1}{10}$ ] , so that @xmath312\geq0\cdotp65 $ ] . then for the row player , @xmath317 is a better response than @xmath342 and @xmath316 ( and regarding @xmath349 , some strategy from @xmath349 received the l - h bonus , so we do not claim @xmath349 is suboptimal ) . if @xmath342 is not a best response , so @xmath368=0 $ ] , since @xmath370 we have @xmath353\geq\frac{3}{4}$ ] . then @xmath107 is strictly better than @xmath313 , contradicting @xmath312\geq0\cdotp65 $ ] . alternatively suppose that @xmath312<\frac{1}{10}$ ] , so that @xmath367\geq0\cdotp65 $ ] . then @xmath316 is a better response than @xmath342 and @xmath317 , since @xmath372 ; @xmath373 since @xmath311\leq\frac{1}{4}$ ] by assumption that @xmath370 ; @xmath374 is negative . with @xmath317 and @xmath342 eliminated , @xmath107 is a strictly better response than @xmath73 , contradicting @xmath367>0 $ ] . suppose alternatively that it was strategy @xmath316 that received the l - h bonus . we show first that the row player s strategies satisfy @xmath353\geq\frac{1}{10}$ ] , @xmath368\geq\frac{1}{10}$ ] . suppose @xmath353<\frac{1}{10}$ ] , so that @xmath368\geq0\cdotp65 $ ] . @xmath6 is a better response than @xmath313 and @xmath73 . but from that it follows that @xmath347+\pr[c]=1 $ ] , contradicting @xmath369 . suppose @xmath368<\frac{1}{10}$ ] , so that @xmath353\geq0\cdotp65 $ ] . @xmath107 is a better response than @xmath313 and @xmath73 , so @xmath311+\pr[d]=1 $ ] contradicting @xmath369 . next we show that the column player s strategies satisfy @xmath312\geq\frac{1}{10}$ ] , @xmath367\geq\frac{1}{10}$ ] . suppose @xmath367<\frac{1}{10}$ ] , so @xmath312>0\cdotp65 $ ] . @xmath317 s payoff is greater than @xmath375 while @xmath349 and @xmath342 have payoff at most @xmath376 so @xmath317 is a better response than @xmath349 and @xmath342 , contradicting @xmath370 . suppose @xmath312<\frac{1}{10}$ ] , so @xmath367>0\cdotp65 $ ] . @xmath316 is a better response than @xmath349 and @xmath342 even ignoring the l - h bonus . at @xmath335 we have that at least one of @xmath311 $ ] , @xmath347 $ ] , @xmath318 $ ] , @xmath319 $ ] is at least @xmath377 , while at the end of the browder path , we know that all these quantities are at most @xmath378 . we set the switch threshold probability to be somewhere between these , but we have to use lower bounds on @xmath312 $ ] , @xmath367 $ ] , @xmath353 $ ] , @xmath368 $ ] at @xmath4 and upper bounds on these at @xmath369 ( as well as lower bounds on these at @xmath369 to ensure that a nash equilibrium of the `` biased game '' is being encoded ) . finally , we need to show that there exists @xmath307 such that the bonus from at least one switch strategy in @xmath326 changes continuously above @xmath307 to below it , while the bonus for the other switch strategy ends up below @xmath307 , thus initially , at least one value of @xmath305 $ ] and @xmath306 $ ] is zero , but at the end both evaluate to 1 . this needs to take into account the variable amount of probability allocated to the strategies in @xmath379 , since that affects the impact of the bonuses on @xmath329 . for any @xmath380 $ ] the weight assigned by each player to @xmath326 s strategies is at least @xmath381 by lemma [ lem : lbweight ] , so that the bonus for player @xmath327 to play @xmath329 , falls by a larger factor than the probability that @xmath379 is played . that means that @xmath307 can indeed be chosen as required . should a more general result be obtainable ? for example , perhaps it should be possible to identify general classes of `` path - following algorithms '' that include the ones we analyzed here , for which it is -complete to compute their output . acknowledgementssupported by epsrc grant ep / g069239/1 `` efficient decentralised approaches in algorithmic game theory '' a potential obstacle is that such a general result may subsume the question of whether the 2-dimensional analogue of oeotl is -complete ( i.e. consider the ppad - complete problem 2d - sperner @xcite ; suppose we ask for the trichromatic triangle identified in the proof of sperner s lemma . ) in 2 dimensions , the gadget that is used to allow `` edges '' to cross each other , rearranges the structure of those edges , such that the corresponding solutions to end of the line are the same , but not the unique solution to oeotl . generally , there are many ways to modify the edges of a given @xmath54-graph so that the degree-1 vertices are unchanged , but the structure of the graph is in other respects completely different . von stengel et al @xcite use a tracing procedure to solve extensive two - person games , and they obtain a _ normal form perfect _ equilibrium by starting from a completely mixed starting vector . what is the complexity of computing a normal form perfect solution using this ( or other ) methods ? ( they note ( @xcite , p. 707 ) that on strategic - form games this procedure mimics the linear tracing procedure of @xcite . ) 1 y. balasko . economic equilibrium and catastrophe theory : an introduction . _ econometrica _ , * 46*(3 ) 557 - 569 . ( 1978 ) f.e . browder . on continuity of fixed points under deformations of continuous mappings . _ summa brasiliensis math _ * 4 * , 183 - 191 . ( 1960 ) x. chen and x. deng . on the complexity of 2d discrete fixed point problem . _ proceedings of 33rd icalp conference _ , lncs 4051 , 489 - 500 . ( 2006 ) x. chen , x. deng and s - h teng . settling the complexity of computing two - player nash equilibria . _ journal of the acm _ , * 56*(3 ) , 1 - 57 . ( 2009 ) p. crescenzi and c. h. papadimitriou . reversible simulation of space - bounded computations . _ theoretical computer science _ * 143*(1 ) , 159 - 165 . ( 1995 ) c. daskalakis , p.w . goldberg and c.h . the complexity of computing a nash equilibrium . _ siam journal on computing _ , * 39*(1 ) , 195 - 259 . ( 2009 ) b.c . eaves . homotopies for computation of fixed points . _ mathematical programming _ , * 3*(1 ) , 1 - 22 . ( 1972 ) k. etessami and m. yannakakis . on the complexity of nash equilibria and other fixed points . _ siam journal on computing _ , * 39*(6 ) , 2531 - 2597 . ( 2010 ) d. fudenberg and d.k . _ the theory of learning in games_. mit press . ( 1998 ) j.c . the tracing procedure : a bayesian approach to defining a solution for @xmath45-person noncooperative games . _ international journal of game theory _ , * 4 * , 61 - 95 . ( 1975 ) j.c . harsanyi and r. selten . _ a general theory of equilibrium selection in games . _ cambridge : mit press . ( 1988 ) p.j - j . herings . two simple proofs of the feasibility of the linear tracing procedure . _ economic theory _ , * 15 * , 485 - 490 . ( 2000 ) p.j - j . herings and a. van den elzen . computation of the nash equilibrium selected by the tracing procedure in @xmath382-person games . _ games and economic behavior _ , * 38 * , 89 - 117 . ( 2002 ) p.j - j herings and r.j.a.p . peeters . a differentiable homotopy to compute nash equilibria of @xmath45-person games . _ economic theory _ , * 18*(1 ) , 159 - 185 . ( 2001 ) p.j - j . herings and r. peeters . homotopy methods to compute equilibria in game theory . _ economic theory _ , * 42*(1 ) , 119 - 156 . ( 2010 ) m. d. hirsch , c. h. papadimitriou and s. a. vavasis . exponential lower bounds for finding brouwer fixed points _ journal of complexity _ , * 5*(4 ) , 379 - 416 . ( 1989 ) m. kearns , m. littman and s. singh . graphical models for game theory . _ proceedings of uai _ , 253 - 260 . ( 2001 ) d. m. kreps . _ a course in microeconomic theory . _ princeton univ . press . ( 1990 ) c. e lemke and j. t. howson , jr . equilibrium points of bimatrix games . _ siam j. appl . math _ , * 12*(2 ) , 413 - 423 . ( 1964 ) r. b. myerson . nash equilibrium and the history of economic theory . _ journal of economic literature _ , * 37*(3 ) , 1067 - 1082 . ( 1999 ) j. nash . noncooperative games . _ annals of mathematics _ , * 54*(2 ) , 289 - 295 . ( 1951 ) m.j . osborne and a. rubinstein . _ a course in game theory_. mit press . ( 1994 ) c.h . papadimitriou . on the complexity of the parity argument and other inefficient proofs of existence . _ j. comput . * 48*(3 ) , 498 - 532 . ( 1994 ) r. savani and b. von stengel . hard - to - solve bimatrix games . _ econometrica _ * 74*(2 ) , 397 - 429 . ( 2006 ) h. e scarf and t. hansen . _ the computation of economic equilibria_. yale univ . press . ( 1973 ) b. von stengel , a. van den elzen and d. talman . computing normal form perfect equilibria for extensive two - person games , _ econometrica _ * 70*(2 ) , 693 - 715 . we give an outline of how to modify our first construction , so as to show that in following a sequence of approximate fixpoints of brouwer functions , or equilibria of games , @xmath2 may have to change direction exponentially many times , and furthermore , oscillate between values whose difference is bounded away from zero . @xmath18 shall be the same as in theorem [ thm : pspace - brouwer ] . we construct a modified form of @xmath19 , which we will call @xmath383 , as follows . we identify two subsets of the cubelets @xmath68 , @xmath384 and @xmath385 , defined as @xmath386 , and @xmath387 . these subsets are `` thin layers '' of cubelets that are perpendicular to the @xmath112-axis . now we define how @xmath383 behaves on points at the centers of cubelets . for @xmath27 at the center of @xmath68 , @xmath388 for points not at the centers of cubelets , @xmath383 shall interpolate between the values at the nearest cubelet centers , using the same general approach as @xmath19 . let @xmath389 . let @xmath6 be the set of cubelets @xmath68 for which @xmath390 , so @xmath6 is the region where fixpoints of @xmath391 may exist . let @xmath392 be the connected component of @xmath6 which contains the cubelet @xmath393 , so @xmath392 is the region within which fixpoints of @xmath391 will occur on the homotopy path . by construction , @xmath392 has the property that @xmath394 ( and similarly @xmath395 ) has exponentially many connected components . this follows from the construction of @xcite from which @xmath6 is derived ; @xmath6 simulates a @xmath54-graph by , for each edge of the @xmath54-graph , including a long sequence of cubelets that passes through both @xmath384 and @xmath385 . now consider points in @xmath396 . the claim is that fixpoints @xmath27 of @xmath391 for which @xmath397 , satisfy @xmath398 , and that for fixpoints @xmath27 of @xmath391 with @xmath399 we have @xmath400 . the general idea ( in the first case ; the second case is similar ) is that @xmath401 will map points @xmath112 in @xmath384 to points @xmath402 for which at least one component of @xmath403 is greater than @xmath404 . hence for @xmath405 , @xmath391 raises the value of this component ( the positive contribution from @xmath383 exceeds the negative contribution from @xmath18 ) and prevents it from being a fixpoint . the homotopy path must pass through this long sequence of regions that require @xmath406 or else @xmath407 . moreover , the two types of regions alternate , so we establish the following result : for continuous functions defined using arithmetic circuits , the sequence of fixpoints along the path given by the linear homotopy @xmath408 has exponentially many alternations of the value of @xmath2 . we obtain the following corollary : for graphical or two - player games , suppose @xmath1 is a game that assigns each player a dominating strategy , and @xmath0 is an arbitrary game . the linear tracing procedure for the homotopy @xmath165 will , in the worst case , have exponentially many reversals of @xmath2 . the corollary follows since , the way we represent brouwer functions parameterized by @xmath2 in terms of games parameterized by @xmath2 , does not change the value of @xmath2 . since we did not change @xmath18 , the associated game @xmath1 is the same `` dominating strategy '' game of section [ sec : lt ] . we can use the machinery of chen et al . @xcite so that when we talk about the hardness of finding an @xmath29-fixpoint , @xmath29 is allowed to be inverse polynomial rather than inverse exponential . this is achieved by using the snake embeddings of @xcite . snake - embeddings a snake embedding reduces a low - dimensional brouwer - mapping function having @xmath44 of cubelets in each dimension , to a @xmath409 dimensional bmf having @xmath410 cubelets in each dimension , in such a way that panchromatic vertices of the high - dimensional bmf efficiently encode panchromatic vertices of the low - dimensional bmf . the reduction can be decomposed into a sequence of @xmath409 iterations , in which at each iteration , the number of cubelets along some axis is reduced by a constant factor , and we acquire an additional axis having @xmath410 cubelets ( in @xcite it is in fact 8 cubelets ) . ( intuitively , the space is folded a constant number of times and gains thickness along the new dimension . ) a specific snake - embedding , and some notation we consider a snake - embedding of a 3-dimensional bmf @xmath89 of the type of proposition [ prop : graph ] . initially the colors are @xmath75 ; let @xmath411 denote the new color at the @xmath134-th iteration and let @xmath212 be the number of iterations required to reduce to 8 the number of cubelets along each axis . let @xmath413 be the unit @xmath45-dimensional cube . partition @xmath413 into cubelets @xmath414 where @xmath415 represents a cubelet of edge length @xmath416 . a _ brouwer - mapping circuit _ maps each such cubelet to one of the colors @xmath417 . again , a bmf should be polynomial - time computable , and map exterior cubelets to color @xmath134 for cubelets whose @xmath134-th coordinate contains the first 0 ( when @xmath418 contains a 0 ) , otherwise color 0 . other types of bmf that correspond to definition [ def : bmf ] are defined analogously . the high - dimensional bmf can be computed by a brouwer - mapping circuit @xmath342 that is polynomial in the size of @xmath73 . let @xmath419 be the function computed . the challenge is to implement @xmath419 using an arithmetic circuit that is polynomial in the size of the circuit that computes the bmf , and computes a lipschitz continuous function . the simplicial - decomposition technique of theorem [ thm : bmf2circ ] no longer works when we move to non - constant dimension , since the number of simplices per cubelet is exponential in the dimension . the cubelets having colors @xmath421 are mapped to cubelets in the @xmath420-dimensional instance in such a way as to have the same neighborhood structure , but with some duplication at the folds of the embedding . the continuous implementation define @xmath422 as follows . if @xmath27 lies at the center of a cubelet ( of length @xmath416 ) , letting @xmath423 , @xmath424 , where @xmath425 if @xmath426 is the background color @xmath427 , otherwise @xmath428 where the position of the non - zero entry depends on the color @xmath426 , and is chosen to satisfy the boundary conditions of a bmf . if @xmath27 does not lie at the center of a cubelet , we claim that for each color @xmath426 , we can efficiently compute the @xmath113 distance from @xmath27 to the closest center of a cubelet having color @xmath426 , using a linear arithmetic circuit . let @xmath429 $ ] be this distance . let @xmath430 . then define @xmath431 . in that expression for @xmath432 , @xmath433 is a constant vector , so we are not multiplying two computed quantities together ( which is disallowed in a linear arithmetic circuit ) . why it works if we are not within distance @xmath381 of a panchromatic vertex , then @xmath434 where @xmath426 is one of the missing colors . however , there is some @xmath435 for which @xmath436 . the choice of the vectors @xmath433 ensure that @xmath437 . consequently any approximate fixpoint of @xmath438 is close to a panchromatic vertex of @xmath419 . to show that it is close enough to permit that panchromatic vertex to be efficiently reconstructed from the coordinates of the fixpoint , it is easiest to assume that there are 24 rather than 8 cubelets along each axis , with the original cubelets having been divided into 27 smaller ones all having the same color . then an approximate fixpoint can be assumed to lie within @xmath439 of a panchromatic vertex . we also need to point out that the high - dimensional bmf @xmath419 has a path of @xmath440-chromatic cubelets which simulates the path of @xmath90-chromatic cubelets in the bmf @xmath89 . thus , the one obtained by following the path in @xmath419 , encodes the one obtained by following the corresponding path in @xmath89 . by way of a final remark , it is necessary for us to make a snake embedding of our 3d graph into higher dimension , rather than ( as in @xcite ) reduce from the 2d version of the problem @xcite . this is because the ppad - completeness of 2d sperner @xcite is a reduction that alters the structure of the end of the line graph being encoded , and so would not ( in an obvious way ) apply in a reduction from oeotl .

the rapid neutron capture process ( r - process ) is one of main processes to synthesize elements heavier than the iron group . the dominant r - process element source in the universe is still matter of debate . two possible sites have been proposed ; core - collapse supernovae ( ccsne , e.g. , * ? ? ? * ; * ? ? ? * ) , and coalescence of binaries of double neutron star ( ns - ns ) and black hole - neutron star ( bh - ns ) systems ( e.g. , * ? ? ? * ; * ? ? ? we refer to scenarios based on chemical evolution models of r - process elements exclusively supplied from binary neutron star mergers ( nsms ) as the nsm scenarios in the following . one promising way to understand the enrichment history of heavy elements in the universe is observations of extremely metal - poor ( emp ) stars . these stars are the living fossils born in the early universe and can be probes for the galactic chemical evolution . europium ( eu ) is commonly used as an element representative of r - process elements but we also discuss with barium ( ba ) abundance as another representative element since it is more easily observed in emp stars than eu . though ba in the solar system is mainly produced by s - process in intermediate mass stars on the asymptotic giant branch ( agb ) , the contribution from s - process is thought to be negligible for emp stars since their formation epoch is in prior to the period when the first intermediate mass stars in the universe evolve to agb stars . we note that the measured abundance ratios between eu and ba in emp stars are similar to the ratio of the r - process components of these elements in the solar system , except for carbon - enhanced emp stars , of which the surface abundances have probably been changed by binary mass transfer . it is known that the r - process abundances of emp stars with @xmath0}\lesssim -2.5 $ ] show a very large star - to - star scatter of @xmath1 ( e.g. , * ? ? ? * ; * ? ? ? the average abundance of ba shows a decreasing trend as metallicity decreases in the metallicity range of @xmath2 } < -2.6 $ ] , while it seems to reach a plateau at @xmath0 } < -3.3 $ ] @xcite . based on this trend , @xcite and @xcite pointed out that we can detect ba on almost all the emp stars . it has been pointed out from the viewpoint of chemical evolution that r - process elements originating from neutron star mergers ( nsms ) have difficulties in realizing the observed abundance features in emp stars . @xcite argued that the model with nsm as the dominant r - process source predicts too large abundance scatter since nsms have a very low event rate ( @xmath3 in the milky way ( mw ) , * ? ? ? * ) and very large r - process yield ( @xmath4 , e.g. , * ? ? ? * ) per event . in addition , it has been argued that the long delay - time of nsm ( @xmath5yr ) is incompatible with the presence of r - process enhanced stars at @xmath0 } < -2.5 $ ] . on the other hand , from other points of view , there is growing evidence that the nsms are major r - process sources . theoretical studies of nucleosynthesis in ccsne show difficulties in synthesizing elements of the second and third r - process peaks , in the framework of neutrino driven explosion ( e.g. , * ? ? ? * ; * ? ? ? * ) , while these elements are successfully synthesized in the nsm ejecta ( e.g. , * ? ? ? * ; * ? ? ? observationally , a short gamma - ray burst grb 130603b showed enhancement of infrared radiation in its afterglow @xcite . this event agrees with the predicted properties of a kilonova " , radiation from matter heated by radioactive decay of r - process ejecta from a nsm @xcite . @xcite pointed out that the discrepancy in abundances of the radioactive r - process element @xmath6pu between the early solar system and the current solar vicinity may indicate a very low event rate for the r - process source . the discovery of an r - process rich ultra - faint dwarf galaxy @xcite also favors the r - process source with a large yield and rare event rate . recently , @xcite pointed out that the ejecta from a nsm can spread into very large volume since the nsm ejecta expand at very large velocities of @xmath7 of the speed of light ( e.g. , * ? ? ? * ; * ? ? ? the stopping length , @xmath8 , of @xmath9eu with speed of 0.2@xmath10 is @xmath11kpc , which can be larger than the radius of a dwarf spheroidal galaxy . ( see section 2.3.1 for the derivation of @xmath8 . note that @xcite evaluate this value as @xmath12 , which is in error by a factor of 153 . ) we investigate effects of this large - scale spreading of the nsm ejecta on the galactic chemical evolution . recent chemical evolution studies confirmed that the emergence of eu around @xmath0}\sim -3 $ ] can not be explained by the nsm scenario , unless binary nss coalesce on very short timescales of 10 myr or less @xcite , while both observations of binary pulsars @xcite and models of binary population synthesis @xcite indicate average coalescence timescales beyond 100 myr . in addition , even with the short delay time , the secondary r - process source such as ccsne is invoked to reproduce the abundance distribution at @xmath0 } < -3 $ ] @xcite . @xcite show that the appearance of r - process elements at @xmath0}\sim -3 $ ] can be explained by the combination with the coalescence timescale of 100 myr for the majority ( 95% ) of nsms and 1 myr for the rest ( 5% ) , if the star formation efficiencies are lower for less - massive sub - halos . but they use a one - zone model and do not account for the scatter of @xmath13 . @xcite follow the enrichment history of r - process elements by chemo - dynamical simulations but only for a dwarf galaxy . @xcite and @xcite compute the formation of the mw by chemo - dynamical simulations but their models lack the resolution to correctly follow the formation of pop iii and early pop ii stars . we note that the results of these simulation studies are dependent on assumptions about the sub - grid mixing process and numerical resolution . all these previous studies do not take into account the difference of the propagation process between nsm ejecta and sn ejecta . in our previous study , we investigated the enrichment of r - process elements using a chemical evolution model with merging history of proto - galaxies , but without considering a difference in propagation @xcite . we pointed out that the observed very low frequency of ba lacking stars is in contradiction with the nsm scenario even when we assume a significant surface pollution . in this paper , we revisit this problem by considering the possibility that nsm ejecta can escape from its host proto - galaxy and pollute neighboring proto - galaxies and intergalactic medium ( igm ) . the structure of this paper is as follows . next section describes our model and its ingredients . section [ results ] presents results and compares them with observations . section [ summarys ] summarizes the paper . we use the hierarchical chemical evolution model presented in @xcite . in the following , we give a brief summary of the model and changes from the previous works . our model treats individual metal poor stars in the framework of the hierarchical galaxy formation . this model also considers gas infall and outflow , metal pre - enrichment in igm , and surface pollution of stars by accretion of interstellar matter ( ism ) . we build merger trees of dark - matter halos based on the extended press - schechter theory @xcite . the first stars are formed in halos with @xmath14 , where @xmath15 is the virial temperature , and @xmath16 corresponds to the halo mass of @xmath17 at a redshift of @xmath18 . we refer to the baryonic component in the branches of the merger tree as proto - galaxies , which are low - mass galaxies as building blocks of the milky way . each halo hosts one proto - galaxy at the beginning . we assume that when two dark halos merge , the proto - galaxy in the more massive halo is placed at the center of the newly formed halo and the proto - galaxy in the smaller halo becomes a satellite in a sub - halo and that the mass of the smaller halo remains unchanged . we neglect tidal stripping and ram pressure stripping . the orbital energy of the satellite decays through dynamical friction @xcite . the satellite merges with the central galaxy on the dynamical friction timescale , @xmath19 , given by the standard chandrasekhar formula , and stars in the satellite are added to the stellar halo of the central galaxy ( e.g. , * ? ? ? the star formation rate , @xmath20 , is expressed as the product of the gas mass , @xmath21 , and the star formation efficiency ( sfe ) , @xmath22 . the masses of individual stars are set randomly following the initial mass function ( imf ) described in section [ s : imf ] . metal yields of supernovae ( sne ) are adopted from theoretical results of @xcite for type ii sne and @xcite for type ia sne . we assume homogeneous mixing of metal in each proto - galaxy . we consider gas outflow triggered by individual sne . the first sn in a proto - galaxy form a galactic wind , and subsequent sne add gas and energy to the wind . for the outflow energy , @xmath23 , we use the following formula as a function of the sn kinetic energy , @xmath24 , and the gas binding energy , @xmath25 , of a proto - galaxy , @xmath26 which supposes that the outflow carries away all the sn kinetic energy if @xmath27 but a fraction , @xmath28 , of the kinetic energy turns to the outflow if @xmath29 . we compute inhomogeneous metal enrichment in the igm by following the evolution of the winds from proto - galaxies assuming the momentum conserving snow - plow shell model . in the fiducial model , we adopt the star formation efficiency depending on the mass of a halo , @xmath30 , motivated by the observational mass - metallicity relation , while a constant efficiency was adopted in our previous studies . the observed correlation between stellar masses , @xmath31 , and average metallicities , @xmath32}\right>$ ] , of galaxies is expressed as @xmath33}\right > } \propto m_*^{0.3}$ ] @xcite , and may indicate a lower star formation efficiency in a lower mass galaxy @xcite . we use @xmath34 instead of @xmath35 to avoid @xmath36 at the beginning . measurements of rotation curves of low mass star forming galaxies showed that the halo mass is almost proportional to the stellar mass @xcite . we also present results of the model with a constant sfe but with a higher outflow rate . we adopt @xmath37 in the constant sfe model rather than @xmath38 as the fiducial value in our previous study . this model yields a slope of mass - metallicity relation very similar to the fiducial model with @xmath39 and @xmath38 . we evaluate the proportional constant of the sfe to yield the solar metallicity in the current mw , @xmath40 in the fiducial model and @xmath41 in the constant sfe model . for the initial mass function ( imf ) , @xmath42 , we use the log - normal function with a power - law tail at higher mass part as @xcite adopted , @xmath43 @xcite yielded @xmath44 , @xmath45 , and @xmath46 for the galactic spheroid . for emp stars , however , we have shown that the low - mass peaked imf obtains a much smaller frequency of carbon enhanced stars and more numerous emp stars than observed by the he s survey @xcite . a high or intermediate - mass peaked imf shows a better consistency with observations . here , we compute models with the imfs peaked at @xmath47 , and @xmath48 ( @xmath49 is also shifted in proportion to @xmath50 ) . the binary fraction is set to be 0.5 , and we adopt the flat mass ratio distribution . we note that the predicted r - process abundance distribution is not sensitive to the choice of the peak mass . ( in previous studies , we may have overestimated the volume observed by the he s survey due to the assumption of @xmath51 for giant stars and applying an incorrect limiting magnitude . the correction slightly lower the estimated value of @xmath50 but it still remains in the intermediate mass range . ) we consider that the surface of a star is polluted by accreting ism . we trace changes of surface abundance of stars with @xmath0 } < -2.5 $ ] using the bondi accretion formula . the surface pollution can be significant on the surface abundances of emp stars , as shown in @xcite . } $ ] of proto - galaxies against redshift , with the gas mass of proto - galaxies color coded . bottom panel : metallicity distribution functions ( mdfs ) of metal - poor stars in the mw halo . the green , red , and blue lines show the predicted number of giant stars in the survey volume of the he s survey for models with @xmath52 , respectively . the dashed red line shows the mdf after surface pollution . the shaded histogram is the raw ( dark - gray ) and bias - corrected ( light - gray ) mdfs by the he s survey @xcite . the solid black line is the high - resolution sample biased toward lower metallicity by the saga database @xcite and the dashed black line is the sdss / segue sample by @xcite , scaled to fit the he s data . the color scale shows the predicted number density and the black crosses are the observational sample from the saga database . , title="fig : " ] + } $ ] of proto - galaxies against redshift , with the gas mass of proto - galaxies color coded . bottom panel : metallicity distribution functions ( mdfs ) of metal - poor stars in the mw halo . the green , red , and blue lines show the predicted number of giant stars in the survey volume of the he s survey for models with @xmath52 , respectively . the dashed red line shows the mdf after surface pollution . the shaded histogram is the raw ( dark - gray ) and bias - corrected ( light - gray ) mdfs by the he s survey @xcite . the solid black line is the high - resolution sample biased toward lower metallicity by the saga database @xcite and the dashed black line is the sdss / segue sample by @xcite , scaled to fit the he s data . the color scale shows the predicted number density and the black crosses are the observational sample from the saga database . , title="fig : " ] + the top panel of figure [ mdf ] shows the metal enrichment history in our fiducial model . we plot @xmath0}$ ] of sampled proto - galaxies against redshift . color denotes the gas mass of proto - galaxies . in our model , @xmath0}$ ] can decrease when a proto - galaxy merges with a more metal deficient proto - galaxy . most of stars with @xmath0}<-3 $ ] are formed at @xmath53 , and stars with @xmath54 } < -2 $ ] are formed at @xmath55 . we show the resultant metallicity distribution functions ( mdfs ) in the bottom panel . the green , red , and blue lines show the predicted mdfs in the cases of @xmath52 , respectively . the solid and dashed red lines are the intrinsic metallicity and the surface metallicity after pollution , respectively . shaded histograms show the raw ( dark - shaded ) and bias - corrected ( light - shaded ) mdfs of the he s sample @xcite . we scale the predicted mdfs considering the field of view ( @xmath56 ) and the magnitude limit ( @xmath57 ) of the he s survey @xcite . since most of the sample stars in @xcite are around the base of the red giant branch , we count the predicted number of giant stars in the survey volume assuming that the typical luminosity is @xmath58 . we employ the de vaucouleur density profile for the mw halo . we also show the mdf of the sdss / segue sample ( * ? ? ? * dashed black line ) scaled to match the he s data and the mdf by the saga database ( solid black line ) , which collects stellar abundances by high - resolution spectroscopy of metal - poor stars from literatures @xcite . as seen in the bottom panel , though all the three imfs lead to a similar form of the mdf , the imf with @xmath59 solely reproduces the total number . the imf with a higher @xmath50 produces a fewer number of low - mass stars which can survive to date , and at the same time form more numerous massive stars producing iron to shorten the timescale for low - metallicity star formation . we show results of the models with @xmath59 in section [ results ] . the main results about r - process elements remain true for the other two @xmath50 . the mdf with @xmath59 slightly underestimates the number of stars at @xmath0}\gtrsim -2.5 $ ] . it may be due to the change of the imf to the low - mass peaked one @xcite . our model obtains the abundance distribution of alpha elements consistent with observations . we present the correlation of the abundance ratios of magnesium to iron with metallicity in figure [ mgfe ] , and oxygen and silicon shows the similar abundance distributions . the mass range for progenitor stars of nss is @xmath60 . we assume that @xmath61 of binary systems in this mass range form ns binaries which coalesce within the age of the universe . this yields one coalescing ns binary out of 1000 massive stars . we adopt the delay time distribution of the nsm , @xmath62 for @xmath63 following results of binary population synthesis by @xcite . in this case , the averaged delay time is @xmath64 . the nsm event rate is proportional to the following integration , @xmath65 the ejecta mass of a nsm is assumed to be @xmath66 , based on the estimation of ejecta mass of the observed kilonova candidate grb 130603b @xcite . we assume the scaled solar r - process abundance pattern @xcite for nsm ejecta . the elemental yields of ba and eu are @xmath67 and @xmath68 , respectively . we do not distinguish ns - ns mergers and bh - ns mergers in our computation , while recent numerical studies show that a bh - ns merger can eject lager mass than a ns - ns merger @xcite . in this paper , we do not consider other heavy element sources such as s - process in intermediate massive stars or r - process in sne ii . a significant fraction of the nsm ejecta can escape from proto - galaxies . nearby proto - galaxies can capture the escaped r - process elements . the rest of r - process ejecta travel away into the intergalactic space with high velocities . the r - process elements cool and are coupled with igm , and are incorporated into galaxies along gas infall as described in the following . r - process elements from a nsm propagate until they lose kinetic energies by interaction with ism . the dominant energy deposition process is the ionization of neutral hydrogen in hi gas , and the coulomb scattering in the ionized gas . the energy loss rate is @xmath69 \nonumber \\ & \times n_{\rm hi } ( 1 + 0.0185 \ln \beta ) \beta^{-1 } \hbox { ev s}^{-1 } \end{aligned}\ ] ] in the neutral gas and @xmath70 in the ionized plasma @xcite , where @xmath71 and @xmath72 is the number density of neutral hydrogen and free electron , respectively , @xmath73 is the velocity in units of the speed of light and set at @xmath74 in this study , and @xmath75 is the atomic number . the stopping length , @xmath76 for @xmath77 in the neutral gas , where @xmath78 is the proton mass , @xmath79 is the mass number , and @xmath80 is the lorentz factor . we assume that r - process elements ejected from an nsm travel the stopping length of @xmath8 on a straight line at a constant velocity . the ejected particles stop rather suddenly once they travels the stopping length since the energy loss rate increases as the energy decreases . the fraction , @xmath81 , of nsm ejecta escaping from the proto - galaxy is estimated as @xmath82 where @xmath83 is the radius of the host proto - galaxy . for simplicity , we assume that a proto - galaxy is a homogeneous sphere of neutral gas , and an nsm takes place at the center of the proto - galaxy . we give the gas density and radius of proto - galaxies assuming isobaric cooling of the virialized gas @xcite , i.e. , @xmath84 , and @xmath85 , where @xmath86 is the averaged density of the virialized dark - matter halos as a function of redshift , @xmath87 is the mean molecular weight , @xmath88 is the virial radius of the host halo . gas can be cooled to @xmath89 k in the mw but @xmath90 k in the primordial mini - halo , and @xmath91 k in relic hii regions without metal . the gas temperature , @xmath92 , is set at the maximum of the temperature of the cosmic microwave background or @xmath93 . these assumptions yield @xmath94 and @xmath95 pc for the primordial proto - galaxy with @xmath96 at @xmath97 , and @xmath98 and @xmath99 kpc for the current mw ( @xmath100 ) . in this study , we assume isotropic mass ejection from a nsm @xcite . the fraction of escaped nsm ejecta encountering with a proto - galaxy with radius @xmath101 is @xmath102 , where @xmath103 is the distance to the proto - galaxy . for a particle which runs into a spherical proto - galaxy with an impact parameter of @xmath104 , the length of the pass crossing the proto - galaxy is @xmath105 . under the same assumptions as in eq . ( [ eq : fesc ] ) , the fraction of particles that are not captured by the proto - galaxy is @xmath106 . the fraction of captured high - energy r - process elements out of all particles encountering the proto - galaxy is @xmath107 we mention that , in the limit of @xmath108 , the capture rate approaches to @xmath109 and the total amount of captured particles is proportional to @xmath110 since @xmath111 ( eq . [ eq : ls ] ) , i.e. , the capture rate is simply proportional to the gas mass but independent of radius and density . we estimate the distance between the @xmath112-th and the @xmath113-th proto - galaxies using the following equation , d_n , m(t ) = r(t|m_t , n , m , t_c , n , m ) [ eq : dhalo ] + ( t < t_c , n , m ) + d_sat , m(t ) = r_vir , n ( 1- ) [ eq : dsat ] + ( t_c , n , m t < t_c , n , m + _ df ) + [ eq : d2sat ] + in the case of two proto - galaxies in different dark halos ( eq . [ eq : dhalo ] ) , we estimate the distance between them under the spherical collapse approximation of halos following @xcite . here two halos are supposed to merge at time @xmath114 to become a single halo with the mass of @xmath115 , and @xmath104 is the distance between the two halos ( see section 4.1 in @xcite for details ) . after the two dark halos merge ( eq . [ eq : dsat ] ) , the proto - galaxy in the more massive halo ( labeled @xmath112 ) is placed at the center of the merged halo and the other one ( labeled @xmath113 ) becomes a satellite . the satellite galaxy is assumed to approach the center as the distance , @xmath116 , from the center linearly decreases as a function of time . @xmath117 is the merger timescale by the dynamical friction and we use the formula given in @xcite . at @xmath118 , the satellite merges to the central galaxy . in the case of two satellites in the same dark halo ( eq . [ eq : d2sat ] ) , we estimate the distance assuming the random distribution for the orbital angles , @xmath119 , between the satellite galaxies . the energy loss timescale , @xmath120 , of @xmath9eu propagating in the igm is given as @xmath121 where we use the energy loss rate for neutral gas ( eq . [ eq : omeganeutral ] ) before reionization and one for ionized gas ( eq . [ eq : omegaplasma ] ) after that . we assume the reionization redshift @xmath122 . since the stopping length in the igm , @xmath123 , is larger than the computation box , we assume homogeneous mixing of the escaped r - process elements in the intergalactic space , while we follow the inhomogeneous metal pollution of the igm for elements provided by sne as mentioned above . r - process elements having escaped from the host galaxy lose their kinetic energies on the timescale of @xmath124 unless they are captured by another proto - galaxy . such r - process elements are eventually settled in the igm and some of them are attracted by the gravity of proto - galaxies and accreted onto them along with gas infall . we can describe the evolution of mass @xmath125 of the high - energy particle in the intergalactic space and the abundance , @xmath126 , of element @xmath127 in the igm as follows , @xmath128 @xmath129 where @xmath130 is the elemental yields from one nsm event , @xmath131 is the time when the @xmath132-th nsm event takes place in the @xmath112-th proto - galaxy . the second term in the right hand side of eq . ( [ eq : he ] ) describes capture of particles incoming from outside of the computation box , and @xmath133 is the volume of the box , which is a comoving sphere enclosing the total mass of @xmath134 . @xmath135 is the gas accretion rate onto the @xmath112-th proto - galaxy . here , we assume that the escaping of the high energy r - process elements from the simulation box balances the incoming from galaxies outside the box . we discuss the validity of this assumption in appendix . in summary , evolution of r - process abundances of a proto - galaxy is described as follows , @xmath136 where the first term on the right - hand side is the contribution from nsms in the proto - galaxy , the second term is the increment by capturing ejecta from nsm events in other proto - galaxies in the simulation box , the third term is high - energy r - process background from galaxies outside the computation box , and the fourth term is the infall of igm . of proto - galaxies , with the number of nsm events in them color coded . we randomly sample @xmath137 of proto - galaxies and plot their abundances . the dash - dotted blue line denote the abundance of high - energy eu in intergalactic space , @xmath138 , and the dashed black line is the low - energy eu abundance in igm . ] figure [ ce ] shows the enrichment history of eu abundances for sampled proto - galaxies and the igm ( dashed black line ) . the number of nsm events in each proto - galaxy is color coded . the dash - dotted blue line denotes the abundance of high energy r - process elements in intergalactic space , @xmath139 . the first nsm in this computation run takes place at @xmath140 . most of the r - process ejecta escape from its host proto - galaxy to enhance @xmath138 to @xmath141 , \end{aligned}\ ] ] where @xmath142 is the solar eu abundance @xcite , and @xmath143 and @xmath144 are densities of matter and baryon relative to the critical density , respectively . in the limit of @xmath108 , the capture rate of high energy r - process particles is proportional to the total gas mass and independent of the density as mentioned , i.e. , the capture rate per unit mass is almost the same for all proto - galaxies with @xmath108 and for the igm . of the igm ( black line ) and proto - galaxies ( cyan lines ) increase in a similar way by capturing high - energy r - process elements at @xmath145 . at such a high redshift , majority of baryon is still in the igm , and most of the escaped nsm ejecta are mixed with the igm on the timescale of @xmath146 . at @xmath122 , @xmath138 suddenly decreases because the energy loss rate in the igm increases due to cosmic reionization . the capture rate per unit mass by the ionized igm becomes several tens of times higher than the one by neutral proto - galaxies . therefore , the abundance of proto - galaxies without nsm events ( cyan lines ) is @xmath147 dex or more smaller than the igm abundance at @xmath148 . after the reionization , pre - enrichment of igm is mainly responsible for the eu abundance of proto - galaxies without nsm events . at @xmath149 , @xmath138 overwhelms @xmath150 because the energy loss timescale in the igm increases as the igm density decreases . a single nsm event enriches its host proto - galaxy with r - process elements to @xmath151 \end{aligned}\ ] ] in the limit of @xmath152 ( @xmath108 ) . when we adopt assumptions for @xmath112 and @xmath101 in section [ s : fesc ] , we can describe this equation as a function of @xmath153 and the cosmological parameters as follows , @xmath154 , \ ] ] where @xmath155 is the overdensity of the virialized halo relative to the critical density , and @xmath156 is the dimensionless hubble parameter . these proto - galaxies become much more r - process rich than the intergalactic space , and the contribution from other proto - galaxies is negligible for them . at @xmath157 , the most massive photo - galaxy shows the highest r - process abundance because of a large event rate and a small escape probability . figures [ rfe ] and [ outflow ] show the r - process abundance of stars against metallicity for the fiducial model and the constant sfe model , respectively . the left and right panels denote results without escape of nsm ejecta ( @xmath158 ) and with escape , respectively . the constant sfe significantly delays the enrichment of r - process elements compared to iron and the predicted abundances of emp stars do not match observational results , as mentioned in previous chemical evolution studies . the predicted abundances of r - process elements are significantly lower than observations at @xmath0}\lesssim -2 $ ] , and r - ii stars with @xmath159 around @xmath0}\sim -3 $ ] is not reproduced . on the other hand , if the sfe is lower at lower mass proto - galaxies , r - process elements emerge at @xmath0}\sim -3 $ ] , even with the long average delay time ( @xmath160 yr ) . this is because that enrichment timescale of iron is longer for low - mass proto - galaxies in which emp stars were formed and some nsms can take place within the formation epoch of emp stars . these are consistent with the result of the one zone model by @xcite . when we do not consider the escape of nsm ejecta from the host proto - galaxies , however , the majority of stars with @xmath0 } < -3 $ ] show very low r - process abundances with [ ba / fe ] @xmath161 , which is apparently at odds with observations ( see the top left panel of fig . 3 ) . on the other hand , our escape model predicts @xmath162 or @xmath163 for the majority of stars at @xmath0}\lesssim -3 $ ] , and well reproduces observations of halo stars , as shown in the right top panel . proto - galaxies lacking nsm events are polluted by r - process elements ejected from nsms in other proto - galaxies . as mentioned , igm pre - enrichment is mostly responsible for the r - process elements in these stars , and capture of high - energy r - process elements also contribute at @xmath164 . stars in proto - galaxies without a nsm are distributed at @xmath165 . nsm events of short delay time , @xmath166 , are the dominant sources of r - process elements for these stars . on the other hand , proto - galaxies in which nsm events take place form stars with @xmath167 . surface pollution can play an important role for stars in proto - galaxies without a nsm . without surface pollution , there emerges a depression in the number density of stars at @xmath0}\sim -3 $ ] in the range of @xmath168 where the first nsm occurs . the surface pollution alleviates the gap . in the model with @xmath158 , the surfaces of most emp stars have a tiny amount of ba such as @xmath169 even after surface pollution . at @xmath0 } > -2 $ ] , the r - process abundance is averaged . the predicted ba abundance is lower than observations since we neglect the s - process contribution . at @xmath0}\gtrsim -1 $ ] , while we can see a clear decreasing trend of @xmath170 as metallicity increases , we do not see a similar knee " in the predicted r - process abundance distributions . this is because the increase of the nsm rate due to the long delay time compensates the increasing production rate of iron enhanced by sne ia . one possible explanation for the observed depression of @xmath13 is a lower frequency of nsms with high metallicity progenitor stars , although we assume constant frequency in the computation for simplicity . the event rate of nsms in this model is @xmath171 in the current mw , as shown in figure [ nsmrate ] . it is also slightly higher than the estimation based on the observations of binary pulsars in the mw ( @xmath172 , * ? ? ? both the binary pulsars and the abundances of the metal rich stars indicate a factor of @xmath173 smaller nsm fraction than used in this paper at @xmath174 . at the right axis of figure [ nsmrate ] , the event rate in the mw is converted into the expected detection rate of the gravitational waves ( gws ) by the advanced ligo - virgo network . ongoing observations of gws will provide better constraints on the nsm rate . for the metallicity dependence of the nsm rate , studies of binary population synthesis predict a lower ns - ns merger rate for lower metallicity stars @xcite . for the bh - ns merger rate , on the other hand , @xcite predict that a rate for stars with the metallicity of @xmath175 is higher than the population i ( pop i ) stars while @xcite predict that a rate for pop iii stars is lower than that for pop i stars . further studies are required for the event rate , delay time distribution , and r - process yield of nsms , and their metallicity dependences . in this paper , we assume that high - energy r - process elements go straight until they lose energies by interaction with gas particles . if proto - galaxies have magnetic field , however , particles can travel along a winding path , which decreases the escape probability and increases the capture rate . on the other hand , the escape probability can be higher in a disk - like gas density profile than in a gas sphere assumed in this work . in addition , nsm events at the outer region of a proto - galaxy will effectively transfer r - process elements into intergalactic space , although we only consider nsm events at the center of proto - galaxies . furthermore , a ns binary itself can go away from a proto - galaxy if it can get enough velocity from sn explosions . the velocity of a ns binary is depending on its binary orbit and sn kick . although we still have no reliable estimation for the velocity distribution of ns binaries , two observed binary pulsars show a proper motion of more than @xmath176 km / s @xcite , which exceeds the typical escape velocity of a proto - galaxy hosting emp stars . the igm pre - enrichment and transfer of r - process elements between proto - galaxies are depending on these details , but a model taking them into account is beyond the scope of this paper . coalescence of ns binaries is one of the most plausible sites for r - process nucleosynthesis but results of previous chemical evolution models with nsms as major r - process sources is inconsistent with the observed r - process abundances of metal - poor stars . we revisit the galactic chemical evolution of r - process elements following the nsm scenario . the novelty of this paper is in considering the effect of propagation of nsm ejecta across proto - galaxies . as pointed out by @xcite , r - process elements from nsms have very large stopping lengths due to their high velocities . we compute the escape of r - process elements from proto - galaxies , pre - enrichment of igm , and capture of the escaped elements by other proto - galaxies in the chemical evolution model based on the framework of the hierarchical galaxy formation . we have confirmed that the nsm scenario can reproduce the emergence of r - process elements at @xmath0}\sim -3 $ ] in the case of lower sfe for less massive proto - galaxies , while the constant sfe delays the enrichment of r - process elements , which results in abundance correlations incompatible with observations . while previous chemical evolution models with the nsm scenario predict many stars lacking r - process elements at very low metallicities , our model with the long - distance propagation of nsm ejecta predicts @xmath165 for stars formed in proto - galaxies without a nsm . our model succeeds in reproducing the r - process abundance distribution of emp stars . this result indicates that the nsm scenario is not rejected by the galactic chemical evolution , when we consider the propagation of nsm ejecta beyond the proto - galaxies . our current model predicts a slightly higher event rate of nsms and results in @xmath177 higher than observations in the current mw while the good consistency with observations at very low metallicity . it may indicate the metallicity dependence of the nsm rate . future observations of gws and their electromagnetic counterparts will give us information about the nsm rate and its contribution to the galactic chemical evolution . we adopt a periodic boundary condition which assumes that the high energy r - process elements incoming from outside the simulation box balance the elements escaping from the box . this seems to be a reasonable assumption at very high redshift since large scale density fluctuation tends to grow at low redshift . for example , at @xmath97 , only @xmath178 fluctuation of @xmath179 collapses , and the box size is much larger . on the other hand , it is not a good approximation at low redshift since the environment around the current mw is significantly different from the average of the universe . in the current universe , there are @xmath180 mw equivalent galaxies ( mwegs ) on average @xcite , but the number density in our computation box corresponds to @xmath181 ( @xmath182 ) . it may indicate that the production rate of high energy r - process elements ( the first term of eq . [ eq : he ] ) may be overestimated by factor @xmath183 at @xmath174 . the capture rate by proto - galaxies ( the second term of eq . [ eq : he ] ) may also be overestimated since the capture rate in this computation box is dominated by the mw at low redshift . in order to mimic this difference , we compute the chemical evolution model with replacing the eq . ( [ eq : he ] ) to the following equation , @xmath184 where @xmath185 is the relative number density of mwegs . but for the different boundary condition . incoming rate of high energy r - process particles from boundaries are smaller than outgoing rate by proportional to the relative density of milky way equivalent galaxies . see text for details . ] figure [ boundary ] show the result . the predicted abundance distribution is indistinguishable from the fiducial model at high metallicity or high r - process abundance ( @xmath167 ) . this is because r - process elements of proto - galaxies in which these stars are formed are dominated by their own nsms . on the other hand , the abundance of proto - galaxies without nsm events decreases since the r - process element abundances of the igm decrease . the abundance of emp stars at @xmath186 formed before the first nsm in their host proto - galaxies decreases by @xmath187 dex . we note that the difference can be overestimated at low metallicity since the region around the mw should be more similar to the average of the universe at high redshift . abadie , j. , abbott , b. p. , abbott , r. , et al . , 2010 , cqgra , 27 , 3001 andrievsky , s. m. , spite , m. , korotin , s. a. , 2009 , , 494 , 1083 arlandini , c. , kappeler , f. , wisshak , k. , gallino , r. , lugaro , m. , busso , m. , & straniero , o. 1999 , , 525 , 886 argast , d. , samland , m , thielemann , f .- k . , & qian , y .- z . 2004 , , 416 , 997 bauswein , a. , goriely , s. , & janka , h .- , 2013 , , 773 , 78 barnes , j. , & kasen , d. , 2013 , , 775 , 18 beniamini , p. , & piran , t. 2016 , , 456 , 4089 belczynski , k. , kalogera , v. , & bulik , t. 2002 , , 572 , 407 berger , e. , fong , w. , chornock , r. 2013 , , 774 , l23 burris , d. l. , pilachowski , c. a. , armandroff , t. e. , sneden , c. , cowan , j. j. , & roe , h. 2000 , , 544 , 302 carollo , d. , beers , t. c. , chiba , m. , et al . 2010 , , 712 , 692 cescutti , g. , romano , d. , matteucci , f. , chiappini , c. , & hirschi , r. 2015 , , 577 , 139 chabrier , g. 2003 , , 115 , 763 christlieb , n. , schrck , t. , frebel , a. , et al . 2008 , , 484 , 721 cole , s. , lacey , c. g. , baugh , c. m. , & frenk , c. s. , 319 , 1 dominik , m. , belczynski , k. , fryer , c. , et al . 2012 , , 759 , 52 goriely , s. , bauswein , a. , & janka , h .- t . 2011 , , 738 , l32 grevesse , n. , noels , a. , & sauval , a. j. 1996 , aspc , 99 , 117 hirai , y. , ishimaru , y. , saitoh , t. r. , fujii , m. s. , hidaka , j. , & kajino , t. 2015 , , 814 , 41 honda , s. , aoki , w. , kajino , t. , ando , h. , beers , t. c. , izumiura , h. , sadakane , k. , takada - hida , m. 2004 , , 607 , 474 hotokezaka , k. , kiuchi , k. , kyutoku , k. , et al . 2013 , phrvd , 87 , 4001 hotokezaka , k. , kyutoku , k. , tanaka , m. , et al . 2013 , , 778 , l16 hotokezaka , k. , piran , t. , & paul , m. natph , 11 , 1042 lacey , c. , & cole , s. 1993 , , 262 , 627 ishimaru , y. , wanajo , s. , & prantzos , n. , 2015 , , 804 , l35 ji , a. p. , frebel , a. , chiti , a. , & simon , j. d. , nature , 531 , 610 kim , c. , perera , b. b. p. , mclaughlin , m. a. , 2015 , , 448 , 928 kinugawa , t. , inayoshi , k. , hotokezaka , k. , et al . 2014 , , 442 , 2963 kirby , e. n. , cohen , j. g. , guhathakurta , p. , cheng , l. , bullock , j. s. , & gallazzi , a. 2013 , , 779 , 102 kobayashi , c. , umeda , h. , nomoto , k. , tominaga , n. , & ohkubo , t. 2006 , , 653 , 1145 komiya , y. , habe , a. , suda , t. , & fujimoto , y. m. 2009 , , 696l , 79 komiya , y. , suda , t. , & fujimoto , y. m. 2009 , , 694 , 1577 komiya , y. , yamada , s. , suda , t. , & fujimoto , m. y. 2014 , , 783 , 132 komiya , y. , suda , t. , & fujimoto , y. m. 2015 , , 808 , l47 komiya , y. , suda , t. , & fujimoto , y. m. 2015 , , accepted . lattimer , j. m. , & schramm , d. n. , 1974 , , 192 , 145 lorimer , d. r. 2008 , llr , 11 , 8 matteucci , f. , romano , d. , arcones , a. , korobkin , o. , & rosswog , s. 2014 , , 438 , 2177 miller , s. h. , ellis , r. s. , newman , a. b. , & benson , a. 2014 , , 782 , 115 nomoto , k. , thielemann , f .- k . , yokoi , k. 1984 , , 286 , 644 roederer , i. u. 2013 , , 145 , 26 roederer , i. u. , karakas , a. i. , pignatari , & herwig , f. , 2016 , , 821 , 37 rosswog , s. , liebendrfer , m. , thielemann , f .- k . , davies , m. b. , benz , w. , & piran , t. 1999 , , 341 , 499 sato , k. , 1974 , pthph , 51 , 726 schlickeiser , r. 2002 , cosmic ray astrophysics ( berlin : springer ) schrck , t. , christlieb , n. , et al . 2009 , , 507 , 817 shen , s. , cooke , r. j. , ramirez - ruiz , e. , madau , p. , mayer , l. , guedes , j. 2015 , , 807 , 115 somerville , r. s. , & kolatt , t. s. 1999 , , 305 , 1 suda , t. , et al . 2008 , pasj , 60 , 1159 suda , t. , et al . 2013 , , 432 , l46 tanaka , m. , & hotokezaka , k. , , 775 , 113 tanvir , n. r. , levan , a. j. , fruchter , a. s. , et al . 2013 , nature , 500 , 547 tsujimoto , t. , & shigeyama , t. 1999 , , 565 , l5 van de voort , f . , quataert , e. , hopkins , p. f. , keres , d. , & faucher - gigure , c - a . , 2015 , , 447 , 140 wanajo , s. , 2013 , , 770 , l22 wanajo , s. , sekiguchi , y. , nishimura , n. , et al . 2014 , , 789 , l39 wanajo , s. , janka , h - t . , & mller , b. 2011 , , 726 , l15 wehmeyer , b. , pignatari , m. , & thielemann , f .- k . , 2015 , , 452 , 1970 woosley , s. e. & hoffman , r. d. 1992 , , 395 , 202 yamada , s. , suda , t. , komiya , y. , aoki , w. , & fujimoto , m. y. , , 436 , 1362

light - cone gauge string field theory provides a useful way to define string theories @xcite@xcite@xcite . given the action , it is possible to define the amplitudes and calculate them perturbatively , although we should check if they are well - defined . since it is a gauge fixed theory , there is no problem in considering a light - cone gauge string field theory in noncritical space - time dimensions . being noncritical , it does not possess the space - time lorentz invariance . it should correspond to a string theory in a lorentz noninvariant background . in other words , we should be able to find a brst invariant worldsheet theory , with a nonstandard @xmath0 part . what we would like to do in this paper is to study this @xmath0 theory . we give the energy - momentum tensor and the action of the theory and calculate the correlation functions . we show that the energy - momentum tensor possesses desired properties , and construct a brst invariant formulation of the worldsheet theory . the reason why we are interested in this theory is that it can be used to regularize the string field theory . the dimensional regularization is one of the most powerful regularizations for ordinary quantum field theory and it may be useful also in string theory . in particular , in the light - cone gauge superstring field theory , unwanted divergences occur even at the tree level , because of the transverse supercurrent insertions at the interaction points of the vertices . in ref . @xcite , we have proposed a dimensional regularization scheme to deal with these divergences . we have found that the divergences of the tree amplitudes can be regularized by shifting the number of the space - time dimensions . we have checked that the results of the first quantized formalism are reproduced without any counterterms for the four - point case . in order to proceed further we need to show that the dimensional regularization preserves important symmetries of the theory . if the light - cone gauge string field theory corresponds to a brst invariant formulation even in noncritical dimensions , it means that the dimensional regularization preserves the brst symmetry . in this paper , we deal with the case of closed bosonic string field theory . we restrict ourselves to the tree amplitude and consider the worldsheet theory on the complex plane . the organization of this paper is as follows . in section [ sec : lccov ] , we present a way to rewrite light - cone gauge string amplitudes in a brst invariant form . we follow the procedure in noncritical dimensions , and show what kind of worldsheet theory of @xmath0 should appear in the end . in section [ sec : xpmcft ] , we study the theory for @xmath3 and show that it is a cft with the right virasoro central charge . namely , combining the cft for @xmath3 constructed here with the worldsheet theory for the light - cone gauge strings in noncritical dimensions , we obtain a cft with the central charge @xmath4 by which we can define a brst invariant worldsheet theory . in section [ sec : brst ] , we show that the tree amplitudes in the noncritical case can be written in a brst invariant form . section [ sec : discussions ] is devoted to discussions . in appendix [ sec : action ] , we present the action of the light - cone gauge bosonic string field theory in @xmath5 dimensions . in appendix [ sec : mandelstam ] , the mandelstam mapping is given . in appendix [ sec : gamma ] , we present a way to calculate @xmath6 $ ] , which is used in sections [ sec : lccov ] and [ sec : xpmcft ] . in appendix [ sec : relations ] , the details of the calculations in section [ sec : xpmcft ] are presented . in order to study the noncritical case , it is useful to consider the relation between light - cone gauge amplitudes and covariant ones for critical strings . one can calculate the amplitudes starting from the light - cone gauge string field theory action given in appendix [ sec : action ] . the tree amplitudes can be expressed by path integral on the light - cone diagrams , on which the complex @xmath7 coordinate is introduced as usual . via the mandelstam mapping @xmath8 , which is given in appendix [ sec : mandelstam ] , one can express them using correlation functions of vertex operators on the @xmath9-plane endowed with the metric @xmath10 the correlation functions we should consider are @xmath11_{\phi } e^{-s_{x^{i } } } \prod_{r=1}^{n}v_{r}^{\mathrm{lc}}\;. \label{eq : correlation}\ ] ] here @xmath12 denotes the free action for the @xmath13 transverse coordinates @xmath14 and @xmath15 denotes the vertex operator . we assume that the @xmath16-th external state is of the form @xmath17 and the vertex operator should be @xmath18 where @xmath19 is defined in eq.([eq : rho - zir ] ) , @xmath20 is the coordinate of the unit disk of the @xmath16-th external string given in eq.([eq : rho - wr ] ) and @xmath21 . the on - shell and the level - matching conditions require that @xmath22 the path integral measure @xmath23_{\phi}$ ] should be defined using the metric ( [ eq : metric ] ) . it is related to the measure @xmath23 $ ] which is defined with the flat metric @xmath24 as @xmath25_{\phi } & \propto & \left[dx^{i}\right]e^{-\gamma\left[\phi\right]}\;,\end{aligned}\ ] ] roughly speaking , @xmath26 $ ] is given by the liouville action @xmath27 \sim -\frac{24}{24\pi}\int d^{2}z \ , \partial \phi \bar{\partial}\phi ~ , \label{eq : liouville}\ ] ] where @xmath28 . since @xmath29 diverges at the poles and zeros of @xmath30 by the definition ( [ eq : metric ] ) , the liouville action is not well - defined at these points . $ ] can be evaluated by regularizing the divergences and carefully taking various effects into account @xcite@xcite . we present an alternative derivation of @xmath31 $ ] in appendix [ sec : gamma ] and give an explicit form in eq.([eq : gamma - c ] ) . , this form of @xmath31 $ ] is only up to a constant which can be fixed by factorization . with the string field action in eq.([eq : sftaction ] ) and the @xmath31 $ ] in eq.([eq : gamma - c ] ) , one can see that the factor is fixed as @xcite @xmath25_{\phi } & \sim & \left[dx^{i}\right ] \mathrm{sgn}\left(\prod_{r=1}^n\alpha_r\right ) e^{-\gamma\left[\phi\right]}\;,\end{aligned}\ ] ] up to a numerical factor . we ignore the phase factor @xmath32 in the following , because it does not play any important roles . ] thus eq.([eq : correlation ] ) can be rewritten as @xmath33 e^{-s_{x^{i}}-\gamma[\phi ] } \prod_{r}v_{r}^{\mathrm{lc}}\;. \label{eq : correlation2}\ ] ] in order to covariantize the amplitudes , we need to introduce the longitudinal coordinates . the light - cone gauge condition implies that @xmath34 equals the lorentzian time on the light - cone diagram , which means on the @xmath9-plane , @xmath35 therefore we introduce the variable @xmath34 with the delta function @xmath36 , which can be expressed as @xmath37 e^{-\frac{1}{\pi}\int d^{2}z x^{\prime -}\partial\bar{\partial}x^+ } \prod_{r=1}^{n } e^{-ip_{r}^{+}x^{\prime-}}\left(z_{r},\bar{z}_{r}\right ) \nonumber \\ & & \hspace{5 mm } \times \int \left [ db^{\prime}dc^{\prime}d\tilde{b}^{\prime}d\tilde{c}^{\prime } \right ] e^{-\frac{1}{\pi}\int d^{2}z \left ( b^{\prime}\bar{\partial}c^{\prime } + \tilde{b}^{\prime } \partial \tilde{c}^{\prime}\right ) } c'(\infty ) \tilde{c}'(\infty ) ~. \label{eq : x+delta}\end{aligned}\ ] ] eq.([eq : x+delta ] ) should be considered as the formal euclideanized version of a lorentzian path integral . the grassmann odd fields @xmath38 of conformal weights @xmath39 , @xmath40 , @xmath41 , @xmath42 are introduced to cancel the determinant factor @xmath43 . with these variables , we can rewrite the right hand side of eq.([eq : correlation2 ] ) as @xmath44 e^{-s_{x^{i}}-s_\pm^\prime -s_{b^{\prime}c^{\prime } } } \nonumber\\ & & \hspace{10em } \times c^\prime ( \infty ) \tilde{c}^\prime ( \infty ) \prod_{r=1}^{n } \left(v_{r}^{\mathrm{lc}}e^{-ip_{r}^{+}x^{\prime- } } \left(z_{r},\bar{z}_{r}\right ) \right ) \ ; , \label{eq : correlation3}\end{aligned}\ ] ] where @xmath45 \;. \label{eq : xprimeaction}\end{aligned}\ ] ] we consider @xmath46 as the worldsheet action for these variables . the action @xmath47 is with the interaction term which depends only on @xmath34 . this interaction term does not affect the correlation functions with less than two insertions of @xmath48 , and they coincide with those of the free theory . thus one can show the ope @xmath49 from the worldsheet action , the energy - momentum tensor can be obtained as @xmath50 when @xmath51 . here @xmath52 denotes the schwarzian derivative . this term can be derived from the variation of @xmath53 which coincides with the liouville action except for the singular points . the covariant expression for the amplitudes can be obtained by introducing @xcite @xmath54 and their anti - holomorphic counterparts @xmath55 and @xmath56 . the fields @xmath57 and @xmath58 have now weights @xmath59 and @xmath60 respectively . we should also introduce @xmath61 so that the ope s between @xmath62 and the ghosts are regular and the energy - momentum tensor ( [ eq : emtensor ] ) takes the form @xmath63 which coincides with the energy - momentum tensor in the conformal gauge . in the following , we would like to rewrite the correlation function ( [ eq : correlation3 ] ) using these new variables . let us first consider the vertex operator . from eq.([eq : emtensor2 ] ) , one can see that the action for the new variables should be the free action . @xmath64 appears essentially in the same way as @xmath65 does in eq.([eq : x+delta ] ) , and we obtain the delta function . therefore @xmath8 and @xmath66 which appear in the integrand can be replaced by @xmath67 and @xmath68 , where @xmath69 and @xmath70 are the holomorphic and the anti - holomorphic part of @xmath34 respectively . we will denote the equality which holds under this identification by @xmath71 . thus the factor @xmath72 in the definition ( [ eq : vrlc ] ) of @xmath73 can be rewritten as @xmath74 where @xmath75 is the ddf operator given by @xmath76 one can also use @xmath77 for @xmath78 , where @xmath79 is defined in eq.([eq : nbarrr00 ] ) . using these , one can show that @xmath80 in eq.([eq : correlation3 ] ) subject to the on - shell condition ( [ eq : on - shell - cov ] ) can be rewritten as @xmath81 where @xmath82 is the vertex operator corresponding to the ddf state . one subtle point to notice in eq.([eq : defbc ] ) is that the variables @xmath83 should have zeros and poles at the zeros and poles of @xmath84 . since the integration over @xmath62 leads to the identification @xmath85 , the zeros and poles of @xmath84 become those of @xmath30 , namely @xmath86 and @xmath87 . accordingly @xmath57 should be inserted at @xmath86 and @xmath58 should be inserted at @xmath87 , if we wish to rewrite eq.([eq : correlation3 ] ) @xcite@xcite@xcite@xcite . these insertions should come with appropriate factors made from @xmath88 or @xmath34 in order that eq.([eq : correlation3 ] ) is reproduced . using eqs.([eq : lcddf ] ) and ( [ eq : exp - gamma2 ] ) , one can show that eq.([eq : correlation3 ] ) can be rewritten as @xmath89 e^{-s_{x}-s_{bc } } \ , \left|\sum_{r}\alpha_{r}z_{r}\right|^{2 } \left ( \lim_{z\to\infty}\frac{1 } { \left|z\right|^{4 } } c\left(z\right)\tilde{c}\left(\bar{z}\right)\right ) \nonumber \\ & & \hspace{1 cm } \times\prod_{i } \left ( \frac{b}{\partial^{2}\rho}(z_{i } ) \frac{\tilde{b}}{\bar{\partial}^{2}\bar{\rho}}(\bar{z}_{i})\right ) \prod_{r=1}^{n } \left(c\tilde{c } v_{r}^{\mathrm{ddf}}\right ) \left(z_{r},\bar{z}_{r}\right ) \ ; , \label{eq : correlation4}\end{aligned}\ ] ] where @xmath90 and @xmath91 are the free actions for @xmath92 and for @xmath83 respectively . one can show that eq.([eq : correlation4 ] ) yields the expression for the amplitudes in the conformal gauge . using @xmath93 and deforming the contour , one can recast eq.([eq : correlation4 ] ) into the form @xmath89 e^{-s_{x}-s_{bc}}\nonumber \\ & & \hspace{1 cm } \times\prod_{i } \left ( \oint_{c_{i}}\frac{dz}{2\pi i}\frac{b}{\partial\rho}(z ) \oint_{c_{i } } \frac{d\bar{z}}{2\pi i } \frac{\tilde{b}}{\bar{\partial}\bar{\rho } } ( \bar{z } ) \right ) \prod_{r=1}^{n } \left(c\tilde{c}v_{r}^{\mathrm{ddf}}\right ) \left(z_{r},\bar{z}_{r}\right)\ ; , \label{eq : correlation5}\end{aligned}\ ] ] where the integration contour @xmath94 is depicted in fig . [ fig : npt ] in appendix [ sec : gamma ] on the @xmath7-plane . the amplitudes can be obtained by integrating the correlation function @xmath95 over the @xmath96 moduli parameters @xmath97 defined in eq.([eq : mathcalt ] ) as @xmath98 one can see that the antighost insertion @xmath99 corresponds to the quasiconformal vector field associated with the moduli parameter @xmath100 . the form of the amplitude is brst invariant , because the vertex operator @xmath101 is invariant and @xmath102 yields a total derivative with respect to @xmath100 . we can change the integration variables to @xmath103 and @xmath104 with the expression ( [ eq : correlation5 ] ) for @xmath95 , the determinant factor can be combined with the antighost factors as @xmath105 performing the contour integrals around @xmath106 in the last line , we eventually obtain the familiar expression @xmath107 e^{-s_{x}-s_{bc}}\nonumber \\ & & \hspace{1 cm } \times \prod_{s=1,2,n } \left(c\tilde{c}v_{s}^{\mathrm{ddf}}\right ) \left(z_{s},\bar{z}_{s}\right ) \prod_{r=3}^{n-1 } \int d^2z_r v_{r}^{\mathrm{ddf}}\left(z_{r},\bar{z}_{r}\right)\;.\end{aligned}\ ] ] hence the variables @xmath108 can be identified with those in the covariant formulation . from eq.([eq : newvariables ] ) , one can find the ope s of the variables @xmath109 to be @xmath110 and regular otherwise . these can also be deduced from the action ( [ eq : xprimeaction ] ) . using these ope s , one can easily show that the energy - momentum tensor ( [ eq : emtensor ] ) satisfies the virasoro algebra of central charge @xmath111 . the light - cone gauge string field theory in @xmath5 @xmath112 space - time dimensions can be defined with the action given in appendix [ sec : action ] . this time , the correlation functions we should consider is_{\phi } & \sim & \left[dx^{i}\right ] \mathrm{sgn}\left(\prod_{r=1}^n\alpha_r\right ) e^{-\frac{d-2}{24}\gamma\left[\phi\right]}\;,\end{aligned}\ ] ] and we ignore the phase @xmath32 in the following . ] @xmath33 e^{-s_{x^{i}}-\frac{d-2}{24}\gamma[\phi ] } \prod_{r}v_{r}^{\mathrm{lc}}\;. \label{eq : noncriticalf}\ ] ] we can follow the above procedure and introduce the variables @xmath113 without any problem . in this case , eq.([eq : emtensor ] ) should be @xmath114 by using @xmath108 defined in eq.([eq : newvariables ] ) , @xmath115 can be rewritten as @xmath116 and the action @xmath117 should be @xmath118\;. \label{eq : sxpm}\ ] ] therefore in noncritical dimensions the worldsheet theory for @xmath3 is different from the usual free theory , and obviously the lorentz symmetry is broken . what we should study is this theory for @xmath3 . it is a conformal field theory similar to the one for @xmath119 and @xmath65 in the previous subsection . we will study its properties in the next section . before closing this subsection , a comment is in order . for @xmath120 , one can define @xmath121 with @xmath122 using these variables , the energy - momentum tensor ( [ eq : nenergymomentum ] ) can be written as @xmath123 therefore the worldsheet theory is a free theory with ghosts of noninteger spins . we may be able to study the theory using these variables , although we need to figure out the way to deal with the ghost zero - modes . the theory we would like to consider is with the action ( [ eq : sxpm ] ) and the energy - momentum tensor @xmath124 this theory will be well - defined if @xmath84 has a nonvanishing expectation value . as in the case of the theory for @xmath119 and @xmath65 , we always consider this theory in the presence of the insertions of the vertex operators @xmath125 ( @xmath126 ) with @xmath127 , so that the classical equation of motion implies @xmath128 thus the quantities we would like to calculate are the expectation values @xmath129 \prod_{r=1}^{n } e^{-ip_{r}^{+}x^{-}}\left(z_{r } , \bar{z}_{r}\right ) \right\rangle } \nonumber\\ & & \equiv \int\left[dx^{+}dx^{-}\right ] e^{-s_{x^\pm}}f\left[x^{+},x^{-}\right ] \prod_{r=1}^{n } e^{-ip_{r}^{+}x^{-}}\left(z_{r } , \bar{z}_{r}\right),\end{aligned}\ ] ] for functionals @xmath130 $ ] which satisfy @xmath131 = f\left[x^{+},x^{-}\right]~,\ ] ] for arbitrary constants @xmath132 . for functionals @xmath133 $ ] which do not depend on @xmath64 , it is formally possible to perform the path integral as in eq.([eq : x+delta ] ) and obtain @xmath134 \prod_{r=1}^{n } e^{-ip_{r}^{+}x^{-}}\left(z_{r } , \bar{z}_{r}\right ) \right\rangle \sim f\left[-\frac{i}{2}\left(\rho+\bar{\rho}\right)\right ] \exp \left(-\frac{d-26}{24}\gamma \left[\ln\left(\partial \rho \bar{\partial}\bar{\rho}\right)\right ] \right ) ~ , \label{eq : fx+}\ ] ] up to the factor coming from the integration over the zero - modes of @xmath0 . it is convenient to define @xmath135\right\rangle_{\rho } \equiv \frac { \left\langle f\left[x^{+},x^{-}\right ] \prod_{r=1}^{n } e^{-ip_{r}^{+}x^{-}}\left(z_{r } , \bar{z}_{r}\right ) \right\rangle } { \left\langle \prod_{r=1}^{n } e^{-ip_{r}^{+}x^{-}}\left(z_{r } , \bar{z}_{r}\right ) \right\rangle } ~,\ ] ] and we obtain @xmath136\right\rangle _ { \rho } = f\left[-\frac{i}{2}\left(\rho+\bar{\rho}\right)\right]\;. \label{eq : expx+}\ ] ] the manipulation to obtain eq.([eq : fx+ ] ) is rather formal because @xmath138 possesses zeros and poles and we need to specify the regularization procedure to define @xmath139 $ ] . as we mentioned in the last section , this was done in ref . @xcite and @xmath140 $ ] is given for arbitrary mandelstam mapping @xmath7 . assuming that eq.([eq : fx+ ] ) is true with @xmath53 derived in ref . @xcite ( and also in appendix [ sec : gamma ] ) , one can calculate correlation functions which involve @xmath64 . roughly speaking , differentiating the left hand side of eq.([eq : fx+ ] ) with respect to @xmath141 and setting @xmath142 , we obtain @xmath134x^-(z_n,\bar{z}_n ) \prod_{r=1}^{n-1 } e^{-ip_{r}^{+}x^{-}}\left(z_{r } , \bar{z}_{r}\right ) \right\rangle ~,\ ] ] although the momentum conservation condition complicates the procedure a bit . in the same way , we can in principle obtain arbitrary correlation functions , treating the right hand side of eq.([eq : fx+ ] ) as a kind of generating functional . since all the correlation functions on the complex plane are given this way , we define the theory based on eq.([eq : fx+ ] ) , rather than starting from the action ( [ eq : sxpm ] ) . if one define the theory in this way , it is not a priori clear if @xmath143 in eq.([eq : emtensorxpm ] ) can be considered as the energy - momentum tensor . in the rest of this section , we would like to calculate the correlation functions of @xmath143 and examine if it can be regarded as the energy - momentum tensor of the theory . one comment is in order . the operator @xmath144 can be considered to create a hole of length @xmath145 in the light - cone diagram . therefore it is similar to the macroscopic observables in the old matrix models @xcite . these operators are nonlocal objects and give rise to singularities at @xmath146 where no operators are inserted . taking @xmath147 limit to obtain local operators is exactly what was done in the old matrix models . in order to calculate the correlation functions of @xmath143 , we need correlation functions of @xmath64 . let us follow the above - mentioned procedure and calculate them . first we consider the simplest example , @xmath148 . in order to calculate this quantity we start from @xmath149 \right ) ~,\ ] ] where @xmath150 and @xmath151 then the one point function @xmath148 can be given as @xmath152 \right ) \right|_{\alpha_{0}=0 } \;. \label{eq : expdx-}\ ] ] it is straightforward to evaluate the right hand side of eq.([eq : expdx- ] ) by using the expression ( [ eq : gamma - c ] ) for @xmath6 $ ] . in the calculations , one should notice the following points . while there are @xmath153 interaction points @xmath154 for @xmath155 , there are only @xmath156 interaction points @xmath86 for @xmath157 . since @xmath158 what happens in the limit @xmath159 is as follows : one of @xmath154 s , which we denote by @xmath160 , goes to @xmath161 ; one of @xmath154 s , which we denote by @xmath162 , goes to @xmath163 ; the other interaction points @xmath164 go to the interaction points @xmath86 for @xmath7 , which are denoted with the same subscripts . as we mentioned earlier , @xmath165 with finite @xmath166 should be considered as a nonlocal operator , which induces singularities at the interaction points where there are no operator insertions . @xmath167 in the limit @xmath168 is consistent with the fact that @xmath165 tends to a local operator . one subtle point to notice is that if @xmath169 is close to one of @xmath170 and the interaction point nearest to @xmath171 coincides with @xmath160 , the neumann coefficient @xmath172 does not go to @xmath173 in the limit @xmath168 . since @xmath174 $ ] depends on @xmath172 , it implies that @xmath165 has a nonlocal effect even if we take @xmath168 in such a configuration . then we can not expect to get correlation functions of local operators from @xmath174 $ ] . in the following , we will assume that @xmath169 and @xmath163 are not close to any of @xmath170 , to avoid such a situation . namely we assume that @xmath175 and @xmath176 are not in the regions of external propagators in the @xmath7 plane . we calculate @xmath177 for such @xmath169 , and analytically continue the result to the whole complex plane . one can generalize eq.([eq : expdx- ] ) and calculate the correlation functions with insertions of @xmath34 as @xmath180 \right\rangle _ { \rho } \nonumber \\ & & \hspace{5 mm } = \left . \frac { 2i \partial_{z_0 } \partial_{\alpha_0 } \left\langle f\left[x^{+}\right ] \prod_{r=0}^{n+1 } e^{-ip_{r}^{+}x^{-}}\left(z_{r } , \bar{z}_{r}\right ) \right\rangle } { \left\langle \prod_{r=0}^{n+1 } e^{-ip_{r}^{+}x^{-}}\left(z_{r } , \bar{z}_{r}\right ) \right\rangle } \right|_{\alpha_{0}=0 } \nonumber \\ & & \hspace{5 mm } = \left\langle \partial x^{-}\left(z_{0}\right ) \right\rangle _ { \rho } f \left [ -\frac{i}{2 } \left ( \rho+\bar{\rho } \right ) \right ] + \int d^{2}z \ , \frac{1}{z_{0}-z } \left . \frac{\delta f\left[x^{+}\right ] } { \delta x^{+ } ( z,\bar{z } ) } \right|_{x^{+}=-\frac{i}{2}\left(\rho+\bar{\rho}\right)}\;. \label{eq : expxpm}\end{aligned}\ ] ] since we are dealing with the correlation functions with source terms for @xmath64 , we can read off the operator relations from these correlation functions . from eqs.([eq : expx+ ] ) and ( [ eq : expxpm ] ) , we obtain the ope s @xmath181 which are valid if @xmath9 and @xmath182 are away from the singularities @xmath183 . these are consistent with eq.([eq : ope ] ) . the two point function for @xmath184 can be calculated by using eq.([eq : dx- ] ) as @xmath185 here we are interested in the singularity at @xmath186 . it is straightforward to calculate the right hand side of eq.([eq : twopoint ] ) and obtain @xmath187 + \mbox{regular terms}\;. \label{eq : dx-2}\ ] ] from this , we deduce the ope @xmath188 \nonumber\\ & \sim & { } -\frac{d-26}{12 } \left[- \frac{1}{(z - z')^{4 } } \frac{6}{\left(\partial x^{+}(z')\right)^{2 } } - \frac{1}{(z - z')^{3 } } 3 \partial \left ( \frac{1}{\left ( \partial x^{+ } ( z ' ) \right)^{2 } } \right ) \right . \nonumber\\ & & \hspace{5em } \left . { } -\frac{1}{(z - z')^{2 } } \frac{1}{2 } \partial^{2 } \left ( \frac{1}{\left ( \partial x^{+ } ( z ' ) \right)^{2 } } \right ) \right]~ , \label{eq : ope3}\end{aligned}\ ] ] which is valid if @xmath9 and @xmath182 are away from the singularities @xmath183 . this is consistent with eq.([eq : ope ] ) . we can also obtain an expression for the correlation functions with @xmath34 insertions as we did in eq.([eq : expxpm ] ) : @xmath189 \right\rangle _ { \rho } } \nonumber\\ & = & \left\langle \partial x^{-}\left(z\right ) \partial x^{-}\left(w\right)\right\rangle _ { \rho } f\left[-\frac{i}{2}\left(\rho+\bar{\rho}\right)\right ] \nonumber \\ & & \quad { } + \left\langle \partial x^{-}\left(z\right ) \right\rangle _ { \rho } \int d^{2}w ' \ , \frac{1}{w - w ' } \left.\frac{\delta f\left[x^{+}\right ] } { \delta x^{+ } ( w',\bar{w}^\prime ) } \right|_{x^{+}=-\frac{i}{2}\left(\rho+\bar{\rho}\right ) } \nonumber \\ & & \quad { } + \left\langle \partial x^{-}\left(w\right ) \right\rangle _ { \rho } \int d^{2}z ' \ , \frac{1}{z - z ' } \left.\frac{\delta f\left[x^{+}\right ] } { \delta x^{+ } ( z',\bar{z}^\prime ) } \right|_{x^{+}=-\frac{i}{2}\left(\rho+\bar{\rho}\right ) } \nonumber \\ & & \quad { } + \int d^{2}z ' \int d^{2}w ' \ , \frac{1}{z - z ' } \frac{1}{w - w ' } \left.\frac{\delta^{2}f\left[x^{+}\right ] } { \delta x^{+}(z',\bar{z } ' ) \delta x^{+ } ( w',\bar{w } ' ) } \right|_{x^{+}=-\frac{i}{2}\left(\rho+\bar{\rho}\right)}\;. \label{eq : expdx - dx - f}\end{aligned}\ ] ] one can evaluate the right hand side of eq.([eq : vevtpm ] ) and examine how it behaves around the possible singularities : @xmath194 \right ) } { \partial z_{r } } & z\sim z_{r } \\[1ex ] \displaystyle \mathrm{regular } & z\sim z_{i } \\[0.5ex ] \mathcal{o } \left(\frac{1}{z^{4 } } \right ) & z\sim \infty \end{array } \right . . \label{eq : vevtpm2}\ ] ] eqs.([eq : vevtpm2 ] ) and ( [ eq : exptf ] ) imply that @xmath195 is regular at @xmath196 and @xmath197 , if no operators are inserted there . therefore , although @xmath198 is singular at @xmath199 without any operator insertions , the energy momentum tensor is regular and conserved . this property is essential for constructing the brst charge . from eq.([eq : vevtpm2 ] ) for @xmath200 we can read off the ope @xmath201 although @xmath202 is a nonlocal operator , it behaves as a primary field of weight @xmath203 . using the ope s ( [ eq : ope - xpm ] ) and ( [ eq : ope3 ] ) , we can show that @xmath143 satisfies @xmath204 therefore the central charge of the virasoro algebra in the @xmath3 cft is @xmath205 . we can also find that @xmath206 and thus @xmath207 are primary fields of weight @xmath208 . since the worldsheet theory for @xmath0 is a cft with virasoro central charge @xmath205 , with the transverse coordinates @xmath14 added the total central charge of the system for @xmath0 and @xmath14 is @xmath4 . therefore with ghosts @xmath57 and @xmath58 , we can construct a nilpotent brst charge @xmath209 . as we have shown in section [ sec : lccov ] , the amplitude for the light - cone gauge string field theory can be rewritten by using these variables . we start from the correlation function given in eq.([eq : noncriticalf ] ) , where the vertex operator @xmath73 is of the form ( [ eq : vrlc ] ) but with the on - shell condition @xmath210 we can proceed in the same way as in section [ sec : lccov ] and eventually obtain @xmath211 e^{-s_{x^{i}}-s_{x^{\pm}}-s_{bc } } \nonumber\\ & & \hspace{1 cm } \times \prod_{i } \left ( \oint_{c_{i}}\frac{dz}{2\pi i}\frac{b}{\partial\rho}(z ) \oint_{c_{i } } \frac{d\bar{z}}{2\pi i}\frac{\tilde{b}}{\bar{\partial}\bar{\rho } } ( \bar{z } ) \right ) \nonumber \\ & & \hspace{1 cm } \times \prod_{r=1}^{n } \left(c\tilde{c}v_{r}^{\mathrm{ddf } } \exp \left(-i\frac{d-26}{24}\frac{x^+}{p_r^+}\right ) \right ) \left(z_{r},\bar{z}_{r}\right ) \nonumber \\ & & \hspace{1 cm } \times \prod_{r=1}^{n } \exp \left(i\frac{d-26}{24}\frac{x^+}{p_r^+}\right ) \left ( z_i^{(r)},\bar{z}_i^{(r ) } \right)~ , \label{eq : brstinv}\end{aligned}\ ] ] where @xmath212 is defined in the same way as in eq.([eq : vddf ] ) . it is easy to show that @xmath213 is a primary field of weight @xmath214 and @xmath143 is regular even with the insertion @xmath215 . therefore eq.([eq : brstinv ] ) gives a brst invariant expression for the amplitude . @xmath213 may look as a vertex operator with momentum @xmath216 instead of @xmath217 , but the momentum which is conserved is @xmath218 . the conserved momentum can be identified with the operator @xmath219 on the worldsheet , which is conserved at the interaction points with insertions @xmath220 . in this paper , we have constructed a brst invariant worldsheet theory which corresponds to the light - cone gauge string field theory in @xmath221 space - time dimensions . the worldsheet theory for the longitudinal coordinate variables @xmath0 is different from the usual free theory , but it is a cft with @xmath222 . our results provide yet another way to construct string theories in noncritical dimensions . the brst invariant formulation will be useful to study d - branes for such string theories . now that the cft is given , we can at least formally construct the interaction vertices of the string field theory based on this cft through the prescription of ref . @xcite . since we have constructed the cft on the worldsheet of the light - cone string diagram , the gauge unfixed version of the string field theory is supposed to possess the joining - splitting type of interactions . such a theory looks similar to the @xmath223 hikko theory given in ref . @xcite . the results in this paper should be generalized to be used in regularizing string field theory . one should consider the @xmath0 on the riemann surfaces with higher genera , in order to check if it works for regularizing the uv and ir divergences . one should also construct a supersymmetric version of the cft . the light - cone gauge superstring field theory in noncritical dimensions can be used to dimensionally regularize the tree amplitudes of the critical theory , as was discussed @xcite . it is possible to generalize the calculations performed in this paper into the superstring case , although they are much more complicated . we will present these results elsewhere . this work was supported in part by grant - in - aid for scientific research ( c ) ( 20540247 ) and grant - in - aid for young scientists ( b ) ( 19740164 ) from the ministry of education , culture , sports , science and technology ( mext ) . in order to fix the notation , we present the action for the light - cone gauge bosonic string field theory in @xmath5 space - time dimensions . the worldsheet variables are @xmath224 . the action of the string field theory takes the form @xmath225~. \label{eq : sftaction}\end{aligned}\ ] ] here @xmath226 is the zero - mode of the transverse virasoro generators for the @xmath16-th string , @xmath227 is the coupling constant , and @xmath228 denotes the integration measure for the momentum zero - modes of the @xmath16-th string defined as @xmath229 where @xmath230 is the string - length parameter of the @xmath16-th string . @xmath231 is the reflector given by @xmath232 @xmath233 denotes the three - string interaction vertex defined as @xmath234 here @xmath235 is the three - string lpp vertex @xcite for the transverse coordinate variables @xmath14 and @xmath236 . the string field @xmath237 is taken to obey the reality and the level - matching conditions . let @xmath7 be the standard complex coordinate on the @xmath153-string tree diagram with the joining - splitting type interaction . the portion on the @xmath7-plane corresponding to the @xmath16-th external string @xmath238 is mapped to the unit disk , @xmath239 , of the @xmath16-th string as @xmath240 where @xmath241 is the coordinate on the @xmath7-plane at which the @xmath16-th string interacts . the @xmath153-string tree diagram is mapped to the complex @xmath9-plane with @xmath153 punctures by the mandelstam mapping @xcite @xmath242 where the puncture @xmath243 correspond to the origin of the unit disk @xmath244 . the @xmath156 interaction points @xmath245 are determined by @xmath246 . these are related to the interaction points on the @xmath7-plane by @xmath247 where @xmath248 denotes one of @xmath86 at which the @xmath16-th external string interacts . the fact that @xmath86 s are the zeros of @xmath249 yields @xmath250 @xmath6 $ ] can be obtained by evaluating the liouville action for the metric ( [ eq : metric ] ) on the @xmath9-plane @xcite@xcite . here we will present an alternative derivation . @xmath251}$ ] can be regarded as the partition function of the light - cone gauge string theory in @xmath4 space - time dimensions from the view point of the worldsheet cft on the light - cone diagram . therefore the variation @xmath252 $ ] under @xmath253 can be given by using the expectation value of the transverse energy - momentum tensor @xmath254 . we can obtain @xmath255 $ ] by integrating @xmath256 $ ] . the result should be up to a factor which does not change under the variation eq.([eq : deltazr ] ) . such a factor can be fixed by imposing the factorization condition in each case @xcite . before we begin the calculation , two comments are in order . because of the constraint @xmath257 , all the @xmath258 s can not be treated as independent variations . here we think of @xmath258 with @xmath259 as independent of each other and @xmath260 as being determined by the relation @xmath261 we also note that @xmath255 $ ] involves divergence originating from the infinite length of the external lines . we regularize it by cutting off the @xmath16-th external line so that its length becomes @xmath262 , where @xmath263 is a large positive constant . by doing so , the contribution from the @xmath16-th external line to the partition function becomes @xmath264 , and the divergences of @xmath255 $ ] become independent of the parameters @xmath265 . on the @xmath9-plane , this corresponds to cutting a hole of radius @xmath266 around @xmath171 , where @xmath266 and @xmath263 are related as @xmath267 the variations ( [ eq : deltazr ] ) correspond to the variations of the following parameters of the light - cone diagram : @xmath269 the moduli parameters @xmath270 @xmath271 the heights @xmath272 of the external cylinders ; @xmath273 the circumferences @xmath274 of the external cylinders . thus the variation @xmath275)$ ] is expressed as @xmath276 \right ) = ( \mathrm{i } ) + ( \mathrm{ii } ) + ( \mathrm{iii})~ , \label{eq : variation0}\ ] ] where @xmath277 + \mathrm{c.c.}~ , \nonumber\\ ( \mathrm{iii } ) & = & \sum_{r=1}^{n-1 } i 2\pi \delta \alpha_{r } \int_{l_{rn } } \frac{d\rho}{2\pi i } \left\langle t_{x^{i } } ( \rho ) \right\rangle + \mathrm{c.c.}~. \label{eq : variation1}\end{aligned}\ ] ] here @xmath278 stands for the complex conjugate . the integration contour @xmath94 of the term @xmath269 lies between the consecutive interaction points @xmath279 and @xmath280 as depicted in fig . [ fig : npt ] . the integration path @xmath281 of term @xmath273 in eq.([eq : variation1 ] ) is a line stretching from the asymptotic region of the @xmath16-th external string to that of the @xmath153-th string on the @xmath7-plane . as an example , the path @xmath282 is depicted in fig . [ fig : npt ] . on the @xmath9-plane , the path @xmath281 becomes a segment connecting the two punctures @xmath87 and @xmath283 with the orientation from @xmath87 to @xmath283 . -string tree diagram with @xmath284 . the contours @xmath94 and the path @xmath282 on the @xmath7-plane of the integrals in eq.([eq : variation1 ] ) are depicted for this case . ] the expectation value of @xmath285 can be evaluated by going to the @xmath9-plane . since @xmath286 we obtain @xmath287 using eq.([eq : vevt ] ) , one can rearrange the term @xmath269 in eq.([eq : variation1 ] ) as @xmath288 where @xmath289 denotes the interaction point closest to @xmath290 . from eq.([eq : delta - alphan ] ) , one can easily find that term @xmath271 becomes @xmath291 we can recast the @xmath16-th term of @xmath273 in eq.([eq : variation1 ] ) into @xmath292 here we have taken the cut of the function @xmath293 in the integrand on the right hand side to be the segment @xmath281 and the integration path @xmath294 to be the sum of the two oriented segments connecting two punctures @xmath87 and @xmath283 as depicted in fig . [ fig : cut ] . deforming the contours to rearrange eq.([eq : log - int ] ) further and summing over @xmath16 , we obtain @xmath295 where @xmath296 and @xmath297 are depicted in fig . [ fig : cut ] . consists of the two oriented segments described by the arrowed bold lines . the dashed line denotes the cut of the integrand on the right hand side of eq.([eq : log - int ] ) . ] eventually we obtain the expression @xmath298 \right ) & = & \left [ { } -\sum_{r=1}^{n } \int_{c_{r } } \frac{dz}{2\pi i } \frac{\delta \rho ( z ) - \delta \rho ( z^{(r)}_{i } ) + \delta \alpha_{r } \ell_{r } } { \partial \rho ( z ) } \left(- 2 \ { \rho , z\ } \right ) \right . \nonumber\\ & & \quad { } - \sum_{i } \oint_{z_{i } } \frac{dz}{2\pi i } \frac{\delta \rho ( z ) - \delta \rho ( z_{i } ) } { \partial \rho ( z ) } \left(- 2 \ { \rho , z\ } \right ) \nonumber\\ & & \left . \quad { } - \oint_{\infty } \frac{dz}{2\pi i } \frac{\delta \rho ( z ) - \delta \rho ( z^{\infty}_{i } ) } { \partial \rho ( z ) } \left ( - 2 \ { \rho , z\ } \right ) \right ] + \mathrm{c.c.}~. \label{eq : variation2}\end{aligned}\ ] ] let us evaluate the contour integrals on the right hand side of eq.([eq : variation2 ] ) . for @xmath299 , the schwarzian derivative @xmath300 behaves as @xmath301 \displaystyle \frac{3}{(z - z_{i})^{2 } } + \frac{1}{z - z_{i } } \frac{\partial \left(-w\right)}{\partial z_{i } } + \mathcal{o } \left ( ( z - z_{i})^{0 } \right ) & z \sim z_{i } \\[2ex ] \mathcal{o } \left(\frac{1}{z^{4 } } \right ) & z \sim \infty \end{array } \right.~ , \label{eq : schwarzian - around}\ ] ] where @xmath179 is a function of @xmath302 defined as @xmath303 for the contour integral along @xmath296 , one should notice that @xmath304 involves a term @xmath305 . this part can be integrated as @xmath306 for @xmath307 . the other terms can be evaluated by taking the residues of the simple poles and one can show @xmath308 where @xmath79 is a neumann coefficient given by @xmath309 and @xmath310 is defined in eq.([eq : rho - zir ] ) . for @xmath311 , one can show @xmath312 using @xmath313 which is derived by varying the equation @xmath314 under eq.([eq : deltazr ] ) . the contour integral around @xmath146 can be obtained as @xmath315 it is easy to see that the integral around @xmath197 vanishes . putting all the pieces together , we obtain @xmath316 = - w -2 \sum_{r=1}^{n } \mathop{\mathrm{re } } \bar{n}^{rr}_{00 } -3\sum_{i}\ln \left| \partial^{2 } \rho ( z_{i } ) \right| ~. \label{eq : gamma - c}\ ] ] this form of @xmath6 $ ] is useful in the calculations in section [ sec : xpmcft ] . we note that @xmath317 can be described as @xmath318 which follows from eqs.([eq : deldelrho ] ) and ( [ eq : zr - zs ] ) . this yields @xmath319 } = \prod_{r=1}^{n } \left| \alpha_{r } \right|^{-2 } \left| \sum_{s=1}^{n } \alpha_{s } z_{s } \right|^{4 } e^{-2 \sum_{r=1}^{n } \mathop{\mathrm{re } } \bar{n}^{rr}_{00 } } \prod_{i } \left| \partial^{2 } \rho ( z_{i } ) \right|^{-1}~. \label{eq : exp - gamma}\ ] ] this expression is the one obtained by the method in ref . we also note that @xmath319 } = \prod_{r=1}^{n } \left| \alpha_{r } \right|^{-1 } e^{-\frac{1}{2}w } \left| \sum_{s=1}^{n } \alpha_{s } z_{s } \right|^{2 } e^{-2 \sum_{r=1}^{n } \mathop{\mathrm{re } } \bar{n}^{rr}_{00 } } \prod_{i } \left| \partial^{2 } \rho ( z_{i } ) \right|^{-2}~ , \label{eq : exp - gamma2}\ ] ] which is used in sections [ sec : lccov ] and [ sec : brst ] . in this appendix , we present the details of the calculations to obtain eqs.([eq : dx- ] ) and ( [ eq : dx-2 ] ) . let us consider the interaction points @xmath321 and @xmath162 for @xmath322 defined in eq.([eq : drhohat ] ) , which tend to @xmath323 and @xmath163 as @xmath159 respectively . from eq.([eq : drhohat ] ) , we can obtain the expansion of @xmath324 and @xmath162 in terms of @xmath325 , @xmath326~ , \nonumber \\ & & z'_{i^{(n+1)}}-z_{n+1 } \nonumber\\ & & \qquad \quad = \frac{\alpha_{0}}{\partial\rho\left(z_{n+1}\right ) } \left[1 -\left ( \frac{\partial^{2}\rho\left(z_{n+1}\right ) } { \left(\partial\rho\left(z_{n+1}\right)\right)^{2 } } + \frac{1}{\partial\rho\left(z_{n+1}\right ) } \frac{1}{z_{n+1}-z_{0 } } \right ) \alpha_{0 } + \mathcal{o } \left ( \alpha_{0}^{2 } \right)\right].~~~~~~~ \label{eq : zi0-z0}\end{aligned}\ ] ] the neumann coefficients @xmath327 for the mandelstam mapping @xmath155 behave as @xmath328 using eqs.([eq : delrho2 ] ) and ( [ eq : zi0-z0 ] ) , we obtain @xmath329\alpha_0 + \mathcal{o}\left ( \alpha_{0}^2 \right),\end{aligned}\ ] ] and @xmath330 \ ; , \nonumber\\ \partial^{2 } \rho ' \left(z'_{i^{\left(n+1\right)}}\right ) & = & \frac{\left(\partial\rho\left(z_{n+1}\right)\right)^{2 } } { \alpha_{0 } } \nonumber\\ & & \ \times \left [ 1 + \left ( \frac{3\partial^{2}\rho\left(z_{n+1}\right ) } { \left(\partial\rho\left(z_{n+1}\right)\right)^{2 } } + \frac{2}{\partial\rho\left(z_{n+1}\right)\left(z_{n+1}-z_{0}\right ) } \right ) \alpha_0 + \mathcal{o } \left(\alpha_{0}^2 \right)\right]~.~~~~~ \ ] ] gathering all the relations obtained above , we have @xmath331 = 6 \ln |\alpha_0 | - 4 + \gamma [ \ln\left(\partial\rho\bar{\partial}\bar{\rho}\right ) ] } \nonumber\\ & & \hspace{4em}{}+ 2 \mathop{\mathrm{re } } \left [ \sum_{r=1}^n \frac{1}{\alpha_r } \ln \left|\frac{\left(z_i^{(r ) } - z_0\right ) \left(z_r- z_{n+1}\right ) } { \left(z_r - z_0\right ) \left(z_i^{(r ) } - z_{n+1}\right ) } \right|\right . \nonumber\\ & & \hspace{8em } { } -\sum_{i } \frac{1 } { \partial^{2 } \rho ( z_{i } ) } \left ( \frac{1}{z_{i } -z_{0 } } - \frac{1}{z_{i } - z_{n+1 } } \right ) \frac{\partial w } { \partial z_{i } } \nonumber\\ & & \hspace{8em } { } -\frac{3}{2}\sum_{i } \frac{1}{\partial^2 \rho(z_i ) } \left ( \frac{1}{\left(z_{i}-z_{0}\right)^{2 } } -\frac{1}{\left(z_{i}-z_{n+1}\right)^{2 } } \right ) \nonumber\\ & & \hspace{8em } \left . { } -\frac{3}{2}\sum_{i } \frac{\partial^3 \rho\left(z_i\right ) } { \left ( \partial^{2}\rho\left(z_{i}\right ) \right)^2 } \left ( \frac{1}{z_{i}-z_{0}}-\frac{1}{z_{i}-z_{n+1 } } \right ) \right]\alpha_0 + \mathcal{o}\left ( \alpha_{0}^2 \right).~~~~\end{aligned}\ ] ] we can see that @xmath332 $ ] is divergent and does not coincide with @xmath333 $ ] . this singularity can be avoided by modifying @xmath334 , which corresponds to a renormalization of the operator @xmath335 . such a renormalization is irrelevant to the calculation of the right hand side of eq.([eq : expdx- ] ) and we obtain @xmath336 \right|_{\alpha_0=0 } = \partial_{z_0 } \left [ \sum_{r=1}^n \frac{1}{\alpha_r } \ln \left(\frac{z_i^{(r ) } - z_0 } { z_r - z_0 } \right)\right . -\sum_{i } \frac{1 } { \partial^{2 } \rho ( z_{i } ) } \frac{1}{z_{i } -z_{0 } } \frac{\partial w } { \partial z_{i } } } \nonumber\\ & & { } \hspace{12em } \left.-\frac{3}{2}\sum_{i } \frac{1}{\partial^2 \rho(z_i ) } \left ( \frac{1}{\left(z_{i}-z_{0}\right)^{2 } } + \frac{\partial^3 \rho\left(z_i\right ) } { \partial^{2}\rho\left(z_{i}\right ) } \frac{1}{z_{i}-z_{0 } } \right ) \right ] . \label{eq : dgamma}\end{aligned}\ ] ] from this equation , we can compute the right hand side of eq.([eq : expdx- ] ) and obtain eq.([eq : dx- ] ) . let us evaluate the right hand side of eq.([eq : twopoint ] ) . this can be evaluated by using eq.([eq : dgamma ] ) with @xmath161 , @xmath337 and @xmath338 replaced by @xmath9 , @xmath339 and @xmath340 respectively and with the range of the index @xmath16 taken to be from @xmath203 to @xmath341 . the terms in which we are interested are the @xmath342 contribution of the first term and the @xmath343 case of the second and the third terms in the square brackets on the right hand side of eq.([eq : dgamma ] ) with the replacements mentioned above . in order to evaluate the first term , we use @xmath344\alpha_{0 } \nonumber \\ & & \qquad { } + \mathcal{o } \left ( \alpha_{0}^{2 } \right)\ ; , \label{eq : for - rprime0}\end{aligned}\ ] ] which follows from eq.([eq : zi0-z0 ] ) . for the computation of the second and the third terms , we use @xmath345 \alpha_{0}^{2 } + \mathcal{o } \left ( \alpha_{0}^{3 } \right)\ ; , \nonumber \\ \frac{1}{z - z'_{i^{(0 ) } } } & = & \frac{1}{z - z_{0 } } - \frac{1}{\partial\rho ( z_{0 } ) } \frac{1 } { ( z - z_{0})^{2 } } \alpha_{0 } + \mathcal{o } \left ( \alpha_{0}^{2 } \right)\ ; , \nonumber \\ \frac{\partial w^{\prime } } { \partial z'_{i^{(0 ) } } } & = & - \frac{\partial^3\rho'(z'_{i^{(0)}})}{\partial^2\rho'(z'_{i^{(0 ) } } ) } = - 2 \partial \rho ( z_{0 } ) \frac{1}{\alpha_{0 } } + \frac{2}{z_{0}-z_{n+1 } } + \mathcal{o } ( \alpha_{0})~. \label{eq : for - i0}\end{aligned}\ ] ] combining these relations , we obtain eq.([eq : dx-2 ] ) . m. kaku and k. kikkawa , `` the field theory of relativistic strings , pt . 1 . trees , '' phys . d * 10 * , 1110 ( 1974 ) ; `` the field theory of relativistic strings . 2 . loops and pomerons , '' phys . d * 10 * , 1823 ( 1974 ) . y. baba , n. ishibashi and k. murakami , `` light - cone gauge superstring field theory and dimensional regularization , '' arxiv:0906.3577 [ hep - th ] . s. mandelstam , `` the interacting string picture and functional integration , '' in _ workshop on unified string theories , _ eds . green and d.j . gross ( world scientific , singapore , 1986 ) , p46 . m. b. green , j. h. schwarz and e. witten , `` superstring theory . 2 : loop amplitudes , anomalies and phenomenology , '' _ cambridge , uk : univ . pr . ( 1987 ) 596 p. ( cambridge monographs on mathematical physics ) _ e. dhoker and s. b. giddings , `` unitary of the closed bosonic polyakov string , '' nucl . b * 291 * , 90 ( 1987 ) . h. kunitomo and k. suehiro , `` a new method of constructing gauge string field theory , '' nucl . b * 289 * , 157 ( 1987 ) . h. sonoda , `` functional determinants on punctured riemann surfaces and their application to string theory , '' nucl . b * 294 * , 157 ( 1987 ) . e. dhoker and d. h. phong , `` functional determinants on mandelstam diagrams , '' commun . * 124 * , 629 ( 1989 ) . t. banks , m. r. douglas , n. seiberg and s. h. shenker , `` microscopic and macroscopic loops in nonperturbative two - dimensional gravity , '' phys . lett . b * 238 * , 279 ( 1990 ) .

in feynman s statistical mechanics textbook , it is stated that the many - body ground state wavefunctions of bosons are _ positive - definite _ in the coordinate representation provided no external rotation is applied and interactions are short - ranged @xcite . the proof is very intuitive : due to the time - reversal ( tr ) symmetry , the ground state wavefunction @xmath0 can be chosen as real . if it is not positive - definite , _ i.e. _ , it has nodes , the following surgery can be done to lower its energy . we first take its absolute value of @xmath1 , whose energy expectation value is exactly the same as that of @xmath0 . however , such a wavefunction has kinks at node points . further smoothing the kinks results in a positive - definite wavefunction , and the kinetic energy is lowered by softening the gradients of the wavefunction . although the single body potential energy and the two - body interaction energy increase , they are small costs of a high order compared to the gain of the kinetic energy . we can further conclude that the ground state is non - degenerate because two degenerate positive - definite wavefunctions can not be orthogonal to each other . this `` no - node '' theory is a very general statement , which applies to almost all the well - known ground states of bosons , including the superfluid , mott - insulating , density - wave , and even super - solid ground states . it is also a very strong statement , which reduces the generally speaking complex - valued many - body wavefunctions to be positive - definite distributions . this is why the ground state properties of bosonic systems , such as @xmath2he , can in principle be exactly simulated by the quantum monte - carlo method free of the sign problem . furthermore , this statement implies that the ordinary ground states of bosons , including bose - einstein condensations ( bec ) and mott - insulating states , can not spontaneously break time - reversal ( tr ) symmetry , since tr transformation for the single component bosons is simply the operation of the complex conjugation . it would be exciting to search for exotic emergent states of bosons beyond this `` no - node '' paradigm , whose many - body wavefunctions can be complex - valued with spontaneous tr symmetry breaking . since properties of complex - valued functions are much richer than those of real - valued ones , we expect such states can exhibit more intriguing properties than the ordinary ground states of bosons . for this purpose , we have recently made much progress with two different ways to bypass feynman s argument , including the meta - stable states of bosons in the high orbital bands of optical lattices @xcite and multi - component bosons with spin - orbit coupling linearly dependent on momentum @xcite . clearly , the `` no - node '' theorem is a ground state property which does not apply to the excited states of bosons . the recent rapid development of optical lattices with ultra - cold bosons provides a wonderful opportunity to investigate the meta - stable states of bosons pumped into high orbital bands . due to the lack of dissipation channels , the life time can be long enough to develop inter - site coherence @xcite . we have shown that the interaction among orbital bosons are characterized by the `` _ orbital hund s rule _ '' @xcite , which gives rise to a class of complex superfluid states by developing the on - site orbital angular momentum ( oam ) moments . in the lattice systems , the inter - site hoppings of bosons lock the oam polarization into regular patterns depending the concrete lattice structures . furthermore , the `` no - node '' theorem even does not apply to the ground states of bosons if their hamiltonians linearly depend on momentum , i.e. , the gradient operator . a trivial example is the formation of vortices in bose condensates with the external rotation in which the coriolis force is represented as the vector potential linearly coupled to momentum . however , in this case tr symmetry is explicitly broken . a non - trivial example is that bosons with spin - orbit ( so ) coupling , whose hamiltonians also linearly depend on momentum and are tr invariant . the invalidity of the `` no - node '' theorem can also give rise to complex - valued ground state wavefunctions . although @xmath2he is spinless and most bosonic alkali atoms are too heavy to exhibit the relativistic so coupling in their center - of - mass motion , so coupling can be important in exciton systems in two dimensional quantum wells . we have shown that the rashba so coupling in the conduction electron bands induces the same type of so coupling in the center - of - mass motion of excitons . in a harmonic trap , the ground state condensate wavefunction can spontaneously develop half - quantum vortex structure and the skyrmion type of the spin texture configuration @xcite . on the other hand , effective so coupling in boson systems can be induced by laser beams , which has been investigated in several publications by other groups in refs @xcite . in the following , we will review our work in both directions outlined above including many new results unpublished before . in section [ sect : boson ] , we explain the characteristic feature of interacting bosons in high orbital bands , the `` orbital hund s rule '' , and the consequential complex - superfluid states with the ordering of on - site oam moments . the ordering of oam moments in the mott - insulating states is also investigated . in section [ sect : spinorbit ] , we review the tr symmetry breaking states of spinful bosons with spin - orbit coupling . interesting properties including half - quantum vortex , and the skyrmion - like spin - textures are studied . conclusions are made in section [ sect : conclusion ] . due to the limit of space , we will not cover interesting related topics of orbital bosons , such as the nematic superfluid state @xcite and the algebraic superfluid state @xcite , and the orbital physics with cold fermions @xcite , which has also aroused much research attention recently . in this section , we will review the `` complex - condensation '' with tr symmetry breaking , which is a new state of the @xmath3-orbital bosons . this is an example of novel orbital physics in optical lattice with cold bosons . below let us give a brief general introduction to orbital physics for the general audience . orbital is a degree of freedom independent of charge and spin . it plays important roles in magnetism , superconductivity , and transport in transition metal oxides @xcite . the key features of orbital physics are orbital degeneracy and spatial anisotropy . optical lattices bring new features to orbital physics which are not easily accessible in solid state orbital systems . first , optical lattices are rigid lattices and free from the jahn - teller distortion , thus orbital degeneracy is robust . second , the meta - stable bosons pumped into high orbital bands exhibit novel superfluidity beyond feynman s `` no - node '' theory @xcite . third , @xmath3-orbitals have stronger spatial anisotropy than that of @xmath4 and @xmath5-orbitals , while correlation effects in @xmath3-orbital solid state systems ( e.g. semiconductors ) are not that strong . in contrast , interaction strength in optical lattices is tunable . we can integrate strong correlation with spatial anisotropy more closely than ever in @xmath3-orbital optical lattice systems @xcite . recently , orbital physics with cold atoms has been attracting a great deal of attention@xcite . for orbital bosons , a series of theory works have been done @xcite , including the illustration of the ferro - orbital nature of interactions @xcite , orbital superfluidity with spontaneous time - reversal symmetry breaking @xcite , the nematic superfluidity@xcite . the theory work on the @xmath3-orbital fermions is also exciting including the flat band and associated strong correlation physics in the honeycomb lattice @xcite , orbital exchange and frustration @xcite , and topological insulators @xcite . the experimental progress has been truly exciting @xcite . _ @xcite have realized the meta - stable @xmath3-orbital boson systems by using the stimulated raman transition . the spatially anisotropic phase coherence pattern has been observed in the time of flight experiments . sebby - strabley _ _ @xcite have successfully pumped bosons into excited bands in the double - well lattice . in addition , @xmath3-orbital bose - einstein condensation ( bec ) has also been observed in quasi one - dimensional exciton - polariton lattice systems @xcite . below we illustrate the important feature of interactions between orbital bosons , the `` orbital hund s rule '' , which results in tr symmetry breaking . the most remarkable feature of interacting bosons in high orbital bands is that they favor to maximize their onsite orbital angular momentum ( oam ) as explained below @xcite . let us illustrate this hund s rule type physics through the simplest example : a single site problem with two degenerate @xmath6-orbitals filled with two spinless bosons . because bosons are indistinguishable , the hilbert space for the two - body states only contains three states , which can be classified according to their oam : an oam singlet as @xmath7 depicted in fig [ fig : pboson ] . a , and a pair of oam doublets @xmath8 , the interaction energy of the former is calculated as @xmath9 while that of the latter is @xmath10 with the definition of u = gd r^2 |_p_x ( r)|^4= gd r^2 |_p_y ( r)|^4 . in the oam singlet state , bosons occupy polar ( real ) orbitals ( e.g. @xmath11 ) , whose angular distribution in real space is narrower than that of the axial ( complex ) orbitals ( e.g. @xmath12 ) for the oam doublets . by occupying the same axial orbital and therefore maximizing oam , two bosons enjoy more room to avoid each other . this `` ferro - orbital '' interaction is captured by the following multi - band hubbard hamiltonian for the @xmath3-orbital bosons as h_int=_r \{n^2_r- l_z^2 } , [ eq : bshamint ] where @xmath13 and @xmath14 . the first term is the ordinary hubbard interaction and the second term arises because of the orbital degree of freedom . in three dimensional systems in which all of the three @xmath3-orbitals are present , we only need to replace @xmath15 with @xmath16 and @xmath17 defined as @xmath18 , @xmath19 . when more than two bosons occupy a single site , bosons prefer to go to the same single particle state by their statistical properties . again going into the same axial state and thus maximizing oam can reduce their repulsive interaction energy . this is an analogy to the hund s rule of electron s filling in atomic shells . the first hund s rule of electrons maximizes electron total spin to antisymmetrize their wavefunction , and the second hund s rule further maximize their oam to extend the spatial volume of wavefunction . the key feature is that electrons want to avoid each other as possible as they can . for the orbital physics of spinless bosons , the same spirit applies with the maximizing of oam . now we have understood the single site physics in which bosons develop rotation . from the symmetry point of view , it is similar to the @xmath20 superconductors . the @xmath3-orbital hamiltonian within the tight - binding approximation can be written as h_0= t__i j [ p^_i , e_ij p_j , e_ij+h.c.]-t__i j [ p^_i , f_ij p_j , f_ij + h.c . ] , [ eq : bandsquare ] where the unit vector @xmath21 is along the bond orientation between two neighboring sites @xmath22 and @xmath23 and @xmath24 is perpendicular to @xmath21 . @xmath25 and @xmath26 are the projections of @xmath3-orbitals along ( perpendicular to ) the bond direction respectively as defined below p_e_ij= ( p_x e_x + p_y e_y ) e_ij , p_f_ij= ( p_x e_x + p_y e_y ) f_ij . the @xmath27-bonding @xmath28 and the @xmath29-bonding @xmath30 describe the hoppings along and perpendicular to the bond direction as depicted in fig . [ fig : pboson ] c and d , respectively . the opposite signs of the @xmath27 and @xmath29-bondings are due to the odd parity of @xmath3-orbitals . @xmath30 is usually much smaller than @xmath28 in strong periodical potentials . [ eq : bandsquare ] exhibits degenerate band minima in both the square and triangular lattice@xcite . in the square lattice , the brillouin zone ( bz ) is a square with the edge length of @xmath31 where @xmath32 is the length between the nearest neighbors . the spectra read @xmath33 and @xmath34 . as depicted in fig . [ fig : band ] . a , the band minima are located at @xmath35 for the @xmath36-band and @xmath37 for the @xmath38-band , respectively . in the triangular lattice , we set the @xmath29-bonding to zero which does not change the qualitative band structures . we take unit vectors from one site to its six neighbors as @xmath39 with @xmath40 , @xmath41 . the brillouin zone takes the shape of a regular hexagon with the edge length @xmath42 . the energy spectrum of @xmath43 is e(k)= t_\ { f_k } , with f_k= _ i=1 ^ 3 ( k e_i ) , g_k = _ 3i > j1 ( k e_i ) ( k e_j ) . the spectrum contains three degenerate minima located at the three non - equivalent middle points of the edges as @xmath44 , the factor @xmath45 takes the value of @xmath46 uniformly in each horizontal row but alternating in adjacent rows . if the above pattern is rotated at angles of @xmath47 , then we arrive at the patterns of @xmath48 . each eigenvector is a 2-component superposition vector of @xmath36 and @xmath38 orbitals . the eigenvectors at energy minima are @xmath49 . @xmath50 can be obtained by rotating @xmath51 at angles of @xmath52 respectively as depicted in fig . [ fig : band ] b. if there were no interactions , any linear superposition of the above band minima would be a valid condensate wavefunctions . however , interactions will select a particular type combination which exhibits tr symmetry breaking and agrees with the above picture of `` orbital hund s rule '' . in the square lattice , we take the condensate wavefunction as @xmath53 under the constraint of @xmath54 . any choice of @xmath55 minimizes the kinetic energy . however , the interaction @xmath56-terms break the degeneracy . the degenerate perturbation theory shows that the ground state values of @xmath55 take @xmath57 or its equivalent tr partner of @xmath58 . the mean field condensate can be described as \ { ( ^_k_p_x + i ^_k_p_y)}^n_0 |0 , [ eq : square ] where @xmath59 is the particle number in the condensate . for better insight , we transform the above momentum space condensate to the real space . the orbital configuration on each site reads e^i_r(|p_x+i_r |p_y ) , where the @xmath60 phase @xmath61 is specified at the right lobe of the @xmath3-orbital . the ising variable @xmath62 denotes the direction of the oam , and is represented by the anti - clockwise ( clockwise ) arrow on each site in fig . [ fig : square ] b. each site exhibits a nonzero oam moment and breaks tr symmetry . the condensate wavefunction of eq . [ eq : square ] describes the staggered ordering of @xmath63 . we check that the phase difference is zero along each bond , and thus no inter - site bond current exists . the condensate described in eq . [ eq : square ] breaks both tr reversal and translational symmetries , and thus corresponds to a bec at non - zero momenta . this feature should exhibit itself in the time - of - flight experiments which have been widely used to probe the momentum distribution of cold atoms . the coherence peaks of eq . [ eq : square ] are not located at integer values of the reciprocal lattice vectors but at @xmath64 , and @xmath65 as depicted in fig . [ fig : square ] a. furthermore unlike the @xmath66-orbital condensate the @xmath3-wave wannier function superposes a _ non - trivial profile _ on the height of density peaks . as a result , the highest peaks are shifted from the origin a standard for the @xmath66-wave peak to the reciprocal lattice vectors whose magnitude is around @xmath67 where @xmath68 is the characteristic length scale of the harmonic potential of each optical site . a detailed calculation of form factors is given in ref . @xcite . the elementary excitations in @xmath3-orbital becs consist of both the gapless phonon mode and the gapped orbital mode . the former corresponds to the goldstone mode related to the @xmath60 symmetry breaking , and the latter corresponds to the flipping of the direction of oam moments . in the following , we will take the 2d square lattice as an example . we assume the condensate as @xmath69 $ ] with @xmath70 and @xmath71 . the boson operators take the expectation values as = ( -1)^r_x , = i(-1)^r_y . from minimizing the onsite part of the free energy respect to the condensate order parameter @xmath72 f =- n+ ( n^2- l_z^2 ) , we have @xmath73 respect to the band minima . the fluctuation around the expectation value is defined as p_x(r)=+p_x , p_y(r)=+p_y . then the interaction terms in eq . [ eq : bshamint ] are expanded as h_int&= & ||^4 + [ p^_x p_x + p^_y p_y ] ||^2 + ^*,2 [ p_x p_x -p_y p_y ] + & + & ^2 [ p_x^p_x^-p_y^p_y^ ] + ( -)^r_x+r_y [ ^2 p_x^p_y^- ^*,2 p_x p_y ] . combined with the free part , we arrive the mean field hamiltonian as h_mf=_k , a , b^^(k)_a m_ab(k ) _ b(k ) where @xmath74 $ ] , @xmath75 , and the summation is only over half of the brillouin zone . the matrix kernel reads m(k)= ( cccc _ x(k)+||^2 & 0 & ^2 & i ^2 + 0 & _ y ( k + q)+ ||^2 & i ^2 & ^2 + u ^*,2 & - u ^*,2 & _ x(k)+||^2&0 + - u ^*,2 & u ^*,2 & 0 & _ y ( k + q)+ ||^2 ) . [ eq : kernel ] the spectra are reduced to generalized eigenvalue problem of x^-1 \{(1,1,-1,-1 ) m ( k ) } x = ( e_1,e_2,-e_3,-e_4 ) , where @xmath76 are excitation eigenvalues , and @xmath77 contains the eigenvectors . let us consider the excitation close to the condensation wavevector @xmath78 , and then @xmath79 . at small value of @xmath80 , we obtain the excitation spectra as e_1,3(k)&= & , e_2,4(k)= |(q ) + u ||^2 . \approx \frac{t_\pp + t_\perp}{2 } ( q_x^2 + q_y^2)$ ] . clearly , the gapless mode with the linear dispersion relation describes the superfluid phase fluctuations ; the gapped mode describes the orbital excitations corresponding to the flipping of orbital angular momenta . in the triangular lattice , the oam moments instead form a stripe ordering , i.e. , the oam moments along one row polarizes along the @xmath82-axis and those along the neighouring rows polarizes with the opposite direction . this can be intuitively understood as follows . in the superfluid state , the oam moments behave like vortices whose interactions are long range . the above stripe configuration of positive and negative vortices is the optimal configuration to minimize the globe vorticity . let us first examine the weak coupling limit . again we write a general form for condensation wavefunction as a linear superposition of the three band minima _ c(r)= ( c_1 _ k_1 + c_2 _ k_2 + c_3 _ k_3 ) , where @xmath83 are the locations of band minima defined in sect . [ subsect : band ] . without loss of generality , we set @xmath84 , and define @xmath85 , @xmath86 and @xmath87 and @xmath88 , then the normalization factor @xmath89 . the interaction energy persite is calculated as e_int&= & \ { ( x_1 ^ 2+x_2 ^ 2-y_1 ^ 2-y_2 ^ 2)^2 + 8 x_1 ^ 2 + 8 y_1 ^ 2 } + 1 + & - & \ { 4 ( x_1 y_2 -x_2 y_1)^2 + 2x_2 ^ 2 + 2 y_2 ^ 2}. the terms in the first line are from the density - density interaction which can be minimized by setting @xmath90 and @xmath91 . this means only one of @xmath92 is nonzero and purely imaginary . in this case , the particle number on each site is uniform . the terms in the second line can also be minimized at this condition with a further requirement of @xmath93 . if we take @xmath94 nonzero , then @xmath95 . thus the mean field condensate can be expressed as @xmath96 with @xmath97 the vacuum state and @xmath59 the particle number in the condensate . this state breaks the @xmath60 gauge symmetry , as well as tr and lattice rotation symmetries , thus the ground state manifold is @xmath98 . this state also breaks lattice translation symmetry , which is , however , equivalent to suitable combinations of @xmath60 and lattice rotation operations . again we transform the above momentum space condensate to the real space , whose orbital configuration takes e^i_r(|p_x+i_r |p_y ) [ eq : triangular ] with @xmath99 as @xmath100 . the general configuration of @xmath101 is depicted in fig . [ fig : triangular ] b for later convenience . at @xmath100 , @xmath6 are not equally populated , and the moment per particle is @xmath102 . this does not fully optimize @xmath103 which requires @xmath104 . however , it fully optimizes @xmath43 which dominates over @xmath103 in the weak coupling limit . we check that the phase difference is zero along each bond , and thus no inter - site bond current exists . interestingly , as depicted in fig . [ fig : triangular ] b , oam moments form a stripe order along each horizontal row . this stripe ordering in the weak coupling limit is robust at small values of the @xmath29-bonding @xmath30 because it does not change the location of the band minima and the corresponding eigenfunctions of @xmath105 at all . the driving force for this stripe formation in the sf regime is the kinetic energy , i.e. , the phase coherence between bosons in each site . by contrast , the stripe formation in high t@xmath106 cuprates is driven by the competition between long range repulsion and the short range attraction in the interaction terms @xcite . this stripe phase should manifest itself in the time of flight ( tof ) signal as depicted in fig . [ fig : triangular ] a. in the superfluid state , we assume the stripe ordering wavevector @xmath107 , and the corresponding condensation wavevectors at @xmath108 . as a result , the tof density peak position after a fight time of @xmath109 is shifted from the reciprocal lattice vectors @xmath110 as follows _ t & & _ g\ { |_2 ( , k)|^2 ^2(k - k_2- g ) + |_3 ( , k)|^2 ^2(k - k_3-g ) } , where @xmath111 ; @xmath112 is the fourier transform of the wannier @xmath3-orbital wavefunction @xmath113 , and @xmath114 $ ] with @xmath115 integers . thus bragg peaks should occur at @xmath116 $ ] . due to the form factors of the @xmath3-wave wannier orbit wavefunction @xmath117 , the locations of the highest peaks is not located at the origin but around @xmath118 . due to the breaking of lattice rotation symmetry , the pattern of bragg peaks can be rotated at angles of @xmath52 . in this subsection , the ordering of the oam moments in the strong coupling superfluid regime is examined . we will employ the lattice gauge theory formalism developed by moore and lee in ref . @xcite in the context of the @xmath119 josephson junction array systems . in the superfluid regime , each site @xmath22 is denoted by a @xmath60 variable @xmath120 which denotes the superfluid phase along the @xmath121-direction , and by an ising variable @xmath122 for the direction of the oam moment as depicted in fig . [ fig : strong ] . along a general direction of the azimuthal angle @xmath123 , the superfluid phase is @xmath124 . the intersite @xmath27 and @xmath29-bonding become the intersite josephson coupling as h_ij =- nt_(_i-_j - a_(i , j ) ) -nt_(_i-_j - a _ ( i , j ) ) , [ eq : josephson ] where @xmath125 is the average particle number per site . the phase differences in eq . [ eq : josephson ] takes into account the geometric orientation of the bond @xmath126 as captured by the gauge fields @xmath127 and @xmath128 with the definition that @xmath129 , @xmath130 , and @xmath131 are the azimuth angles relative to the @xmath121-axis defined in fig . [ fig : strong ] . let us first consider the case of @xmath132 and write down an effective hamiltonian for the ising variables @xmath133 . in order to minimize both the josephson couplings of the @xmath27 and @xmath29-bonds , we need @xmath134 and thus @xmath135 . otherwise , if @xmath136 , then @xmath137 . this costs an energy of @xmath138 , and leads to an effective antiferro - orbital interaction between the ising variables as h_eff = n t__ij _ z , i _ z , j . thus the antiferro - orbital ordering of oam moments in the square lattice remains valid in the strong coupling limit at non - vanishing @xmath30 , which enforces that both the @xmath27 and @xmath29-bonds achieve phase coherence as depicted in fig . [ fig : square ] . b. on the other hand , in the limit of a vanishing @xmath30 , the leading order effect involves multiple site interaction around a plaquette . we follow the method described in ref . @xcite , and perform the duality transformation for the @xmath60 phase variables @xmath72 under the background of the geometric gauge potential @xmath127 . we separate the contributions from the phonon part and the vortex part as z&=&z_ph _ m^\{- _ x ( m(x ) - _ x)^2 + & + & nt__xx^ ( m(x ) -_x ) ( m ( x^ ) -_x^ ) } , [ eq : vortex ] where @xmath139 is the phonon contribution in the vortex - free configuration ; @xmath121 marks the dual lattice site ( or the plaquette index in the original lattice ) ; @xmath140 is the vortex charge ; @xmath141 is the external flux through the plaquette defined as _ x=_ij a _ , ij . in the square lattice , the geometric gauge flux around each plaquette @xmath121 is calculated as _ x=- ( _ i + _ i+e_x+_i+e_x+e_y + _ i+e_y ) . if @xmath141 is integer - valued , it can be absorbed by shifting the zero of the vortex charge as @xmath142 which remains integer - valued , thus there is no cost for energy . the feature can be captured by the effective hamiltonian in ref . @xcite as h_eff=- k _ ijkl _ i _ i _ k _ l , [ eq : z2 ] where @xmath143 are four sites around a plaquette centered at @xmath121 ; @xmath144 is the energy scale of the @xmath27-bonding . this model has a sub - extensive @xmath145 symmetry investigated in ref . @xcite , i.e. , flipping the sign of @xmath133 along each row or column leave eq . [ eq : z2 ] invariant . thus it can not develop the ordering of oam at any finite temperature . let us come back to the superfluid sector @xmath72 , its most relevant topological defect is the half - quantum vortices because @xmath141 can take the values of @xmath146 . the kosterlitz - thouless transition is associated with the unbinding of half - quantum vortices . as a result , the low temperature phase is the quasi - long range ordering of pairing of bosons @xcite . next we move to the strong coupling theory in the triangular lattice . we start from the limit of @xmath147 . the geometrical gauge flux becomes _ x=- ( _ 1 + _ 2+_3 ) , [ eq : triagflux ] where @xmath148 are three sites around a triangular plaquette . there is no way to form an integer flux . the smallest vorticity per plaquette is @xmath149 which corresponds to either two @xmath150 s and one @xmath151 , or two @xmath151 s and one @xmath150 to minimize the vortex core energy . in such a dense vortex configuration , eq . [ eq : vortex ] rigorously speaking does not apply because its validity relies on the assumption of small vortex fugacity . however , the structure of interactions among vortices still implies that vortices form a regular lattice with alternating positive and negative vorticity . the dual lattice ( the center of the triangular plaquette ) is the bipartite honeycomb lattice . it is tempting to assign @xmath149 alternatively to each plaquette , but actually it is not possible due to the following reason . consider a plaquette with vorticity @xmath152 , thus its three vertices are with two @xmath150 s and one @xmath151 . the neighboring plaquette sharing the edge with two @xmath150 s must have the same vorticity , and merges with the former one to form a rhombic plaquette with vorticity @xmath153 as depicted in fig . [ fig : strong ] b. thus the ground state should exhibit a staggered pattern of rhombic plaquette with vorticity of @xmath154 . this arrangement precisely corresponds to the stripe order of the ising variables . this configuration breaks both the lattice rotational and translational symmetries , which is six - fold degenerate . if we turn on the small @xmath30 term , it results in an antiferromagnetic ising coupling with nearest bonds , the stripe configuration also satisfies its ground state requirement . thus we believe that @xmath30 term does not change the stripe configuration in the strong coupling limit either . having established the stripe order for the oam moments , it is straightforward to further optimize the @xmath60 phase variable @xmath72 . the result is depicted in fig . [ fig : triangular ] b with @xmath72 marked on the right lobe of the @xmath155 orbit . each horizontal bond has perfect phase match , while each of tilted bond has a phase difference of @xmath156 . thus around each rhombic plaquette , the phase winding is @xmath157 and gives rise to the josephson supercurrent along the bonds as j= , [ eq : pqcurrent ] with @xmath158 and the directions specified by arrows in fig . [ fig : triangular ] b. in other words , in addition to the stripe order of the ising variables which corresponds to the onsite oam moments , there exists a staggered plaquette bond current order . this feature does not appear in the square lattice @xcite . the orbital plaquette current ordering has been studied in the strongly correlated electron systems , such as the circulating orbital current phase @xcite and the d - density wave states in the high t@xmath106 compounds @xcite . it is amazing that in spite of very different microscopic mechanism and energy scales , the two completely different systems exhibit a similar phenomenology . we have shown that in the triangular lattice the stripe ordering of the oam moments exist in both weak and strong coupling superfluid states . the orbital configurations are slightly different : the orbital mixing angle @xmath101 defined in eq . [ eq : triangular ] equals to @xmath159 in the weak coupling limit , while it equals to @xmath160 in the strong coupling limit . this arises from the competition between the kinetic energy and the onsite hund s rule . below we will see that as @xmath161 goes from small to large , the stripe ordering remains the same with a smooth evolution of @xmath101 from @xmath159 to @xmath160 as depicted in fig . [ fig : triangular ] b. we use a gutzwiller type mean field theory for a @xmath162 lattice under the periodic boundary condition . we explicitly checked for three sets of parameters @xmath163 marked as points of @xmath164 and @xmath165 in fig . [ fig : triang - phase ] b. the stripe - ordered ground state with a @xmath166 unit cell depicted in fig . [ fig : triangular ] a is found stable against small random perturbations in all the three cases . then we further apply the gutzwiller type theory assuming a @xmath166 unit cell and obtain the phase diagram depicted in fig . [ fig : triang - phase ] a which includes both the stripe ordered superfluid phase and mott - insulating phases . so far , we have only discussed the orbital ordering in the superfluid states . since the oam moment is a different degree of freedom from the superfluid phase , we expect that its ordering can survive even inside the mott - insulating phases . in this subsection , we continue to study the exchange physics of orbital bosons and the related ordering of oam moments in the absence of the superfluidity order . for simplicity , we will use the triangular lattice as an example . we consider the mott - insulating phases with @xmath125 spinless bosons per - site and two degenerate orbitals of @xmath178 and @xmath38 . we define the tr doublets of all the particles in the states of @xmath179 as the eigenstates of the ising operator @xmath133 with the eigenvalues of @xmath46 . the ising part of the effective exchange hamilton occurs at the level of the second order perturbation theory , while the ising variable flipping process occurs at the @xmath180-th order perturbation theory . we consider the large-@xmath125 case in which the system is deeply inside the ising anisotropy class , and no orbital - flip process occurs at the leading order . in the following , we will study the physics of the @xmath181-site exchange processes . let us consider the virtual hopping processes in the mott - insulating states along the bond depicted in fig . [ fig : strong ] a. both the @xmath27-bonding @xmath182 and @xmath29-bonding @xmath30 are kept . with the definition of @xmath183 we can express the hopping as h_t&=&-_^ \ { t_p^_(j ) p_^ ( i ) e^-i(_j-^_i ) + t_p^_(j ) p_^ ( i ) e^-i(-^ ) ( _ i + ) + h.c . } + & = & _ \{p^_j , p_i+h.c . } + _ \{p^_j , p_i , - e^-2 i _ i+h.c . } , [ eq : motthop ] where the definition of angles @xmath184 follows the convention depicted in fig . [ fig : strong ] a. we calculate the energy shifts in both the ferro - orbital configurations ( @xmath136 ) and the antiferro - orbital configurations ( @xmath185 ) within the second order perturbation theory . the energy difference between these two configurations is e_fo&= & - - + e_afo&= & - - , where @xmath186 and @xmath187 . the antiferro - orbital configuration has lower energy , which arises from the fact that the orbital - flip process has a larger amplitude than that of the orbital non - flip process described in eq . [ eq : motthop ] . this is because that the @xmath27-bonding and @xmath29-bonding amplitudes have a @xmath29-phase shift . the ising coupling @xmath188 reads h= j_afo _ ij _ z ( i ) _ z ( j ) [ eq:2ptcl ] where @xmath189 with only the leading order contribution kept . if @xmath30 is set to zero , @xmath188 varnishes . this is clear from the fact that we can flip the sign of the @xmath3-orbit component perpendicular to the bond direction . this operation changes the value of @xmath133 , but has no effect on the energy . in this case , we need to further study the multi - site virtual hopping processes . the @xmath27-bonding term by itself gives non - zero ring exchange terms for the multi - site processes , thus we neglect the contribution from the @xmath29-bonding part . in the following , we only keep the leading order virtual process proportional to @xmath190 . we consider a triangular plaquette with three sites @xmath191 each of which is denoted by the particle number and the ising variables as @xmath192 , respectively . there are 12 different ring - hopping processes whose contribution is at the order of @xmath190 . we enumerate 4 of them explicitly as & & ( n,_z , i ) ( n,_z , j ) ( n,_z , k ) ( n-1,_z , i ) ( n+1,_z , j ) ( n,_z , k ) + & & ( n-1,_z , i ) ( n,_z , j ) ( n+1,_z , k ) ( n,_z , i ) ( n,_z , j ) ( n,_z , k ) + & & ( n,_z , i ) ( n,_z , j ) ( n,_z , k ) ( n+1,_z , i ) ( n-1,_z , j ) ( n,_z , k ) + & & ( n,_z , i ) ( n-1,_z , j ) ( n+1,_z , k ) ( n,_z , i ) ( n,_z , j ) ( n,_z , k ) + & & ( n,_z , i ) ( n,_z , j ) ( n,_z , k ) ( n-1,_z , i ) ( n+1,_z , j ) ( n,_z , k ) + & & ( n,_z , i ) ( n+1,_z , j ) ( n-1,_z , k ) ( n,_z , i ) ( n,_z , j ) ( n,_z , k ) + & & ( n,_z , i ) ( n,_z , j ) ( n,_z , k ) ( n+1,_z , i ) ( n-1,_z , j ) ( n,_z , k ) + & & ( n-1,_z , i ) ( n,_z , j ) ( n+1,_z , k ) ( n,_z , i ) ( n,_z , j ) ( n,_z , k ) . the other @xmath193 processes can be obtained by a cyclic permutation @xmath194 . after sum over all these processes , the corresponding energy shift reads e ( _ i , _ j , _ k)6 ( ) ^3 ( n ( n+1))^3/2 & & = -j_3 ( 2_3 ) , [ eq:3ptcl ] where @xmath195 and @xmath196 ; @xmath197 is defined as before in the superfluid case . again for each plaquette , @xmath197 takes the value of @xmath198 which correspond to @xmath133s taking the values of two @xmath199s and one @xmath151 or two @xmath151s and one @xmath199 . . [ eq:3ptcl ] plays a similar role to vortex core energy in eq . [ eq : vortex ] . [ eq:3ptcl ] by itself has a huge ground state degeneracy because the requirement is the same as that of the antiferromagnetic ising model . we need to add the four - site process to lift the degeneracy . the four - site process mimics the interaction between two adjacent vortices . by the spirit of perturbation theory , the 4-particle process is further suppressed by a factor @xmath200 , thus the interaction between vortices should be of short range . this makes sense in the mott insulating state due to the condensation of vortices and the resulting screening effect . again we only consider the effect from the @xmath27-bonding term , and only keep the leading order contribution at the level of @xmath201 . similarly , we have 48 different virtual hopping processes . for a plaquette with vertices @xmath202 , the energy shift is & & e(_i,_j,_k,_l ) -j_4 ( 2_4 ) , [ eq:4ptcl ] where @xmath203 . in the square lattice , @xmath204 . if @xmath147 , eq . [ eq:4ptcl ] is the leading order contribution , thus the effective hamiltonian looks the same as that in the superfluid case but with a reduced coupling constant . if @xmath132 , then the two - site exchange of eq . [ eq:2ptcl ] results in the long range antiferro - orbital order . in the triangular lattice , the @xmath205 of the four - site plaquette @xmath202 just equals the sum of the flux of the two triangular plaquettes . since each triangular plaquette takes the flux value of @xmath149 , the rhombic four - site plaquette can only take the flux value of @xmath206 , @xmath172 . the stripe ordering pattern gives the maximum number of the zero - flux four - site plaquettes , and thus is selected as the ground state . the stripe ordering appears at temperatures @xmath207 and disappears at @xmath208 in which the three - site exchange process dominates the physics . in this section , we discuss another route for `` complex - condensation '' of bosons , i.e. , spinful bosons with spin - orbit ( so ) coupling . the proof of feynman s `` no - node '' theorem crucially relies on the fact that the kinetic energy of bosons only contains even power of momentum , thus so coupling which linearly depends on momentum invalidates feynman s proof . in the following , we consider the simplest example of the two - component bosons with the rashba - like so coupling and investigate their condensate in the inhomogeneous harmonic traps . a related work on the bec of two - component bosons has also been studied by stanescu _ @xcite . we consider bosons with two - components denoted as pseudospin up and down . the many - body hamiltonian for interacting bosons is written as h&=&d^2 r ^_\ { -- + v_ext(r ) } _ + & + & _ r ^ _ ( -i _ y _ x + i _ x _ y)_+^ _ ^ _ _ _ , [ eq : mnbdy ] where @xmath209 is the boson operator ; @xmath101 refers to boson pseudospin @xmath210 and @xmath211 ; @xmath212 is the external potential ; @xmath213 describes the @xmath66-wave scattering interaction which is assumed to be spin - independent for simplicity ; @xmath214 is the rashba so coupling strength . although eq . [ eq : mnbdy ] is of bosons , it satisfies a suitably defined tr symmetry like fermion systems as @xmath215 with @xmath216 where @xmath217 is the complex conjugate and @xmath218 operates on the pseudospin doublet . in the homogeneous system , the single particle states are the helicity eigenstates of @xmath219 with the dispersion relations of @xmath220 where @xmath221 . the energy minima are located at the lower branch along a ring with radius @xmath222 . the corresponding two - component wavefunction @xmath223 with @xmath224 can be solved as @xmath225 , where @xmath226 is the azimuth angle of @xmath227 . instead of discussing the condensation in the homogeneous system in which frustrations occur due to the degeneracy of the single particle ground states , we consider a 2d system with an external harmonic trap with @xmath228 . we define the characteristic so energy scale for the harmonic trap as @xmath229 where @xmath230 is the length scale of the trap , and correspondingly a dimensionless parameter as @xmath231 . the characteristic interaction energy scale is defined as @xmath232 where @xmath59 is the total particle number and the dimensionless parameter @xmath233 . the gross - pitaevskii equation can be obtained as the saddle - point equation of eq . [ eq : mnbdy ] as \ { - & + & _ r ( -i _ y _ x , + i _ x _ y , ) + g ( ^ * _ _ ) + m ^2_t r^2 } _ ( r , ) + & = & e _ ( r , ) . [ eq : gp ] due to the 2d rotational symmetry , the ground state condensate wavefunctions can be denoted by the total angular momentum @xmath234 , which can be represented in the polar coordinate as c f(r ) + g(r ) e^i ) , |_-= ( c -g(r ) e^i + f(r ) ) , where @xmath235 and @xmath72 are the radius and the azimuthal angle , respectively . both @xmath236 and @xmath237 are real functions . [ eq : gp ] is numerically solved in the harmonic trap with parameters of @xmath238 and @xmath239 . we choose the condensate as one of the tr doublet @xmath240 , and present its radial density profiles of both spin components @xmath241 and @xmath242 at @xmath238 in fig . [ fig : trap ] a. further in this strong so coupling case with @xmath243 , the condensate wavefunction has nearly equal weight in the spin up and down components , i.e. @xmath244 , thus the average spin moment along the @xmath82-axis equals to zero . the total angular momentum per particle @xmath245 is mainly from the orbital angular momentum polarization , i.e. , one spin component stays in the @xmath66-state and the other one in the @xmath246-state . this is an example of half - quantum vortex configuration @xcite , thus spontanously breaking tr symmetry . clearly this is a `` complex - valued '' ground state wavefunction beyond the `` no - node '' theorem . this condensate wavefunction exhibits interesting spin density distributions in real space as skyrmion - like spin textures . the radial wavefunction in pseudo - spin up and down components @xmath236 and @xmath237 exhibit oscillations with an approximate wavevector of @xmath247 , which originates from the ring structure of the low energy states in momentum space , thus are analogous to friedel oscillations in fermion systems . the pseudo - spin up component is @xmath66-wave like , thus @xmath236 reaches the maximum at @xmath248 ; while the down component is of the @xmath3-wave , thus @xmath249 at @xmath248 . in other words , approximately there is a relative phase shift of @xmath250 between the oscillations of @xmath236 and @xmath237 . in fig . [ fig : trap ] a , @xmath241 and @xmath242 are plotted . the spin density distribution @xmath251 can be expanded as s_z ( r , ) & = & ( |f(r)|^2 -|g(r)|^2 ) , s_x ( r , ) = f(r ) g(r ) , + s_y ( r , ) & = & f(r ) g(r ) . along the @xmath121-axis , the spin density lies in the @xmath82-@xmath121 plane as depicted in fig . [ fig : trap ] b. because the circulating supercurrent is along the tangential direction , the spin density distribution is along the radial direction and exhibits an interesting topological texture configuration which spirals in the @xmath82-@xmath121 plane at the pitch value of the density oscillations . the distribution in the whole space can be obtained through a rotation around the @xmath82-axis . this spin texture configuration is of the skyrmion - like . next we discuss the possible realization of the above exotic bec in exciton systems . excitons are composite objects between conduction electrons and valence holes @xcite . in particular , the recently progress on the indirect exciton systems greatly enhances the lift - time @xcite , which provides a wonderful opportunity to investigate the exotic state of matter of the exciton condensation @xcite . the ordinary bosons are too heavy to exhibit the relativistic so coupling in their center - of - motion . due to the small effective mass of excitons , so coupling can result in important consequences , including anisotropic electron - hole pairing @xcite , spin hall effect of the center - of - mass motion of excitons @xcite , and the berry phase effect on exciton condensation @xcite . we will show that the rashba so coupling in the electron band also survives in the center - of - mass motion of excitions . we begin with the hamiltonian of indirect excitons h_e&=&- ( _ e , x^2 + _ e , y^2 ) + i _ r , el ( _ e , x _ y - _ e , y _ x ) , [ eq : hamrashba ] + h_hh&=&- ( _ h , x^2 + _ h , y^2 ) , [ eq : hamhh ] + h_e - hh&= & - , where @xmath252 is the effective mass of conduction electrons ; @xmath253 is the rashba so coupling strength of the conduction electron , @xmath254 is the dielectric constant ; @xmath4 is the thickness of the barrier . for small exciton concentrations , we only need to consider the heavy hole ( @xmath255 ) band with the effective mass @xmath256 and @xmath257 , which is separated from the light hole band with a gap of the order of 10 mev . we consider the center - of - mass motion in the bec limit of excitons , which can be separated from the relative motion in @xmath258 and @xmath259 . similarly to ref . @xcite , the effective hamiltonian of the 4-component @xmath255 excitons denoted as @xmath260 can be represented by the matrix form as [ eq : hamexciton ] where @xmath227 is the center - of - mass momentum ; @xmath261 is the total mass of the exciton ; @xmath262 ; @xmath263 ; @xmath264 is the exchange integral and @xmath265 has the @xmath4-wave structure as @xmath266 , both of which are exponentially suppressed by the tunneling barrier for indirect excitons , and will be neglected below . consequentially , @xmath267 becomes block - diagonalized . we consider using circularly polarized light to pump the exciton of @xmath268 , and then focus on the left - up @xmath269 block of the @xmath267 matrix with heavy hole spin @xmath270 . if we use the electron spin number @xmath210 and @xmath211 as the exciton component , we will arrive the eq . [ eq : mnbdy ] with the renormalized so coupling strength @xmath271 . we next justify the above choice of the values of parameters based on experimental situations . the effective masses of electrons and holes in gaas / algaas quantum wells are @xmath272 and @xmath273 in ref . so coupling strength @xmath274 can reach @xmath275 ev.m in ref . thus we can estimate a reasonable value of @xmath276 @xmath277 . for a harmonic trap with @xmath278 , @xmath279 and @xmath280 mk . in two dimensional harmonic traps , the critical condensation temperature @xmath281 @xcite . if we take the exciton density @xmath282 @xmath283 and the effective area @xmath284 , we arrive at @xmath285 mk , which is an experimentally available temperature scale @xcite . the average interaction energy per exciton in the typical density regime of @xmath286@xmath277 is estimated around 2mev in ref . @xcite , thus we take the interaction parameter @xmath239 in the calculation above . the spatial periodicity of the spin texture is about @xmath287 and , thus , is detectable by using optical methods . we have reviewed the exotic condensations of bosons whose many - body wavefunctions are complex - valued in the coordinate representation , thus they go beyond the well - known paradigm of the `` _ no - node _ '' theorem . we studied two possible ways to bypass this theorem to achieve unconventional condensations , i.e. , meta - stable states of bosons in the high orbital bands in optical lattices , and spinful bosons with so coupling . the first mechanism of orbital bosons is essentially an interaction effect which is characterized by the hund s rule . with the orbital degeneracy , bosons favor to enlarge their spatial extension to reduce the inter - particle repulsion , which results in the maximization of their onsite oam . we reviewed the ordering of the oam moments in both the square and triangular lattices due to the inter - site coupling in both the weak and strong coupling limits . the low energy excitations include both the gapless phonon mode and the gapped orbital - flip mode . the survival of the oam ordering in the soft mott - insulating regime is also discussed . the second mechanism employing so coupling , which linearly depends on momentum , is a kinetic energy effect . in this case , taking the absolute value of a non positive - definite wavefunction will change its energy and thus invalidate feynman s proof . we have shown that the condensate wavefunction in a harmonic trap becomes a half - quantum vortex and develops skyrmion - like spin textures . in both cases , tr symmetry is spontaneously broken which are not possible in the conventional bec . there are a number of interesting open issues for future study . for example , the ordering of oam moments in various different lattices and in three dimensions generates a variety of challenging problems of the lattice gauge theory @xcite . more importantly , in order to facilitate the communication between the theory work with cold atom experiments , detailed calculation and optimization of the life time of orbital bosons in different lattices are highly desired . furthermore , the dynamics of bosons in high orbital bands in another challenging topic . taking into account the tremendous progress of cold atom physics , it would be great if this research can stimulate general interests on the unconventional condensates beyond the `` no - node '' theorem . i thank l. balents , d. bergman , h. h. hung , w. c. lee , j. moore , i. mordragon - shem , w. v. liu , s. das sarma , v. stajonovic , and s. z. zhang for fruitful collaborations on this topic . in particular , i am grateful to w. v. liu for his introducing me to this research direction , and s. das sarma for his guidance in the research . i also thank i. bloch , l. butov , h. deng , l. m. duan , m. fogler , j. hirsch , t. l. ho , n. kim , z. nussinov , l. sham , y. yamamoto , s. c. zhang and f. zhou for helpful discussions . this work is supported by the nsf grant no . dmr-0804775 and sloan research foundation .

a tendency towards spin and charge ordering in cuprates has been observed in results of several neutron scattering experiments . low frequency spin fluctuations observed by many groups at incommensurate wave vectors as relatively sharp peeks in the magnetic structure factor @xcite have been interpreted as being due to stripe fluctuations @xcite . anisotropy observed in resistivity @xcite also points at stripe formation . stripe order appears to have an impact on phonon mediated heat transport @xcite . measurements performed by means of nmr and nqr techniques demonstrate the emergence of slow spin fluctuations whose appearance is correlated with pinning of charge - modulations @xcite . the distribution of nearest neighbor bond lengths deduced from neutron powder diffraction data @xcite and measured by means of extended x - ray - absorption fine structure spectroscopy @xcite agrees with expectations based on a scenario of lattice response to local charge - stripe order . an indication of stripe order may by also found by analyzing the shape of arpes spectra and their evolution with doping . for example , it is natural to expect the fermi surface to be flat in the stripe phase in the anti - nodal region , near the vectors @xmath2 and @xmath3 @xcite . in addition , quasi - one - dimensionality of the system should bring about the depletion of spectral weight in the nodal regions , near the brillouin zone diagonals . much research has been done to check if these hypotheses are true @xcite . it seems that some general structure of spectra may be definitely attributed to an underlying quasi - one dimensional electronic structure @xcite . in this paper we will concentrate on doped @xmath4 ( lsco ) compounds , slightly above the doping level @xmath5 at which , it is believed , a tendency towards nanoscale phase separation seems to be evident in the the arpes results @xcite . in arpes intensity maps obtained by integrating over 30 @xmath6 at the fermi energy the spectra from 15% doped @xmath7 ( nd - lsco ) @xcite , patches with high intensity may be seen in anti - nodal regions . appreciable spectral weight is also detected at the fermi energy in nodal regions . a similar pattern of the spectral density at the fermi energy was predicted by a phenomenological theory of disordered charge stripes and anti - phase spin domains @xcite . unfortunately , this theory does not discuss the origin of renormalized hopping terms in the effective one - body hamiltonian and does not explain the relation of this renormalization with the underlying magnetic structure . the evolution of the spectral weight as a function of doping has been analyzed for the stripe phase by means of the cluster perturbation technique ( cpt ) in the framework of the microscopic @xmath8-@xmath9 and hubbard models @xcite . this theory captures quite well the general trend of this development . nevertheless , it seems that the spectral weight maps derived within this approach for the doping level at and above 12.5% do not show continuous well developed high - intensity straight patches bridging anti - nodal regions . such structures are experimentally seen in nodal regions @xcite . in addition , that theory does not provide us with much information about the magnetic structure of stripes . the remarks made above seem to suggest that some understanding of the relation between the single particle spectral weight of nd - lsco at the filling level about @xmath5 and the formation of the stripe phase is missing . a phase which we may expect to emerge in a natural way in weakly doped afs is a bond - ordered state @xcite . recently , an exact diagonalization of the @xmath8-@xmath9 model ( @xmath10 m ) at a finite cluster has been performed to study stripe formation . it has been shown that the cluster geometry change , from the standard tilted square form of the 20 sites cluster to the rectangular one , @xmath11 cluster induces the formation of a ground state with pronounced stripe - like charge inhomogeneities @xcite . the distribution of peaks in the single - particle spectral weight , which has been calculated by means of the same method @xcite , resembles experimental arpes spectra from @xmath12 @xcite . in particular , the theory captures quite well the strong dispersion along the @xmath13 - @xmath14 line . the spreading of the quasiparticle - peaks in the single - particle spectrum obtained by means of the exact diagonalization is in good agreement with results of an additional calculation performed by means of a different method which is the bond operator theory for the spin - peierls phase . unfortunately , neither the numerical approach nor the analytical method reproduce the flattening of the experimental band near the point @xmath14 and both of them fail to explain the emergence of the spectral weight at the nodal region , in the form of a straight patch which is observed at the fermi energy in the experiments . another analysis of the spectral weight in doped afs , based on the exact diagonalization of a small cluster concerns the @xmath10 m with inhomogeneous terms locally breaking the translational invariance and the spin - rotational su(2 ) symmetry @xcite . the pattern of the spectral weight at the fermi energy obtained by means of this method resembles arpes spectra from nd - lsco at the doping level @xmath15 and does not show enhanced intensity at nodal regions . this enhancement observed in experiments is the manifestation of remnant two - dimensionality ( 2d ) in the stripe system . thus , also this theory seems not account for the shape of arpes spectra from nd - lsco or lsco at the doping level slightly higher than @xmath5 . results of inelastic neutron scattering ( ins ) experiments indicate that pronounced af correlations exist in the stripe phase @xcite . on the other hand , the relevance of bond - order has been recently demonstrated by means of the same experimental method @xcite . in a recent work @xcite we have suggested that the coexistence of bond - order with long range af order may take place in the stripe phase . the scenario of nanoscale phase separation is realized by means of this coexistence . we have shown in the framework of this concept that above the doping level @xmath5 , when the distance between stripe axes is 4 lattice spacing a bond - centered stripe with spin - peierls order on these bonds is more stable than a site - centered stripe . in addition it is also known about the site - centered stripe that it is more stable than a homogeneous system of holes created in the homogeneous af @xcite . the mechanism of coexistence between af order and spin - peierls order in the stripe phase is based on lowering the kinetic energy of holes moving without confinement in hole - rich stripes which are formed in the form of two - leg ladder - like domain walls ( dws ) between af hole - poor domains in which the exchange energy decreases @xcite . in the next section , in the framework of a scenario for nanoscale phase separation and coexistence of af long range order with bond - order , we will derive an effective tight - binding hamiltonian describing a quasiparticle propagating in such a spin background . next we will calculate the part of the spectral function which is accessible to measurements in photoemission experiments . finally we will discuss results of the calculation . they seem to show characteristic features that may be seen in arpes spectra of nd - lsco slightly above the doping level @xmath5 . the @xmath10 m in the framework of which we perform the calculation is @xmath16 the states in which any site is doubly occupied have been excluded by definition from the hilbert space in which that model acts . @xmath17 represents pairs of nearest neighbor ( nn ) sites . @xmath18 and @xmath19 denote the operators of electron spin and density at site @xmath20 respectively . @xmath8 represents the hopping matrix element @xmath21 between nn sites in the square lattice on which the @xmath10 m is defined , @xmath22- the hopping matrix element @xmath21 between second nn sites , and @xmath23 - the hopping matrix element between third nn sites . the rest of @xmath21 is zero . we concentrate on the doping level about @xmath5 at which the distance between axes of nearest stripes is 4 lattice spacings and af correlations seem to be of long range @xcite . our previous analysis has provided convincing arguments that in such a case the stripe takes the form of a two - leg ladder - like dw which separates hole - poor af domains @xcite . the phase of sublattice magnetization in domains changes by @xmath0 across a dw . each dw is in the spin - peierls state with singlets formed on legs . the underlying spin structure of the stripe system at the doping level about @xmath5 has been presented in fig.[fig1 ] . a natural question arises , what is mechanism which gives rise to long range af correlations , if af domains are separated by dws which consist of singlets ? the structure depicted in fig.[fig1 ] emerges in the presence of holes only . that presence is induced by doping . the creation of a hole gives rise to appearance of an uncompensated spin in the dw . this spin may be parallel or antiparallel to a nearest spin in the neighbor domain . the weight of states in which these spins are antiparallel , fig.[fig2](a ) , ( b ) , is higher , because such a configuration is preferred by af coupling between nn sites . states depicted by fig.[fig2](a ) and ( b ) are coupled by the hopping term in the hamiltonian . since their weight is higher than the weight of states in which an uncompensated spin in the dw and the nearest spin in the domain are parallel , the hopping term which transforms the state depicted by fig.[fig2 ] ( a ) into the state depicted by fig.[fig2](b ) , mediates effective ferromagnetic ( fm ) coupling between sites @xmath20 and @xmath24 . some quantum fluctuations will be present in the underlying spin background of the stripe system . they may take the form of triplet excitations on bonds in dws and multimagnon excitations in domains . fig.[fig4 ] in the previous work @xcite contains some examples of quantum fluctuations in the spin background . they contribute a lot to the energy of the system . on the other hand , it seems that quantum fluctuations merely renormalize the shape of quasiparticle dispersion for a given spin background and do not change the picture qualitatively . this has been demonstrated in the case of a hole propagating in the af spin background @xcite , as well as in the case of bond ordered two - leg ladders @xcite and bond - ordered 2d systems @xcite . when it is necessary we will take into account the influence of quantum fluctuations on the distribution of spectral weight . a scenario which underlies the calculation which we are going to outline next , is based on the assumption that hole motion inside ladder - like dws is governed by the exchange of positions between a hole - fermion pair on a bond and a singlet on a nearby bond @xcite . this exchange is mediated by the hopping terms in the initial hamiltonian ( [ hamorg ] ) . we also assume that a hole propagates in the af spin background as a spin polaron @xcite . in the calculation we take into account a simplest form of coupling which moves a hole between a bond ordered dw and an af domain . this form of coupling originates in the hopping term of the initial hamiltonian . before we proceed to construct an effective hamiltonian which describes the motion of a quasiparticle in the underlying spin background , we will present some formulas which will be useful for that purpose . a singlet on two sites `` @xmath25 '' and `` @xmath26 '' , is created in the vacuum by the operator @xmath27_{\alpha \beta } c^\dagger_{l \alpha } c^\dagger_{u \beta}. \label{sin}\ ] ] @xmath28 is the two - dimensional identity matrix and @xmath29 , @xmath30 , are pauli matrices . the summation over repeating greek indices is assumed . three operators which are components of the vector @xmath31 create three triplet states on sites @xmath25 and @xmath26 , @xmath32_{\alpha \beta } c^\dagger_{l \alpha } c^\dagger_{u \beta}. \label{trip}\ ] ] a formula which we will often use later is @xmath33 the left side of ( [ hsex ] ) together with the right - side term which contains a singlet at the pair of sites @xmath34 and @xmath35 represent the exchange of a singlet and a hole - fermion pair between two bonds , which is mediated by the hopping term in the initial hamiltonian . as we have already stated before , we will neglect the creation of triplets on bonds , however this process also takes place during hole hopping . the creation of a triplet is represented by the second term at the right side of ( [ hsex ] ) . we also need formulas which represent the hopping of a hole from a single site @xmath20 which belongs to an af domain to a site which belongs to a bond occupied by a singlet and vice versa @xmath36_{\alpha \beta } \mathbf{t}^\dagger_{lu } ) |0\rangle . \label{defs2}\end{aligned}\ ] ] when we start to analyze contributions from quantum fluctuations in the ground state of the system to the spectral function we will need followings formulas @xmath37 we begin the derivation of the effective hamiltonian describing hole propagation in the spin background depicted by fig.[fig1 ] with outlining the mechanism of hole propagation inside an af domain . a hole created and moving in the nel background shifts spins between different sublattices and creates defects in the af structure , fig.[fig3 ] ( a)-(c ) . such a process gives rise to an increase of the ising energy . this rise is roughly speaking proportional to the length of a path along which the hole has traveled , which means that a tendency towards hole confinement appears @xcite . in order to take into account such a tendency we will analyze hole motion in the framework of a basis which consists of states representing holes confined in the af background by linear defects ( strings ) left behind by moving holes on their way . we call these states spin polarons . a wave function representing a confined spin polaron at a site @xmath20 in an af domain is a combination of states which are created from the state @xmath38 by the nn hopping term when a hole created at the site @xmath20 starts to move , @xmath39 @xmath40 is the nel state in the domain , @xmath41 denotes a state obtained by hopping along a path @xmath42 of a hole created at the site @xmath20 . @xmath43 is the amplitude of this state . we have assumed for simplicity that @xmath43 depends solely on the length @xmath44 of the path @xmath42 . the length of a path or of a string state is defined as the number of hops needed to form a given string state from a state representing a hole created in the af spin medium . at the first stage of the analysis we take into account only the hopping between nn sites because @xmath45 . processes related to hopping between further neighbors and processes related to swapping antiparallel spins by the transversal term in the exchange interaction will be considered later as a perturbation . a hole moving inside a domain may make its first step in @xmath46 directions . @xmath47=4 is the coordination number of the square lattice . there are in principle @xmath48 direction choices of each next hop , if the hole moves without retracing inside the domain . on the square lattice there are @xmath47 choices for the direction of the first step and @xmath46 for the direction of further hops during the non - retractable motion . these number gets reduced by one for the domain formed by two chains of sites . thus if we neglect some details , as for example path crossing , we may write , @xmath49 each prefactor in ( [ norm ] ) at the square @xmath50 represents the number of different paths with the length @xmath51 . we calculate the energy of the spin polaron state @xmath52 , @xmath53 , which is given by the expectation value of a trial hamiltonian @xmath54 , @xmath55 , with the assumption , that the motion of a hole which has started from the site @xmath20 , is restricted to the interior of the domain , and that the contribution from the interaction term is restricted to the ising part @xmath56 , @xmath57\frac{j}{2 } \nonumber\\ & + & 2(z-1)\sum_{\mu=0}(z-2)^\mu \alpha_\mu \alpha_{\mu+1}t . \label{ph0p}\end{aligned}\ ] ] the first term in ( [ ph0p ] ) basically counts the number of `` broken bonds '' , that are not occupied by a pair of antiparallel spins , in which case the ising contribution to the energy of that bond is higher by @xmath58 than in the case if it were occupied by a pair of antiparallel spins . the second term in ( [ ph0p ] ) is the contribution from the hopping operator to the spin polaron energy . the prefactors appearing in this term represent the number of paths with a given length multiplied by the number of directions in which these paths may be extended . the appearance of the factor 2 is related to the fact that the hopping which couples paths of length @xmath51 and @xmath59 may take place forth and back . the values of these paths @xmath60 can be found by minimizing @xmath61 under the constraint @xmath62 after we have constructed spin polarons which are formed in af domains we are able to present the full basis of single - particle states . the underlying spin background which has been presented in fig.[fig1 ] plays the role of the vacuum @xmath63 for hole - like quasiparticles which propagate in this background . @xmath63 has been obtained by acting on the absolute vacuum @xmath64 , in which no particles are present , with a product of operators like @xmath65 creating singlets on bonds connecting sites @xmath25 and @xmath26 and operators @xmath66 creating spins in domains according to the pattern shown in fig.[fig1 ] . new fermionic positively charged operators @xmath67 create single particle hole - like states from the vacuum for holes @xmath68 . the action of the operator @xmath69 on @xmath68 , where the site @xmath70 belongs to a ladder - like dw , exchanges the operator @xmath65 in the product defining @xmath68 by the operator @xmath71 , which means that instead of a singlet on the bond connecting sites @xmath25 and @xmath26 there is a hole on the site @xmath70 and spin @xmath72 on the site @xmath73 . the action of the operator @xmath67 on a site @xmath20 which belongs to an af domain , creates a spin polaron @xmath52 in that domain . the spin polaron is a combination of some states , the amplitudes of which are given by prefactors @xmath60 . these states include @xmath74 and states obtained by applying consecutively the nn inside - domain hopping term to the state @xmath74 . during this process , the hole hops without retraces and @xmath51 is @xmath44 , the length of the path @xmath42 , along which the hole has traveled to form a given component state of @xmath52 . a label which we will use to mark the fermionic operator creating either a bond hole or a hole - like spin polaron at a give site is @xmath75 . @xmath76 refers to the column number of the unitary cell to which belongs the site where the hole - like particle has been created , @xmath77 refers to the row number of the unitary cell , and @xmath20 ( column ) , @xmath24 ( row ) are indices representing the position of that site inside the unitary cell . @xmath20 and @xmath24 run from 0 to 7 and from 0 to 1 , respectively . now we start to explain with some details the origin of contributions to the effective hamiltonian for a single quasiparticle propagating in the spin background depicted by fig.[fig1 ] which is an exemplification of coexistence between af and bond orders . we concentrate on the case of a spin - up quasiparticle . within the approximation which we use , operators @xmath78 create in the vacuum @xmath68 eigenstates of the unperturbed hamiltonian @xmath54 , @xmath79 where @xmath17 are pairs of sites on which singlets depicted in fig.[fig1 ] have been formed and @xmath80 are pairs of nn sites belonging to af domains . action of @xmath54 is restricted to the space containing states in which none of the sites is doubly occupied . this hamiltonian contains the ising part of the exchange energy of links inside domains . it also drives hole hopping between nn sites inside each domain . since all matrix elements which may give rise to deconfinement of a hole , have been by definition removed from @xmath54 , polaron states @xmath52 are its eigenstates . processes which bring about deconfinement of holes will be treated at the latter stage of the calculation as a perturbation . the exchange energy and the hopping between sites belonging to bonds occupied by singlets in fig.[fig1 ] are the only contribution to the hamiltonian @xmath54 from dws . terms in the hamiltonian of the @xmath10 m which couple a site belonging to a ladder - like dw with a site belonging to a domain do not contribute to @xmath54 . it is clear that the vacuum state @xmath68 and the single particle states @xmath81 are eigenstates of @xmath54 . within the first order approximation the on - site energy of a quasiparticle with spin up created at the site @xmath82 which belongs to a dw is @xmath83 . from now on , the reference value of the energy is the energy of the vacuum state @xmath68 . the contribution from a destroyed singlet to the on - site energy of a hole - like quasiparticle created at the site @xmath82 is @xmath84 . a hole - like quasiparticle occupying the site @xmath82 with spin up is by definition the same as a single fermion with spin up which occupies the site @xmath85 belonging to the bond @xmath85-@xmath82 . since spins on the sites @xmath85 and @xmath86 are in this case parallel , the lowest order contribution to the exchange energy of the bond between these two sites additionally increases by @xmath87 compared to the contribution from this bond in the vacuum state @xmath68 . in the presence of a hole , the contribution from the potential term @xmath88 in the initial hamiltonian of the @xmath10 m is higher by @xmath9 . by adding all partial contribution we get the value @xmath83 of the total on - site energy of the spin - up quasiparticle at the site @xmath82 . the same on - site energy have spin up quasiparticles created in dws at all sites which are nn of sites in domains occupied by spins pointing down . finally we deduce that in the effective single - particle hamiltonian appears a term @xmath89 . \label{d1heff}\end{aligned}\ ] ] the fermionic operators @xmath90 and @xmath91 transform the underlying vacuum @xmath68 into the single particle state described above and the single particle state into @xmath68 , respectively . the notation for indices which we use seems not to be very short , but such a form of it is basically unavoidable because the elementary cell of the underlying spin background , fig.[fig1 ] , is rather big . that notation also helps to trace easily at the map which is fig.[fig1 ] , the results of hopping events mediated by the hamiltonian @xmath92 . an analogous term will appear in the hamiltonian representing a propagating spin - down quasiparticle , but with a different set of indices labelling sites in the elementary cell . the creation of the spin down quasiparticle at these sites , @xmath13 , @xmath93 , @xmath94 , @xmath95 induces the formation at the ladder leg of an uncompensated spin , the direction of which is parallel to the direction of the nearest spin in the domain . for the sake of simplicity we will concentrate in this paper on the propagation of the spin - up quasiparticle . such a simplification is possible because despite the breakdown of the time reversal symmetry by the underlying spin structure , the energy of the spin - up and the spin - down quasiparticles is degenerate . later we will user this observation in the calculation . the on - site energy of the hole - like quasiparticle is lower for a group of sites by @xmath58 then for sites to which the contribution ( [ d1heff ] ) refers because spin of the bond - fermion which appears after a hole has been created at a bond initially occupied by a singlet may be antiparallel to the nearest spin in one of domains , @xmath96 . \label{d2heff}\end{aligned}\ ] ] let us concentrate now on the on - site energy of quasiparticles created at sites belonging to domains . a spin - up hole - like spin polaron can by definition be created exclusively at sites which have been initially occupied by a spin - down fermion . an obvious contribution to the on - site energy of a hole - like quasiparticle energy in domains is @xmath53 , the minimum value of the matrix element ( [ ph0p ] ) obtained under the constraint @xmath97 . during the construction of spin polarons we have considered only hopping between nn sites , which is governed by the hopping with the highest integral @xmath8 . we have also neglected some `` high order '' processes related to path crossing or to the action of the xy term in the heisenberg model . in the analysis which we start now , we will discuss in the framework of the first order perturbation theory the contribution of some neglected processes to the effective hamiltonian @xmath92 . states depicted by figs.4(b ) and ( c ) are string states of length 1 and are components of the spin - polaron wave - function @xmath98 at the site @xmath20 . a single vertical or horizontal hop of the hole created in the af background at the site @xmath20 gives rise to the states depicted by figs.[fig4](b ) and ( c ) respectively . the hopping term to next nearest neighbors ( nnn ) which we treat as a perturbation couples states represented by figs.4(b ) and ( c ) , which brings about a contribution to the matrix element @xmath99 , where @xmath100 . this contribution is @xmath101 we recognize in ( [ con1 ] ) a product of amplitudes with which string states of length 1 appear in the definition ( [ polar ] ) of the spin polaron state . the factor 4 which appears in ( [ con1 ] ) is related to the fact that hopping which couples states ( b ) and ( c ) may occur in both directions and that analogous coupling takes place between the state depicted by fig.[fig4](c ) and the state depicted by fig.[fig4](d ) . the state depicted by fig.[fig4](d ) has been obtained by a single downward hop of a hole created in the af domain at the site @xmath20 . the contribution @xmath102 to @xmath103 originates in a similar way with coupling between states depicted in fig.[fig4](c ) and ( d ) by the hopping term to third nearest neighbor ( tnn ) sites . the coupling between longer string states which are components of the same spin - polaron wave - function may also contribute to the renormalization of the spin - polaron on - site energy . to be more specific , the coupling between states as depicted by figs.[fig4](e ) and ( f ) and between their reflections in the horizontal line running through the site @xmath20 gives rise to the correction @xmath104 \label{con3}\ ] ] to @xmath103 , while the coupling between the state depicted by fig.[fig4](g ) and its reflection in the horizontal line running through the site @xmath20 together with some similar process in which longer strings are involved brings about the correction @xmath105 . \label{con4}\ ] ] the states represented by figs.[fig4](e ) , ( f ) and ( g ) have been obtained by means of three different sequences of hole moves . the hole has been created at the site @xmath20 and these sequences are upwards - upwards , upwards - left and left - upwards respectively . longer strings pinned to the site @xmath20 may be coupled in a very similar way , which also gives rise to a change of the on - site energy . this change has actually already been incorporated into parameters @xmath106 and @xmath107 . finally , by collecting all terms we may infer that in the effective hamiltonian @xmath92 appears the following term referring to the on - site energy of spin - polarons @xmath108 . \label{d3heff}\end{aligned}\ ] ] the additional term @xmath87 in the prefactor is related to the fact that the contribution from the contact interaction -@xmath109 between a site @xmath20 in a domain and a site @xmath24 in a dw is higher by @xmath87 , when a hole is created in the domain at the site i. this interaction was neglected , when we were calculating the eigenenergy @xmath53 of the spin polaron and needs to be taken into account now . ( [ d3heff ] ) is the last on - site term in the effective hamiltonian . the nn hopping integral for the quasiparticle , the propagation of which we describe in this paper is the same as the bare hopping integral provided that the sites between which the quasiparticle moves belong to the same initially ordered bond inside a dw @xcite . thus , the term which describes the quasiparticle hopping between nn sites , on which singlets have been formed in the underlying spin background , takes the following form @xmath110 + h.c.\}. \nonumber\\ \label{d4heff}\end{aligned}\ ] ] the nn hopping integral for a pair of sites , which belong to different singlets in the underlying insulating state is @xmath111 instead of -@xmath8 . this change of sign and this reduction of size may be deduced from the form of the first term on the right side of eq.([hsex ] ) . therefore , we may write the following expression for the contribution to the effective hamiltonian describing quasiparticle hopping between nn sites belonging to different bonds on which singlets has been formed in the underlying spin background @xmath112 + h.c.\}. \nonumber\\ \label{d5heff}\end{aligned}\ ] ] the value of the nn hopping integral , between a site which belongs to a domain and a site which belongs a dw , can be inferred from the first part of the right side in eq.([defs2 ] ) , @xmath113 + h.c.\}. \label{d6heff}\end{aligned}\ ] ] since the time reversal symmetry and the translational symmetry are broken inside af domains , the propagating quasiparticle can not move between different sublattices and any term related to nn hopping inside domains is not generated in the effective hamiltonian @xmath92 . nnn quasiparticle hopping inside dws and hopping between sites which belong to a dw and a domain is a first order process that is mediated by the nnn hopping term in the bare hamiltonian . the explicit forms of related contributions to @xmath92 may be deduced from formulas ( [ hsex ] ) , ( [ defs1 ] ) and ( [ defs2 ] ) , @xmath114 + h.c.\ } , \label{d7heff } \\ \delta_8 h_{eff}&= & -\frac{t'}{\sqrt{2 } } \alpha_0 \sum_{m , n } \ { [ h^\dagger_{\genfrac{(}{)}{0pt}{}{m-1,n}{7,1 } \uparrow } h_{\genfrac{(}{)}{0pt}{}{m , n}{0,0 } \uparrow } \nonumber\\ & + & h^\dagger_{\genfrac{(}{)}{0pt}{}{m-1,n-1}{7,1 } \uparrow } h_{\genfrac{(}{)}{0pt}{}{m , n}{0,0 } \uparrow } + h^\dagger_{\genfrac{(}{)}{0pt}{}{m , n}{2,1 } \uparrow } h_{\genfrac{(}{)}{0pt}{}{m , n}{0,0 } \uparrow } \nonumber\\ & + & h^\dagger_{\genfrac{(}{)}{0pt}{}{m , n-1}{2,1 } \uparrow } h_{\genfrac{(}{)}{0pt}{}{m , n}{0,0 } \uparrow } + h^\dagger_{\genfrac{(}{)}{0pt}{}{m , n}{3,0 } \uparrow } h_{\genfrac{(}{)}{0pt}{}{m , n}{4,1 } \uparrow } \nonumber\\ & + & h^\dagger_{\genfrac{(}{)}{0pt}{}{m , n+1}{3,0 } \uparrow } h_{\genfrac{(}{)}{0pt}{}{m , n}{4,1 } \uparrow } + h^\dagger_{\genfrac{(}{)}{0pt}{}{m , n}{6,0 } \uparrow } h_{\genfrac{(}{)}{0pt}{}{m , n}{5,1 } \uparrow } \nonumber\\ & + & h^\dagger_{\genfrac{(}{)}{0pt}{}{m , n+1}{6,0 } \uparrow } h_{\genfrac{(}{)}{0pt}{}{m , n}{5,1 } \uparrow } ] + h.c.\}. \label{d8heff}\end{aligned}\ ] ] the task of finding the formula for the term describing nnn hopping inside af domains is little bit more tedious . for example the coupling by the xy term in the heisenberg model between string states depicted by fig.[fig4](f ) and ( h ) gives rise to hopping terms in @xmath92 which shift a spin polaron from the site @xmath20 to the site @xmath24 and vice versa . we analyze now the xy term , because it has been neglected during the first stage of the analysis , when quasi - confined spin polaron states have been constructed . since the string state depicted by fig.[fig4](f ) is a component of the wave function @xmath52 for the spin polaron created at the site @xmath20 and the string state depicted by fig.[fig4](h ) is evidently a component of the wave function for the spin polaron created at the site @xmath24 , the coupling between these components gives rise to the coupling between spin polaron states @xmath52 and @xmath115 . this brings about a contribution to the matrix element @xmath116 and to the hopping term in the effective hamiltonian @xmath92 . it is probably useful to remind now that @xmath92 is an approximation to the initial hamiltonian @xmath117 in ( [ hamorg ] ) . this approximation is expressed in terms of the operators creating and annihilating spin polarons . also the coupling between the states depicted by fig.[fig4](g ) and ( h ) contributes to the hopping term between the sites @xmath20 and @xmath24 in the effective hamiltonian @xmath92 . this coupling is mediated by the xy term . we have already discussed the action of the xy term which by removing two defects in the af structure transforms string states of length 2 , tails of which are pinned to the site @xmath20 , into the state representing a hole created at the site @xmath24 in the af ordered domain . by a string tail we mean its end opposite the end at which sits a hole . the xy term may also transform a state representing a hole created in the domain at the site @xmath20 into a string state of length 2 pinned at the tail to the site @xmath24 . this additional coupling between components of the spin polaron states at the sites @xmath20 and @xmath24 doubles the value of the hopping integral between these sites in the effective hamiltonian . thus , after a little thought we may deduce that the integral for the nnn hopping inside domains is @xmath118 the first term in the sum presented above refers to coupling between strings of length 0 and 2 . we have just outlined its origin in detail . other terms appear in the sum ( [ tau1 ] ) because longer strings , which are created when a hole moves further from the site @xmath24 in figs.[fig4](f ) , ( g ) and ( h ) , are also coupled by the xy term in the heisenberg model . the hop left of the hole depicted by fig.[fig4](g ) gives rise to the string state of length three , fig.[fig4](i ) , which is pinned to the site @xmath24 . since we neglected the possibility of path crossing when we were constructing the quasiconfined spin - polaron states , some corrections need to be made now . for example we did not consider before , that by applying the nn hopping term to the state depicted by fig.[fig4](i ) we may create the state depicted by fig.[fig4](j ) . since the latter state is a string - like component of the spin - polaron at the site @xmath24 obtained by hopping downward and left of a hole created at that site , we deduce that the process described above generates the nnn hopping term in the effective hamiltonian with the amplitude @xmath119 the factor 2 originates with the fact that the motion of a hole around a plaque in the square lattice may take place clockwise and anti - clockwise . we also recognize @xmath120 and @xmath121 as amplitudes of strings , which have the length 3 and 2 , respectively . exactly such strings which are components of the spin - polaron states at the sites @xmath20 and @xmath24 in figs.[fig4](i ) , ( j ) are coupled by the nn hopping term in the initial hamiltonian . the nnn term in the initial hamiltonian generates coupling between states representing `` bare '' holes created in the af background of domains . since these states are also string components with the length 0 of some spin polaron wave functions , coupling between the latter is also generated . an example of such coupled string states are figs.[fig4](h ) and ( k ) . the contribution to the nnn hopping integral in @xmath92 is @xmath122 the nnn term in the bare hamiltonian also couples states depicted by figs.[fig4](f ) and ( l ) which are string - like components with length 2 of spin - polaron states at the sites @xmath20 and @xmath24 , respectively . the coupling amplitude is @xmath123 the appearance of the factor 2 is related to the fact that states which are created when holes move between the sites @xmath20 and @xmath24 in opposite directions around the plaque , than in the case of the states depicted by figs.[fig4](f ) and ( l ) are also coupled by the bare nnn hopping . the state in fig.[fig4](l ) has been obtained by hopping left and downward of a hole which has been initially created at the site @xmath24 . by collecting all contributions , which we have discussed above , we see that the new term in to @xmath92 is , @xmath124 + h.c.\}. \nonumber\\ \label{d9heff}\end{aligned}\ ] ] by means of a similar analysis , as for the hopping between nnn sites we may find the tnn hopping term in the effective hamiltonian . for the operator representing the quasiparticle hopping inside dws we get @xmath125 + h.c.\}. \label{d10heff}\end{aligned}\ ] ] the tnn hopping of the quasiparticle between domains and dws is governed by the following term @xmath126 + h.c.\}. \label{d11heff}\end{aligned}\ ] ] to tnn hopping term between sites in domains contribute : a ) the mechanism which is based on the shortening of strings by the action of the xy term , b ) the coupling of strings with length 0 by the tnn hopping term in the initial hamiltonian and c ) the exchange between the head and the tail of a straight string with length 2 . the last process is mediated by the tnn hopping term in the initial hamiltonian ( [ hamorg ] ) . the contributions to the hopping integral in these three cases are @xmath127 since the origin of these couplings is the same as for the nnn hopping term in @xmath92 we do not discuss them in detail . thus , the tnn hopping operator , which is the last contribution to @xmath92 discussed by us at the approximation level , that we have assumed , takes the form , @xmath128 + h.c.\}. \label{d12heff}\end{aligned}\ ] ] fig.[pardisp ] depicts the electronic structure which we have obtained by solving the hamiltonian @xmath92 . it represents the energy dispersion e(*p * ) of all bands along the line @xmath13-@xmath14 -@xmath129-@xmath13 and the line obtained by performing the rotation by @xmath130 around the point @xmath13 . such a combination of dispersion curves is justified by conditions in which experiments are performed , because it seems that stripes in some regions of the sample may run vertically , while in other regions they may run horizontally . the reason for this mixing can be , for example , sample twinning . nondispersing parts of bands are related to some obstacles for hole propagation in the directions perpendicular to stripes . that kind of motion seems to be blocked , which is clear because @xmath92 by definition can not mediate quasiparticle motion occurring exclusively in the direction perpendicular to stripes . it seems that the lack of the energy dispersion in some directions accounts for straight patches of the spectral weight which appear in the density maps obtained by means of arpes measurements @xcite . the coordination frame applied to draw these maps is the same as we have used to obtain fig.[pardisp ] ( momentum - energy ) , while the second derivative of arpes spectra plays the role of a density - like parameter which marks regions with the high spectral weight @xcite . we shall make a more detailed comparison with arpes spectra , after we calculate the spectral weight by means of the approach applied by us . that approach is based , as we have already mentioned several times , on a combination of bond and string formalisms . arpes probes the one - particle spectral function @xmath131 . we neglect in the further analysis of arpes spectra the influence of temperature . we also assume that the description of the photoelectric effect in terms of fermi s golden rule is sufficient and finally we omit , in the calculation , electromagnetic dipole matrix elements between the wave function of photoelectron and the wave functions of the electrons in the initial states . furthermore , we apply the single particle approximation , in the calculation of the spectral function @xmath131 , which is natural , because the hamiltonian @xmath92 by definition does not contain the interaction terms . since electrons are emitted by the photoelectric effect , at @xmath132 information is gathered only about the one - electron removal part of the spectral function , which takes the form @xmath133 due to the formation of the stripe structure the shape of which is determined by the underlying spin background depicted by fig.[fig1 ] , the 1bz gets reduced by a factor of 8 in the horizontal direction and by a factor of 2 in the vertical direction . the hamiltonian @xmath92 may be written in the diagonal form in terms of operators @xmath134 , and their hermitian conjugates , which are determined by the form of hamiltonian eigenstates , @xmath135 where @xmath136 belongs to the reduced 1bz , @xmath137 labels the band number , @xmath138 is the length of the elementary supercell , @xmath139 is the width of the elementary supercell and @xmath140 is the system size . thus , within the single - particle approach , the one - electron removal part of the spectral function is approximated as @xmath141 where @xmath142 denotes the vector @xmath143 reduced to the 1bz of the superlattice , the elementary cell of which is depicted by fig.[fig1 ] and @xmath144 is the energy of the @xmath137-th band in fig.[pardisp ] . in order to evaluate ( [ spsfap ] ) we need to find matrix elements , @xmath145 a scheme showing how to do this in the framework of the spin polaron ( string ) approach was developed before @xcite . for example , the removal of the spin down electron at the site @xmath146 in the elementary cell of the underlying spin background depicted by fig.[fig1 ] gives rise to a state which is component of the wave function for the polaron created at this site . thus the matrix element ( [ matel ] ) for @xmath147 and @xmath148 is @xmath149 . in the appendix we list all sites labelled by the pairs of numbers @xmath150 , @xmath151 , @xmath152 , for which a nonvanishing matrix element ( [ matel ] ) exists . the removal of a spin down electron from a site which belongs to ladder - like dws in the state depicted by fig.[fig1 ] gives rise to the hole - like quasiparticle @xmath153 at this site . since this removal may be mediated by the electron annihilation operator @xmath154 in ( [ matel ] ) , a nonvanishing diagonal matrix element @xmath155 is generated . quantum spin fluctuation which are generated in the underlying spin background schematically depicted by fig.[fig1 ] can not be neglected when we evaluate the spectral function . within the first order perturbation theory the admixture @xmath156 of quantum fluctuations to the groundstate @xmath157 of @xmath54 brought about by the perturbation @xmath158 is @xmath159 where @xmath160 are excited eigenstates of @xmath54 with energies @xmath161 and @xmath162 is the groundstate energy . the action of the exchange interaction between sites depicted by fig.[qnfns](a ) , @xmath70 belonging to a dw and @xmath163 belonging to an af domain , may transform the singlet on sites @xmath25 and @xmath26 into a triplet . within our approach we treat this part of the exchange interaction as the perturbation . the transformed state is an excited eigenstate of @xmath54 with the energy @xmath9 and also a quantum fluctuation that , according to the formula ( [ 1order ] ) , contributes to the underlying vacuum state @xmath68 , about which we assume that it is an eigenstate of the hamiltonian ( [ hamorg ] ) . @xmath68 is schematically depicted by fig.[fig1 ] in which quantum corrections have not been taken into account even within the framework of the first order perturbation theory . the electron annihilation operator @xmath154 that acts on a site belonging to a bond , which has been excited to the triplet state , transforms it into the state representing the hole - like quasiparticle @xmath153 created at that site . this process gives rise an addendum to the matrix element ( [ matel ] ) . this addendum may be attributed to the existence of quantum corrections to the vacuum state @xmath68 . for example the value of this addendum is -@xmath164 for the matrix element ( [ matel ] ) labelled by the indices @xmath165 . the xy part of the exchange interaction which we treat as a perturbation , creates in @xmath68 an excitation that takes the form of two nn spins turned upside down , fig.[qnfns](b ) , with respect to the underlying spin structure , fig.[fig1 ] . according to the recipe ( [ 1order ] ) this excitation contributes a correction to the underlying vacuum state @xmath68 . if the annihilation operator @xmath154 removes the spin down fermion at the site @xmath24 in fig.[qnfns](b ) , the configuration depicted by fig.[qnfns](c ) will be created , which is a string state , a component of the wave function @xmath166 for the spin polaron at the site @xmath20 . since this spin polaron state is by definition created by the fermionic operator @xmath167 a contribution -@xmath168 to some diagonal matrix elements of the form ( [ matel ] ) is generated . we discuss now the last category of processes which give rise to new terms in the sum that appears in ( [ spsfap ] ) . the electron removal from the site @xmath169 , fig.[fig4](b ) , gives rise to a string state of length 2 , fig.[fig4](d ) , which is a component of the wave function for the spin polaron at the site @xmath20 . since the spin - polaron state is created by the operator @xmath167 , a nonvanishing matrix element is generated . its value is -@xmath170 , which gets multiplied by a factor of 2 , because there are two strings of length two , connecting sites @xmath20 and @xmath169 . by collecting all contributions to the sum which appears in ( [ spsfap ] ) the one - electron removal part of the spectral function @xmath171 can be evaluated . we will apply in the numerical evaluation some lorentzian broadening of the dirac delta function , which is justified because experimental measurements , with which we are going to compare our results , have finite resolution and also some averaging procedure is often applied to present experimental data . it seems that a most complete set of data to compare with our theoretical analysis , provide arpes measurements of the lsco and nd - lsco systems not for the doping level @xmath172 but for the doping level @xmath173 @xcite . we will present our results in a way similar to the method of presentation , which was used in experimental papers reporting these measurements . the nn hopping parameter @xmath8 defines a unit in which energy is measured . we choose @xmath174 , @xmath175 , @xmath176 . similar parameters have been applied in a recent theoretical analysis of arpes spectra in the 2d @xmath10 m , based on the exact diagonalization of the @xmath177 cluster and in a separate calculation performed by means of the bond operator formalism applied to the columnar bond order underlying spin structure @xcite . this coincidence helps to make comparison between results of the calculation presented in our paper and results of the earlier analysis . the choice of the hamiltonian parameters is ever to some extent arbitrary . on the other hand a theoretical analysis of arpes spectra from lsco systems , based on the tight binding approach indicates that the ratios @xmath178 and @xmath179 are lower for these systems than for other members of cuprate family @xcite . parameters suggested by the authors of the analysis based on a tight binding approach @xcite are basically the same as parameters applied in our calculation . the position of the fermi energy in the band structure at the doping level @xmath1 has been determined in our calculation by counting the number of hole - like states . it has been marked as a narrow straight line in fig.[pardisp ] . fig.[parintmap ] depicts the intensity of the one - electron removal spectrum function presented in the coordination frame momentum - energy . contributions from vertical and horizontal stripes have been summed in order to account for presumed coexistence of these structures in different parts of the sample . in some agreement with the experimental result obtained at the doping level @xmath1 @xcite we notice in the calculated spectrum a strongly dispersing band between points @xmath13 and @xmath14/@xmath180 which approaches the fermi level , @xmath181 , near the zone boundary , where it joins a flat patch formed by the region of high spectral intensity . after passing the anti - nodal region as a straight narrow strip , the band - like region of high intensity disappears somewhere between @xmath14/@xmath180 and @xmath129 points . in the calculated spectrum we also notice two cusps near @xmath182 and @xmath183 which are absent in the experimental spectrum . this discrepancy may be attributed to the fact that to this high - intensity patch which looks like a single band in the plot , actually contribute two of many bands which may be seen in fig.[pardisp ] . the band which appears in the results of the exact diagonalization performed for the @xmath10 m and the results of the calculation based on the scenario of the columnar spin - peierls order @xcite has also the strong dispersion but does not flatten in the anti - nodal region . despite that neither the spectral weight of the quasiparticle propagating in the bond ordered spin background nor the spectral weight of the quasiparticle propagating in the af background @xcite resembles arpes spectra from doped lsco and doped nd - lsco in the stripe phase , spectral properties of the model in which these two phases coexist gives rise to qualitative agreement with experimental data . this agreement may be attributed to the fact that the band structure of the model of nano - scale phase separation which we discuss here has some features which are present in band structures of both pure homogeneous phases . for example in our results we actually see a remnant of a strip formed by high intensity which takes the shape of a strongly dispersing band along the line between points @xmath13 and @xmath14/@xmath180 and resembles the band that is formed in the columnar bond - ordered phase . this band - like structure actually does not flatten and reaches the maximum near the points @xmath184/@xmath185 . on the other hand it is likely that the presence of this maximum may be attributed to the brute force method of sewing the bond - ordered parts of the system with the parts which are af - ally ordered . one thing is for sure , that the lack of the straight high intensity patch at the anti - nodal region in results of the theory based on the scenario of the columnar spin - peierls phase demonstrates that the spectral function calculated within this picture does not agree in some details with measured arpes spectra and suggests that it is necessary to take into account also the long range af correlations in order to formulate a theory which accounts for the spectral properties of the cuprates in the stripe phase . in the region between the zone center and antinodal points @xmath14/@xmath180 similar agreement has been observed between the experimental data and results of a calculation based on the dynamical mean field theory ( dmft ) applied to the hubbard model @xcite . along the line connecting the points @xmath13 and @xmath129 both in arpes spectra and in our results we see a band in the vicinity of the fermi surface . this band has a maximum at the points @xmath186 , bends downward and disappears near the zone corner . a similar feature may be observed in the results of an exact diagonalization performed for the @xmath8-@xmath22-@xmath23-@xmath9 model and in the results of the calculation performed within the scenario of the bond - ordered columnar phase @xcite . the dmft of the stripe phase in the hubbard model at the doping level @xmath1 gives rise to a slightly different result , namely it seems that the band crosses the fermi level near the point @xmath186 . the intensity map of the one - electron removal spectral function @xmath187 at the fermi energy obtained in the framework of the scenario which assumes coexistence of bond and af orders is depicted by fig.[parflcut ] . fig.[parflcut](a ) refers to results of the calculation in which we have assumed that bond order is formed on legs in ladder - like dws . that figure also shows some agreement with arpes data from lsco and nd - lsco @xcite obtained for the doping level @xmath1 . we see well developed spectral weight in the nodal regions . these regions are bridged by high - intensity continuous almost straight high - intensity patches . it seems that the agreement between the experimental results and the results of the theory based on the mixture of the bond formalism and the spin polaron approach is better in this respect than the agreement with results of previous calculations based on the cpt applied to microscopic models which are the @xmath10 m and the hubbard model @xcite . fig.[parflcut](b ) depicts the spectral weight at the fermi energy obtained in a separate calculation for the underlying spin background with bond order on rungs . we know from the results of a previous paper @xcite that such a structure is less stable and do not expect much similarity with experimental results . such a lack of similarity may be seen , indeed . for example the high - intensity patches does not form a shape resembling the fermi surface obtained by means of calculations based on the local density approach . such a shape may be observed both in experimental spectra and in fig.[parflcut](a ) . in fig.[parflcut ] apart from patches of high spectral density we notice additional regular structures formed by regions of enhanced intensity . it seems that the origin of thouse structures may be attributed to simplicity of our approach within which fluctuations of the underlying spin structure depicted in fig.[fig1 ] are neglected to a great extent . it is natural to expect that such fluctuations smear out the contribution to the spectral function from excitations which may be classified as incoherent background and only the dominating quasiparticle contributions , which may be seen as bright patches in fig.[parflcut ] , are preserved in the real system . calculations based on the phenomenological approach to disordered charge stripes and antiphase spin domains give rise to a pattern formed by regions of high spectral intensity in the 1bz which strongly resembles arpes spectra @xcite . unfortunately no microscopic justification has been provided , of phenomenological one - body hamiltonians which has been applied to derive the spectral density by means of calculations based on this scenario . our calculation is based on the microscopic @xmath8-@xmath22-@xmath23-@xmath9 model . on the other hand it seems that disorder may give rise spreading of spectral weight over the whole anti - nodal region . such a spreading is not observed in the results of our calculation . in a previous paper we have demonstrated that the magnetic structure of the stripe which we have considered here , is likely to have lowest energy at and above the doping level @xmath188 , if the distance between axes of nearest stripes is 4 lattice spacings , as it has been suggested by experiments @xcite . bond order parallel to stripe axes and long range af order coexist in this magnetic structure . it has been suggested that the shape of the fermi surface seen in arpes spectra from nd - lsco and lsco at the doping level in the range @xmath189 may be also explained in the framework of more conventional band calculations which neglect the formation of nano - scale inhomogeneities @xcite . it seems to be hard to reconcile such a way of thinking with the evidence for stripes forming in these systems . in conclusion , motivated by results of a previous calculation @xcite indicating that , at the doping level @xmath5 and above , the stripe structure , which consists of a ) hole - filled two - leg ladder - like dws with the spin - peierls order formed on legs and b ) af domains of width 2 lattice spacings and with the changing phase of the sublattice magnetization by @xmath0 across each dw , is stable , we have performed the calculation of the single - particle spectral density which is generated in such a system . our analysis has been made in the framework of the @xmath8-@xmath22-@xmath23-@xmath9 model , with parameter values in the range suggested by comparison between band structure calculations and the fermi surface of overdoped lsco systems . the calculation which we have performed is a combination of the bond fermion method and the spin polaron approach . we observe pronounced spectral weight both in the anti - nodal and nodal regions . very similar features may be seen in arpes spectra from lsco and nd - lsco at the filling level @xmath1 , which is exactly the same as we have assumed in the calculation . this similarity is not trivial , because different optional structures of bond order in dws , as the spin - peierls order on rungs in the ladder - like dws or the bond order which takes the shape of two layers in a brick - wall , give rise to spectra at the fermi level which have completely different forms . we consider the observed agreement between experiment and theory as an argument for the scenario of coexisting bond and long range af orders in the stripe phase of doped cuprates . we list here contributions to matrix elements ( [ matel ] ) . the contribution to ( [ matel ] ) is @xmath149 for the following values of indices @xmath150 , @xmath151 , @xmath152 : ( 2,1 ) , ( 0,0 ) , ( 2,1 ) ; ( 3,0 ) , ( 0,0 ) , ( 3,0 ) ; ( 4,0 ) , ( 0,0 ) , ( 4,0 ) ; ( 5,1 ) , ( 0,0 ) , ( 5,1 ) . the contribution is -@xmath190 for ( 0,0 ) , ( 0,0 ) , ( 0,0 ) ; ( 0,1 ) , ( 0,0 ) , ( 0,1 ) ; ( 1,0 ) , ( 0,0 ) , ( 1,0 ) ; ( 1,1 ) , ( 0,0 ) , ( 1,1 ) ; ( 4,0 ) , ( 0,0 ) , ( 4,0 ) ; ( 4,1 ) , ( 0,0 ) , ( 4,1 ) ; ( 5,0 ) , ( 0,0 ) , ( 5,0 ) ; ( 5,1 ) , ( 0,0 ) , ( 5,1 ) ; -@xmath191 for ( 0,0 ) , ( 0,0 ) , ( 0,0 ) ; ( 1,0 ) , ( 0,0 ) , ( 1,0 ) ; ( 4,1 ) , ( 0,0 ) , ( 4,1 ) ; ( 5,1 ) , ( 0,0 ) , ( 5,1 ) ; @xmath191 for ( 0,1 ) , ( 0,0 ) , ( 0,1 ) ; ( 1,1 ) , ( 0,0 ) , ( 1,1 ) ; ( 4,0 ) , ( 0,0 ) , ( 4,0 ) ; ( 5,0 ) , ( 0,0 ) , ( 5,0 ) ; -@xmath192 for ( 2,0 ) , ( 0,0 ) , ( 2,1 ) ; ( 2,0 ) , ( 0,0 ) , ( 3,0 ) ; ( 2,0 ) , ( 0,-1 ) , ( 2,1 ) ; ( 3,1 ) , ( 0,0 ) , ( 2,1 ) ; ( 3,1 ) , ( 0,0 ) , ( 3,0 ) ; ( 3,1 ) , ( 0,1 ) , ( 3,0 ) ; ( 6,1 ) , ( 0,0 ) , ( 6,0 ) ; ( 6,1 ) , ( 0,0 ) , ( 7,1 ) ; ( 6,1 ) , ( 0,1 ) , ( 6,0 ) ; ( 7,0 ) , ( 0,0 ) , ( 6,0 ) ; ( 7,0 ) , ( 0,0 ) , ( 7,1 ) ; ( 7,0 ) , ( 0,-1 ) , ( 7,1 ) ; -@xmath193 for ( 2,1 ) , ( 0,0 ) , ( 3,0 ) ; ( 2,1 ) , ( 0,1 ) , ( 3,0 ) ; ( 3,0 ) , ( 0,0 ) , ( 2,1 ) ; ( 3,0 ) , ( 0,-1 ) , ( 2,1 ) ; ( 6,0 ) , ( 0,0 ) , ( 7,1 ) ; ( 6,0 ) , ( 0,-1 ) , ( 7,1 ) ; ( 7,1 ) , ( 0,0 ) , ( 6,0 ) ; ( 7,1 ) , ( 0,1 ) , ( 6,0 ) ; -@xmath194 for ( 2,1 ) , ( 0,1 ) , ( 2,1 ) ; ( 2,1 ) , ( 0,-1 ) , ( 2,1 ) ; ( 3,0 ) , ( 0,1 ) , ( 3,0 ) ; ( 3,0 ) , ( 0,-1 ) , ( 3,0 ) ; ( 6,0 ) , ( 0,1 ) , ( 6,0 ) ; ( 6,0 ) , ( 0,-1 ) , ( 6,0 ) ; ( 7,1 ) , ( 0,1 ) , ( 7,1 ) ; ( 7,1 ) , ( 0,-1 ) , ( 7,1 ) . t. r. thurston , r. j. birgeneau , m. a. kastner , n. w. preyer , g. shirane , y. fujii , k. yamada , y. endoh , k. kakuri , m. matsuda , y. hidaka , and t. murakami , phys . b * 40 * , 4585 ( 1989 ) ; s .- w . cheong , g. aeppli , t. e. mason , h. mook , s. m. hyden , p. c. canfield , z. fisk , k. n. clausen , and j. l. martinez , phys . lett . * 67 * , 1791 ( 1991 ) ; t. r. thurston , p. m. gehring , g. shirane , r. j. birgeneau , m. a. kastner , y. endoh , m. matsuda , k. yamada , k. kojima , and i. tanaka , phys . rev . b * 46 * , 9128 ( 1992 ) . j. m. tranquada , b. j. sternlieb , j. d. axe , y. nakamura , s. uchida , nature * 375 * , 561 ( 1995 ) . m. matsuda , r. j. birgeneau , h. chou , y. endoh , m. a. kastner , h. kojima , k. kuroda , g. shirane , i. tanaka , and k. yamada , j. phys . . jpn . * 62 * , 443 ( 1993 ) ; k. yamada , s. wakimoto , g. shirane , c. h. lee , m. a. kastner , s. hosoya , m. greven , y. endoh , and r. j. birgeneau , phys . lett . * 75 * , 1626 ( 1995 ) ; k. hirota , k. yamada , i. tanaka , and h. kojima , physica b * 241 - 243 * , 817 ( 1998 ) ; h. kimura , et al , phys . b * 59 * , 6517 ( 1999 ) ; j. m. tranquada , n. ichikawa , k. kakurai , and s. uchida , j. phys . solids * 60 * , 1019 ( 1999 ) ; j. m. tranquada , n. ichikawa , and s. uchida , phys . b * 59 * , 14712 ( 1999 ) . s. a. kivelson , i. p. bindloss , e. fradkin , v. oganesyan , j. m. tranquada , a. kapitulnik , and c. howald , rev . phys . * 75 * , 1201 ( 2003 ) . y. ando , k. segawa , s. komiya , and a. n. lavrov , phys . . lett . * 88 * , 137005 ( 2002 ) . c. hess , b. bchner , m. hcker , r. gross , and s - w . cheong , phys . b * 59 * , 10397 ( 1999 ) . a. w. hunt , p. m. singer , k. r. thurber , and t. imai , phys . lett . * 82 * , 4300 ( 1999 ) ; m .- h . julien , f. borsa , p. carreta , m. horvati , c. berthier , and c. t. lin , phys . lett . * 83 * , 604 ( 1999 ) ; p. m. singer , a. w. hunt , a. f. cederstrm , and t. imai , phys . b * 60 * , 15345 ( 1999 ) ; b. j. suh , p. c. hammel , m. hcker , and b. bchner , phys . rev . b * 59 * , r3952 ( 1999 ) ; n. j. curro , p. c. hammel , b. j. suh , m. hcker , b. bchner , a. ammerahl , and a. revcolevschi , phys . lett . * 85 * , 642 ( 2000 ) ; b. j. suh , p. c. hammel , m. hcker , b. bchner , u. ammerahl , and a. revcolevschi , phys . rev . b * 61 * , r9265 ( 2000 ) ; g. b. teitelbaum , b. bchner , and h. de gronckel , phys . rev . lett . * 84 * , 2949 ( 2000 ) , a. w. hunt , p. m. singer , a. f. cederstrm , and t. imai , phys . b * 64 * , 134525 ( 2001 ) ; h .- h . julien , a. campana , a. rigamonti , p. carretta , f. borsa , p. kuhns , a. p. reyes , w. g. moulton , m. horvati , c. berthier , a. vietkin , and a. revcolevschi , phys . b * 63 * , 144508 ( 2001 ) ; g. b. teitelbaum , i. m. abu - shiekah , o. bakharev , h. b. brom , and j. zaanen , phys . rev . b * 63 * , 020507 ( 2001 ) ; b. simovi , p. c. hammel , m. hcker , b. bchner , and a. revcoleschi , phys . rev . b * 68 * , 012415 ( 2003 ) . e. s. boin , g. h. kwei , h. takagi , and s. j. l. billinge , phys . lett . * 84 * , 5856 ( 2000 ) . m. braden , m. meven , w. reichardt , l. pintschovius , m. t. fernandez - diaz , g. heger , f. nakamura , and t. fujita , phys . b * 63 * , 140510 ( 2001 ) . a. bianconi , n. l. saini , a. lanzara , m. missori , t. r. h. oyanagi , h. yamaguchi , k. oka , and t. ito , phys lett . * 76 * , 3412 ( 1996 ) ; a. lanzara , n. l. saini , t. rossetti , a. bianconi , h. oyanagi , h. yamaguchi , and y. maeno , solid . state commun . * 97 * , 93 ( 1996 ) ; n. l. saini , a. lanzara , h. oyanagi , h. yamaguchi , k. oka , t. ito , and a. bianconi , phys . b * 55 * , 12759 ( 1997 ) . a. v. fedorov , t. valla , p. d. johnson , q. li , g. d. gu , and , n. koshizuka , phys . lett . * 82 * , 2179 ( 1999 ) ; a. ino , c. kim , t. mizokawa , z. x. shen , a. fujimori , m. takaba , k. tamasaku , h. eisaki , and s. uchida , j. phys . . jpn . * 68 * , 1496 ( 1999 ) ; t. valla , a. v. fedorov , p. d. johnson , b. o. wells , s. l. hulbert , q. li , g. d. gu , and n. koshizuka , science * 285 * , 2110 ( 1999 ) ; p. j. white , z. x. shen , d. l. feng , c. kim , m. z. hasan , j. m. harris , a. g. loeser , h. ikeda , r. yoshikazi , g. d. gu , and n. koshizuka , preprint cond - mat/9901349 . x. j. zhou , p. bogdanov , s. a. kellar , t. noda , h. eisaki , s. uchida , z. hussain , z .- x . shen , science * 286 * , 268 ( 1999 ) . x. j. zhou , t. yoshida , s. a. kellar , p. v. bogdanov , e. d. lu , a. lanzara , m. nakamura , t. noda , t. kakeshita , h. eisaki , s. uchida , a. fujimori , _ et al . 86 * , 5578 ( 2001 ) . j. c. campuzano , m. r. norman , and m. randeria , in _ physic of conventional and unconventional superconductors _ , edited by k. h. bennemann , and j. b. ketterson ( springer , berlin , 2003 ) . a. damascelli , a. z. hussain , and z .- x . shen , rev . phys . * 75 * , 473 ( 2003 ) . m. ichioka , and k. machida , j. phys . . jpn . * 68 * , 4020 ( 1999 ) . m. fleck , a.i . lichtenstein , e. pavarini , and a. m. ole , phys . * 84 * , 4962 ( 2000 ) ; m. fleck , a.i . lichtenstein , and a. m. ole , phys . b * 64 * , 134528 ( 2001 ) . r. s. markiewicz , phys . b * 62 * , 1252 ( 2000 ) ; p. wrbel , and r. eder , phys . rev . b * 62 * , 4048 ( 2000 ) . p. wrbel , and r. eder , int . b * 14 * , 3759 ( 2000 ) . m. g. zacher , r. eder , e. e. arrigoni , and w. hanke , phys . lett . * 85 * , 2585 ( 2000 ) . e. w. carlson , d. orgad , s. a. kivelson , and v. j. emery , phys . b * 62 * , 3422 ( 2001 ) . j. eroles , g. ortiz , a. v. balatsky , and a. r. bishop , phys . b * 64 * , 174510 ( 2001 ) . m. granath , v. oganesyan , s. a. kivelson , e. fradkin , and v. j. emery , phys . lett . * 87 * , 167011 ( 2001 ) ; g .- h . gweon , j. d. denlinger , j. w. allen , r. claesson , c. g. olson , h. hoechst , j. marcus , c. schlenker , and l. f. schneemeyer , j. electron spectrosc . relat . phenom . * 117 - 118 * , 481 ( 2001 ) ; d. orgad , s. a. kivelson , e. w. carlson , v. j. emery , x. j. zhou , and z. x. shen , phys . lett . * 86 * , 4362 ( 2001 ) . m. granath , v. oganesyan , d. orgad , and s. a. kivelson , phys . b * 65 * , 184501 ( 2002 ) . m. g. zacher , r. eder , e. e. arrigoni , and w. hanke , phys . b * 65 * , 045109 ( 2002 ) . e. w. carlson , v. j. emery , s. a. kivelson , and d. orgad , in _ the physics of conventional and unconventional superconductors _ , edited by k. h. bennemann , and j. b. ketterson ( springer , berlin , 2003 ) ; m. granath , preprint cond - mat/0401063 . a. ino , c. kim , m. nakamura , t. yoshida , t. mizokawa , a. fujimori , t. kakeshita , h. eisaki , and s. uchida , phys . b * 62 * , 4137 ( 2000 ) . m. i. salkola , v. j. emery , and s. a. kivelson , phys . lett * 77 * , 155 ( 1996 ) . anderson , science * 235 * , 1196 ( 1987 ) ; d. s. rokshar , and s. a. kivelson , phys . 61 * , 2376 ( 1988 ) ; t. dombre , and g. kotliar , phys . b * 39 * , 855 ( 1989 ) ; e. fradkin , and s. a. kivelson , mod . phys b * 4 * , 225 ( 1990 ) ; n. read , and s. sachdev , phys . b * 42 * , 4568 ( 1990 ) ; s. sachdev , and n. read , phys . rev . lett . * 77 * , 4800 ( 1996 ) ; n. read , and s. sachdev , nucl . b * 316 * , 609 ( 1998 ) ; m. vojta , and s. sachdev , phys . * 83 * , 3916 ( 1999 ) ; e. altman , and a. auerbach , phys . b * 65 * , 104508 ( 2002 ) ; s. sachdev , and k. park , ann . phys . ( n.y . ) * 298 * , 58 ( 2002 ) ; m. vojta , phys . b * 66 * , 104505 ( 2002 ) ; s. sachdev , rev . phys . * 75 * , 913 ( 2003 ) . r. eder , and y. ohta , phys . b * 69 * , 094433 ( 2004 ) . r. eder , and y. ohta , phys . b * 69 * , 100502(r ) ( 2004 ) . s. m hayden , h. a. mook , p. dai , t. g. perring , and f. doan , nature * 429 * , 531 ( 2004 ) ; j. m. tranquada , h. woo , t. g. perring , h. goka , g. d. gu , g. xu , m. fujita , and k. yamada , nature * 429 * , 534 ( 2004 ) . p.wrbel , a. maciag , and r. eder , preprint cond - mat/0408578 . a. l. chernyshev , a. h. castro neto , and a. r. bishop , phys . lett . * 84 * , 4922 ( 2000 ) ; a. l. chernyshev , s.r . white , and a. h. castro neto , phys . b * 65 * , 214527 ( 2002 ) . k. w. becker , r.eder , and h. won , phys . b * 45 * , 4864 ( 1992 ) . r. eder , phys . b * 57 * , 12832 ( 1998 ) . c. jurecka and w. brenig phys . rev . b * 61 * , 14307 ( 2000 ) ; w. brenig , and k. w. becker , phys . rev . b * 64 * , 214413 ( 2001 ) ; c. hess , c. baumann , u. ammerahl , b. bchner , f. heidrich - meisner , w. brenig , and a. revcolevschi , phys . rev . b * 64 * , 184305 ( 2001 ) ; c. jurecka , and w. brenig , phys . rev . b * 63 * , 094409 ( 2001 ) ; f. heidrich - meisner , a. honecker , d. c. cabra , and w. brenig , phys . rev . b * 68 * , 134436 ( 2003 ) . chernyshev , and r.f . wood , in _ models and methods of high - tc superconductivity : some frontal aspects _ , vol . 1 , edited by j. k. srivastava and s. m. rao ( nova science publishers , inc . , hauppauge ny , 2003 ) ; and refererences therein . bulaevskii , e.l . nagaev , and d.i . khomskii , zh . . fiz . * 54 * , 1562 ( 1968 ) ; w. f. brinkman and t. m. rice , phys . b * 2 * , 1324 ( 1970 ) ; c. l. kane , p. a. lee , and n. read , phys . b * 39 * , 6880 ( 1988 ) ; f. marsiglio , a. e. ruckenstein , s. schmitt - rink , and c. m. varma , phys . rev . b * 43 * 10882 ( 1991 ) ; s. schmitt - rink , c. m. varma , and a. e. ruckenstein , phys . lett . * 60 * , 2793 ( 1988 ) ; b. shraiman , and e. siggia , phys . lett . * 60 * , 740 ( 1988 ) ; s. trugman , phys . b * 37 * , 1597 ( 1988 ) , * 41 * , 892 ( 1990 ) ; see also j. r. schrieffer , x .- g . wen , and s .- c . zhang , phys . lett . * 60 * , 944 ( 1988 ) ; p. bran , d. poilblanc , and r.b . laughlin , nucl . b * 473 * , 707 ( 1996 ) ; r. b. laughlin , phys . rev . lett . * 79 * , 1726 ( 1997 ) . a. ino , c. kim , m. nakamura , t. yoshida , t. mizokawa , a. fujimori , z .- x . shen , t. kakeshita , h. eisaki , and s. uchida , phys . b * 65 * , 094504 ( 2002 ) . r. eder , and k. w. becker , phys . b * 44 * , 6982 ( 1991 ) ; r. eder and k. w. becker , z. phys . b * 78 * , 219 ( 1990 ) ; r. eder , k. w. becker and w. stephan , z. phys . b * 81 * , 33 ( 1990 ) . t. tohyama , s. nagai , y. shibata , and s. maekawa , phys . lett . * 82 * , 4910 ( 1999 ) ; t. tohyama , s. maekawa , supercond . * 13 * , r17 ( 2000 ) . b. o. wells , z .- x . shen , a. matsuura , d. m. king , m. a. kastner , m. greven , and r. j. birgeneau , phys . lett . * 74 * , 964 ( 1995 ) a. bansil and r.s . markiewicz ( private communication ) .

perovskite materials are of considerable interest both for fundamental reasons and for their many actual and potential technological applications . the great fascination of the cubic perovskite structure is that it can readily display a variety of structural phase transitions , ranging from non - polar antiferrodistortive ( afd ) to ferroelectric ( fe ) and antiferroelectric ( afe ) in nature.@xcite the competition between these different instabilities evidently plays itself out in a variety of ways , depending on the chemical species involved , leading to the unusual variety and richness of the observed structural phase diagrams . for example , as temperature is reduced , batio@xmath0 undergoes a series of fe phase transitions , while srtio@xmath0 has a single afd transition . more extreme examples are nanbo@xmath0 and bazro@xmath0 ; the former has a series of six transitions , while the latter stays cubic down to zero temperature . another appealing property of these cubic perovskites is that all of the structural phase transitions involve only small distortions from the ideal cubic structure , the typical distortion being less than 5% of the lattice constant . this simplifies the theoretical treatment considerably . the ample experimental data on these compounds also provide many insights and opportunities for checking the accuracy of theoretical calculations . it is no wonder that there have been many theoretical attempts to study these compounds . previous phenomenological model hamiltonian approaches@xcite have largely been limited by oversimplification and ambiguities in interpretation of experiment , while empirical @xcite and non - empirical pair - potential methods @xcite have not offered high enough accuracy . recently , advances in density - functional techniques have made possible first - principles investigations of such perovskite compounds . such calculations have proven capable of providing accurate structural properties and fe distortions for perovskites at zero temperature.@xcite recently , a thermodynamic theory based upon such first - principles calculations was developed to study the finite - temperature properties of batio@xmath1@xcite and predicted the correct transition sequence and fairly accurate transition temperatures . this thermodynamic approach involves three steps : ( i ) constructing an effective hamiltonian to describe the important degrees of freedom of the system ; ( ii ) determining all the parameters of this effective hamiltonian from high - accuracy _ ab - initio _ lda calculations ; and ( iii ) performing monte carlo simulations to determine the phase transformation behavior of the resulting system . a similar approach was also successfully applied to pbtio@xmath0 by rabe and waghmare.@xcite the construction of the effective hamiltonian is carried out in view of the special structure properties of cubic perovskite compounds . at higher temperature , the cubic perovskite compounds abo@xmath0 have a simple cubic structure with o atoms at the face centers and metal atoms a and b at the cube corner and body center , respectively . the two most common instabilities result from the softening of either a polar zone - center phonon mode , leading to a fe phase , or the softening of a non - polar zone - boundary mode involving rotations of oxygen octahedra , leading to an afd phase . ( in some cases a zone - boundary polar mode may also occur , leading to an afe phase . ) in our previous thermodynamic theory for batio@xmath0 , we assumed fe and strain distortions would be the only important degrees of freedom of the system . in other words , all other distortions are assumed to be much higher in energy . this is true for batio@xmath0 , but not true for cubic perovskites in general . as shown in our recent first - principle calculations,@xcite most cubic perovskite compounds may also undergo afd transitions . to study these compounds , we need to extend our theory to include afd distortions among the low - energy distortions . this extended theory would also allow us to study the interaction between fe and afd instabilities . however , because the afd distortion is a zone - boundary distortion without a clear corresponding zone - center mode , the extension is not trivial . the rest of the paper is organized as follows . in sec . ii , we go through the detailed procedure for the construction of the effective hamiltonian with the afd distortion included . in sec . iii , we describe our first - principles calculations and the determination of the expansion parameters for the three compounds srtio@xmath0 , catio@xmath0 , and nanbo@xmath0 . in sec . iv , we report our calculated transition temperatures , phase sequences , and order parameters for those three compounds . we also identify the differences between the correlation functions of the fe and afd local modes in srtio@xmath0 . sec . v concludes the paper . in our previous development,@xcite we argued that the total energy of a cubic perovskite can be well approximated by a low - order taylor expansion over all the relevant low - energy distortions , specifically fe distortions and strain . the fe distortions are represented by local modes , whose arrangement will reproduce the fe soft phonon modes throughout the brillouin zone ( bz ) . to extend the theory to include the afd distortions , we need to construct a new set of local modes to represent the lowest afd modes over the whole bz , or at least over the portion of the bz where the energy change due to the afd distortions is either negative , or positive but small . the afd mode typically has the lowest energy at the zone - boundary r ( 0.5 , 0.5 , 0.5)@xmath2 and m [ ( 0.5 , 0.5 , 0)@xmath2 , etc . ] points , while near the zone center @xmath3 the energy is very high . so it is necessary to choose local modes that will accurately reproduce the potentially soft modes in the vicinity of the r and m points . the rotation of an isolated oxygen octahedron can be represented by a pseudovector passing through its center . assuming the origin of coordinates at the center of the octahedron , a pseudovector with polarization @xmath4 involves displacements @xmath5 for oxygen atoms at ( @xmath6/2,0,0 ) and displacements @xmath7 for atoms at ( 0,@xmath8/2,0 ) . here @xmath9 is the lattice constant of the ideal cubic perovskite . in the case of the abo@xmath0 perovskite crystal , we can represent octahedral rotation using pseudovectors sitting on the center of each octahedron , i.e. , on the b atoms . however , neighboring octahedra share oxygen atoms , so that some continuity conditions would have to be imposed if we were to insist that the displacement of a given oxygen be consistently described by both neighboring pseudovectors . with such constraints the neighboring pseudovectors would no longer be independent of one another , leading to potential problems in the implementation of the monte carlo simulations . to avoid such problems , we simply construct a set of `` virtual '' pseudovectors @xmath10 which are independent of each other , and let the actual oxygen displacements be the superposition of the displacements that would result from these . to be precise , let @xmath11 denote the pseudovector centered on the b atom of unit cell @xmath12 ( position vector @xmath13 ) , so that each oxygen atom is shared by two pseudovectors . the physical displacement of the oxygen atom shared by @xmath10 and @xmath14 is then given by @xmath15 where @xmath16 , @xmath17 . the afd soft modes of interest at r and m are then easily represented by the corresponding pattern of pseudovectors . for example , choosing @xmath18 reproduces one of the m - point modes polarized along @xmath4 . ( here , @xmath19 , etc . ) other possible choices of the pattern of pseudovectors correspond to other modes which are probably higher in energy , but possibly still relevant for some materials . for example , choosing @xmath20 corresponds to an x - point mode that can be regarded as either of afd character polarized along @xmath21 , or of afe character polarized along @xmath4 . finally , note that choosing @xmath22 constant ( @xmath3-point arrangement ) gives rise to no displacements whatever . in view of this last point , it is important realize that the @xmath23 themselves do not have direct physical meaning ; only differences between adjacent @xmath10 s are physical . adding a constant to all @xmath10 s will not change the physical configuration of the system . so any physical distortions can be mapped to infinitely many pseudovector arrangements , but any pseudovector arrangement only corresponds to one specific physical distortion . because only the pseudovector differences between sites have physical meaning , the hamiltonian should be expanded in terms of these differences , not the pseudovectors themselves . using this approach , we reduce the number of degrees of freedom associated with oxygen displacements perpendicular to the o b bonds from six to three , and raise the symmetry of the system considerably . as a result , the taylor expansion is significantly simplified . the two other low - energy distortions , the fe and elastic distortions , are treated as in ref .. briefly , for each unit cell of the abo@xmath0 perovskite structure , we define a fe local - mode centered on the b site , and a displacement local mode centered on the a site . the former is chosen in such a way that a uniform superposition of fe local modes reproduces the soft to mode obtained by diagonalizing the force - constant matrix at the brillouin zone center . the quantities @xmath24 and @xmath25 are the vector amplitudes of the fe and translational local modes , respectively , in the @xmath12th unit cell . ( note the difference in notation between ref . and this paper.@xcite ) inhomogeneous strains @xmath26 are expressed in terms of differences between @xmath25 in neighboring cells , and we add six extra degrees of freedom to describe the homogeneous strain . thus , the total energy @xmath27 depends on the set of variables \{@xmath28 } and the homogeneous strain , and is expanded in a taylor series in terms of these quantities . the expansion terms can be divided into four kinds , those involving the fe local modes @xmath29 alone , the afd modes @xmath23 alone , the strain variable @xmath25 alone , and the coupling between them , @xmath30 the part of the energy involving the fe local modes alone , @xmath31 , contains the on - site self energy , dipole - dipole interactions , and short - range residual interactions . their forms have been given by equations in ref . ( 3 ) in sec . ii.b , eq . ( 7 ) in sec . ii.c , and eq . ( 9 ) in sec . ii.d , respectively.@xcite the energy due to @xmath25 alone , @xmath32 , is just the elastic energy , and its form has been given by eq . ( 11 ) of ref .. also , the energy terms representing coupling between the @xmath29 and @xmath25 have been given in eq . ( 14 ) of sec . ii.f of ref .. it remains to present here the energy terms involving solely the afd local modes @xmath10 and those representing the coupling of the @xmath10 to the @xmath29 and @xmath25 . their expressions are presented in the following . the afd local modes @xmath10 are nonpolar and involve no dipole moment , so long - range dipole - dipole interactions need not be considered , unlike for fe local modes . recalling that only the differences of the @xmath10 between neighboring sites are physical , it is not appropriate to separate energy contributions into on - site and inter - site interactions , as we did for the @xmath33.@xcite instead , we separate the interaction into harmonic and higher - order contributions , @xmath34 in principle , all the afd energy terms should be expanded in terms of the @xmath35 expressed through eq . ( [ ro ] ) . for intersite interactions , this would become very complicated because of the low crystal symmetry at the o sites . however , for harmonic terms , the expression can be simplified by expansion in terms of the @xmath10 directly with certain conditions enforced . in this case , we can write @xmath36 here , @xmath37 and @xmath38 denote cartesian components , and @xmath39 is a function of @xmath40 and should decay very fast with increasing @xmath41 . we need to impose conditions on the @xmath42 reflecting the fact that the dependence of the energy on the @xmath10 is only through differences between neighboring sites [ eq . ( [ ro ] ) ] . the appropriate conditions are @xmath43 where the sum is over sites @xmath44 such that @xmath45 . this reflects the fact that if we make a change @xmath46 for all the pseudovectors in an @xmath47-@xmath48 plane at a distance of @xmath49 unit cells away from the site @xmath12 , the resulting the `` force '' on the pseudovector on site @xmath12 should vanish . it can be shown that with these conditions enforced , the interaction energy is only related to pseudovector difference between adjacent sites . the description in terms of the @xmath10 directly makes it possible to simplify the interaction matrix @xmath39 by symmetry , since @xmath10 is centered on the high - symmetry sites . for a cubic lattice , we have @xmath50\delta_{\alpha\beta } \ ; , \nonumber\\ { \rm 2nd \ ; nn : \quad } & j^{\rm a}_{ij,\alpha\beta } & = [ j^{\rm a}_4 + ( j^{\rm a}_3-j^{\rm a}_4 ) \theta_{ij,\alpha } ] \delta_{\alpha\beta } \nonumber\\ & & \qquad + j^{\rm a}_5 \theta_{ij,\alpha } \theta_{ij,\beta } ( 1-\delta_{\alpha\beta } ) \ ; , \nonumber\\ { \rm 3rd \ ; nn : \quad } & j^{\rm a}_{ij,\alpha\beta } & = j^{\rm a}_6 \delta_{\alpha\beta } + j^{\rm a}_7 \theta_{ij,\alpha } \theta_{ij,\beta } ( 1-\delta_{\alpha\beta } ) \ ; , \nonumber\\ { \rm 4th \ ; nn : \quad } & j^{\rm a}_{ij,\alpha\beta } & = j^{\rm a}_8 \theta_{ij,\alpha } \delta_{\alpha\beta } \ ; , \label{bigeq}\end{aligned}\ ] ] where @xmath51=1 if @xmath52 has a non - zero @xmath37 component and 0 otherwise . we include in - plane interactions ( @xmath53 ) to 4th neighbor , since this kind of interaction is much stronger than other interactions . the conditions eq . ( [ cond1 ] ) can then be simplified to @xmath54 thus , the complicated harmonic intersite interaction matrix for afd local distortions can be determined from seven independent interaction parameters . the structural phase transition problem is intrinsically an anharmonic problem . since the harmonic modes may be unstable , it is necessary to introduce higher order terms . for simplicity , we first only consider on - site anharmonic contributions associated with oxygen atoms . because of the tetragonal symmetry on the o sites , the lowest anharmonic terms are of fourth order . since each oxygen involves two nearest neighbor afd pseudovectors , this quartic term will take the form @xmath55 ^ 4 + [ a_z ( { \bf r}_i ) -a_z({\bf r}_i+ { \bf d})]^4 \right\ } \nonumber\\ & + & \sum_i \sum_{{\bf d}=\pm { \bf x } } \gamma^{\rm a } [ a_y ( { \bf r}_i ) -a_y({\bf r}_i + { \bf x } ) ] ^2 [ a_z ( { \bf r}_i ) -a_z({\bf r}_i + { \bf x } ) ] ^2 \nonumber\\ & + & { \rm \;cyclic\;permutations\;}.\end{aligned}\ ] ] here , @xmath56 , and @xmath57 and @xmath58 are parameters to be determined from first - principles calculations . in our previous work on batio@xmath0 the intersite fe interactions have been expanded only up to harmonic order . for afd interactions the corresponding approximation would be to truncate the interactions between the afd - induced displacements of different oxygen atoms to harmonic order . [ such terms are already included in eq . ( [ bigeq ] ) . ] we find this approximation to be satisfactory for those compounds with weak distortions , as in the case of batio@xmath0 or srtio@xmath0 . for catio@xmath0 , the afd distortion is very large and the transition temperature is around 2000k . in this case , we find it necessary to include more complicated anharmonic terms , such as third - order intersite interaction terms , for the afd distortions . in fact , such terms turn out to be responsible for inducing a displacement component corresponding to an x - point phonon ( with both o and ca character ) in catio@xmath0 . for nanbo@xmath0 , although the distortion is not as strong ( the highest transition temperature is around 700k ) , there are many structures with very close free energy . we find that inclusion of the third - order afd terms does have a noticeable effect for these compounds , so we include these third order interaction terms for catio@xmath0 and nanbo@xmath0 . we consider only those third - order interactions between afd modes on two or three neighboring lattice sites . we can follow the treatment of the harmonic intersite interactions by listing all the possible interactions and using symmetry arguments to eliminate forbidden terms . following this approach leads to three kind of terms , and we would need three more parameters to fully specify the hamiltonian . since the third - order terms are relatively weak and the exact determination of the three parameters is costly , we investigate the relations between these three parameters , and use a simple argument to combine the three terms to form a single new term with only one free parameter to determine . the afd interactions involve only the displacement of oxygen atoms . the strongest energy difference is associated with the distortion of oxygen octahedra , or the change of the length of a nearest - neighbor o o bond @xmath59 . we can start by analyzing @xmath59 for two oxygens at ( 0,0,@xmath60/2 ) and ( @xmath60/2,0,0 ) . we approximate the total - energy change as solely due to @xmath59 . expanding it as a function of the rotation vectors @xmath61 up to the third order , we obtain the desired third - order intersite terms . using the short - hand notations @xmath62 the cubic coupling term involving one o o bond is @xmath63[a'_x(z)+a'_z(x)]^2 \ ; .\ ] ] the other 11 nearest - neighbor o o bonds will give rise to 11 other terms and the overall total - energy contribution can be expressed as @xmath64 , where @xmath65[a'_y(z)+a'_z(y)]^2 \nonumber \\ & & - [ a'_x ( -y ) - a'_x(z)][a'_y(z)-a'_z(-y)]^2 \nonumber \\ & & - [ a'_x ( y ) - a'_x(-z)][a'_y(-z)-a'_z(y)]^2 \nonumber \\ & & + [ a'_x(-y ) -a'_x ( -z)][a'_y(-z)+a'_z(-y)]^2 \nonumber \\ & & \qquad + \ ; \hbox{\rm cyclic permutations } \,\ } \ ; .\end{aligned}\ ] ] this assumption that @xmath59 is solely responsible for the cubic intersite interactions significantly simplifies the energy expression and reduces the number of expansion parameters from three to one . there are three kind of coupling energy terms : those between fe and elasticity , between fe and afd , and between afd and elasticity , @xmath66 for simplicity , we consider only on - site couplings . the coupling between @xmath29 and @xmath25 ( @xmath67 ) has been given by eq . ( 14 ) in sec . ii.f of ref .. @xcite here , we expressions for @xmath68 and @xmath69 . the coupling between elasticity and afd modes at lowest order can be written as @xmath70 \nonumber\\ & & \qquad + \ ; \hbox{\rm cyclic permutations } \;,\end{aligned}\ ] ] where @xmath71 and @xmath72 is the six - component local strain tensor in voigt notation ( @xmath73 , @xmath74 ) . @xmath72 can be expressed as a function of @xmath25 following sec . ii.f of ref .. @xcite @xmath75 is a high order coupling tensor . because of the symmetry , there are only four independent coupling constants in @xmath75 , @xmath76 all other elements are zero . the lowest - order coupling between @xmath10 and @xmath29 is linear with both @xmath10 and @xmath29 . it takes the form @xmath77 \nonumber\\ & & \qquad + \ ; \hbox{\rm cyclic permutations } \;.\end{aligned}\ ] ] however , in the afd state , this term is zero . to account correctly for the coupling between @xmath10 and @xmath29 , it is necessary to include higher - order terms the lowest non - zero term in the afd state is quadratic in both @xmath10 and @xmath29 . defining @xmath78 by @xmath79\ ] ] and similarly for @xmath80 and @xmath81 , we can write @xmath82 \;.\end{aligned}\ ] ] ( in principle a term @xmath83 could also be included , but for practical reasons we have not done so in this work . ) in summary , up to the fourth - order terms , the coupling between @xmath10 and @xmath29 is expressed as @xmath84 the expansion parameters in the model hamiltonian can be obtained from a set of first - principles calculations . we use density - functional theory within the local density approximation ( lda ) . the technical details and convergence tests of the calculations can be found in ref . . the use of vanderbilt ultra - soft pseudopotentials@xcite allows a low - energy plane - wave cutoff to be used for first - row elements , and also allows inclusion of semicore shells of the metal atoms . this makes high - accuracy large - scale calculations of materials involving oxygen and @xmath85 transition - metal atoms affordable . a generalized kohn - sham functional is directly minimized using a preconditioned conjugate - gradient method.@xcite we use a ( 6,6,6 ) monkhorst - pack k - point mesh@xcite for single - cell calculations ( 216 k - points in the full brillouin zone ) , and the corresponding smaller sets of mapped k - points for supercell calculations . .fe soft - mode eigenvectors for @xmath86o@xmath0 cubic perovskites srtio@xmath0 , catio@xmath0 , and nanbo@xmath0 . @xmath87 and @xmath88 indicate oxygen displacement parallel and perpendicular to o@xmath89 bond , respectively . [ t_fe ] [ cols= " < , > , > , > " , ] the calculation of the afd expansion parameters follows a similar procedure as for the fe ones . the afd eigenvector itself does not need to be calculated , since it is determined by symmetry . the lda total energy vs. afd distortion , with polarization along @xmath47 and @xmath90 , and at k - points x=(@xmath91,0,0 ) , m=(@xmath91,@xmath91,0 ) , and r=(@xmath91,@xmath91,@xmath91 ) , are calculated . the arrangements of the afd local modes are the same as for the fe - mode calculations as shown in figs . 3(a)(f ) of ref . however , the arrangements at the @xmath3 point ( fig . 3(a ) of ref . ) and at the x point ( fig . 3(b ) of ref . ) involve no actual distortions . so the work reduces to four 10-atom cell calculations and two 20-atom cell calculations needed to determine @xmath92 , @xmath93 , @xmath94 , @xmath95 , @xmath96 , and @xmath97 . the decomposition of @xmath98 and @xmath99 follows the same argument as for the fe case . it is difficult to perform sufficient lda calculations to carry out the decomposition , and probably not very important to do so . instead , we rely on the heuristic that the interaction between two afd local modes should be minimal ( in practice , zero ) when they are so arranged that reversing one of them induces a minimum bond - length change . for the afd case , this leads to @xmath100 , allowing @xmath98 and @xmath99 to be obtained separately . the fourth - neighbor interaction parameter @xmath101 is obtained from lda calculations on a 15-atom supercell . as mentioned before , we need to include the effect of third - order intersite coupling in the effective hamiltonian in some compounds having large afd distortions . this kind of interaction generates a coupling among three distortions : an r - point mode with polarization ( 110 ) , an m - point ( @xmath91,@xmath91,0 ) mode polarized along ( 001 ) , and an x - point ( 0,0,@xmath91 ) mode polarized along ( 110 ) . to determine the strength of this coupling , we carry out a calculation with the above r - point and m - point distortions frozen in ( 20-atom supercell ) , and calculate the forces of x - point character . we find that the force on the a - metal atom ( ca or na ) is non - zero , and opposite in sign to the force on the oxygen atom . this is in qualitative agreement with the experimentally observed displacements in the low - temperature phase , which are also opposite in sign . we then calculate the projection of these forces onto the x - point mode , under the simplifying assumption that the latter consists of equal - amplitude out - of - phase motion of the two atoms . this projection determines our third - order intersite coupling parameter @xmath102 . the calculation of the anharmonic coefficient @xmath103 is performed with an r - point distortion polarized along ( 001 ) . typically a set of eight calculations are performed for each compound . the resulting lda total energy is fitted to a polynomial @xmath104 using standard least - squares methods to extract @xmath105 and @xmath103 . a single r - point calculation with polarization in the ( 111 ) direction is used to extract @xmath106 . the four parameters describing the coupling between afd modes and elastic strains are obtained by performing four more 10-atom supercell calculations : at the m point , @xmath107 , with an isotopic strain @xmath108 ; at the m point , @xmath107 , with strain @xmath109 ; at the x point , @xmath107 , with strain @xmath109 ; and at the r point , @xmath110 , with strain @xmath111 . extra care has been taken to ensure cancellation of errors due to k - point sampling and basis - size differences for the different unit cells involved . finally , the couplings between fe and afd modes are determined as follows . the harmonic coupling @xmath112 is determined by considering a geometry in which the primitive cell has been tripled along the @xmath47 direction , and for which @xmath113 is non - zero in two primitive cells and @xmath114 is non - zero in the third . the anharmonic couplings @xmath115 and @xmath116 are determined from a series of calculations on a doubled - cell configuration in which @xmath117 @xmath118 ( 100 ) and @xmath119 @xmath118 ( 100 ) or ( 010 ) . the calculated parameters for all three compounds are listed in table [ t_para ] . we note that the intersite interaction parameters between afd local modes have a much stronger anisotropy than those between fe modes . for fe , the @xmath44 s show no marked anisotropy . ( of course , when the coulomb interaction is included , the actual interaction between fe modes are quite anisotropic ) . on the other hand , for the first - neighbor afd couplings , @xmath92 is more than one order of magnitude larger than @xmath93 , which is reasonable since @xmath93 involves no distortion of oxygen octahedra . for second neighbors , @xmath95 is again much larger than the others , confirming that the distortion of oxygen octahedra dominates the energy for afd distortions . this observation that the in - plane interaction parameters are much stronger than the out - of - plane ones is what prompted us to include also the fourth - neighbor in - plane afd interactions in the effective hamiltonian . after the expansion parameters have been determined from first - principles calculations , the finite - temperature properties of the compounds can be calculated using metropolis monte carlo ( mc ) simulations.@xcite the details of the mc simulations involving fe and elastic distortions have been described in sec . iv of ref . . with the afd distortions included , the number of degrees of freedom is increased from 6 to 9 per unit cell . all the details are very similar , the main difference being that the afd degrees of freedom introduce many more possibilities for modes which may go soft . the primary candidates for soft afd modes are three modes at the r point and one mode at each of the three m points , all of them involving only rigid rotations of oxygen octahedra . we thus have to consider a rather complicated set of order parameters , and we anticipate that complex phases may form . the results for the three different compounds srtio@xmath0 , catio@xmath0 , and nanbo@xmath0 will be presented in the three following subsections . because of the strong sensitivity of structural phase transition temperatures to the lattice constant and the well - known @xmath120 lda underestimate of lattice constants , we concentrate on presenting calculated transition temperatures and transition sequences at the experimental lattice constants . we thus implicitly apply a negative fictitious pressure to the simulation cell , as explained in sec . iv of ref . . = 4.5 in thermodynamic properties for this compound have been calculated and published in ref . . a pressure @xmath121gpa is applied to restore the experimental lattice constant . a transition from the cubic phase to a tetragonal afd structure at 130k , and two further fe phase transitions at 70k and 10k , were predicted . a later quantum path - integral mc simulation revealed that quantum fluctuations suppress the fe phases entirely , and reduce the afd phase transition temperature to 110k.@xcite this gives excellent agreement with experiment , which reveals a single afd phase transition at 105k,@xcite and no unambiguous phase transition ( but the presence of `` quantum paraelectric '' behavior @xcite ) at low temperature . our calculated pressure - temperature phase diagram showed that the fe and afd instabilities have opposite trends with pressure , and fe and afd instabilities tend to suppress each other . we have performed some further simulations to investigate the behavior of the afd and fe local modes , and in particular the nature of the intersite correlations for fe and afd local modes in the cubic phase but just above the phase transition temperature . the m - point afd modes do not appear to be important for srtio@xmath0 , so we focus on the two vector order parameters @xmath117 and @xmath122 associated with the zone - center fe modes and the zone - corner afd modes . since the order parameters are vectors , the correlation functions are second - rank tensors . these can be calculated from our simulations as @xmath123 where the average is taken over all the sites @xmath124 in the mc simulation cell and over all mc sweeps @xmath125 . here the @xmath126 denote the components of the fe or afd order parameters [ @xmath127 for fe , @xmath128 for afd , where @xmath129 . we can get a good picture of the nature of the correlation by investigating the diagonal elements ( @xmath130 ) only . since the three cartesian directions are equivalent , it suffices to present @xmath131 . in fig . [ corr](a ) , we show the calculated afd correlation function @xmath132 in the two planes @xmath47-@xmath90 and @xmath47-@xmath48 for srtio@xmath0 at t=150k , in the cubic phase but just above the afd phase transition temperature of 130k . we find that the correlations are quite strong in @xmath47-@xmath48 plane , with a correlation length of about three lattice constants . along the @xmath90 direction , even the nearest - neighbor vectors are almost completely uncorrelated . thus , the shape of the `` equal correlation surface '' for afd local modes is disc - like . this is easy to understand on the basis of the rum picture.@xcite since the afd local modes involve a rotation of oxygen octahedra , and any distortion of the oxygen octahedra involves a large energy cost ( as shown by the large magnitude of @xmath133 , @xmath134 , and @xmath135 ) , the afd octahedral rotations about @xmath4 correlate strongly in the @xmath47-@xmath48 direction . on the other hand , the rigidness of the octahedra does not impose any relation between @xmath90-polarized afd modes in different @xmath90 planes ( as reflected in the very small @xmath136 ) . thus , the pancake - like correlation naturally results . [ corr](b ) shows the corresponding fe correlation function @xmath137 for srtio@xmath0 at t=150k ( the fe phase transition occurs at 70k ) . its behavior is just the reverse of the afd modes , being strong along the @xmath90 direction and weak in the @xmath47-@xmath48 plane , and resulting in a needle - shaped `` equal correlation surface . '' this behavior is a direct consequence of the anomalously large mode effective charges in the cubic perovskite compounds@xcite which strongly suppress the longitudinal fe fluctuations and leads to the strong correlation of @xmath138 in the @xmath90 direction . on the other hand , the transverse fe modes can easily go soft , resulting in a short correlation length for @xmath138 in the @xmath47-@xmath48 plane . the above picture of the correlation functions for afd and fe local modes are presumably quite general for the cubic perovskite compounds . for the fe modes , we decided to repeat the calculations for the case of batio@xmath0 , where the afd instability does not intervene . we show in fig . [ corr](c ) the correlation function calculated in the cubic phase at t=320k , about 20k above the fe phase transition temperature . we can see the behavior of the correlations is again needle - like and even more pronounced than for srtio@xmath0 , presumably because we are closer to the fe transition temperature . ( an elongated supercell was used to accomodate the correlations in this case . ) = 3.0 in as was done for the fe modes in batio@xmath0,@xcite it is revealing to compute the equilibrium distribution of one cartesian component of the afd order parameter @xmath122 in the cubic phase just above the phase transition temperature in srtio@xmath0 . this is shown in fig . [ sr_dos ] , where it can be seen that the distribution looks approximately maxwellian . this is indicative of a transition having a character much closer to the displacive than to the order - disorder limit . catio@xmath0 is one of the more complicated perovskites . experimentally , it is found to have two stable phases , an orthorhombic phase at lower @xmath139 and a cubic phase above @xmath140k.@xcite some recent experiments suggest that the transition is to a highly disordered cubic phase.@xcite the room - temperature orthorhombic phase has a very complicated structure with a 20-atom unit cell . the displacements of all the atoms away from their idea positions have been determined in ref . . the refined structure as a function of temperature has also been determined recently using x - ray diffraction.@xcite this complicated structure can be decomposed into a simultaneous freezing in of three afd modes : an r - point mode polarized along ( 1@xmath1410 ) with rotation angle 0.20 ( angles in radians ) , an m - point mode polarized along ( 001 ) with rotation angle 0.14 , and an x - point mode polarized along ( 1@xmath1410 ) . the x - point mode involves not only the rotation of oxygen octahedra , but also an associated displacement of ca atoms . the ratio of o and ca displacement is about @xmath142 and the oxygen octahedral rotation angle is only about 0.03 . for such a complicated structure , even a complete first - principles determination of its @xmath143 structure would be very difficult . however , we can arrive at a partial understanding of this structure as follows . our calculations show that the reference cubic structure is unstable towards either the r - point or the m - point afd mode individually ( negative @xmath144 ) , whereas it is _ stable _ with respect to the x - point mode individually ( positive @xmath144 , no double - well behavior ) . in fact , the x - point mode involves a strong distortion of the oxygen octahedron , and thus is far from soft . however , the symmetry of the crystal is such that if both the r - point and m - point afd mode distortions are already simultaneously present , then the ca and o atoms experience forces in the pattern of the x - point mode , as a result of the cubic anharmonic interactions discussed in sec . thus , under these conditions the crystal would necessarily acquire some x - point mode distortion . we therefor conclude that the appearance of the x - point mode distortion must be the result of the third - order coupling between r- , m- , and x - point afd modes . this is a major reason why we chose to include the third - order coupling in our model hamiltonian . the strain coupling is expected to be important in the determination of the actual magnitude of the x - mode distortion , complicating the problem and making a complete quantitative lda determination of the x - mode amplitude difficult . = 3.0 in to obtain the phase transition sequence , we start the mc simulation at a high temperature ( @xmath145k ) and equilibrate the system for 10,000 mc sweeps . an isotropic pressure @xmath146gpa is imposed to restore the experimental lattice constant , and all the subsequent simulations are done under this pressure . the temperature is reduced in small steps ( as small as 10k around the transition temperature ) with 30,000 mc sweeps at each @xmath139 to ensure equilibration . the order parameters are accumulated over the last 20,000 mc sweeps , after checking that they do not vary significantly over this period . in our simulation , we find that except for the r - point afd order parameters @xmath122 , all other fe or afd order parameters are zero throughout the simulation . so the only phase transition we observe is associated with @xmath122 . [ corr](a ) shows the calculated order parameters @xmath147 as a function of temperature . ( what we actually plot are the averaged maximum , intermediate , and minimum absolute values of the order - parameter components . ) at high temperature ( @xmath148k ) the system is in the cubic structure , where all the components are zero . the material goes through a phase transition at 1750k , where all three order parameter components increase simultaneously . our phase transition temperature is close to the experimentally measured 1530k , but we obtain the wrong low - temperature structure . ours is rhombohedral with a 10-atom cell , instead of orthorhombic with a 20-atom cell as observed experimentally . the difference between our theoretical result and experiment is not quite as dramatic as it might seem . both the structure and the structural energy are very similar for the r - point and the m - point afd mode polarized along ( 001 ) . note that in the observed structure , the amplitude of the m - point mode is about @xmath149 times the the amplitude of the r - point mode polarized in the ( 1@xmath1410 ) direction . thus , we can say that the main difference between our calculated structure and the observed one is that one component of the r - point afd mode is replaced by an m - point mode in the observed structure . as argued above , the additional presence of the x mode is just a result of third - order anharmonic coupling . in fact , we find that if we artificially increase the third - order coupling constant @xmath102 by a factor of 5 , we recover the experimental @xmath143 structure in our mc simulations . clearly , the fact that the structure is strongly affected by relatively weak anharmonic intersite interactions makes the determination of the correct low - temperature phase very difficult in catio@xmath0 . it is possible that a more careful treatment of the cubic intersite interactions ( for example , an independent determination of the coupling constants associated with all three of the cubic anharmonic invariants discussed in sec . ii.b , or three - site or further - neighbor terms ) might bring a better agreement with experiment , although one should not rule out the possibility that neglect of quantum fluctuations@xcite or intrinsic limitations of the lda might be at fault . experimentally , nanbo@xmath0 is probably the most complex cubic perovskite known . the high - temperature phase is the simple prototype cubic structure as in the other cubic perovskites . below 910k , a whole series of structural phase transitions has been found and at least six more phases have been identified . as the temperature decreases , the compound first goes through a cubic tetragonal transition at 910k with freezing in of @xmath150 modes polarized along one axis . there are then three orthorhombic phases present in the temperature range 845638k , the most complicated having a unit cell containing 24 nanbo@xmath0 formula units . all of these phases can be regarded as given by rigid rotations of oxygen octahedra , accompanied by small induced x - point distortions . from 638k down to at least 170k , nanbo@xmath0 is antiferroelectric with an orthorhombic unit cell containing eight formula units . at even lower temperature , the crystal has been reported to transform into either a rhombohedral @xcite or monoclinic@xcite structure . the complexity of the structural phase - transition sequence suggests the presence of several competing structural instabilities with very similar free energies . in principle , all the distortions involved in the observed structures of nanbo@xmath0 are included in our model . however , it is not realistic to expect that the calculated structural energies and free energies will be in exactly the right order , given the complexity of the problem and the level of accuracy of current first - principles based approaches . nevertheless , we believe a first - principles study of nanbo@xmath0 is still important in identifying the most prominent distortions , as well as for demonstrating the limitations of such approaches . the determination of the structure is done using mc simulations on a cubic @xmath151 simulation supercell . an isotropic pressure @xmath152gpa is imposed to restore the experimental lattice constant , and all the subsequent simulations are done under this pressure . we start the simulation at very high temperature and equilibrate . the temperature is reduced in small steps ranging from 10k to 50k depending on proximity to a phase transition . at each temperature step , at least 40,000 mc sweeps are used to ensure that equilibrium is reached . the order parameters are accumulated over the last 30,000 mc sweeps . the calculated averages of order parameters @xmath153 and @xmath117 are shown in fig . [ corr](b ) as a function of temperature . all other modes are found to be zero throughout the simulation . as was the case for catio@xmath0 , the averaged maximum , intermediate , and minimum absolute - value components are plotted . at high temperature ( @xmath154k ) , the system is in the cubic structure with all the order - parameter components close to zero . as @xmath139 decreases to about 700k , one afd component increases rapidly and becomes significantly non - zero , and the structure transforms from cubic to tetragonal . with further decrease of temperature , a second component became non - zero , indicating the occurrence of an orthorhombic phase . below 560k , a third afd component grows and the structure becomes rhombohedral . at very low temperature ( below 50k ) , the simulation also apparently shows a sequence of three ferroelectric transitions , and the compound ends up in a rhombohedral ferroelectric structure at very low temperature . our first cubic tetragonal phase transition compares favorably with experiment ; we obtain the correct structure and underestimate the transition temperature by only @xmath15520% . in the orthorhombic phase , however , the calculated structure is much simpler than the observed one . only one orthorhombic phase seems to occur in our simulation . however , fig . [ corr](b ) shows signs of fluctuations occurring in the vicinity of this phase ( these fluctuations persist even if the number of mc sweeps is increased significantly ) . this indicates that the orthorhombic phase is not very stable , and may involve a mixing of different phases . moreover , the transitions do not appear very distinct in fig . [ corr](b ) as a result of finite - size broadening , so increasing the lattice size may help to resolve the different phases . however , the computational load increases rapidly with increasing lattice size , and it becomes impractical to carry out simulations at much larger size . our inability to get the correct afe phase at room temperature is probably the most significant failure of our approach . our zero - temperature structure is ferroelectric , but since the fe phases occur only at such low temperatures , it is likely that quantum fluctuations would need to be included to determine the actual low - temperature structure.@xcite in this and previous studies , we perform a series of _ ab initio _ studies of the thermodynamic properties of perovskite compounds . without introducing any adjustable parameters , we have calculated structural transition sequences , transition temperatures , phase diagrams , and other thermodynamic properties based on first - principles calculations . for compounds with simple phase transitions , like batio@xmath0 and srtio@xmath0 , our calculated thermodynamic properties agree very well with experiment observations . for more complicated compounds like catio@xmath0 and nanbo@xmath0 , our results are less satisfying . there are two major sources of errors , the inaccuracy of lda calculations and the imperfection of our models . our lda calculations have been carefully performed to avoid possible errors , and convergence has been carefully tested . as for the intrinsic accuracy of lda , our calculated structural parameters and energies are within a few percent of experimental values . although this is the usual high accuracy observed generally for the lda , it is unfortunately not enough for truly accurate determination of the thermodynamic properties of perovskites . for example , it is embarrassing that we are forced to choose between carrying out the calculations at the theoretical equilibrium lattice constant or the experimental lattice constant ( negative fictitious pressure ) ; this choice can affect phase transition temperature by @xmath155100% . we regard this as being the most important probable source of error in our calculations . it is also possible to improve our model hamiltonian . for example , our restricted assumption for the form of the third - order intersite interactions may be lifted , resulting in a significantly more complicated model hamiltonian . also , other higher - order terms can be included in the hamiltonian . it would also be possible to include more degrees of freedom per cell , or include eigenvector information from more k - points of the brillouin zone when defining the local - mode vectors , to treat other phonon excitations more accurately.@xcite however , in view of the current accuracy of first - principles calculations , we are not sure that these modifications would dramatically improve our results . finally , we emphasize that all of the mc simulations reported here treat the atomic motion purely classically . as mentioned above , we have recently reported results of quantum path - integral mc simulations showing that quantum fluctuations of the atomic coordinates ( i.e. , zero - point motion ) can shift transition temperatures by tens of degrees , and in some cases even eliminate delicate phases.@xcite certainly this remains an important avenue of investigation for catio@xmath0 and nanbo@xmath0 , but we nevertheless think it unlikely that inclusion of quantum fluctuations would immediately resolve the discrepancies with the experimental phase diagrams for these compounds . in conclusion , we have extended our previous first - principles theory for perovskite ferroelectric phase transitions to treat also antiferrodistortive transitions . we apply this approach to the three cubic perovskite compounds srtio@xmath0 , catio@xmath0 , and nanbo@xmath0 , and calculate their thermodynamic properties including phase transition sequences and transition temperatures . for srtio@xmath0 , our calculated results are in good agreement with experiment . for catio@xmath0 and nanbo@xmath0 , our calculated structural transitions have the correct general trend and the transition temperatures are in rough agreement with experiment , but the calculated transition sequences are not correct in detail . we attribute this to the larger distortions and many multiple competing instabilities in these compounds . for srtio@xmath0 , we also analyzed the intersite correlations for both fe and afd local modes , finding needle - like and pancake - like correlations respectively for fe and afd modes as expected on physical grounds .

solar flares are sudden , brief , and powerful outbursts , releasing an energy of @xmath0 ergs on time scales of a second to several tens of minutes . in solar flares , both protons and electrons are accelerated to non - thermal energies . although these energetic charged particles are strongly affected by the solar magnetic fields and can not usually be detected , we can utilize neutral secondary particles , neutrons and photons , to probe them . among them , hard x - ray to gamma - ray continua , produced via bremsstrahlung process , provide the best diagnostics of the flare - accelerated electrons . although high - spatial - resolution images of flares in the hard x - ray and gamma - ray energies have become available by the yohkoh hxt @xcite and rhessi @xcite respectively , we still depend heavily on the spectral data when attempting to study the non - thermal electrons . actually , to calculate back the electron spectrum from the observed gamma - ray spectra has been an important issue in solar physics . as we understand , flare - accelerated electrons having a power - law distribution penetrate the solar chromosphere , and produce bremsstrahlung photons via `` thick - target '' emission . the thick - target emission is a process wherein non - thermal electrons impinge on a thick matter , and lose almost all their energies through repeated coulomb collisions while emitting x - ray / gamma - ray photons via bremsstrahlung . in the thick - target condition , the spectrum of x - ray emitting electrons is formed by an equilibrium between the injection of newly accelerated electrons and the loss of their energies predominantly through collisions . if the injected electron number spectrum is given by a power - law as @xmath1 ( electrons kev@xmath2 s@xmath2 ) , with @xmath3 the electron kinetic energy and @xmath4 a constant called electron index , then we expect the x - ray photon - number spectrum also to take a power - law from as @xmath5 where @xmath6 is a quantity called photon index . in 1974 , brown found an approximate relation as @xmath7 in non - relativistic regime , assuming that all emitted photons are collected . the photon spectrum thus becomes relatively flat , because of `` loss - flattening '' effect working on the electrons ; as they penetrate deeper into the target material , the spectrum gradually hardens because lower - energy electrons have shorter time scales of energy loss through coulomb collisions . besides the thick - target condition , another extreme case called thin - target condition ( brown 1971 ) was studied extensively . this represents a situation wherein a bunch of energetic electrons pass through a thin medium while emitting bremsstrahlung photons , but they leave the emission region with their energies almost unchanged . in the thin - target condition , the energy distribution of the incident electron is not significantly affected by collisions with the target . the emergent photon spectrum is steeper , because the electrons are free from the loss flattening effect and is characterized as @xcite @xmath8 since then , there have been many attempts to improve equations ( [ eq.brown ] ) and ( [ eq.brownthin ] ) , particularly toward relativistic regime . some utilized purely analytic calculations of various probability distributions , while others employed numerical solutions to the fokker - planck equation , or monte carlo techniques . in electron energies up to 10 mev , @xcite investigated angular dependence of the bremsstrahlung in the hard x - ray range , and discussed the associated electron transport . @xcite dealt with the particle transport by employing a simple thick target emission model , assuming an isotropic electron distribution . @xcite conducted a detailed study of the directional bremsstrahlung emission by anisotropically accelerated electrons . @xcite made a calculation including pitch angle scattering of electrons by plasma waves . these previous works have indicated that the gamma - ray continua of solar flares can be generally explained in terms of chromospheric thick - target emission , produced by electrons having a power - law energy distribution with @xmath9 . in spite of these extensive studies , the estimation of electron spectra from flare gamma - ray data would not be self - contained unless we take into account another important physical process ; namely , compton scattering . the relativistic bremsstrahlung from precipitating electrons must be significantly forward - beamed , so that a majority of gamma - ray photons must be compton back - scattered by the solar materials in order to reach us . this process will strongly degrade the gamma - ray energies , and affect the photon spectrum to be observed . to treat this kind of physical condition which involve multiple compton scatterings , we clearly need to use monte carlo techniques . the effect of compton scattering has been studied by several authors , beginning , e.g. , with @xcite who used monte - carlo simulations in such early days . kontar et al . ( 2006 ) considered this issue using a semi - analytic green s function by magdzias & zdziarski ( 1995 ) , and calculated `` primary '' photon spectra from observed ones . these works have clarified that the compton back - scattering can significantly modify the bremsstrahlung spectra . however , these pioneering works have treated electrons with sub - mev kinetic energies , leaving higher energies less studied . in the present paper , we calculate not only the electron transport and bremsstrahlung emission , but also the photon propagation via compton process . for this purpose , we employ a monte carlo simulation toolkit named geant4 , which is widely used in experimental high energy physics . our results imply that the compton scattering has significant effects on the observed gamma - ray spectra . together with full - relativistic effects , this is expected to require revision to the relation such as equation([eq.brown ] ) and equation([eq.brownthin ] ) . we show the relevance of these results to actual flares , by comparing the simulated gamma - ray spectra with those observed with yohkoh from a sample of solar flares . the geant4 monte carlo simulation toolkit @xcite , developed in experimental high energy physicists , has four important features . firstly , it allows a very flexible construction of `` geometry '' in which particles interact . secondly , it can `` track '' individual particles while monitoring their physical quantities such as the energy , momentum , and position ; the name geant in fact comes from `` geometry and tracking '' . thirdly , it allows us to incorporate the desired interactions ; in the present case , coulomb scattering and bremsstrahlung for electrons , and compton scattering for photons . finally , the toolkit has been tested and calibrated extensively by many investigators under different conditions . in our monte carlo simulation using geant4 , we basically shoot a parallel beam of a large number of electrons , either vertically or slantly , into our `` solar atmosphere '' to be explained later . the electrons are assumed either to be mono - energetic , or to have a power - law energy distribution with an index @xmath4 . in the latter case , the initial electron energies are assumed to be in the range of 1100 mev , so as to simulate gamma - ray flares . we implemented the following physics processes which we require our particles to obey in our simulation : ionization , multiple scattering , and bremsstrahlung for electrons , while photo - absorption , compton scattering and electron - positron pair creation for photons . we suppressed the production of secondary electrons for simplicity . because those secondary electrons ( typically less than 100 kev ) have much lower energies than the primary ones , they can be considered negligible . one of the essential features of geant4 is step - by - step tracking of each individual particle , considering all the implemented elementary processes . the particle tracking algorithm roughly consists of the following 4 steps . 1 ) the initial particle velocity is calculated . 2 ) each physical process calculates the mean - free - path of the specified particle under the relevant interaction , and generates a random number around the mean . then it proposes this random number as a step length . 3 ) among various competing processes , the interaction which proposes the shortest physical length is adopted . 4 ) track properties , such as the kinetic energy and momentum , of the current particle are updated . among the implemented processes , the coulomb scattering of electrons need special consideration , because its mean free path is much smaller than those of the other processes ( e.g. , @xmath10 of that of bremsstrahlung for a 10 mev electron ) . as a result , its full monte carlo treatment would demand too much computing times . therefore , the expected total energy loss of an electron due to multiple coulomb scatterings , in each step which is determined by the processes other than the coulomb scattering itself , is computed based on semi - empirical formulae by @xcite . fluctuations around the mean energy loss is expressed by generating random numbers after @xcite . similarly , the total spatial displacement caused by the multiple scattering in each step is calculated based on lewis method @xcite , which solves a diffusion equation of electrons in the matter . fluctuations around the expected mean is again represented by random numbers @xcite . thus , geant4 simulates the effect of coulomb interactions after a given step as a statistical sum of a large number of scatterings , instead of treating them one by one . in this way , the injected electrons and the bremsstrahlung - generated photons are tracked , until they escape out of the model boundary , or they become less energetic than a certain threshold energy which we set at 1 kev . we collect these photons under a specified condition , and make their spectra . the simulation of flare - accelerated electrons , precipitating onto the chromosphere , requires yet another physical process to be implemented : electron gyration around the solar magnetic fields . its exact treatment , however , would make the simulation extremely time - consuming , because the gyration radius of a @xmath11 mev electron in a @xmath11 gauss field is as small as @xmath12 m. therefore , we have decided not to take gyration into our simulation , but instead , to consider slant injections ; as illustrated in figure [ fig.coordinate ] , injecting electrons at a polar angle @xmath13 , and collecting the emergent photons at another polar angle @xmath14 but all over the azimuth angle @xmath15 , are equivalent to simulating electrons gyrating with a constant pitch angle @xmath13 , as long as the field lines are close to normal to the photosphere . the case of tilted magnetic fields can be reproduced by a superposition of different injection angles , @xmath13 . before actually simulating the electron transport and gamma - ray emission employing a realistic model for the solar atmosphere , we carried out simpler simulations with two purposes in mind ; to validate our simulation code by examining whether each of the relevant physical processes is correctly reproduced , and to grasp the essence of the physics to be investigated . we accordingly approximated the sun as a hydrogen gas box of @xmath16 , having a uniform density of 3.2@xmath17 g @xmath18 which is the density at the solar photosphere . hereafter we call this model `` simple solar atmosphere model '' ; it is one of three uniform target models employed in the present paper . we vertically injected @xmath19 mono - energetic electrons with an initial kinetic energy of 50 mev into our `` simple solar atmosphere '' . in order to monitor how the coulomb interaction affects the spectrum and angular distribution of the electrons , we prepared imaginary boundaries at different depths , 100 , 200 , and 300 km from the injection surface , and collected electrons at each boundary . figure [ fig.monoelectronspectrum ] a shows the electron spectra obtained in this way , while figure [ fig.monoelectronspectrum]b the associated angular distributions . from these , we can see that the maximum energy of the spectrum decreases as the electrons penetrate deeper . since a 100 km thick slab in our `` solar atmosphere '' has a column density of 3.2 g @xmath18 , an electron of energy @xmath20 50 mev passing through it is expected to lose @xmath20 15 mev , as we can easily calculate from the bethe - bloch formula . the results in figure [ fig.monoelectronspectrum]a indeed meet this expectation . the geometrical thickness ( 10,000 km ) of the model is thus large enough to make it physically thick to coulomb loss , for electrons with initial energies up to @xmath201.5 gev . furthermore , the electron energy distribution is seen to broaden as the energy loss proceeds . figure [ fig.monoelectronspectrum]b reveals that the injected electrons , initially forming a parallel beam , are gradually deflected due to coulomb interactions with ambient atoms . the simulated angular distributions approximately consist of two components : a gaussian component caused by small - angle multiple scatterings , and a power - law tail due to large - angle rutherford scatterings . in a similar way , we simulated the transport of energetic electrons which has a power - law distribution over the 1 - 100 mev range with @xmath21 . the spectra and angular distributions , measured at depths of 100 , 200 , 300 , 400 , and 500 km , are presented in figure [ fig.powelectronspectrum ] . the maximum electron energy ( initially at 100 mev ) thus decreases according to the bethe - bloch formula . furthermore , the low - energy part of the spectrum gradually flattens , because the energy loss per unit length is roughly independent of the electron energy , and hence lower - energy electrons suffer a larger _ fractional _ energy loss than higher - energy ones . this is essentially the same as the `` loss flattening '' effect mentioned in section [ sec.introduction ] . the electron index we employed here , @xmath21 , is significantly smaller ( flatter ) than would be achieved in the standard diffusive shock acceleration process @xcite . this is motivated by two reasons . one is that the effects of compton scattering are more clearly observed as @xmath4 gets smaller . the other is that such a flat electron distribution could be realized in actual solar flares , via , e.g. , direct electric acceleration . hereafter , we hence consider both flat ( @xmath22 ) and steep ( @xmath23 ) electron distributions . we investigated photon spectra created by the bremsstrahlung process under the two representative conditions , namely the thick - target and thin - target conditions explained in section [ sec.introduction ] . the geant4-simulated gamma - ray spectra are compared with analytic formulae , which are fully relativistic unlike the non - relativistic calculations of eq.([eq.brown ] ) and eq.([eq.brownthin ] ) . to validate the bremsstrahlung process under the thin - target condition which is simpler of the two , we let @xmath24 mono - energetic electrons vertically penetrate a second simplified model atmosphere , namely a uniform hydrogen gas with a size of ( 1000 km)@xmath25 , having a density of @xmath26 g @xmath27 ; this is henceforth called `` thin atmosphere model '' . the model has a column density of 1.0 mg @xmath18 , so that the injected electrons lose their energies by no larger than 1 % , in agreement with the thin - target condition . we collected the information of each bremsstrahlung photon when it was created , and obtained the spectra shown in figure [ fig.bremsmonoenergyspectrum ] . also shown in figure [ fig.bremsmonoenergyspectrum ] are analytically calculated spectra , using a fully relativistic formula @xcite , as @xmath28 \right\ } \label{km3bse}\end{aligned}\ ] ] with @xmath29 here , @xmath30 is a cross section for an electron of total energy @xmath31 ( including the rest mass energy ) to interact with a target atom of charge @xmath32 , and emit a bremsstrahlung photon of energy @xmath33 , thus achieving a final energy of @xmath34 . the classical electron radius is denoted @xmath35 . as seen in figure [ fig.bremsmonoenergyspectrum ] , the difference between our simulation and the analytical calculation is less than 10 % , indicating that the bremsstrahlung process has been correctly implemented . in the fully relativistic regime(@xmath36,@xmath37 ) , in particular , the emitted gamma - ray spectrum has a power - law shape with a photon index @xmath38 , or an energy index of @xmath39 , up to the maximum photon energy that is identical to the initial photon energy . this is reasonable , because eq.([km3bse ] ) predicts @xmath40 when @xmath41 and hence @xmath42 . figure [ fig.powerthinspectrum ] shows the bremsstrahlung photon spectra calculated under the same thin - target condition , but when the electrons are assumed to have a power - law spectrum distributed from 1 to 100 mev , with @xmath43 or 1.5 . the simulated spectra again exhibit approximately power - law like shapes , with a steeper slope than in the case of mono - energetic electrons because softer electrons can obviously emit only lower - energy photons . in figure [ fig.powerthinspectrum ] , we also show photon spectra calculated analytically using equation ( [ km3bse ] ) as @xmath44 where @xmath45 is the photon number spectrum , @xmath46 is the injected electron spectrum ( @xmath47 in the present case ) , @xmath48 is the target ion density , and @xmath49 is the velocity of an electron having an energy @xmath50 . the simulation agrees with the analytic results within @xmath51 . thus , the geant4 simulation and the fully - relativistic analytic formalism consistently indicate @xmath52 for @xmath53 and @xmath54 for @xmath55 , both measured in a typical gamma - ray energy range of 1 to 10 mev . these photon indicies are significantly smaller ( flatter ) , particularly for @xmath55 , when compared to the conventional non - relativistic relation of equation ( [ eq.brownthin ] ) . in this way , we repeated the geant4 simulation with difference values of the electron index , and calculated the bremsstrahlung photon index over an energy of 1 - 10 mev . the obtained relation between the electron index and the photon index is shown in figure [ fig.thinthick ] , in comparison with the prediction by equation ( [ eq.brownthin ] ) . we see that the formula of equation ( [ eq.brownthin ] ) in the non - relativistic regime , namely @xmath56 , is no longer valid in the present relativistic regime , with the discrepancy increasing toward larger electron indicies . the cross section of bremsstrahlung as a function of emitted photon energy and angle is given by @xcite and @xcite as @xmath57\ln m(y)\right\ } , \label{km2bs}\ ] ] where `` reduced photon angle '' @xmath58 , together with its function @xmath59 , is given by @xmath60 and @xmath61 is the initial angle of an emitted photon . figure [ fig.bremsangdistvalid ] shows equation ( [ km2bs ] ) as a function of @xmath62 , where the results of the simulation using the thin atmosphere model are superposed . thus , the whole emission becomes concentrated roughly within @xmath63 ; the emission is strongly forward - peaked to within @xmath64 . the thick - target emission is significantly more complex than the thin - target case , because it results from electrons while their energies change continuously due to coulomb scattering . moreover , it has a higher practical importance , because the flare hard x - rays and gamma - rays are thought to be produced mainly through this process in the solar chromospheric and denser regions . under the thick - target condition , we first investigated the emission from mono - energetic electrons , as we did for the thin - target case . we shot @xmath65 electrons with an initial kinetic energy of 10 mev or 50 mev , into a hydrogen target of which the size is ( 10@xmath66 km)@xmath25 ; this is our third simplified model atmosphere called `` thick atmosphere model '' . figure [ fig.monothicksim ] shows the spectra produced by collecting all photons emitted by each electron while its energy gradually decreases down to @xmath67 kev . compared with the thin - target emission from mono - energetic electrons , the photon spectra are somewhat steeper , and roll over more prominently toward the maximum energy . obviously , this is because the electrons keep losing their energies while radiating . in figure [ fig.monothicksim ] , we also show the results of analytic calculations , which are formulated as @xmath68 where @xmath69 is the time , @xmath70 is the initial total energy of an electron , @xmath71 is its energy loss rate , @xmath72 is the instantaneous electron velocity , @xmath48 is the ion density of the target plasma , and @xmath73 refers to equation ( [ fig.bremsmonoenergyspectrum ] ) . again , the simulation and analytic calculation agrees within @xmath74% . we investigated the thick - target emission from power - law distributed electrons , with @xmath43 or @xmath75 , over the 1 - 100 mev energy range . employing @xmath76 electrons , we collected information of the photons in the same way as in the preceding simulations , and obtained figure [ fig.powthicksim ] . designating the electron injection rate ( electrons @xmath77 ) by @xmath78 and the target area by @xmath79 , and employing equation ( [ fig.bremsmonoenergyspectrum ] ) , the photon production rate is analytically given by @xmath80 this prediction is again show on the figure . the simulation agrees within @xmath81% with the analytic calculation via eq.([eq.powbremsthick ] ) . as can be seen from figure [ fig.powthicksim ] , the photon spectra emitted by power - law distributed electrons under the thick - target condition again exhibit power - law shape up to @xmath12 mev , like in the case of the thin - target emission ( figure [ fig.powerthinspectrum ] ) . the obtained photon index is @xmath82 for @xmath83 and @xmath84 for @xmath85 . in contrast , the hardest portion of the spectra exhibits rather opposite trend , and can not become as hard as implied by equation.([eq.brown ] ) , since the maximum energy of electrons decreases as they propagate deeper through the thick target . figure [ fig.thinthick ] ( right ) is the same comparison as figure [ fig.thinthick ] ( left ) , but for the thick - target condition . photon indicies in this case are generally smaller compared with thin target emission , because of the loss - flattening effect ( section [ sec.introduction ] ) . in addition , @xmath6 depends more weakly on @xmath4 than in the thin - target case . in the previous section , we treated the solar atmosphere as a uniform gas slab . this is basically reasonable , because the energy losses of electrons and photons are primary determined by the column density of matter along its path . the actual solar atmosphere , with the strong vertical density gradient , could be transformed into such a uniform zone , as long as all the particles are traveling vertically . however , particle trajectories with significant transverse components can no longer be conserved by such a transform , and hence we need to construct a more realistic solar model taking into account its vertical density gradients . we here refer to `` harvard - smithsonian reference atmosphere '' @xcite for the solar mass - density profile , and approximate this numerical table analytically by an exponential function as @xmath86 here , @xmath87 is the vertical coordinate measured from the photosphere , @xmath88 is the mass density in the unit of g @xmath27 , and @xmath89 is the scale height which is @xmath90 km for @xmath91 ( the solar interior ) and @xmath92 km for @xmath93 ( the coronal region ) . we construct our new solar atmosphere model as a series of boxes ( or `` zones '' ) stacked in the vertical direction . individual zones are defined to have different densities but to share approximately the same column density , and the density within each of them is assumed to be uniform . each zone is assumed to have a lateral extent of @xmath94 km , to simulate the cross section of a typical magnetic loop . on this transverse scale , the curvature of the solar surface is negligible . as illustrated in figure [ fig.zones ] , we divide the coronal ( @xmath93 ) region into 5 zones , each having approximately the same column density of 0.70 g @xmath18 . according to the harvard smithonian model , the total column density above @xmath95 is 3.5 g @xmath18 , which corresponds to a depth of @xmath11 km in our simple atmosphere model . also , the sum of the 5 coronal zones in our new model gives an overall column density of 3.5 g @xmath18 , thus faithfully representing the actual solar atmosphere for @xmath93 . while traversing each zone , a 100 mev electron loses @xmath96 mev in total , of which only @xmath81 % is in radiation whle the rest is in coulombic loss . in simulating the solar interior ( @xmath91 ) , we use 10 thicker zones having 3.5 g @xmath18 each ; again , this corresponds to a depth of @xmath11 km in our simple atmosphere model . with the overall column density of 35 g @xmath18 below @xmath97 , our new model goes down to a depth of @xmath98 km , below the solar photosphere in comparison with equation ( [ eq.harvarddensity ] ) . this depth is considered sufficient , because a 100 mev electron , injected at the top of our model , would lose @xmath11 % of its initial energy by the time it reaches the bottom of the deepest zone . furthermore , the thicker 10 zones , when summed up , has a compton optical depth of @xmath12 , for a vertically precipitating photons ; few photons would penetrate deeper . although our `` solar atmosphere '' is comprised of pure hydrogen , the actual solar atmosphere contains helium to @xmath99 by number , or @xmath100 by mass . since the ionization loss of electrons is proportional to the electron number density , the helium would decrease the ionization loss per unit mass column density by @xmath99 . bremsstrahlung per unit mass should remain the same , because a helium atom has 4 times higher mass than a proton but 4 times higher bremsstrahlung cross section at the same time . the compton scattering depends only on the number of electrons , and hence would decrease by @xmath99 if helium was included . to examine the flare photon spectra , we vertically injected @xmath76 electrons into our solar atmospheric model from a height of @xmath101 km , and tracked them from that height to @xmath102 km , as they lose energy via coulomb scattering and emit bremsstrahlung photons . the electrons are assumed to be distributed initially in the 1 - 100 mev range with a power - law index @xmath21 . the electrons , initially moving in the vertical direction , gets gradually deflected as indicated in figure [ fig.powelectronspectrum ] ; in actual configuration , this corresponds to pitch - angle scattering . nevertheless , as represented by figure [ fig.powelectronspectrum ] in red , still 98 % of the electrons are contained within @xmath103 of the vertical direction when they have reached the photosphere . therefore , the vertical injection employed here is valid as long as the magnetic fields are sufficiently normal to the solar photosphere , the electrons pitch angle is initially concentrated near @xmath104 , and non - coulombic ( e.g. , magnetohydro - dynamic , or due to magnetic mirroring ) pitch - angle scatterings can be neglected . the bremsstrahlung photons emitted by the almost vertically moving electrons are strongly forward collimated , as indicated by figure [ fig.bremsangdistvalid ] . therefore , most of them are expected to enter the atmosphere and undergo compton scattering or photoabsorption . although some of these photons will die , others will be scattered back to escape out of the photosphere . we collected these outcoming photons at different viewing angles @xmath14 ( figure [ fig.coordinate ] ) , and averaged the results over @xmath105 . figure [ fig.comptongamma0deg ] shows the photon spectra obtained in this way for equal intervals in @xmath106 ( hence over an equal solid angle ) . in figure [ fig.comptongamma0deg ] , spectra with small @xmath14 , simulating disk - center flares , exhibit a clear `` knee '' , which represents the effect of single compton scatterings . in the case of @xmath107 ( blue in figure [ fig.comptongamma0deg ] ) , the knee energy becomes @xmath108 , because of the strong energy degradation in the single compton back - scatterings . photons with energies above the knee result either directly from the bremsstrahlung source , or through multiple compton scatterings with a relatively small energy loss in each step . the spectrum in this region has a much steeper slope than in the region below the knee . as @xmath14 gets larger in figure [ fig.comptongamma0deg ] , the knee energy increases , because the angle needed in the single compton scattering decreases nevertheless , the spectrum remains much softer ( e.g. , @xmath109 in the 0.2 - 0.6 mev range ) than is expected when the compton effects are not taken into account ( @xmath110 ; figure [ fig.thinthick]b ) . the observed photon flux below the knee energy decreases as @xmath14 increases and approach @xmath111 , due to the same limb darkening effect as in visible light . as we have confirmed in figure [ fig.comptongamma0deg ] , bremsstrahlung photons back - scattered from the solar photosphere exhibit very steep spectra as a result of the compton degradation . in order to explain the origin of the observed flare spectra often extending beyond 1 mev , we must then considered photons which reach us after compton scatterings with much smaller angles . for this purpose , we have to consider electrons penetrating into the atmosphere with shallow angles ( i.e. , large @xmath13 ) , namely `` slant injection '' . we then examined the case of slant injection , by shooting @xmath76 power - law electrons ( in the 1 - 100 mev energy range ) , into our realistic solar model . specifically , we repeated the geant4 simulation under the same condition , but assuming this time an injection angle of @xmath112 ( figure [ fig.coordinate ] ) , instead of the vertical injection ( @xmath113 ) considered in the previous subsection . we now inject the electrons from a height of 100 km ( i.e. , neglecting the uppermost two zones in figure [ fig.zones ] ) , to avoid them escaping sideways before arriving at thicker regions . figure [ fig.comptongamma80deg ] shows the results of this simulation for the same three intervals of @xmath14 as figure [ fig.comptongamma0deg ] , again averaged over the observer s azimuth angle @xmath15 . the spectrum becomes considerably harder than in the case of vertical electron injection , and extends much beyond @xmath114 mev when @xmath14 is rather large ( @xmath115 ) . in this case , electrons make grazing incidence on the atmosphere , and some of their bremsstrahlung photons get nearly forward scattered to leave the zone with a small energy loss , thus producing the hard power - law continuum . in this way , we have monte - carlo simulated slant injections for a family of electron spectra with various indicies @xmath4 ( again distributed over 1100 mev ) , assuming two representative injection angles of @xmath116 and @xmath117 . figure [ fig.deltagamma ] summarizes the photon index @xmath6 of 110 mev gamma - rays , collected over the most favorable viewing angle of @xmath118 ( i.e. , rather grazing to the solar photosphere ) and integrated over the observer s azimuth . thus , even with the extreme assumption of @xmath119 , the simulated gamma - ray spectra become significantly softer ( typically by @xmath114 in @xmath6 ) than the original bremsstrahlung spectra before the photons are compton scattered ( reproduced in figure [ fig.deltagamma ] in blue ) . moreover , @xmath6 further increases by @xmath120 as @xmath13 decreases from @xmath121 to @xmath122 . these results imply that the compton scattering significantly soften the emergent photon spectra , of which the effect depends sensitively on @xmath13 and @xmath14 . an interesting prediction of figure [ fig.deltagamma ] is that the observed gamma - ray photon index should appear in a relatively narrow range of @xmath123 , regardless of the electron index @xmath4 . of course , we would observe considerably larger values of @xmath6 when either @xmath13 or @xmath14 is rather small . however , such a spectral component would not be easily detected in the mev energy region . although our simulations of the slant injection neglected the uppermost two coronal zones , their effects may not be necessarily negligible , because the large injection angle @xmath124 ( or @xmath122 ) effectively increases the electron path length in each zone by @xmath125 [ or @xmath126 times . since the electron energy losses in these two regions , if properly taken into account , would reduce the maximum electron energies while flatten the low - energy electron slope ( figure 3 ) , the emergent photon flux above @xmath20 1 mev would further be suppressed if the electron injection in actual flares takes place at a height much exceeding the assumed 100 km . in other word , our simplification employed in simulating the slant injection makes the case more conservative . in order to investigate the effects of compton scatterings on the solar gamma - ray spectra , we have conducted monte carlo simulations of energetic electrons , impinging on the solar atmosphere , using geant4 as our basic toolkit . we modeled the solar atmosphere as vertically stacked parallel hydrogen zones , each assumed to be uniform . the density gradient was expressed by assigning higher densities to lower zones . we injected electrons which are power - law distributed over the 1 - 100 mev range , changing their spectral index @xmath4 and incident angle @xmath13 . the results of the present simulations can be summarized as follows . 1 . the observed photons hardly reach mev energies if the electrons are injected vertically , because the bremsstrahlung - produced gamma - rays must be compton back - scattered to reach the observer , and hence their energies are subject to strong compton degradation . the emergent gamma - ray spectrum , integrated over the observer s azimuth , gradually hardens as the injection angle @xmath13 increases ( becoming more `` slant '' ) , and also as the observer s polar angle @xmath14 gets larger . 3 . as long as @xmath13 and @xmath14 are both relatively large , the azimuthally - averaged 1 - 10 mev gamma - ray spectra exhibit @xmath127 over a relatively wide range of @xmath4 . these results generally agree with , and further extend , the previous works by , e.g. , @xcite and @xcite . there are some effects which are not considered in our simulation . among them , those of the secondary electrons and helium were already described in section 2.2 and section 3.1 , respectively . as the electron energy increases , their synchrotron energy loss becomes significant . however , in the chromospheric region in which we are interested , the magnetic field intensity is typically @xmath11 g , and hence the rate of synchrotron loss of 100 mev electrons is only @xmath114 % of their coulomb loss , and only @xmath128 % of their bremsstrahlung loss . this is because the coulomb and bremsstrahlung losses are rather large due to the high matter density . thus , the synchrotron process can be neglected . we also ignored the inverse compton process between flare - accelerated electrons and ambient photons , because electrons with the maximum energy assumed , 100 mev , can produce at most hard x - rays up to @xmath11 kev by scattering off the solar visible photons . however , if the electrons have a very flat spectrum well extending to energies above 350 mev , the inverse compton process may become a key process in creating gamma - ray photons . the size of the world used in our simulation , 10000 km @xmath129 10000 km @xmath129 20000 km , simulates a typical size of a foot point of a magnetic tube . the number of photons which escaped from the boundaries below the photosphere is less than @xmath130 of that of produced photons , and energies of these escaped photons are less than 100 kev each . hence we ignored those photons in the simulation . possible effects of the relatively low heights of our slant injection was already considered in subsection 3.3 . in the present simulation , we truncated the power - law electron spectrum at 100 mev . in order to examine whether this abrupt cutoff produces any artifact , we repeated some of the simulations by changing the maximum cutoff energy to 1 gev . q then we have found that @xmath6 decreases only slightly , typically 0.1 or less , if @xmath4 is larger than 2 . finally , a separate gamma - ray component could be produced without the compoton effect , simply if the electron spectrum consists of two power - law components with different slopes . even so , the good agreement between the predicted range of power - law indices and the _ yohkoh _ data suggests that the compton scattering is playing a significant role ( see the next section ) . in order to evaluate the significance of the results of our simulation , let us compare them with actual observations of hard x - ray and gamma - ray emission from solar flares . for that purpose , we have also to relate the simulation parameters @xmath13 and @xmath14 to observational parameters , including pitch - angle distributions of electrons , the tilt of magnetic field lines , and the flare longitude on the solar disk . as the observational data , we utilize 40 flares that show significant gamma - ray emissions , selected out of 2788 x - ray flares observed from 1991 october to 2001 december by yohkoh @xcite . yohkoh , the japanese solar satellite launched in 1991 , was able to take four - color hard x - ray flare images with the hard x - ray telescope(hxt ) @xcite and perform spectro - photometry over the @xmath131 mev range with the gamma - ray spectrometer ( grs ) @xcite . as the simplest case , we may consider a condition where the magnetic field lines are perpendicular to the photosphere . in this case , we can identify @xmath13 with the pitch - angle of an electron , and the effect of electron gyration around the magnetic field lines is fully represented by our photon collection method which takes an average over @xmath15 . if the electrons have a narrow pitch - angle distribution around 0 , the strong compton degradation , described in section 3.2 and shown in figure [ fig.comptongamma0deg ] , will make the observed photon spectrum very steep , and will prevent it from reaching mev energies . if , on the other hand , the pitch - angle distribution is rather broad , electrons with larger values of @xmath13 will emit harder spectra , as represented by figure [ fig.comptongamma80deg ] . conversely , as long as the field lines are close to vertical , flares with significant gamma - ray emission must involve broad pitch - angle distributions of electrons . analyzing the 40 gamma - ray flares detected by yohkoh , @xcite found that their gamma - ray ( typically 1 mev ) to hard x - ray ( @xmath20 70 kev ) flux ratios scatter largely ( by an order of magnitude ) , with no correlations to their hard x - ray spectral slopes . hence they conclude that there is a hidden parameter which causes the gamma - ray flux to vary significantly relative to the hard x - ray extrapolation , in such a way that the gamma - ray component often appears as a separate spectral hard tail ( see their figure [ fig.thinthick ] ) . given the present results , this parameter is very likely to be the electron pitch - angle distribution , with broader distributions yielding higher gamma - ray to hard x - ray flux ratios . in addition to the electron injection angle @xmath13 , the viewing angle @xmath14 has been confirmed to strongly affect the flare gamma - ray spectra ( section [ sec.solarsimulation ] ) . assuming again the vertical field geometry , we can identify @xmath14 with the angle @xmath132 which our line - of - sight to the flare makes relative to the normal to the photosphere . hereafter , we simply call @xmath132 `` flare - viewing angle '' . using the flare longitude @xmath133 and latitude @xmath134 , it can be calculated as @xmath135 to examine the effect of @xmath14 , we plotted in figure [ fig.longitude ] the gamma - ray to hard x - ray hardness ratio of the 40 yohkoh flares , as a function of @xmath132 calculated via equation ( [ eq.theta_v ] ) . it is similar to figure 7 of @xcite , but differs from it in that @xmath132 is used instead of @xmath136 . as predicted by our simulation , the hardness ratio clearly increases with @xmath132 ; this effect was already known as `` gamma - ray limb brightening '' in solar flares using _ data @xcite . @xcite interpreted this effect in terms of the radiation anisotropy : supposing that the electrons have a pitch - angle distribution which increases with angle from the outward normal , then higher - energy continua are expected to be more strongly beamed parallel to the photosphere , because harder bremsstrahlung photons are more strongly forward - beamed than softer ones even when emitted by electrons of the same energy . we presume that this mechanism can explain at least part of the observed limb - hardening effects , while the compton back - scattering process must be enhancing them . in particular , our scenario can explain limb hardening even when the electrons are mostly directed downwards . for a more quantitative comparison of our simulation with the yohkoh observations , let us limit ourselves to limb flares which have absolute longitudes larger than @xmath137 , because gamma - ray data of disk flares are of poor photon statistics than those of limb flares as is clear from figure [ fig.longitude ] , and because these disk flares are more contaminated by nuclear gamma - ray lines @xcite which are unrelated to the accelerated electrons . figure [ fig.color_color2 ] shows a scatter plot between two hardness ratios of the 40 sample flares , one calculated in the hard x - ray range ( 5793 kev vs 3357 kev ) while the other in the gamma - ray range ( 1.436.21 mev vs 0.221.43 mev ) . in this `` color - color '' plot , the dashed line represents the condition that the two colors imply the same photon index @xmath6 , of which values are given in the figure . the hard x - ray ( or gamma - ray ) photon index of a flare can be found by drawing a vertical ( or horizontal ) line from the data point , and read its intersection with the dashed line . from figure [ fig.color_color2 ] , we can see that the gamma - ray spectral index of the 40 flares is distributed in a typical range of 1.7 - 2.5 . this agrees very well with the prediction by our monte - carlo simulations ( figure 13 ; subsection 3.3 ) , on condition that the electrons in these flares have broad pitch - angle distributions beyond @xmath138 . further , in figure [ fig.color_color2 ] , a majority of flares are distributed to the left side of the single power - law locus . such a concave - shaped wide - band spectrum can be naturally explained as a consequence of a broad pitch - angle distribution of electrons , because electrons with small pitch angles are expected to emit steep - spectrum photons , while those with large pitch angles emit much harder spectra . a superposition of these spectral components with different slopes will result naturally in a concave spectrum just as is observed . thus , as long as the magnetic lines are assumed perpendicular to the photosphere and the flare viewing angle @xmath132 is large , flares with significant gamma - ray signals are inferred to have rather broad pitch - angle distributions . as a more general case , we may briefly consider tilted magnetic - field conditions . in this case , it is somewhat difficult to directly compare the results of our simulation with the observational data , because the electron gyration can no longer be faithfully emulated by our method of taking an average over @xmath15 . nevertheless , we may employ the present results to predict some aspects of that case in a qualitative way . let us first consider limb flares as an extreme case . in this case , a field - line tilt on the sky plane is considered to have little effects , because the most efficient gamma - ray emission ( as seen from the observer ) occurs over the gyration phase wherein the electrons are approaching us , and this phase is the same even the filed lines are tilted in that way . in contrast , a field - line tilt along our line - of - site is expected to have a prominent effect : if the field lines are tilted away from us , the observed spectra tend to reach higher energies , e.g. a few mev , even if the electrons have a rather narrow pitch - angle distribution . if the lines are tilted toward us , the gamma - ray spectra are expected to soften , on the contrary . then , the observed scatter in the gamma - ray to hard x - ray intensity ratio for limb flares ( figure [ fig.longitude ] , with @xmath139 ) may be contributed by the field - line tilts , in addition to the pitch - angle distribution . the other extreme case is a disk - centered flare , where the field - line tilt is specified only by its angle @xmath140 with respect to the photosphere normal ( i.e. , our line - of - sight ) without depending on the tilt direction . then , an electron with a fixed pitch angle , @xmath141 is expected to sweep a range of injection angle , from @xmath142 to @xmath143 . therefore , a larger value of @xmath140 is favorable in delivering strong gamma - ray signals to the observer , even though @xmath14 is still limited to the unfavorable value of @xmath144 . if @xmath141 and @xmath140 are both rather large so that @xmath145 , the electron is implied to be approaching us over a certain phase of its gyration , thus allowing a considerable fraction of the bremsstrahlung photons to directly reach us without being compton scattered . therefore , a disk flare may become a strong gamma - ray source under a limited condition , that the electrons have a broad pitch angle and the magnetic field lines are significantly tilted . now that broad pitch - angle distributions of accelerated electrons are considered to be an important element of observing intense gamma - ray emission , it may be imperative to simulate such a case , and compare the results with gamma - ray spectra from actual solar flares . therefore , we repeated our solar simulation , assuming that the electrons have a continuous incident - angle distribution as @xmath146 which corresponds to @xmath147 in @xcite except for covering only the @xmath148 range . when the magnetic fields are vertical , equation ( [ eq.dermerramaty1986 ] ) becomes identical to the pitch - angle distribution of the accelerated electrons . under this angular distribution , we injected @xmath65 electrons in the same way as in section [ sec.solarsimulation ] , assuming the incident electron spectrum to have a canonical index of @xmath149 , and to extend again over 1100 mev . collecting the produced gamma - rays over a viewing angle range of @xmath150 , and convolving the results with an approximate response of the yohkoh grs @xcite , we have obtained the spectrum shown in figure [ fig.grs_sim_comp ] in red . to compare with this simulation , we have chosen a goes x3.7 class flare of 1998 november 22 , 06h37 m ut , which is one of the gamma - ray brightest events among the 40 yohkoh flares @xcite . since this is an extreme limb flare located at s32w90 , our viewing angle @xmath14 should be close to @xmath151 , in agreement with the simulation condition . actually , the hxt highest - band images of this flare reveal intense hard x - ray emission from a pair of magnetic loop foot points , both located right on the solar west limb @xcite . the grs data , accumulated over a period of 06:37:2206:40:22 , are shown in figure [ fig.grs_sim_comp ] in blue , after subtracting background ( accumulated for 06:56:1406:59:02 ) , but without removing the detector response . thus , the spectrum is featureless , and the 2.23 mev neutron capture line is undetectable with its equivalent width being @xmath152 kev @xcite . therefore , this flare is considered to be a typical `` electron dominated '' event . in figure [ fig.grs_sim_comp ] , the simulation and the actual data agree reasonably well , at least over the 110 mev band . when the grs response is considered , these spectra can be approximated both by @xmath153 @xcite , which is typical in figure [ fig.color_color2 ] . incidentally , we would obtain from figure [ fig.deltagamma ] @xmath154 for the assumed @xmath149 , if we were observing direct thick - target emission . we should instead find @xmath155 ( for @xmath117 ) to @xmath156 ( for @xmath116 ) , if the compton back - scattering dominates . therefore , the values of @xmath157 found in figure [ fig.grs_sim_comp ] are in between these two cases . indeed , an inspection of the monte - carlo photons revealed that @xmath158 of them are compton - scattered gamma - rays , while the remaining @xmath159 are direct bremsstrahlung events . given the good agreement between the two data sets , it is possible that the actual gamma - rays from the november 22 flare are also a mixture of the direct and scattered photons . in order to quantitatively estimate the spectra of energetic electrons generated in solar flares based on the observed gamma - ray spectra , we numerically studied the electron transport and gamma - ray emission in the solar atmosphere . as elementary processes , we mainly considered coulomb scattering , bremsstrahlung and compton scattering . we modeled the solar atmosphere with a vertical stack of parallel zones with different densities , and neglected the magnetic fields . as a result of our simulation , we have found that the thick - target gamma - ray spectra emitted from magnetic loop foot - point can hardly reach mev energies , because of heavy compton degradation , unless the electrons make rather grazing angles to the photosphere and we are observing forward - scattered gamma - rays . when this compton effect is taken into account , the traditional @xmath4 vs. @xmath4 relation is drastically modified . the pitch - angle distribution of flare - accelerated electrons are suggested to be broad and extend to large angles for those flares of which the emitted photon spectra reach mev energies . actually , a gamma - ray spectrum simulated under electron injection with a broad pitch angle distribution reproduced reasonably well the yohkoh spectrum of the 1998 november 22 flare . 1 . when the magnetic field lines are perpendicular to the photosphere , and the electrons have a narrow pitch - angle distribution around 0 , the gamma - ray spectra are very soft and can not reach mev energies . 2 . even in the same field configuration , the gamma - ray spectra become harder if the electrons have larger pitch angles . therefore , the electrons producing gamma - rays are inferred to have relatively large pitch angles . limb flares are expected to have harder spectra than disk flares , because of their smaller angles of compton scattering . this agrees with the observed fact that the gamma - ray to hard x - ray intensity ratio increases as the flare longitude gets closer to the solar limb . 4 . in case of a broad pitch angle distribution , electrons with small pitch angles emit steep hard x - rays , while those with large angles emit photons with flatter spectra . this can explain another yohkoh result , that the gamma - ray slope tends to be flatter than the contemporaneous hard x - ray slope . 5 . under the conditions of @xmath160 and @xmath161 , the values of @xmath6 obtained in our simulations in the 110 mev range are concentrated over 1.72.5 . this agrees again with the yohkoh observations . a disk flare may become a strong gamma - ray source under a limited condition , that the electrons have a broad pitch - angle and the magnetic field lines are significantly tilted . the gamma - ray spectra of actual solar flares are expected to be an appropriate mixture of the direct and compton - scattered photons . , s. , allison , j. , amako , k. , apostolakis , j. , araujo , h. , arce , p. , asai , m. , axen , d. , banerjee , s. , & barrand , g. 2003 , nuclear instruments and methods in physics research section a : accelerators , spectrometers , detectors and associated equipment , 506 , 250 , r. p. , dennis , b. , hurford , g. , smith , d. m. , & zehnder , a. 2004 , in telescopes and instrumentation for solar astrophysics . edited by fineschi , silvano ; gummin , mark a. proceedings of the spie , volume 5171 , pp . 38 - 52 ( 2004 ) . , 3852

there is a strange omission in the traditional version of the second law of thermodynamics ( slt ) . it does not take gravity into account . however the existence of the arrow of time is usually explained with reference to slt . and most discussions of the origin of the arrow of time appeal ultimately to the initial condition and the evolution of the universe . gravity plays an essential role for this evolution . as pointed out by p.c.w . davies @xcite there seems to be a paradox that the material contents of the universe began in a condition of thermodynamic equilibrium , whereas the universe today is far from equilibrium . hence the thermodynamic entropy has been reduced in conflict with slt which says that the entropy of the universe is an increasing function of cosmic time . this is due to the tendency of self - gravitating systems irreversibly to grow inhomogeneous . in order to include this gravitational effect into a generalized version of slt , one has to define a gravitational entropy . several tentative definitions have been given based on penrose s weyl curvature hypothesis @xcite . the weyl curvature tensor vanishes identically in the homogeneous and isotropic friedmann - robertson - walker universe models . it was suggested by penrose @xcite to use this tensor as a measure of inhomogeneities of the universe models . the weyl curvature scalar is expressed by scalars constructed from the riemann curvature tensor as follows c^c_=r^r_-2r^r_+13 r^2 penrose has given several formulations of the conjecture . a version that emphasizes the entropy aspect was given in ref . @xcite p.178 . it seems that in some way the weyl tensor gives a measure of the entropy in the space- time geometry . the initial curvature singularity would then be one with large ricci tensor and vanishing weyl tensor ( zero entropy in the geometry ) ; the final curvature singularity would have weyl tensor much larger than ricci tensor ( large entropy in the geometry ) . in order to quantify penrose s conjecture wainwright and collaborators @xcite have suggested that the quantity p^2= [ defp]may represent a `` gravitational entropy '' , at least in a cosmological context with a non - vanishing ricci - tensor . we could of course ask ourself whether the entropic behavior of the weyl curvature invariant ( or an invariant composed thereof ) results because the weyl tensor is _ directly related _ to the gravitational entropy , or whether it is only a side - effect because the universe evolves towards a state of maximal entropy . hence , even though the results may show that the weyl curvature invariant has an entropic behaviour , it is by no means a proof that the weyl tensor should be identified as the `` gravitational entropy '' . in an earlier article @xcite we discussed the weyl curvature tensor in a homogeneous but anisotropic model as well as in an inhomogeneous model . our survey will be summarized in this and the next sections , as well as bringing new arguments into the discussion . as a basis for our study , we used two different models which induce weyl curvature effects in two conceptually different ways . one was the anisotropic bianchi type i model , the other was the inhomogeneous lematre - tolman model . the weyl curvature conjecture has been investigated in the szekeres cosmological model , that generalize the lematre - tolman model , by w.b . bonnor @xcite . inhomogeneous modes are in general local modes for general relativity while anisotropic modes are global modes . the global topology and geometry of our universe has significant consequences for the possibilities for anisotropic modes @xcite . the role of inhomogeneous modes can be considered as more local in origin , but the effect on the global geometry and topology are not known in detail . hence , it is important to study both an inhomogeneous model as well as a homogeneous model to get a more complete picture of the behaviour of the weyl tensor in generic cosmological models . the bianchi type i model is the simplest anisotropic generalisation of the frw models . it has flat spatial sections , and hence , can be a good candidate for the universe we live in . the metric for this model can be written as ds^2=-dt^2+e^2_ijdx^idx^j where @xmath0 . in @xcite we solved the einstein field equations for dust and a cosmological constant @xmath1 . by introducing the volume element @xmath2 , the einstein field equations gives the following equation for @xmath3 : ^2=3v^2 + 3mv+a^2 where @xmath4 is the total mass of the dust , and @xmath5 is an anisotropy parameter . the equations for @xmath6 are in terms of the volume element _ = where the constants @xmath7 are related to @xmath5 via @xmath8 . it is useful to define an angular variable @xmath9 by @xmath10 and @xmath11 . an important special case of the bianchi type i is the kasner vacuum solutions . the kasner solutions are characterised only by the angular variable @xmath9 . these solutions have a weyl scalar ( c^c_)_i= ( 1 - 2z3+z^2 ) [ weylbti]where @xmath12 and @xmath13 for the kasner solutions which have vanishing ricci tensor . so in a sense , the kasner solutions are the anisotropic counterpart to the inhomogeneous schwarzschild solution . the weyl tensor will decay as the volume expands , even at late times . one should expect that this entity would increase if it represents gravitational entropy . but is decreases monotonically , and hence , it is doubtful that it is the correct measure in these models . inserting dust and a cosmological constant will not have a significant effect on this decreasing behaviour . if @xmath4 is the mass of the dust inside the volume @xmath3 , then eq . ( [ weylbti ] ) still holds but with z=. hence , except for the special case is somewhat interesting . for @xmath14 it can be shown that the kasner solution is just a special part of minkowski spacetime , which has no singularities . it seems a bit odd that for any @xmath15 we will have an initial singularity , while for @xmath14 we do not have one . however , if we compactify the spatial sections this oddity disappears . the model has a singularity at @xmath16 even for the case @xmath14 ! ] @xmath14 , the weyl tensor diverges as @xmath17 as @xmath18 . for the bianchi type i universe models the entity @xmath19 defined in eq . ( [ defp ] ) , turns out to be ( p^2)_i = . also this measure for the weyl entropy diverges as @xmath20 . let us now consider the inhomogeneous lematre - tolman ( lt ) models . the line element for the lt models can be written as [ ltmetric ] ds^2=-dt^2+q^2dr^2+r^2(d^2+^2d^2 ) where @xmath21 and @xmath22 . in @xcite we used this model and investigated the solutions of the einstein field equations where the spacetime contains a cosmological constant @xmath1 and dust . the equations then turn into @xmath23 where @xmath24 is an arbitrary function , and [ meq ] r^2+(1 -f^2)r - r^3=m . [ masseq]the function @xmath25 is given by the integral m(r)=^r_0 4r^2 rdr , where prime denotes derivative with respect to @xmath26 and @xmath27 is the dust density . hence , @xmath28 can be interpreted as the total mass of the dust inside the spherical shell of coordinate radius @xmath26 . inverting the equation for @xmath28 , the dust density can be written in terms of @xmath29 and @xmath28 : 4= [ rhoeq ] it is also useful to define the _ mean dust density function _ @xmath30 by the relation m(r)= |r^3 . [ rhobareq]interestingly , the weyl curvature scalar can now be written quite elegantly as ( c^c_)_lt= ^2(|-)^2 . [ rhoweyl]this relation provides us with a physical interpretation of the weyl scalar in the lt models . it is just the difference between the mean dust density and the actual dust density . also , the weyl tensor is everywhere zero , if and only if @xmath31 . in this case the lt models turn into the frw models with homogeneous spatial sections . the frw models are conformally flat , hence they have a zero weyl tensor . let us return to the equations of motion for the lt model . the mass equation ( [ masseq ] ) can be written as an `` energy '' equation : ^2+v(r , r)=e(r ) [ energyeq]where the `` potential '' @xmath32 and the `` energy '' @xmath33 are given by v&= & --r^2 + e&=&-(1-f^2 ) this energy equation may be integrated and solved exactly . a summary of the results and a qualitative description of the physical meaning of the solutions is given in @xcite . the solutions may be written in terms of the weierstrass elliptic functions @xcite . the actual expressions are not very informative unless the reader has massive knowledge of these elliptic functions . however a lot of qualitative information can be extracted from simple classical considerations . the classical solutions will move on level curves of the energy function @xmath34 , since the total energy @xmath33 is independent of @xmath35 . in figure [ levelcurves ] the level curves of a typical energy function are drawn . if @xmath36 there will exist a saddle - point of the energy function . this saddle point will be at @xmath37 where the energy function will have the value @xmath38 . the saddle point solution is static , and is as a matter of fact the einstein static universe . this can be done since we have to assume that @xmath39 on physical grounds . in the case @xmath40 , the metric become degenerate . ] . if @xmath41 , the solutions fall into two distinct classes : 1 . schwarzschild - like solutions : these solutions expands , but they do not possess enough energy to escape the gravitational collapse , so they end as black holes . if @xmath42 is constant these solutions are those of a schwarzschild black hole in lematre coordinates . de sitter - like solutions : solutions where the universe evolves approximately as that of de sitter solutions with positively curved hypersurfaces . if @xmath43 the ( test ) matter has enough energy to escape the gravitational collapse ( expanding solutions ) or enough energy to prevent the gravitational repulsion from the cosmological constant ( contracting solutions ) . near the initial singularity , both @xmath44 and @xmath27 diverge . unless they are identically equal , the weyl tensor will diverge near the initial singularity . explicitly we have near the initial singularity 4(|-)= where we have used that near the initial singularity , we can approximate the solutions with @xmath45 the free function @xmath46 is the big bang time . to avoid intersecting world - lines we have to assume @xmath47@xcite , thus @xmath48 . the weyl scalar will diverge as @xmath49 unless @xmath50 . also the entity @xmath51 will diverge in general near the initial singularity . @xmath51 is given by @xmath52 which near the initial singularity can be approximated by ( p^2)_lt()^2 . hence , @xmath51 diverges as @xmath53 unless @xmath50 , i.e. unless the big bang is homogeneous . this does not prove to be a very promising behaviour for the wcc . the curvature scalars diverge near the initial singularity , and hence , come in conflict with the wcc . but let us analyse the situation in the lt a bit more carefully . as the universe expands , both @xmath44 and @xmath27 will decrease . they both decrease from an infinite value at the initial singularity . as the universe expands the value of @xmath44 decrease as @xmath54 while @xmath27 decreases as @xmath55 close to the initial singularity . hence , the value of @xmath44 approaches the value of @xmath27 by a factor or @xmath55 , but since they both diverge , the weyl tensor and @xmath51 diverge as well . however , even though r. penrose and s. hawking @xcite showed that according to the general theory of relativity the big bang must have started in a singularity , it has later been emphasized that the physical universe must obey not only the general relativistic laws of nature , but also the quantum mechanical laws . hence , the initial singularity is a fiction which not corresponds to physical reality . the classical laws can be applied only after the planck time . we should therefore study the behaviour of @xmath19 not only in the limit @xmath56 , but rather at a very small cosmic time . choosing the origin of cosmic time @xmath57 at @xmath58 we introduce @xmath59 . hence close to @xmath60 @xmath61_{t\lra 0}<0 $ ] . bonnor @xcite considered the model with @xmath62 and a vanishing cosmological constant , and found the opposite result . however he restricted his investigation to models with homogeneous initial singularity , i.e. @xmath63 . as we will show this is an exceptional case . the behaviour of more general models with @xmath64 is different . to see what is actually happening if we push @xmath65 towards zero , we utilize that close to @xmath60 the models with @xmath62 and @xmath66 behave according to r^3=92 mt^2(1-(r)t^23 ) [ smallr3]where @xmath67 is assumed to be a small function of @xmath26 . from eqs . ( [ rhoeq ] ) , ( [ rhobareq ] ) and ( [ rhoweyl ] ) with @xmath66 we get |p|=(3 - 1 ) . using ( [ smallr3 ] ) we obtain |p|=. this function is plotted in fig . [ pperturb ] . note that this function diverges for @xmath68 , obtains a minimum for some small @xmath57 and increases thereafter . as @xmath69 is driven towards zero this minimum goes towards @xmath60 . let us consider a finite 3-volume @xmath32 comoving with the cosmic gas in our space time . according to the first law of thermodynamics the matter entropy @xmath70 and the internal energy @xmath71 will evolve as : tds_m = du+pdv . if the matter content is dust then @xmath72 . there are a couple of things to note . firstly , in a dense dust cloud , we expect the internal energy of the dust to be large , thus we expect the entropy to be large . secondly , the entropy is an increasing function of the volume . let us therefore consider a co - moving volume @xmath32 in our spacetime . what could the expression for a _ gravitational entropy _ be ? from the previous sections we noticed that an lt model with @xmath31 is homogeneous . the quantity @xmath73 is an inhomogeneity measure in the lt models . in the absence of a cosmological constant we notice that p= ( ) . [ eq : pdiff]the sign of @xmath51 is here chosen so that the configuration @xmath74 is associated with @xmath75 while the more realistic configuration @xmath76 has @xmath77 corresponding to eq . ( [ eq : pdiff ] ) positive . let us consider the entity defined by : = _ v p dv . introducing co - moving coordinates @xmath78 we write @xmath79 where @xmath80 is the 3-volume element . the integration range is now constant as a function of time and if we integrate over a unit coordinate volume which is so small that the integrand is approximately constant , we may write = _ v p dvp [ defs ] this formulation of the gravitational entropy is an intuitive one , but it is not a covariant formulation . to find a covariant formulation we define the entropy current vector by @xcite = s*u*+ where @xmath81 and @xmath82 are orthogonal , @xmath83 is the entropy density , @xmath81 is the material flow vector ( @xmath84 ) and @xmath82 is the entropy flux . the second law of thermodynamics can now be expressed as _ ; 0 . writing this in a local coordinate system , the divergence is _ ; = _ ( ^ ) . in our case we take the entropy density proportional to @xmath51 and assume a vanishing entropy flux . hence , p*u*. in comoving coordinates @xmath85 so that _ ; _ t(p)=. for the `` second law to hold '' , we have to check whether the entity @xmath86 is increasing . if we take the corresponding one - form of the entropy vector , and take the dual of this form ( hodge dual ) , we obtain the entropy three - form which is given by the contraction of the space - time volume form @xmath87 with the entropy current current vector = i_. in our case , this three - form is simply = p*dx * hence , if the coordinates are comoving , then it is the component of this three - form that is increasing as long at the matter and fields obeys the sec . this explains why the entropy should scale as the volume . it is usually this entropy three - form we are thinking of . it has a more intuitive behaviour than the entropy current vector . we will therefore use @xmath88 in the further to study the wcc . the square of the weyl tensor and the entity @xmath51 do not , as we have seen , capture the entropic behaviour properly . they both diverge at the initial singularity and thus , can not be a proper measure of the gravitational entropy . the entity @xmath88 on the other hand , seems to be a more promising candidate for the weyl entropy . let us start with the lt model . the motivation for studying the entity @xmath88 given in ( [ defs ] ) came from the lt models . therefore it is natural to start with this model . in all the lt models , we can approximate the behaviour near the initial singularity , @xmath49 with s_lt=2(m(t - t_0)-2t_0 ) . [ eqsmallt ] for physically realistic spacetimes , @xmath89 , so as @xmath90 , @xmath88 is positive and finite . note also that @xmath88 is increasing , as suggested by the wcc . in the absence of a cosmological constant and if @xmath91 , this will be the exact expression for @xmath88 . if a cosmological constant is present , the universe will eventually go into a de sitter phase if the universe is allowed to expand for ever . for the sake of illustration , it is useful to investigate a specific case . let us choose @xmath92 . the energy equation eq . ( [ energyeq ] ) , can now be solved to yield r=()^13^ 23[eqexactlt]where @xmath93 . in general , the late time behaviour of the lt model ( if the universe has one ) is that of a de sitter universe . this exact solution provides us with a solution which connects the initial singularity with the late time de sitter era . the initial behaviour of @xmath94 is the same as in eq . ( [ smallimit ] ) . the late time behaviour of @xmath29 is r=()^13e^h(t - t_0(r ) ) . hence , at late times the entity @xmath88 is approximately constant : s=. [ eq : latetimedesitter]at late times , @xmath88 approaches a constant value which is inversely proportional to @xmath95 . the larger the value of the cosmological constant is , the smaller the value of the final value of @xmath88 . in fig . [ fig : sevol ] we have plotted , using the exact solution ( [ eqexactlt ] ) , the evolution of @xmath88 for three different values of @xmath1 . the volume under consideration expands at late times exponentially as @xmath96 , where @xmath95 is the hubble parameter . hence , the entropy per unit volume will decrease exponentially : e^-3ht . the entropy in a unit volume will therefore decrease rapidly during a de sitter stage . this is a very important consequence of inflation . at the exit of the inflationary period , the entropy in a unit volume is very low , there are very little inhomogeneities left of the primordial ones . let us summarize the evolution of the entity @xmath88 in the lt models : 1 . * large @xmath1 and ever expanding : * in the initial epoch the dust dominates and @xmath88 is increasing linearly in @xmath35 . the universe is becoming more and more inhomogeneous . after the universe has grown considerably , the cosmological constant becomes dominant , @xmath88 stops growing and if the cosmological constant is large enough , it evolves asymptotically towards a constant value . the universe is smoothened out . this scenario corresponds approximately to the lowest two graphs in fig . [ fig : sevol ] . * small @xmath1 and ever expanding : * again the dust dominates initially . the cosmological constant is too small to make @xmath88 decreasing . the entity @xmath88 is ever increasing but is bounded from above by a relatively large constant value . * zero @xmath1 and ever expanding : * the @xmath88 will again be ever increasing and will asymptotically approach a function @xmath97 where @xmath98 and @xmath99 are constants and @xmath100 iff @xmath101 and @xmath102 iff @xmath92 . * recollapsing universe : * due to the dust term , the final singularity will not be similar to the initial singularity . hence this entity is asymmetric in time for a recollapsing universe . in the lt models we see that @xmath88 behaves in agreement with the wcc . let us investigate more carefully the schwarzschild spacetime , where the whole motivation of gravitational entropy comes from @xcite . the schwarzschild spacetime is a special case of the general lt models ; it has @xmath103 and a vanishing ricci tensor . if we look at the entity @xmath88 in the region outside the schwarzschild singularity , @xmath88 will diverge . this is in some sense the maximal possible value of @xmath88 , the weyl tensor is as large as possible and the ricci tensor is the smallest as possible . thus at this classical level it seems that the schwarzschild spacetime has the largest possible @xmath88 which is a good thing if one wants to connect @xmath88 with the entropy of the gravitational field . outside a real black hole , @xmath104 is probably impossible . even though the classical vacuum has vanishing ricci tensor , the quantum vacuum will probably not have @xmath104 the quantum fields will fluctuate and cause the expectation value of the square of the ricci tensor to be non - zero : @xmath105 . hence , there will probably be an upper bound of how large @xmath88 can be , even in a vacuum . in the bianchi type i with dust , the explicit expression for the entity @xmath88 is s_i = p=()^ where @xmath106 . in a neighbourhood of the singularity we can make a taylor expansion to first order in @xmath3 : @xmath107 we interpret @xmath108 as the dust density in coordinate space @xcite . hence , as we go towards the initial singularity @xmath88 will be finite . for the vacuum case , @xmath109 and @xmath66 , and the solutions are the kasner solutions . the kasner solutions have a vanishing ricci tensor , but a non - zero weyl tensor . hence , as in the schwarzschild case , the entity @xmath88 will diverge . from the taylor expansion we see that , if @xmath110 then @xmath88 will increase immediately after the initial singularity . if @xmath111 then @xmath88 will decrease . this can be understood as follows . if the cosmological constant is to large , the universe will increase too rapidly initially to allow the dust to contribute significantly to the anisotropy . initially the universe will become more and more isotropic . in the presence of a cosmological constant , the universe will eventually enter a de sitter phase . again @xmath88 will evolve towards a constant value given by @xmath112 if the cosmological constant vanishes @xmath88 is monotonically increasing : @xmath113 note that for large @xmath35 we have @xmath114 , which is according to the same power law as the @xmath92 and @xmath66 case of the lt model ( compare with eq . ( [ eqsmallt ] ) ) . to summarize our investigation of @xmath88 in the bianchi type i model we can say that the entity @xmath88 at the initial singularity is a constant determined by the inverse of the dust density . for @xmath110 it will increase immediately after the initial singularity . in most cosmological considerations it is assumed that the cosmological constant is small . the exception is in the inflationary era in which the vacuum energy dominates over all other matter degrees of freedom . the inflationary era will smooth out anisotropies as well as inhomogeneities , and the behavior of @xmath88 in this case is therefore expected . in the de sitter limit @xmath88 will asymptotically move towards a constant . large @xmath1 means small value , while small @xmath1 corresponds to a large value . this is in full agreement with the lt models . it is also interesting that the entity @xmath88 is very sensitive to different matter configurations . this makes it a lot easier to check whether the @xmath88 has the right entropic behaviour . the question now arises : how generic is this behaviour ? does the entity behave in the correct way for all physically realistic models ? we know that our universe today is close to homogeneous on a scale larger than a billion light years . as mentioned earlier , the frw models are conformally flat and hence , they have a zero weyl tensor . so a question would be , at late times when the universe is close to isotropic and homogeneous , does the entity @xmath88 still behave in the correct manner ? does it still increase ? we saw that at late times , both the lt model and the bianchi type i model , did behave correctly for @xmath66 even though they both isotropise and evolve towards homogeneity . let us choose a more general model , a flat model that allows for inhomogeneities and anisotropies . we will assume that the model at late times asymptotically evolve towards a frw model . in @xcite , barrow and maartens investigated the general equations of motion for such models with @xmath66 . in the velocity dependent regime they derived some approximate solutions to the field equation close to a frw model , having both inhomogeneities and anisotropy . also an anisotropic stress of the form @xmath115 where @xmath116 is a constant matrix , was included . their result was that the late time behaviour of such a model is consistent with a perfect fluid where the equation of state parameter @xmath9 for the perfect fluid obeys @xmath117 . for @xmath118 the frw model is unstable at late times . in the latter case one can get for instance different stable anisotropic magnetic solutions @xcite . let us therefore assume that @xmath119 ( so that the matter obeys the sec ) . the ratio of the weyl tensor squared and the ricci tensor squared is in this case p^2=o(t^-2n)+o(t^-3n)+o(t^-4n ) where @xmath120 for @xmath121 . in the radiation case , @xmath122 , we get a logarithmic decay of @xmath19 : p^2_=4/3 . thus in the range of validity of the assumption @xmath123 , this entity will decrease in the future . however , the entity @xmath88 shows more promising behaviour s = p = a^3 pt^-n [ eq : bmsol]for @xmath124 , while the @xmath122 case yields s(t)^-1t^32 . hence , @xmath88 increases as long as 23 . this entity increases in the future ( as any entity describing entropy should do ) and tells us that even if the universe itself asymptotically goes towards isotropy , the entropy of the gravitational field actually increases ( if we should believe the wcc ) . note that this result is slightly different than the case where no anisotropic stress is present . if the anisotropic stress is not present , we get @xmath125 at late times for all @xmath9 . we should also mention a work done by hervik @xcite which investigates the evolution of the weyl curvature invariant for generic solutions of spatially homogeneous models containing a @xmath9-law perfect fluid with @xmath126 . the conclusion was that all spatially homogeneous models , except for sets of measure zero , had an increasing @xmath88 at late times . as long as the matter obeys the strong energy condition ( sec ) , matter will behave more or less attractive . hence , when the sec is fulfilled , we should expect the gravitational entropy to increase . during inflation the sec is violated and gravity is not necessarily attractive . we have already seen how the inclusion of a cosmological constant could alter the behaviour of @xmath88 . the late time behaviour of @xmath88 in the presence of a cosmological constant is that @xmath88 evolves approximately as a constant . the constant itself is a decreasing function of @xmath1 ( see eq . ( [ eq : latetimedesitter ] ) ) . what happens for a more general inflationary fluid ? consider a perfect fluid with equation of state p=(-1)for @xmath127 . this perfect fluid will violate the strong energy condition and will in general cause a power law inflation . note that the case @xmath14 can be considered the same as including a cosmological constant . in the frw cases , all the models will now have a late time behaviour similar to the flat case . the frw models will become inflationary and will be dominated by this fluid at some stage in the future ( provided that the universe is ever - expanding ) . to simplify , we will therefore consider a flat universe , and perturb the frw flat universe model with @xmath128 . following @xcite , the anisotropic stresses will not dominate the shear modes at late times . the shear will under these assumptions decrease as _ t^- while the hubble parameter is the same as in the frw case ( to lowest order ) h = t^-1 . the late time behaviour of @xmath19 is now p^2= o(^2 t^- ) which is decreasing for all @xmath129 . the late time behaviour of @xmath88 is s = p = o(t ) which is increasing linearly in @xmath35 at late times . for @xmath14 we recover the cosmological constant case where @xmath88 is approximately a constant at late times . note also that this result coincide with eq . ( [ eq : bmsol ] ) for @xmath130 . hence , during the inflationary period , @xmath88 increases much slower than for ordinary matter ( @xmath131 for dust and @xmath132 for radiation ) . maybe the closest to a quantum theory of gravitation that have been obtained , is what we call quantum cosmology ( qc ) . we will in this section show how one might be able to determine the likelihood of a certain initial state to occur . qc is perhaps best described as a `` theory of initial conditions '' . we will again use the results of the previous paper to try determine which of the initial states are more probable . in qc , the wave function of the universe satisfies the _ wheeler - dewitt ( wd ) equation _ : ( -g_ijkl+ v(h_ij))=0 where @xmath133 is called dewitt s supermetric , and @xmath134 is the metric on the 3-dimensional spatial hypersurfaces . the potential term @xmath135 consists of the ricci scalar of the three - dimensional hypersurfaces and possibly matter potentials and a cosmological constant . we want now to calculate the expectation value for the weyl scalar , in these models . all of the curvature invariants have to go over to their respective curvature operators . especially , the entity @xmath88 goes over to the curvature operator @xmath136 . the question is now , what is more likely , creation of a universe with large expectation value of @xmath136 or a universe with a low value ? in the paper @xcite we considered semi - classical tunneling wave functions which are solutions of the wd equation . the universe was tunneling from a matter - dominated universe classically confined to a finite size , into a de sitter like universe . after tunneling across the classically forbidden region the universe became @xmath1-dominated , similarly to an inflationary model of the universe . thus we have a `` flow '' from matter - dominated universes towards de sitter like universes . the actual expectation values of entities like @xmath137 , @xmath19 and @xmath88 are not explicitly obtained for these models , because the actual calculations suffer from `` endless '' expressions and highly time - consuming quantities . we will therefore give a more general description of the evolution of the weyl tensor for the lt models . it would be useful first to recapitulate some of the discussion done in @xcite . first of all we discussed tunneling wave functions in the wkb approximation . in the wkb approximation we assume that the wave function has the form @xmath138 , where @xmath88 will to the lowest order satisfy the hamilton - jacobi equation : @xmath139=0\ ] ] here @xmath140 denotes the functional derivative with respect to the function @xmath29 . if we assume that @xmath141 , the resulting equation will be the hamilton - jacobi equation for a point particle with action @xmath142 ( @xmath26 is only a parameter and the functional derivatives turn into ordinary partial derivatives ) . in the hamilton - jacobi equation the functional @xmath143 turns out to be the action at the classical level . since the classical action can be written as an integral over @xmath26 the assumption @xmath141 is therefore reasonable at the lowest order wkb level . we can interpret the action @xmath144 as the action of a point particle moving in a potential @xmath145 $ ] with zero energy . the wkb wave function @xmath146 for the point particle can then be written @xmath147 . the two wkb wave functions can therefore be related by @xmath148 . finding first the wave function @xmath149 we can then relate its wkb approximation to @xmath150 through @xmath148 . let us now ask the question : _ is it more likely for a universe with small weyl tensor to tunnel through the classical barrier than a universe with a large weyl tensor ? _ the question is difficult to answer in general but we shall make some simple considerations in order to shed some light upon it . we assume that the dust density near the origin of the coordinates is larger than further out . we define the homogeneous mass function for a closed universe ( @xmath151 ) as @xmath152 where @xmath153 is a constant . the constant @xmath153 is determined by demanding @xmath154 . if the dust density is larger near the origin of the @xmath26-coordinate than for larger values of @xmath26 then @xmath155 . this will not in general change the size of the universe so we can look at the effects from @xmath28 alone . since @xmath28 is greater in general for an inhomogeneous universe than for a homogeneous universe , we see that the potential barrier will be smaller for an inhomogeneous universe than for a homogeneous universe . thus an inhomogeneous universe will tunnel more easily through the classical barrier than the homogeneous universe . since an inhomogeneous universe will have a larger weyl tensor than an almost homogeneous one , we see that universes with large weyl tensor tunnel more easily than those with a small weyl tensor . if we look at the tunneling amplitude concerning effects from the @xmath1 term , it is evident that larger @xmath1 will yield a larger tunneling probability . in the initial era inhomogeneities will increase the value of @xmath88 . we saw that an inhomogeneous state will tunnel more easily through the potential barrier than a homogeneous state . the largest tunneling probability amplitude thus occurs for universes with a large cosmological constant and large local inhomogeneities . from a classical point of view the value of @xmath88 initially was large ( but increasing thereafter ) , but as the universe entered the inflationary era the cosmological constant had a large value , hence the value of @xmath88 at the end of the inflationary era was relatively small . in the initial epoch the universe is not believed to be dust dominated . the dust does not exert any pressure and dust particles do therefore not interact with each other . a more probable matter content is matter which has internal pressure . even though gravitation tends to make the space inhomogeneous , internal pressure from the matter will try to homogenise the space . since dust is the only matter source in our model , the model only indicates the tendency for gravitation itself to create inhomogeneities . since the universes tunnel into a de sitter - like state , the cosmological constant will rapidly dominate the evolution . the larger the cosmological constant the lower will the entity @xmath88 be after the inflationary era ends . we can write the general solution of the wd - equation for the bianchi type i models as : @xmath156\ ] ] where @xmath157 is a particular solution of the wd equation ( with dust ) , @xmath158 is a `` normalizing constant '' , and @xmath159 is a distribution function in momentum space . this distribution function satisfies the equation : @xmath160 the function @xmath161 is the _ whittaker function_. there is still a factoring problem in turning the classical entities to operators . let us first investigate the expectation value of @xmath137 . in the expression for the weyl tensor squared , there is a term ( p_+^2+p_-^2)^2 where @xmath162 is the conjugated momenta to @xmath6 . upon quantization , the above term is replaced with the operator : ( p_+^2+p_-^2)^2(+ ) ^2 in the expectation value of the square of the weyl tensor , the above term will contribute with @xmath163 ; @xmath164 , the contribution from this term to the weyl curvature invariant will diverge as @xmath17 for small @xmath3 . thus , we have to conclude that , in the small @xmath3 limit the expectation value of @xmath165 goes as : @xmath166 just as in the classical case . investigating the invariant @xmath167 , we notice that things are not so easy . the ricci square also has a term which presumably would contribute with a @xmath168 term . however , looking at the classical expression we see that the ricci square is independent of the anisotropy parameter @xmath5 . this indicates that at the classical level all terms involving the anisotropy parameters , have to cancel exactly . this is not the case quantum mechanically . in the quantum case operators do not necessarily commute . hence there may be contributions from terms which classically would cancel each other . in other words , the fact that the classical vacuum has @xmath104 , does not mean that the quantum vacuum has @xmath169 . we assume that @xmath170 is a set of factor - ordering parameters which represents the `` true '' quantum mechanical system in such a way that @xmath171 represents the classical system . with this parameterization of the factor ordering we would expect the ricci square expectation value to be : @xmath172 where @xmath173 is some function of @xmath3 which has the property : @xmath174 . thus for small @xmath3 and @xmath175 the expectation value would behave as @xmath176 and @xmath177 the weyl square divided by the ricci square is in general finite as @xmath178 for a quantum system . in some sense , the quantum mechanical effects renormalizes the infinity that the classical system possesses at @xmath16 . the expectation value at @xmath16 is , however , strongly dependent on the factor - ordering . as the factor ordering parameters approach zero , the value will diverge . quantum effects in the early epoch are essential for the behaviour of this entity near the initial singularity . as @xmath178 we expect the quantum effect to be considerable , thus expecting the factor - ordering parameters @xmath175 to be large . as indicated in the above discussion , the quantum mechanical expectation value of @xmath179 will be lower and presumably finite at the initial singularity . therefore the expectation value of @xmath180 is also presumed to be considerably lower in the initial stages than its classical counterpart . comparing different tunnelling amplitudes in the bianchi type i model is difficult and more speculative because the bianchi type i universe has no classically forbidden region for @xmath181 . this causes the lowest order wkb approximation to be purely oscillatory . the lowest order wkb wave function will therefore be approximately constant . in the paper @xcite we did however construct under some assumptions a wave function which clearly peaked at small values of the anisotropy parameter . thus these wave functions predicts universes that have a relatively low value of @xmath88 . let us now recapitulate how the evolution in the context of the wcc _ might _ have been . the universe was created in a rather arbitrary state . as the time ticked past the planck time @xmath182 s a rather inhomogeneous universe appeared . qc suggests that this state was rather inhomogeneous , but it is more or less a guess how inhomogeneous the universe was at that state . nevertheless , as the universe grew larger , the expectation value of the weyl entropy increased initially . at what rate the weyl entropy increased is highly uncertain , it depends very much on the true nature of our universe . it depends on what matter fields that were present , the topology of the universe , whether it was anisotropic or not , quantum effects etc . nevertheless , at some time very short after the big bang , an enormous effective cosmological constant appeared . the universe was driven unconditionally into an inflationary period . the weyl entropy stopped increasing and began instead to evolve asymptotically towards a constant . whether or not the weyl entropy had earlier a higher value than at the exit of the inflationary regime , is difficult to say . during the inflationary period , the universe was more or less in an adiabatic expanding state , the weyl entropy was more or less constant . since the universe was expanding exponentially during this period , the entropy per unit volume dropped exponentially . if the scale factor increased by a factor of 60 @xmath183-foldings during the inflationary epoch , then the entropy per unit volume of space would have decreased by a factor of ~e^-36010 ^ -78 . this is quite a drastic decrease , and shows how powerful inflation is when it comes to smoothing out the inhomogeneities and anisotropies of our universe . at the exit of the inflationary regime , the weyl entropy had dropped by an enormous factor compared to what is would have been if no inflation had occurred . the weyl entropy was very small , compared to the maximal possible value . the universe was more or less uniform and homogeneous . this homogeneity can be seen in the cmb radiation today . however , small fluctuations in the spectrum can also be seen , reflecting the state of the universe 300 000 years after the big bang . after the inflation , small seeds of inhomogeneities from the quantum fluctuations of quantum fields were the only thing left of the initial inhomogeneities . nevertheless , these seeds were large enough to gradually clump together and form galaxies and stars . the weyl entropy began to increase again after the inflationary regime was over . the radiation- , and later the matter - dominated universe caused the weyl entropy to grow steadily and firmly for almost 15 billion years . during the radiation era , there were only small inhomogeneities and anisotropies left of the primordial fluctuations . using the calculations from the earlier section , the weyl entropy increases as @xmath132 in the radiation era . at about @xmath18410 000 years the radiation become sub - dominant . the universe evolved into a matter dominated era and the weyl entropy increased as @xmath185 . today it is still growing . recent observations suggest that the universe has entered a new era with accelerated expansion . the surprising fact that the universe appears to be in an accelerating state today , can be explained with the presence of a vacuum energy . if this is true , the weyl entropy will increase steadily and asymptotically towards a constant . however , the late time behaviour of our universe is still quite speculative , and whether or not this vacuum dominated period will persist , is very uncertain . it might happen that the period ends like inflation did , and perhaps the vacuum period is followed by a curvature dominated period . if so , the weyl entropy might again rise to new heights , increasing towards a value where all the matter is collected in black holes . sh was founded by the research council of norway . r. penrose , in _ proc . first marcel grossmann meet . ( ictp trieste ) _ , ed . r. ruffini , north - holland ( 1977 ) r. penrose , in _ general relativity , an einstein centenary survey _ , s.w . hawking and w. israel , cambridge univ . press ( 1979 ) . r. penrose , _ j. stat . phys , _ , * 77 * ( 1994 ) 217 j. wainwright and p.j . anderson , _ gen . * 16 * ( 1984 ) 609 . goode and j. wainwright , _ class . quantum grav . _ * 2 * ( 1984 ) 99 . grn and s. hervik , _ class . quantum grav . _ * 18 * ( 2001 ) 601 . bonnor , _ class . grav . _ * 3 * ( 1986 ) 495 . h. kodama , _ prog . * 99 * ( 1998 ) 173 j.d . barrow and h. kodama , _ class . quantum grav . _ * 18 * ( 2001 ) 1753 j.d . barrow and h. kodama , _ int . j. mod . _ * d * * 18 * ( 2002 ) h. kodama , ` gr - qc/0109064 ` s. hervik , _ class . quantum grav . _ * 17 * ( 2000 ) 2765 . s. hervik , _ class . quantum grav . _ * 18 * ( 2001 ) 175 . a. krasiski , _ inhomogeneous cosmological models _ , cambridge university press ( 1997 ) . a. zecca , _ il nuovo cimento _ * 106b * ( 1990 ) 413 . b. miller , _ astroph . j. _ * 208 * ( 1976 ) 275 . s.w . hawking and r. penrose , _ proc . _ * a314 * ( 1970 ) 529 . bonnor , _ phys . lett . _ * 112a * ( 1985 ) 26 . s. hayward , _ private communications _ ( 2000 ) .

neutral hydrogen constitutes a significant component of the interstellar medium of disk galaxies . the study of its abundance and kinematics , as measured by the 21 cm emission line , has seen applications to a variety of astrophysical problems , ranging from studies of the large scale distribution of galaxies to secular evolution in clusters . hi line emission carries information about the distribution of neutral gas , and through tracing of its velocity field , of the total mass in a galaxy . even in the absence of any spatial information about the distribution of the gas , the global hi line profile , contained within the beam of a single dish telescope or integrated over the entire hi distribution where resolved , provides three key indicators : ( 1 ) the observed redshift of the system as a whole ; ( 2 ) the observed hi line flux , which , in the optically thin case , can be related to the hi mass ; and ( 3 ) the observed hi line full width , which , when corrected to face - on viewing , is related to the total rotational velocity of the hi gas . these quantities have been used for diverse applications such as the tracing of large scale structure by gas - rich systems , the comparative hi content of galaxies in differing environments and the derivation of secondary distances using the tully - fisher method ( tully & fisher 1977 ) . hi line systemic velocities , fluxes and characteristic velocity widths have been recorded for more than 20,000 galaxies and presented in large published compilations such as that of huchtmeier & richter ( 1991 ) and the _ third reference catalog of bright galaxies _ ( rc3 , de vaucouleurs _ et al . _ 1991 ) . until recently , the vast majority of extragalactic hi line spectral parameters were associated with observational targets selected from optical catalogs . only recently have large - area hi blind surveys been undertaken , notably the arecibo hi strip survey ( ahiss ; zwaan _ et al . _ 1997 ) , the arecibo dual beam survey ( adbs ; rosenberg & schneider 2000 ) , the hi jodrell all - sky survey ( hijass ; lang _ et al . _ 2003 ) and the hi parkes all - sky survey ( hipass ; barnes _ et al . _ 2001 ; koribalski _ et al . _ 2004 ; meyer _ et al . _ most recently , the blind hi survey conducted with the parkes telescope multifeed array ( hipass ) has catalogued 4315 extragalactic hi line sources ( hipass catalog ; meyer _ et al . _ 2004 ) south of dec . @xmath2 identified purely on the peak flux of their hi line emission profiles . perhaps surprisingly , extremely few systems detected in the hi blind surveys have no optical counterpart ( kilborn _ et al . _ 2000 ; rosenberg & schneider 2002 ) , but understanding the difference between optical and hi selected samples is critical to determining the true census of gas - bearing objects in the nearby universe and the relationship of present day disks to the damped lyman @xmath3 absorbers seen in qso spectra . with the exception of the hipass dataset which has been processed uniformly and can be accessed digitally through a public website ( http://hipass.aus-vo.org/ ) , a significant disadvantage of past compilations arises from heterogenous processing and parameter extraction algorithms , as well as the lack of availability of the digital spectra themselves . in many cases , error or noise estimates are lacking and corrections for instrumental effects not applied or derivable . because of our own on - going work related to the applications of hi spectral data in correlative and secondary studies , we have developed and maintained a digital archive of hi line spectra for some 9000 optically selected galaxies observed by us and our numerous collaborators using various telescopes over the last 20@xmath4 years . two of the telescopes used , the 91 m and 42 m telescopes of the national radio astronomy observatory at green bank , are no longer operational ( in fact , the 91 m telescope no longer exists ) . while hi parameters for most of these datasets have been presented previously , changing scientific objectives and computer capabilities have led to advances in the robustness of parameter extraction algorithms . in order to establish a well - characterized and permanent digital archive of these global hi profiles , we have reanalyzed the global hi line profiles using a single set of processing and parameter estimation algorithms . in this paper , we present a homogeneous compilation of the extracted hi line parameters and provide quantitative corrections and error estimates . these data are being used in a variety of ongoing studies , including a robust estimation of the hi mass function ( springob , haynes , & giovanelli 2005 ) and studies of the local peculiar velocity field . in section 2 , we review the observational circumstances of the archival dataset as well as present new hi observations undertaken with the arecibo 305 meter telescope . section 3 discusses the process of extracting useful parameters from the global hi line profiles and the corrections required to derive physically meaningful quantities . the archive is presented in section 4 , followed by discussion of the characteristics of the data set , its completeness and its limitations in section 5 . a brief summary concludes in section 6 . the majority of the hi line datasets included in the current compilation have been discussed in previous publications . here we review the general characteristics of the telescopes employed to conduct the observations , summarized in table 1 , and also present the results of new observations for a sample of 145 obtained with the upgraded arecibo telescope . lccccl arecibo `` dual circular '' & 305 & 8.0 & 3.3 & 6475 & hg84,a , b , c , d , e , f , g , h , k , m , n , o , p , q , sarecibo `` flat '' & 305 & 6.0 & 3.9 & 20 & hhg , o , qarecibo `` gregorian '' & 305 & 11.0 & 3.5 & 116 & this workeffelsberg & 100 & 1.5 & 8.8 & 66 & mgreen bank 91 m & 91 & 1.1 & 10.0 & 1090 & i , j , m , n , ogreen bank 42 m & 42 & 0.25 & 21.0 & 706 & l , m , rnanay & 300@xmath534.6 & 1.2 & 4@xmath520 & 377 & m since the early 1980 s , we have attempted to maintain a digital archive of the extragalactic hi global profiles we and our collaborators have observed at a number of large single - dish radiotelescopes , namely the 305 m arecibo telescope of the national astronomy and ionosphere center , the late 91 m and 42 m green bank telescopes of the national radio astronomy observatory , the nanay radio telescope of the observatory of paris and the effelsberg 100 m telescope of the max planck institut fr radioastronomie . the basic parameters of those telescopes are summarized in table 1 , along with a compilation of the principal references to the published works which include discussions of the individual observing programs , observational setups and target selection criteria . observations at arecibo employed several diffferent l band feed systems , denoted by their designated names : one of the high gain dual circular polarization feeds ( `` dual circular '' ; two such similar feeds were available , one of which was remotely tunable in frequency ) and a low sidelobe linear polarization l - band feed ( `` flat '' feed ) . during the 1990 s , the arecibo telescope was upgraded and a low noise , broadband secondary and tertiary reflector gregorian optical system was installed ; observations made with it are described in the following subsection . nearly all observations were conducted in total power mode , with a position switching technique used to remove instrumental baseline effects . at each telescope , calibration of the noise diodes was performed by conducting observations of continuum sources from the catalog of bridle _ et al . _ ( 1972 ) , and both the flux and velocity scales were further checked by observations of a sample of strong hi line emitting galaxies . typical on source integration times were dictated by signal to noise requirements and varied from as little as 5 minutes at arecibo to as much as 2 hours with the 42 m and nanay telescopes . final spectra are the accumulated weighted averages of multiple on - off pairs . the main distinctions among the hi datasets arise from differences in the sensitivities achieved with the different telescopes , and differences in the resolution and bandwidth associated with the available spectral backends . in order both to permit the analysis of instrumental effects and to measure widths using the same algorithm , the accumulated and calibrated final spectra were exported from the respective on line data format into the standard analyz galpac `` history file '' format developed by rg at the arecibo observatory and were thus available in digital form for reprocessing using the algorithms described below . all accumulated spectra have been re - smoothed and re - baselined using an interactive process that allows the user to select the degree of smoothing ( typically hanning , but sometimes also a 3 channel boxcar in order to improve signal - to - noise , as noted in table 3 ) , choose the baseline regions and order ( typically 3 ) , flag the boundaries for flux integration and examine the results before they are stored away in the permanent archive . all spectra presented here have been reprocessed using the same parameter extraction algorithms , producing new measurements which supercede those presented previously . in addition to the hi spectra already contained in the digital archive , new hi line observations were conducted with the arecibo 305 meter telescope and its gregorian optics system between september and november 2001 . 116 galaxies were detected , and an additional 29 galaxies for which we have optical heliocentric velocities yielded upper limits . targets were selected from among inclined spirals included in our private database , referred to as the arecibo general catalog ( agc ) . the new observations preferentially targeted those objects which appeared to be good tully - fisher candidates and for which we did not already have hi spectral parameters , in the region of the pisces perseus supercluster , that is @xmath6 r.a . @xmath7 , @xmath8 dec . @xmath9 . the l - band narrow receiver was used in 9-level sampling mode for all observations , feeding each polarization to four correlator boards with 1024 channels and 25 mhz bandwidth ( yielding a velocity resolution of 5.15 km s@xmath1 ) . in most ( but not all ) cases , the redshift was not known in advance so that a search mode was employed , staggering the four boards to cover a total of 85 mhz bandwidth . the first board was centered at the 21 cm line rest frequency of 1420.40578 mhz , while the other three boards overlapped but with center frequencies offset at 20 , 40 , and 60 mhz below this frequency . on all nights , the system temperature was approximately 30 k. all observations were conducted in a total power , position switching mode as a sequence of on source and off - source pairs of 4 minutes each . the final spectrum is the average of the combination of the successive on - off pairs . the total number of on / off observations made for each galaxy varied based on the signal - to - noise ratio , and ranged from one to four . on the first three nights of observations , each board was doppler corrected to heliocentric velocity at the start of each on - off pair . beginning with the fourth night , the same doppler correction was applied to each of the four boards , allowing the spacing between correlators to remain at 20 mhz exactly . beginning on the fifth night , we used the naic `` radar blanker '' , which detects the san juan airport s radar and suspends data - taking for brief intervals around the pulse arrival times , for all observations in which the redshift of the galaxy was either not known a priori or known to be such that the redshifted 1420 mhz hi line would overlap with the 1350 mhz frequency of the airport radar . whereas data reduction for all other arecibo observations presented here was performed using the arecibo analyz - galpac analysis system , the data reduction for these new observations was carried out using the aips++ software package . the reduction algorithms match those carried out for the earlier observations in analyz - galpac ( and discussed in haynes _ et al . _ 1999 and references therein ) exactly , with one exception . we have used the ` setfreqmed ' and ` settimemed ' functions within the aips++ ` autoflag ' package to eliminate individual time - frequency pixels found to be corrupted by radio frequency interference ( rfi ) . the rfi identification algorithm flags pixels that deviate by more than 10 times the median deviation from a sliding median of width 10 pixels in both the time and frequency dimensions . the width of each pixel in the time domain is 1 second . the parameters extracted from global 21 cm line profiles obtained with single dish telescope require correction before physical quantities can be derived from them . here the process for conversion from observed to corrected parameters is described in some detail . modern radio telescopes can be continuum flux calibrated to a small fraction of their system noise , but as discussed by van zee _ et al . _ ( 1997 ) , the accuracy with which hi line fluxes can be measured is much less precise . beyond the continuum flux calibration which can be achieved at the better than 2% level , the accuracy with which line fluxes can be measured depends both on baseline stability and on corrections that must be applied to the observed values . the hi line fluxes @xmath10 presented here have been measured on the smoothed and baseline subtracted profiles by an interactive process in which the user marks the range of correlator channels over which the flux is measured by integrating the emission within these boundaries . the observed line flux @xmath11 must then be corrected for ( 1 ) beam attenuation , ( 2 ) pointing offsets , and ( 3 ) hi self absorption . we discuss each correction separately . since the hi distribution in a galaxy is often partially resolved by the telescope beam , the hi line flux detected by a given telescope underestimates the true flux by an amount that depends on how much the source hi distribution fills , or overfills , the beam . since the hi distribution is generally not known , an estimate of the beam attenuation correction is based on the adoption of a model hi distribution and an average telescope beam power pattern . the power patterns of the arecibo , 42 m , 91 m and 100 m telescopes are assumed to be gaussians with half power beam width ( hpbw ) as given in table 1 . for the nanay telescope , the beam width is assumed to be a two - dimensional gaussian with hpbw of 4 in the r.a . direction and 20 in the dec . the hi distribution is modeled as the sum of two gaussians , one of negative amplitude to produce a central hole ( shostak 1978 ; hewitt , haynes & giovanelli 1983 , hereafter hhg ) . following section iv of hhg , the hi surface brightness distribution @xmath12 is assumed to be characterized by the same axial ratio and to lie along the same position angle as the optical galaxy and is scaled such that @xmath13 , the diameter encompassing 70% of the hi mass , equals @xmath14 , where @xmath15 is the optical major axis , in arcminutes , taken from either the uppsala general catalog ( nilson 1973 , hereafter ugc ) or the eso - uppsala survey of the eso(b ) atlas ( lauberts 1982 , hereafter eso ) , or estimated by us on the poss - i . following hhg , the fraction @xmath16 of the source flux that is detected by the telescope beam is given by @xmath17 } \over { \sum_{j=1,2 } a_j \theta_j^2}}\ ] ] like hhg , we assume a double gaussian @xmath18 of amplitude @xmath19 and relative extent @xmath20 . note that @xmath21 above is not the hpbw but rather the beam extent at which the power pattern falls by @xmath22 , assuming a gaussian response . the inclination @xmath23 of the galaxy is derived from the observed optical axial ratio @xmath24 and an assumed intrinsic axial ratio that varies from 0.10 for the latest spirals to 0.23 for the earliest following section iv.e of haynes & giovanelli ( 1984 ; hg84 ) . the correction factor @xmath25 that accounts for beam attenuation is the inverse of this fraction , @xmath26 and approaches 1 when the hi distribution is significantly smaller than the telescope beam . because the hi extent of hi normal spirals is significantly larger than their optical dimension , the application of this correction is very important for most hi observations conducted at arecibo with its relatively small hpbw ( @xmath27 ) . for example , face - on galaxies with optical sizes equal to about half the hpbw ( i.e. , @xmath28 ) have corrections of about @xmath29 ; a similar edge - on hi disk requires a correction half as large . for an object inclined by @xmath30 and an optical extent equal to the hpbw , the correction amounts to @xmath3160% . because the hi disk can normally be traced well beyond the optical extent , the observed hi line flux in face - on systems as small as 0.8 require corrections of 5% or more . the other telescopes that have been employed have substantially wider beams , and there are few galaxies observed with them for which this correction is significant . additionally , if the source hi distribution is not centered in the telescope beam , @xmath11 will be less than it would be in the absence of such pointing offset . departures from centering arise both from the intrinsic accuracy of a radio telescope to center on the input coordinates and from errors in the target coordinates relative to the true source position , and they are important only when such offsets subtend a substantial fraction of the beam extent . for older observations conducted with the arecibo line feeds , the rms pointing accuracy of the telescope system was @xmath32 in both r.a . and dec . the precise amplitude of this error is not known but has been assumed to be random and of order 5% ( hg84 ) . therefore , all hi lines fluxes derived from spectra obtained at arecibo prior to 1993 have been multipled by a constant factor of @xmath33 to account for this systematic bias . this correction is not applied to other datasets . similarly , older datasets are also affected by the inaccuracy of the coordinates input to position the telesope on source at the time the observations were conducted . in such cases , the flux is also underestimated by an amount which depends on the ratio of the angular displacement of the center of the telecope beam from the true source location to the beam extent and is important for some of the older arecibo line feed data . for spectra in our digital archive , a measure of the amount by which the flux has been underestimated because of coordinate inaccuracy can be calculated because the input coordinates are recorded in the digital spectra . to estimate this correction factor , we have performed a set of numerical simulations mimicking observations offset from the arecibo line feed beam center , by convolving the beam power pattern with template galaxy hi distributions following the method developed in hhg . these simulations confirm that the measured flux in the simulations is relatively insensitive to the position angle of the galaxy ( assumed to follow the optical image ) , but quite sensitive to the angular displacement of the beam center from the center of the hi disk . following again hhg s result that @xmath34 , we relate the angular displacement @xmath35 of the targeted position from the true position and the optical apparent diameter @xmath15 to the flux correction @xmath36 , as shown in figure 1 . for each galaxy observed with the arecibo line feeds for which the pointing offset exceeds 0.1 , we adopt the flux correction associated with the closest match in both optical size and pointing offset shown in figure 1 . the figure illustrates the point that the required flux correction can be very large even for optically small galaxies , if the pointing offset is large . however , there are only 183 galaxies for which this correction boosts the flux by a factor of @xmath371.5 - 2.0 , and only 62 for which the flux correction is more than a factor of @xmath372 . lastly , optical depth effects also can result in a decrease of detected hi line flux if not all the photons escape . in the denser regions of disks , the interstellar medium can be optically thick at 21 cm ( braun 1997 ) , but the filling factor of opaque regions is small in the vast majority of disks . following giovanelli _ _ ( 1994 ) and using the optical axial ratio as the measure of disk inclination , we correct for hi self - absorption by multiplying the measured hi line fluxes by the factor @xmath38 . this effect is small , with less than 30% of the hi line flux self absorbed in the most highly inclined systems . however , there are some galaxies for which this is the dominant correction . because this correction is model - dependent , and the interested reader may wish to apply a different self - absorption correction , we report the self - absorption corrected flux separately . the two values we report for the corrected hi line flux are then @xmath39 , including the corrections for source extent and pointing , and @xmath40 , which also includes the correction for hi self - absorption . @xmath41 is , on average , 25% larger than the observed line flux . it should be noted that all of the corrections increase the value of line flux used in the calculation of hi masses . comparison of values presented here with others in the literature should take into account similar treatments in order to avoid systematic differences in the derived hi masses which might otherwise be interpreted as physical differences . several different sources of error affect the measurement of integrated hi line fluxes . the first is the uncertainty in the absolute flux calibration commonly achieved by firing a noise diode at regular intervals ( van zee _ et al . _ 1997 ) and understanding well the telescope gain and its variation . because modern telescope systems are significantly more stable than older ones , we follow van zee _ et al . _ ( 1997 ) in estimating that the uncertainty introduced by the noise diode calibration is 2% of the observed integrated flux for those observations made with the 42 m green bank telescope and for those made with the arecibo telescope after the installation of its gregorian optical system . for all other observations , we estimate the uncertainty introduced by the noise diode calibration to be 10% of the observed integrated flux . further , we estimate that the uncertainty introduced by the correction coefficients is @xmath42 . the integrated flux error contribution from _ statistical _ uncertainty has been discussed by both fouqe _ et al . _ ( 1990b ) and schneider _ et al . _ ( 1990 ) , both of whom derive analytical expressions for the statistical uncertainty . unlike the fouqe _ et al . _ ( 1990b ) formula , however , the schneider _ et al . _ ( 1990 ) formula does not include any assumptions regarding the shape of the line profile , but _ does _ include the contribution to the total error stemming from uncertainties in the baseline fit . we therefore adopt the schneider _ et al . _ ( 1990 ) formula , with one modification . the formula includes a dependence on @xmath43 , the line width measured at 20% of the peak flux density . we have our own prefered width measurement algorithm which is explained in section 3.2 . because we find that the value of @xmath44 ( where @xmath45 is the width as measured by our favored algorithm ) is , on average , @xmath46 , we replace the @xmath47 term in schneider _ et al . _ ( 1990 ) with @xmath48 . @xmath49 , the statistical component of the integrated line flux error , is then given by @xmath50 ( where @xmath51 is the r.m.s . figure , measured in the signal free part of the spectral baseline and @xmath52 is the velocity resolution of the spectrum ) . our estimate of the total integrated line flux error , @xmath53 , is then the sum in quadrature of the error contributions from statistical noise and baseline uncertainties , noise diode calibration , and correction coefficients . for the vast majority of the galaxies in the archive , this uncertainty is 10 - 15% of the corrected integrated flux @xmath41 . the systemic velocity of the hi line profile is usually derived as the midpoint of the segment that connects two points on opposite edges of the emission profile at a selected intensity level , usually some fraction of the mean or peak flux . the same algorithm is also used to derive the hi line width , and its extraction often dictates the choice of measurement algorithm . over the years , numerous algorithms have been applied to derive width measures that both are robust in the presence of noise and most closely reflect the disk rotational velocity ( lewis 1983 ; schneider _ et al . _ 1986 ; chengalur _ et al . _ algorithms applied originally to hi spectra contained in earlier datasets were optimized to produce consistent measurements in noisy spectra ( e.g. , bicay & giovanelli 1986 ) whereas more recent studies added emphasis on sampling the rotational velocity ( lavezzi & dickey 1997 ) . the algorithm adopted to obtain hi line velocities and line widths here is very similar to that discussed by chengalur _ et al . _ ( 1993 ) and has been previously discussed in haynes _ et al . _ ( 1999 ) . figure 2 illustrates the issues involved in systemic velocity and width extraction from global hi line profiles through simulations of representative cases . after identification of the flux level , @xmath54 , of each of the spectral horns , the rising side of the profile is fit by a polynomial between the levels of 15% and 85% of @xmath55 . the velocity flagged for that side of the profile @xmath56 is that for which the polynomial fit has a flux of 50% of the value of @xmath55 . this process is repeated for the other side of the profile to flag a velocity @xmath57 at the corresponding level on the other horn . the adopted velocity width , hereafter designated @xmath45 , is defined as the difference between these two velocities @xmath58 , while the system velocity is their average @xmath59 . in the vast majority of cases , the polynomial fits are straight lines . in rare cases , however , higher order fits are required and can be selected during the analysis stage . this procedure has a major advantage over width algorithms which measure a width as the difference in velocities between points _ on the spectrum itself _ at n% of the peak flux . under the scheme adopted here , noise effects on either side of the profile are averaged out . additionally , the use of the parameter @xmath55 rather than simply @xmath54 substantially reduces the dependence of the measured width on the signal - to - noise ratio ( snr ) . the most important flaws in the method are encountered when it is difficult to identify the location of the horns , such as in low signal to noise profiles as shown in panel 2(b ) , or when one of the horns is missing as in the strongly asymmetric profile shown in panel 2(c ) ( see , e.g. richter & sancisi 1994 and haynes _ et al . _ 1998 ) . in these cases , the width measurement loses objectivity and accuracy and its derived error may not reflect the true uncertainty . for this reason , we also assign a width `` quality index '' to each result listed in table 3 . to convert @xmath45 into a physical rotational velocity , four corrections must be applied : 1 ) an instrumental correction that accounts for the spectrometer resolution , smoothing applied to the spectrum , and snr , 2 ) a cosmological redshift correction , 3 ) a correction for turbulent motions of the hi gas , and 4 ) a correction for the inclination of the disk to the line of sight . the widths presented here account for the first two effects only . the redshift correction is simply the instrumental - corrected width divided by a factor of @xmath60 , where @xmath61 is the galaxy s redshift . the instrumental correction is described in the following subsection . although we do not apply such a correction , the effect of turbulent motions on the measurement of @xmath45 is discussed in section 3.2.3 . in addition to the width measurement algorithm described above , each spectrum has been measured using four additional algorithms : @xmath62 is the width measured at 50% of the mean flux density , @xmath63 is the width measured at 50% of the value of the peak flux density minus the rms value @xmath55 , @xmath43 is the width measured at 20% of the value of the peak flux density minus the rms value @xmath55 , and @xmath64 is the width measured at 50% of each of the two peaks minus the rms value . we include these values to allow the interested reader to compare our width measurement results to those of other authors . however , we have derived corrections only for @xmath45 , not for any of the additional width measurements . it is common practice to convolve radio spectra obtained with autocorrelation spectrometers with a one channel wide cosine function ( in practice a [ 0.25,0.5,0.25 ] three channel function : hanning ) . this is done for two reasons . first , narrowband terrestrial interference , which occurs often within the observed spectral range even within parts of the spectrum protected by regulatory agreements for radio astronomical use , introduces `` ringing '' associated with the @xmath65 functional response of the spectrometer to single channel wide features . and , second , because of the finite number of correlator channels , the true autocorrelation function is truncated , resulting in its apparent multiplication by a rectangular function which is a multiple of the autocorrelation lag time ; this in turn means that the true power spectrum has been convolved with the fourier transform of this rectangular function , itself a `` sinc function '' . the application of hanning smoothing reduces the spectral resolution of the observed spectrum . in some cases , further spectral smoothing is required by the low snr of the observation or for other reasons ; these measurements tend to be less useful for tf applications . all the hi spectra presented here are hanning smoothed once ( smoothing code h in our data tables , e.g. haynes _ et al . _ 1997 ) ; in a minority of cases , as noted in our tabulations , three channel `` boxcar '' averaging plus hanning is applied instead ( smoothing code b ) . previous width measurement algorithms have often relied on measuring the width by computing the velocity difference between channels on either side of the profile . the channels are identified by either starting from inside the profile and searching outwards or starting outside and searching inwards , flagging the first channel to cross a flux limit that represents a particular fraction of the peak flux . this procedure creates a systematic bias in the width measurement as a function of the snr ( lewis 1983 ) . as described earlier , our width measurement procedure minimizes this effect by instead calculating the velocity difference between _ polynomials that have been fit to several channels_. nevertheless , there remains _ some _ dependence on the snr , defined here as the ratio of the peak hi signal to the rms noise in the line free portion of the spectrum , which we discuss in this section . we have studied the instrumental and noise effect via simulations that incorporate autocorrelator spectrometer channel response and a variety of observed characteristics , such as profile slope and curvature . the simulations involve inputting a high resolution , noiseless profile , then binning the flux into channels , adding noise , and measuring the width of the profile using the same algorithm used also on the real data . this process is applied to a range of simulations adopting channel resolutions ranging from 5 to 11 km s@xmath1 , for the two different smoothing prescriptions , and for a variety of different profile shapes . to account for the spectral noise , the true snr of the spectrum can be characterized by snr @xmath66 . the broadening of the width measurement is denoted by the parameter @xmath67 , which is one half the difference between the measured width and the width that would be measured in the absence of instrumental and noise effects ( measured in channels ) . we can then account for the combined instrumental and noise effects by _ subtracting an offset _ from the measured width , parametrized as @xmath68 where @xmath69 is the spectrometer channel separation in km s@xmath1 assumed to be equal to or very close to the channel width ) and @xmath67 depends on the snr and the type of smoothing . the simulations suggest that @xmath67 is influenced by two competing effects . as our width measurement algorithm measures a best - fit line to the sides of the profile at 50% of the value of @xmath70 , the two parameters in play are 1 ) the value of @xmath70 , and 2 ) the range of spectral values over which the line is fit . if instrumental and noise effects increase ( decrease ) @xmath70 , then @xmath67 will decrease ( increase ) , as a higher ( lower ) 50% value corresponds to a narrower ( broader ) width . even in the absence of noise , the peak flux is depressed by channel binning and smoothing , broadening the measured width . for realistic values of snr , however , the situation is more complicated . one can characterize @xmath70 with the expression @xmath71 where @xmath72 is the flux of channel @xmath23 that one would measure in the absence of noise and @xmath73 is a gaussian random variable describing the noise contribution to channel @xmath23 . we find that profiles with steep inner slopes observed with broad channel spacings tend to produce decreasing values of @xmath70 with increasing @xmath74 . other profile shapes and spectral resolutions give a somewhat more complicated dependence . so the variation of @xmath70 affects the profile width measurement , but whether it becomes broader or narrower depends on the details of the profile shape and resolution . @xmath67 is also influenced by the range over which the line is fit to the sides of the profile . the endpoints of the fit are found by searching outward from each peak , identifying the first channels with flux less than 15% and 85% of @xmath70 . this method introduces a slight bias towards the inclusion of channels whose flux has been _ depressed _ by noise . for noisier profiles , the region over which the line is fit also moves closer to the peak of the profile , where the slope is shallower . both of these effects preferentially select channels with flux that falls _ below _ the best fit line in the noiseless case , leading to narrower width measurements . this effect dominates the dependence of @xmath67 on snr over the range @xmath75 log(snr ) @xmath76 for each of our channel resolutions and for all realistic profile shapes , meaning that @xmath67 increases with snr over that range . the simulations show that the most relevant line profile parameters in determining @xmath67 are the slopes on each side of each of the two peaks and the number of channels leading up to the peaks . while @xmath67 typically varies by @xmath31 0.5 channels for the same profile shape with different values of the snr , the variation can be even greater ( @xmath31 1 channel ) among different profile shapes _ and _ among different noise realizations of the same profile for constant snr , channel width , and smoothing prescription . this variation limits our ability to determine @xmath67 precisely for a real profile . however , we can simply apply our simulation results for a profile with mid - range values for the inner and outer slopes and number of channels leading up to the peaks . the resulting approximate solution gives : @xmath77 @xmath78 @xmath79 the values of @xmath80 , @xmath81 , @xmath82 , and @xmath83 are constants for both @xmath84 km s@xmath1 and @xmath85 km s@xmath1 , but depend linearly on @xmath69 within the range @xmath86 km s@xmath1 . the values of these parameters for each of the channel widths and smoothing prescriptions can be found in table 2 . we stress that both the linear dependence of @xmath67 on log(snr ) in the range @xmath75 log(snr ) @xmath76 and the linear dependence of the @xmath87 s on @xmath88 in the range @xmath89 km s@xmath1 are approximate and completely empirical . we also keep the @xmath87 s constant outside the range @xmath86 km s@xmath1 solely because few of our observations have channel resolutions far outside of that range , and we have done little testing of the width corrections outside that range . rccccc @xmath90 5 km s@xmath1 & h & 0.005 & -0.4685 & 0.785 & 0.3955 - 11 km s@xmath1 & h & 0.037@xmath69 - 0.18 & 0.0527@xmath69 - 0.732 & -0.027@xmath69 + 0.92 & 0.023@xmath69 + 0.28@xmath9111 km s@xmath1 & h & 0.227 & -0.1523 & 0.623 & 0.533 @xmath90 5 km s@xmath1 & b & 0.020 & -0.4705 & 0.820 & 0.4305 - 11 km s@xmath1 & b & 0.052@xmath69 - 0.24 & 0.0397@xmath69 - 0.669 & 0.020@xmath69 + 0.72 & 0.062@xmath69 - 0.12@xmath9111 km s@xmath1 & b & 0.332 & -0.2323 & 0.940 & 0.802 in addition to the instrumental and smoothing corrections , the observed width must also be corrected for redshift stretch , giving : @xmath92 the hi clouds in a disk do not move along strictly circular orbits . even in the absence of large scale anomalies , such as warps , bars and other global asymmetries , deviations from circular motion occur , such as those associated with small scale dispersion and streaming motions near spiral arms . et al . _ ( 1983 ) have suggested that the contribution to the observed width of turbulent motions , or otherwise small scale deviations from circular rotation , could be expressed via a term @xmath93 , additive to the circular width in the form : @xmath94 where @xmath23 is the disk inclination and @xmath95 the maximum rotational width . tully & fouqu ( 1985 ) found that , while a linear correction to the observed width may well account for the effect of turbulent motions on large widths , for low luminosity galaxies a quadratic combination would be more appropriate ( see also fouqu _ et al . _ they thus proposed a formula , their equation 12 , that yields a linear combination at large widths , a quadratic combination at small widths , and a transitional form at intermediate widths . this formula was adopted by a variety of other authors ( e.g. , broeils 1992 ; rhee 1996 ; verheijen 1997 ) who also measure hi widths at n% of the profile peak . these authors measure the value of the @xmath93 parameter by minimizing the scatter between the turbulence - corrected velocity widths and the velocity widths determined from rotation curves of the same galaxies . verheijen ( 1997 ) reviews the results of the aforementioned authors , and notes that the optimal value of @xmath93 strongly depends on the instrumental correction used . since we do not measure velocity widths at n% of the profile peak , and our instrumental correction differs from that of these other authors , we are not justified in adopting any of the previous values of @xmath93 . instead , we derive our own turbulent motion correction by introducing a random , isotropic @xmath96 km s@xmath1 turbulent motion into the simulations described in the previous subsection . we find that this depresses the peak flux , and thus broadens the measured profile width by moving the @xmath97 value farther from the profile center . as in verheijen ( 1997 ) and references therein , the amount of turbulence - induced width broadening is found to vary greatly from galaxy to galaxy , depending on the details of the profile shape . we do not find a discernable trend in the amount of width broadening with profile width or signal - to - noise ratio . for the sake of simplicity , we thus subtract the turbulence correction from the width linearly , with the equation : @xmath98 , where @xmath99 km s@xmath1 . lavezzi & dickey ( 1997 ) likewise discuss the merits of a simple linear subtraction of a turbulence term . as stated earlier , in order to use @xmath100 for applications such as those involving the tully - fisher relation , corrections for both inclination angle and turbulence must be applied as well . because of its uncertainty , the values of corrected width @xmath101 reported here _ do not include that correction _ , nor is viewing angle considered . authors wishing to make full use of the data should apply those corrections themselves . all of the spectra in the archive have been newly reprocessed using the new flux and width corrections . the spectra themselves will be made available through the u.s . national virtual observatory . here , we provide the corrected and uncorrected spectral parameters . in table 3 , we present the results for all 8852 spectra for which we are confident that we have detections . the format , described below , closely resembles that of haynes _ et al . _ ( 1999 ) table 1 : column ( 1 ) . entry number in the ugc , where applicable , or else in our private database , referred to as the arecibo general catalog ( agc ) . column ( 2 ) . ngc or ic designation , or other name , typically from the _ catalog of galaxies and clusters of galaxies _ ( zwicky _ et al . _ 1961 - 1968 ) , eso , or the _ morphological catalog of galaxies _ ( vorontsov - velyaminov & arhipova 1968 ) . where used , the designation in the latter is abbreviated to eight characters . columns ( 3 ) and ( 4 ) . right ascension and declination in the j2000.0 epoch either from the nasa / ipac extragalactic database ( ned ) or measured by us on the poss - i . typically , the listed positions have @xmath102accuracy . column ( 5 ) . the blue major and minor diameters , @xmath103 , in arcminutes , either from the ugc or eso catalogs or as estimated by us on the poss - i . column ( 6 ) . the morphological type code following the rc3 system . classification comes either from the ugc or eso catalogs or from our own visual examination of the poss - i prints . column ( 7 ) . the observed integrated 21 cm hi line flux @xmath104 in jy km s@xmath1 . column ( 8) . the corrected integrated 21 cm hi line flux @xmath105 , also in jy km s@xmath1 , after corrections applied for pointing offsets ( arecibo line / flat feed spectra only ) and source extent following section 3.1 . column ( 9 ) . the self - absorption corrected integrated 21 cm hi line flux @xmath41 , also in jy km s@xmath1 , after corrections applied for pointing offsets ( arecibo line / flat feed spectra only ) , source extent , _ and _ hi self - absorption , following section 3.1 . column ( 10 ) . the uncertainty @xmath53 in the self - absorption corrected integrated flux @xmath41 , following section 3.1.1 . column ( 11 ) . the rms noise per channel of the spectrum , rms , in mjy . column the emission profile signal - to - noise ratio , snr , taken as the ratio of the peak hi line flux to the rms noise within the signal free portion of the spectrum . unlike in section 3.2.2 , we do not subtract the rms from the peak flux before dividing by the rms . column ( 13 ) . the heliocentric velocity @xmath106 , in km s@xmath1 , of the hi line signal , taken as the midpoint of the profile at the 50% level also used to measure the width ( see below ) . column ( 14 - 18 ) . the full velocity width of the hi line in km s@xmath1 , uncorrected for redshift or other effects , using the five measurement algorithms discussed in section 2.2 . widths are @xmath107 and @xmath64 respectively . column ( 19 ) . a corrected velocity width @xmath101 , in km s@xmath1 , which accounts for redshift stretch , instrumental effects , and smoothing . the correction is applied to @xmath45 . note that this width is not rectified for viewing angle or turbulence . column ( 20 ) . the estimated error on @xmath101 , @xmath108 , in km s@xmath1 , taken to be the sum in quadrature of the measurement error and the estimated error in broadening corrections . the estimated error on the velocity itself , @xmath109 , can be approximated as @xmath110 . column ( 21 ) . a code indicating the instrument used : ( aolf ) arecibo , line feed system ; ( aoff ) arecibo , flat feed system ; ( aog ) arecibo gregorian ; ( effs ) effelsberg 100 m ; ( gb300 ) green bank 91 m ; ( gb140 ) green bank 42 m ; ( nanc ) nancay . column ( 22 ) . bandwidth of the spectrum in khz @xmath111 . column ( 23 ) . number of channels in the spectrum @xmath112 . column ( 24 ) . three codes indicating the smoothing , width quality , and inclusion in the 2001 arecibo observations . the first code is the smoothing code : ( h ) single hanning only ; ( b ) hanning plus three - channel boxcar . the second is a qualitative assessment of the quality of the profile : ( g ) good ; ( f ) fair ; ( s ) single peak ; ( p ) poor quality for tf applications ; ( c ) confused ; ( a ) shows absorption . velocity widths measured from profiles classified as good detections should be useful for tf applications ; those measured on profiles denoted `` f '' should be used with caution . because the width measuring algorithm is designed for application to two - horned profiles , the widths measured on single peaked profiles should also be used with caution , as they may underestimate the true rotation width . widths in the last two categories can not be used for tf purposes . confusion is identified only when contamination from another galaxy in the beam is certain . the third code where applicable indicates , with a * , inclusion in the sample observed with the arecibo telescope in the fall of 2001 as described in section 2.2 . in addition to the hi detections presented in table 3 , tables 4 and 5 also summarize useful data about objects for which homogeneous upper limits to the hi flux can be calculated . table 4 contains summary data for 156 galaxies classified as `` marginal '' detections , that is , for those cases in which the signal has not been verified through adequate reobservation . table columns match those of table 3 , except that we omit columns 14 - 20 because of their uncertainty , and column 24 because , in this instance , all spectra have been both hanning and boxcar smoothed . table 4 also adds column 25 , the heliocentric velocity as derived from optical observations , @xmath113 , in km s@xmath1 , taken from the ned . although the parameters extracted from the hi line profile are uncertain , the coincidence of the hi signal with the expected location in frequency corresponding to the optical redshift brings a measure of believability to the possible hi signal . such objects may be deserving of further , deeper hi observations which probe sensitivities significantly greater than those presented here . in table 5 , we present summary results for objects which are classified as the hi line nondetections , except that columns 7 , 8 , 9 , 10 , 12 , and 13 are also omitted . column 26 adds the heliocentric velocity @xmath114 corresponding to the central frequency of the band searched for hi emission . only objects for which the radio frequency band covered by the spectrum included the frequency expected from the optical redshift are listed in this table . upper limits to the hi line flux can be estimated from the rms noise per channel and an assumption of the hi profile shape . a simple estimate of the upper limit to the integrated hi line flux , useful for calculating an upper limit to the hi mass for non - detections , adopts a rectangular profile of height 1.5 rms @xmath115 , where @xmath115 is the full width of the doppler broadened hi line , expected , for example , for a galaxy of a given optical luminosity given by the tully fisher relation . in contrast to a wide area , hi blind survey like hipass , the archival dataset presented here is extremely heterogenous in terms of its sky coverage , depth and completeness . at the same time , it contains hi line spectra for more than twice as many detections as the hipass hicat ( meyer _ et al . _ 2004 ) , with much higher typical sensitivity and probing a greater volume . the sky distribution for all detected galaxies listed in table 3 is shown in figure 3 . the unevenness of sky coverage , with strong preference for that portion of the sky visible to the arecibo telescope , is clearly evident . in fact , 77 % of the objects fall within the arecibo declination limits ( @xmath116 dec . @xmath117 ) . the southern cutoff at dec . @xmath118 represents the limitations of the steerable northern hemisphere telescopes that were used . there is also a strong deficiency of objects in the zone of avoidance , not just because the archive is comprised of objects with known optical counterparts but also because several of the previous surveys have avoided regions of high galactic extinction for reasons of photometric accuracy and completeness at optical wavelengths . the sky distribution of targeted galaxies also reflects the large scale structure within the survey volume , evident both in figure 3 and in the redshift distribution illustrated in figure 4 . the most prominent feature is the pisces - perseus supercluster at @xmath6 r.a . @xmath7 and @xmath119 km s@xmath1which has been the target of several past observing programs . also seen clearly is the virgo region toward the center of the local supercluster at r.a . @xmath120 , @xmath121 km s@xmath1 . the distributions of the rms noise per channel and snr are shown in figures 5 and 6 respectively . because of the superior gain and angular resolution of the arecibo telescope , observations undertaken with it are typically more sensitive than ones conducted with other instruments . for this reason , the fraction of the archival dataset extracted from arecibo observations is indicated separately in those figures . the diversity of the observational programs which have contributed datasets to this hi archive is evident in these distributions . in some instances , the targetting of objects for the explicit purpose of measuring comparative hi content or accurate hi line widths has placed more stringent demands for lower rms noise values , or higher snr , than , for example , those programs that were conducted principally as wide area redshift surveys . the total integrated ( corrected ) and peak hi line flux distributions are shown in figure 7 and 8 respectively . comparing these figures to the corresponding distributions for the hipass / hicat blind hi survey shown in figure 9 of meyer _ et al . _ ( 2004 ) and figure 4 of zwaan _ et al . _ ( 2004 ) , it is readily apparent that the optically targeted archive presented here contains a substantial number of galaxies with both integrated fluxes and peak fluxes well below those included in the hipass catalog . the hicat sample contains very few objects with @xmath122 2.5 jy km s@xmath1 , whereas almost one third of the optically selected sample presented here have fluxes @xmath41 lower than that value . likewise , hicat strives to be peak flux limited and contains almost no sources with peak hi fluxes below 30 mjy , whereas the majority of galaxies in the optically - selected sample have peak fluxes below that value . for an hi flux limited sample , the logarithm of the number of galaxies per unit log @xmath41 should be proportional to @xmath123 ; this relationship should also hold for the peak flux in a peak flux limited sample . the bottom panels of figures 7 and 8 show the log - log distributions of corrected flux @xmath41 and peak hi line flux @xmath124 , with the @xmath125 line superposed for reference . clearly the optically - selected sample is neither peak hi flux nor total hi flux limited as the hi - blind surveys strive to be . however , as shown in springob , haynes , & giovanelli ( 2005 ) , with the addition of supplemental data from the literature , it is possible to extract an optical diameter limited and hi flux limited subsample , restricted to the portion of the sky lying within the arecibo declination limits . figure 9 shows the distribution of instrumentally corrected ( but not turbulence or inclination corrected ) widths @xmath101 for all galaxies in the archive . objects with confused , asymmetric or otherwise oddly behaved spectral profiles ( coded `` c '' , `` p '' or `` f '' in table 3 ) are included in the full distribution , but profiles which are classified as `` g '' or `` s '' are included in the shaded area . in comparison to the hipass / hicat dataset , the galaxies included in this optically - selected sample are preferentially faster rotators , unsuprising given the tully fisher relation and the much larger volume sampled here . this dataset is being used to derive the observationally - based rotational velocity function for hi disk systems ( springob , haynes & giovanelli , in preparation ) . there is one interesting issue for which much of this dataset is unfortunately largely unsuited : the study of global asymmetry in hi profiles ( richter & sancisi 1994 ; haynes _ et al . _ 1998 ) . as discussed in section 3.1 , the shapes of global hi profiles observed with single dish telescopes are affected by both beam attenuation and pointing offsets and errors . prior to its recent upgrade , the arecibo telescope l - band feeds were subject to pointing errors of typically 2030 but sometimes larger , especially when the observations were conducted at large zenith angle . we therefore caution against using the spectra contained in the agc digital archive for studies of asymmetry , with the exception of the very carefully collected green bank 42 m telescope dataset obtained for that very purpose as discussed in haynes _ et al . _ ( 1998 ) . we present here a compilation of hi spectral parameters for more than 9000 optically - targeted galaxies in the local universe which have been observed by us and our collaborators over the last 20@xmath4 years . all of the hi spectra have been reprocessed in a homogenous manner and a single set of algorithms has been used to extract spectral parameters which can then be corrected to physical parameters taking into account the instrument and processing related complications such as spectral resolution , source extent , pointing offsets and signal to noise . we use both empirical evidence and the results of simulated observations to produce recipes for deriving these physical parameters from the observed ones based on all of the information available from the digital spectra themselves . the recent blind hi surveys such as hipass offer the advantage of completeness and homogenous data processing and parameter extraction . by reprocessing all of our existing hi line spectra using software specifically designed to exploit the details of the observing setups and the characteristics of the spectra themselves , we have tried to minimize the effects of instrument and processing related differences among the datasets . at the same time , hi completeness issues remain in optically targeted samples , despite their greater depth and sensitivity . it is well known that optical galaxy catalogs often preferentially miss the low optical surface brightness but hi - rich late type galaxies that are common in the local universe . in contrast , blind hi surveys do not suffer from this strong bias ( rosenberg & schneider 2002 ; koribalski _ et al . _ a more thorough discussion of the completeness of the agc hi compilation in the context of the derivation of the hi mass function is presented in springob , haynes & giovanelli ( 2005 ) . despite the lack of flux completeness , the redshift , emission line flux , and observed rotational widths available from the global hi line profiles collected here provide important measures of fundamental galaxy properties useful to diverse studies of galaxies in the nearby universe , especially when combined with complementary multiwavelength data . in addition to the tabulations contained herein , the actual smoothed and baseline subtracted spectra are available in digital form by contacting the authors . plans are underway to make the entire dataset availble through the us national virtual observatory in conjunction with a wide area blind hi survey to be undertaken at arecibo beginning in 2005 . this _ arecibo legacy fast alfa _ ( alfalfa ) survey will make use of the new 7feed dual polarization , l - band array ( alfa ) to survey over 7000 deg@xmath126 of the sky visible to arecibo , detecting tens of thousands of hi sources , from local , very low hi mass dwarfs to gas - rich massive galaxies seen to @xmath127 . a careful study of the detection statistics of the present optically selected sample in such a sensitive hi blind survey will provide insight into the differential characteristics of galaxies according to optical surface brightness and hi content . only the mining of datasets derived from wide area surveys conducted at multiple wavelengths will , in combination , provide the true census of all galaxies that occupy the local universe . we wish to thank our numerous collaborators , colleagues , students , telescope staff and friends who have participated over many years and in many places and contexts in the acquisition of the original data used to produce this archive . we especially thank barbara catinella for conducting the fall 2001 arecibo observations , and joe mcmullin , bob garwood , karen oneil , and jeremy darling for their work in adapting aips++ for use with arecibo spectral line data . this work has been partially supported by ast-9900695 , ast-0307396 , ast-0307661 , the n.r.a.o / gbt 03b-007 graduate student support grant and the nasa new york state space grant .

a detailed knowledge about the dynamics of spinning compact binaries is of interest to both astrophysicists and theoretical physicists . the astrophysical interest arises from the highly accurate radio observations of relativistic binary pulsars , which can be used to test general relativity in strong field regimes @xcite . moreover , inspiralling spinning compact binaries are promising sources of gravitational radiation for the ground based and proposed space based laser interferometric gravitational - wave detectors @xcite . the theoretical interest is due to the possibility of constructing numerically spinning compact binary configurations in full general relativity [ see ref . @xcite and references therein ] . further , it is of some interest to explore if the dynamics of spinning compact binaries will be deterministic or not @xcite . in many of the above mentioned scenarios , the dynamics of spinning compact binaries can be described in great detail by the post - newtonian ( pn ) approximation to general relativity . the pn approximation allows one to express the equations of motion of a compact binary as corrections to newtonian equations of motion in powers of @xmath1 , where @xmath2 , @xmath3 , and @xmath4 are the characteristic orbital velocity , the total mass and the typical orbital separation of the binary . for spinning compact binaries , the dynamics is determined , not only by the orbital equations of motion for these objects , but also by the precessional equations for the orbital plane and the spin vectors themselves @xcite . note that for spinning compact binaries , spin effects enter the dynamics for the first time through spin - orbit interaction . therefore , the hamiltonian describing the leading order spin - orbit interaction can be added to the pn accurate conservative hamiltonian which gives the dynamics of nonspinning compact binaries . it is highly desirable to obtain keplerian - type parametric solution to the conservative dynamics of spinning compact binaries . this is so as 1pn accurate quasi - keplerian parametrization , obtained by damour and deruelle @xcite , is heavily employed to analyze radio observations of relativistic binary pulsars and to test general relativity in strong field regimes @xcite . moreover , 2pn accurate keplerian - type parametric solution for nonspinning compact binaries , obtained in refs . @xcite , is heavily employed to construct highly efficient `` ready to use '' search templates for nonspinning compact binaries moving in inspiralling eccentric orbits @xcite . recently , 3pn accurate generalized quasi - keplerian parametrization for the conservative dynamics of spinning compact binaries in eccentric orbits was obtained in refs . more precisely , the pn accurate parametric solutions are available for the following two distinct cases : ( i ) the binary is composed of two compact objects of equal mass and two arbitrary spins , and ( ii ) the two objects have unequal masses but only one object has a spin . the pn accurate parametric solution may be used to describe the following two astrophysical situations . as the measured neutron star ( ns ) masses in ns ns binaries usually cluster around @xmath5 , case ( i ) may be used to describe such ns ns binaries . the case ( ii ) may be applicable in the instance of a neutron star orbiting a much larger , rapidly rotating black hole . in that case , the black - hole spin will dominate the orbital precession , and the neutron star spin may be neglected . the above parametrization , for the two cases considered , shows that the dynamics is integrable and can not be chaotic @xcite . the parametric solution will be employed to construct `` ready to use '' search templates of spinning compact binaries in inspiralling orbits @xcite . further , it will also be useful to construct a fully 2pn accurate `` timing formula '' , required to analyze binary pulsar observations by the square kilometre array ( ska ) @xcite . in this paper , we compute , using our parametrization , the observable dimensionless relativistic periastron advance parameter @xmath0 , relevant for binary pulsars . the expression for @xmath0 that we derive will be applicable , like the parametrization computed in ref . @xcite , only for the two distinct cases mentioned above . our expression for @xmath0 is in agreement with the one obtained by damour and schfer @xcite . it should be noted that damour and schfer employed the evolution equation for the laplace - runge - lenz vector , computed from a hamiltonian , to obtain the spin - orbit contributions to the parameter @xmath0 . in this paper , we employ our generalized quasi - keplerian parametrization , derived in ref . @xcite , to compute the parameter @xmath0 . there are two reasons that motivated us to do the current computation . the 2pn accurate expression for @xmath0 , derived by damour and schfer , valid for arbitrary masses and spins , is extremely important due to the discovery of the double pulsar psr j07373039 @xcite . this is because the accurate timing of the double pulsar over a period of few years should lead to the measurement of the moment of inertia of one of the pulsars @xcite . it is from the measurement of @xmath0 to 2pn order , which includes leading order spin - orbit contributions , that one will be able to measure the moment of inertia of the fast rotating pulsar in this double pulsar @xcite . the measurement of the moment of inertia of a neutron star for the first time will lead to strong constrains on its equation of state @xcite . we compute @xmath0 , as mentioned earlier , using keplerian - type parametrization with the aim of presenting an alternative way of obtaining this observable quantity . a second reason for computing @xmath0 is the existence of certain intriguing features appearing in our parametrization . we observed that the factor that modulates the true anomaly , due to spin - orbit interaction , contains a contribution , independent of the spins themselves , and of 1pn order [ see eq . ( 5.6k ) in ref . we want to make sure that this feature will not appear in the observable parameter @xmath0 , computed using our generalized quasi - keplerian parametrization . the paper is structured in the following manner . in sec . [ sec2 ] , we briefly describe the 3pn accurate keplerian - type parametrization for the conservative dynamics of spinning compact binaries , moving in eccentric orbits , when spin effects are restricted to the leading order spin - orbit interaction , derived in ref . @xcite , and point out its salient features . [ sec3 ] and sec . [ sec4 ] deal with the issues related to the derivation of the parameter @xmath0 . we show for the first time how to derive @xmath0 using the parametrization , described in sec . [ sec2 ] , after describing how damour and schfer obtained @xmath0 . we summarize our results in sec . [ secconclusions ] . appendices [ appendix_binary_geometry ] and [ appendix_lrl_vector ] provide details of our computation and consistency checks . recently , 3pn accurate generalized quasi - keplerian parametrization for compact binaries in eccentric orbits with the first - order spin - orbit interaction was obtained in refs . @xcite for the following two distinct cases : ( i ) the binary is composed of two compact objects of equal mass and two arbitrary spins , and ( ii ) the two objects have unequal masses but only one object has a spin . the underlying dynamics is fully specified by the following pn accurate ( reduced ) hamiltonian @xmath6 , written symbolically as @xmath7 where @xmath8 , @xmath9 , @xmath10 , and @xmath11 are , respectively , the newtonian , first , second and third pn contributions to the conservative dynamics of compact binaries , when the spin effects are neglected . the leading order spin - orbit coupling to the binary dynamics is given by @xmath12 . in the above equation , @xmath13 , @xmath14 , and @xmath15 , where @xmath16 and @xmath17 are the relative separation vector and its conjugate momentum vector , respectively . the familiar symbols @xmath3 and @xmath18 have the usual meaning , namely the total mass and the reduced mass . the explicit expressions for pn corrections , associated with the motion of nonspinning compact binaries , were obtained in refs . the spin - orbit contribution , available in refs . @xcite , reads @xmath19 where the effective spin @xmath20 is defined by @xmath21 with @xmath22 the finite mass ratio @xmath23 is given by @xmath24 . the ( reduced ) spin vectors @xmath25 and @xmath26 are related to the individual spins @xmath27 and @xmath28 by @xmath29 and @xmath30 , respectively . we recall that @xmath16 , @xmath17 , @xmath27 , and @xmath28 are canonical variables , such that the orbital variables commute with the spin variables , e.g. , see refs . @xcite . in ref . @xcite , we obtained the keplerian - type parametrization in an orbital triad @xmath31 . in this triad , the unit vector @xmath32 is along the line of nodes , associated with the intersection of the orbital plane with the invariable plane @xmath33 , which is the plane perpendicular to the total ( reduced ) angular momentum @xmath34 , the only constant vector in our analysis . the unit vector @xmath35 is defined by @xmath36 , where @xmath37 is the ( reduced ) orbital angular momentum vector , and hence @xmath35 is perpendicular to the orbital plane ( see fig . [ fig:3fullplanes ] ) . we display below , in its entirety , the parametric solution to the conservative 3pn accurate dynamics of spinning compact binaries moving in eccentric orbits , in adm - type coordinates , with spin effects restricted to the leading order spin - orbit interactions . the dynamical vectors of the problem are given by @xmath38 we note that in the equal - mass case ( i ) the ( reduced ) total spin @xmath39 is the sum of two arbitrary spins @xmath25 and @xmath26 , and in the single - spin case ( ii ) @xmath39 is either @xmath25 _ or _ @xmath26 . in our restricted cases , namely ( i ) and ( ii ) , the ( reduced ) orbital angular momentum @xmath40 and the ( reduced ) total spin @xmath39 precess about the conserved ( reduced ) total angular momentum vector @xmath41 at the same rate @xcite . this configuration is called _ simple precession _ in the literature , e.g. , see ref . @xcite . the time dependent basic vectors @xmath31 are explicitly given by [ eq : ijk_in_the_combined_sol ] @xmath42 the angle @xmath43 denotes the constant precession cone angle of @xmath44 around @xmath34 and is given by [ eq : theta_in_terms_of_slj ] @xmath45 where the magnitude of the total angular momentum is given by @xmath46 and @xmath47 is the constant angle between @xmath40 and @xmath39 . the time evolution for @xmath48 , @xmath49 , and @xmath50 is given in a parametric and pn accurate form , which reads [ eq : finalparamadm_3pn_plus_so ] @xmath51 \,.\end{aligned}\ ] ] the pn accurate expressions for the orbital elements @xmath52 , @xmath53 , @xmath54 , @xmath55 , @xmath56 , and @xmath57 and the pn orbital functions @xmath58 , @xmath59 , @xmath60 , @xmath61 , @xmath62 , @xmath63 , @xmath64 , @xmath65 , @xmath66 , @xmath67 , @xmath68 , and @xmath69 , in terms of @xmath70 , @xmath71 , @xmath72 , @xmath23 , and @xmath47 , are available in ref . @xcite . the mass dependent coupling constant @xmath73 connects the effective spin with the total ( reduced ) spin via @xmath74 and is given by @xmath75 we display below the 3pn accurate expression for the _ internal _ parameter @xmath56 , appearing in eq . [ this parameter was denoted by @xmath0 in ref . @xcite , see eqs . ( 5.5c ) and ( 5.6k ) there ] . it consists of two parts : the first one due to the 3pn accurate conservative dynamics of nonspinning point masses and the second one due to the leading order spin - orbit interaction @xmath76 where [ eq : parameter_k_3pn_and_k_so ] @xmath77 \biggr\ } \ , , \\ \label{eq : just_parameter_k_so } \tilde{k}_\text{so } & = \frac{3}{c^2 l^2 } \biggl\ { - \frac { \chi_\text{so } } { 3 } - \chi_\text{so } \frac{s}{l } \cos\alpha \biggr\ } \,.\end{aligned}\ ] ] it should be noted that @xmath78 gives the observable advance of periastron in the case of nonspinning compact binaries . this results from the construction of @xmath56 [ see sec . v in ref . @xcite for details and ref . however , when spin - orbit interactions are added to the dynamics of compact binaries , @xmath56 does not directly lead to the measurable advance of periastron , relevant for binary pulsars . this parameter @xmath56 is a measure for the advance of periastron _ only inside _ our parametrization and has to be related to the observational one . further , note the presence of the @xmath73 term ( which does not depend on the spins ) , appearing in @xmath79 at the 1pn order . it is due to these points , namely the @xmath73 term that appears at the 1pn order and @xmath56 does not give the observable periastron advance for spinning compact binaries , that we compute the dimensionless observable periastron advance parameter @xmath0 for spinning compact binaries in this paper . note , that we have corrected the sign in front of @xmath80 in @xmath79 , important for the proper nonspinning limit and the following calculation of the observable periastron advance parameter @xmath0 in sec . [ sec4 ] . before we go onto the details of computing @xmath0 , let us list a few salient featurs of the parametrization . the parameter @xmath81 , appearing in eq . , is given by the newtonian contribution to say @xmath82 . note that at newtonian order all three eccentricities @xmath82 , @xmath83 , and @xmath84 , are identical to each other and at pn orders they are related by pn accurate relations [ see eqs . ( 5.7 ) in ref . therefore , it is possible to describe the binary dynamics in terms of one of the eccentricities . we emphasize that , while adapting the parametrization to describe the binary dynamics , care should be taken to restrict orbital elements to the required pn order . we are aware that the above parametrization does not lead to the classic keplerian parametrization simply by putting @xmath85 . the detailed explanation of the nonspinning limit and the apparent lack of a _ simple _ nonspinning limit is given in sec . iv b in ref . @xcite . the above parametric solution is usually refered to as the genaralized quasi - keplerian parametrization for the pn accurate orbital dynamics of spinning compact binaries . it extends the computations of damour , deruelle , schfer , and wex @xcite . finally , we want to emphasize that the parametric solution , given by eqs . , describes the entire conservative pn accurate dynamics of a spinning compact binary in an eccentric orbit , when the spin effects are restricted to the leading order spin - orbit interaction . this means the parametrization consistently describes not only the precessional motion of the orbit inside the orbital plane , but also the precessional motions of the orbital plane and the spins themselves . further , the deterministic nature of the dynamics described by the hamiltonian , given by eq . and treated as a self - contained dynamical system , naturally follows from the above parametric solution . in the next section , we develop the basics required to compute @xmath0 from our parametrization . in this section , we will introduce two new triads , in terms of which it will be easy to obtain @xmath0 . we will also discuss in detail how to connect these triads and hence the parametrization expressed in terms of these triads . as we are interested in computing @xmath0 using our parametrization , it is natural that we introduce a periastron - based triad @xmath86 . later , we will express the relative separation vector @xmath87 , given by eq . , in terms of this new triad . let us introduce this periastron - based triad in a pedagogical way . we start by writing the angle @xmath88 , given by eq . , in the following way @xmath89 remember , @xmath88 is the polar angle inside the orbital plane measured from the line of nodes @xmath32 , associated with the intersection of the orbital plane with the invariable plane @xmath33 , and @xmath2 is the `` true anomaly '' , i.e. , the angle between the ( instantaneous ) periastron and the position of the relative separation vector @xmath87 . the angle @xmath90 may be considered as the `` argument of the periastron '' , i.e. , the angle between the line of nodes @xmath32 and the ( instantaneous ) periastron . by comparing eqs . and , we can easily infer the parametric evolution for @xmath90 . therefore , it is obvious that the angle @xmath90 contains via @xmath56 the secular contributions to the periastron advance . inserting eq . into eq . and making use of familiar trigonometric relations leads to @xmath91 where the new basic vectors are given by @xmath92 now , it is obvious that @xmath93 is the unit vector pointing towards the ( instantaneous ) periastron , as @xmath87 is periodic in the angular variable @xmath2 with period @xmath94 in these new basic vectors @xmath93 and @xmath95 . hence it is straightforward to define the periastron - based orthonormal triad @xmath96 , by rotating the @xmath31 frame around the orbital angular momentum unit vector @xmath35 , by an angle @xmath90 . this gives us @xmath97 where @xmath90 is the angle between @xmath32 and the ( instantaneous ) periastron unit vector @xmath93 , as explained above . the periastron - based vectors @xmath98 are co - rotating with the precessing ( instantaneous ) newtonian ellipse . this is similar to the behaviour of a body fixed reference system usually encountered in classical mechanics @xcite . hence , the set of angles @xmath99 is equivalent to the eulerian angles defined there . in the next subsection , we will introduce an observer - based triad and investigate how to connect the observer and periastron - based triads . the second new triad that we introduce is a fixed one associated with the observer on earth and denoted by @xmath100 . in this fixed triad @xmath101 and @xmath102 are two orthogonal unit vectors in the plane of the sky , i.e. , the plane transverse to the radial direction @xmath103 , pointing from the source ( compact binary ) to the observer ( see fig . [ fig:3fullplanes ] ) . in order to compute @xmath0 using our parametrization , it is natural to expect that we need to connect the periastron - based triad @xmath86 to the observer triad @xmath100 . it is possible to connect these two frames with the help of rotational matrices quite similar to the way in which the @xmath31 frame was connected to the @xmath100 frame in ref . @xcite [ see sec . vi in ref . @xcite for details ] . in this paper , we come from the observer triad @xmath100 to the periastron - based triad @xmath86 with the help of _ four _ successive rotations : first , a rotation around @xmath101 via the _ constant _ angle @xmath104 to come to the reference triad @xmath105 , second , a rotation around @xmath106 by an angle @xmath107 , third , a rotation around the direction of the line of nodes @xmath32 [ associated with the intersection of the orbital plane with the invariable plane @xmath33 ] by an angle @xmath43 , and , finally fourth , a rotation around the orbital angular momentum direction @xmath35 by an angle @xmath90 . in terms of rotational matrices denoted by @xmath109 $ ] we have @xmath110 \ ; { \cal r}[\theta , { \boldsymbol{i}}(t ) ] \ ; { \cal r}[\upsilon(t ) , { \boldsymbol{e}}_{z } ] \nonumber \\ & \quad \times { \cal r}[i_{0 } , { \boldsymbol{p } } ] \begin{pmatrix } { \boldsymbol{p } } \\ { \boldsymbol{q } } \\ { \boldsymbol{n } } \end{pmatrix } \,,\end{aligned}\ ] ] where the first rotation appears at the right - hand side , the second one appears left from the one before , and so on . we note , that the approach to generate the @xmath86 frame via these rotations may be also adapted for scenarios , where the angle @xmath43 depends on time . the constant behaviour of @xmath43 is a direct consequence of choosing two distinct cases ( i ) and ( ii ) , where our parametrization is valid and hence we will use the fact that @xmath43 is constant only at the end of our analysis . it should be noted that there are other possible ways to connect the orthonormal triads @xmath100 and @xmath86 . a second way to generate the @xmath86 frame is described in great detail by damour and schfer in sec . 5 of ref . @xcite and involves _ three _ rotational matrices . this way is closely linked to the observation as it involves rotational matrices defined with the help of the observational longitude of periastron @xmath111 and the orbital inclination angle @xmath112 . damour and schfer connected the observer triad @xmath100 to the periastron - based triad @xmath86 via the following _ three _ successive rotations : first , a rotation around @xmath103 by an angle @xmath113 , second , a rotation around the direction of the line of nodes @xmath114 [ associated with the intersection of the orbital plane with the plane of the sky @xmath115 by an angle @xmath112 , and , finally third , a rotation around the orbital angular momentum direction @xmath35 by an angle @xmath111 . the corresponding rotational matrices read @xmath116 \ ; { \cal r}[i(t ) , { \boldsymbol{i}}_{i}(t ) ] \ ; { \cal r}[\omega(t ) , { \boldsymbol{n } } ] \begin{pmatrix } { \boldsymbol{p } } \\ { \boldsymbol{q } } \\ { \boldsymbol{n } } \end{pmatrix } \,.\end{aligned}\ ] ] we note , that for the unit vector @xmath114 along the line of nodes , associated with the intersection of the orbital plane with the plane of the sky , the subscript @xmath112 is employed as @xmath114 is related to the orbital inclination angle @xmath112 and to avoid any confusion with the unit vector @xmath32 . in the next section , we obtain @xmath0 with the help of the parametrization and the ideas discussed in this section . in this section , we will compute @xmath0 , the observable dimensionless periastron advance parameter due to @xmath56 . following damour and schfer , the pn accurate total observable dimensionless periastron advance parameter @xmath0 can be written as @xmath117 where @xmath118 is the usual pn accurate periastron advance parameter up to 3pn order , corresponding to the nonspinning case [ see refs . @xcite for the 2pn and 3pn contributions , respectively ] , and @xmath119 denotes the contribution due to spin - orbit interaction . it should be kept in mind that the 3pn contribution to @xmath118 will not be required while probing the dynamics of relativistic binary pulsars even with ska . let us first briefly summarize how damour and schfer obtained @xmath0 , especially @xmath119 , in the next subsection . timing of binary pulsars allows via the observed `` roemer time delay '' the direct measurement of the argument of periastron @xmath111 @xcite . therefore , damour and schfer defined the relevant _ secular _ periastron advance as @xmath120 , or equivalently the timing parameter ( not to be confused with the orbital angular - momentum unit vector @xmath35 ) @xmath121 where @xmath54 is the mean orbital motion that appears in the pn accurate kepler equation , given by eq . , and @xmath122 denotes the orbital average . damour and schfer computed the contribution to @xmath123 due to spin - orbit interaction in the following way . first , they computed the precessional equations for the orbital angular momentum vector @xmath124 and the laplace - runge - lenz vector @xmath125 , which tracks the periastron advance , with the help of the 2pn accurate version of eq . . these precessional equations where orbital averaged to obtain the rotational velocity vector , denoted by @xmath126 in ref . the expression for @xmath123 was extracted from @xmath126 by decomposing @xmath126 along three _ non - orthogonal _ directions [ see eq . ( 5.11 ) in ref . @xcite ] . in this way a parametric solution to the dynamics of spinning compact binaries was not required to compute the 2pn accurate measurable dimensionless periastron advance parameter @xmath0 in ref . @xcite , and hence the formula of damour and schfer for @xmath0 is more general . in the next subsection , we will compute @xmath123 using our keplerian - type parametric solution and the two different ways of connecting the @xmath100 frame to the @xmath86 frame . this subsection deals with the computation of @xmath123 , arising due to spin - orbit interaction , using our keplerian - type parametrization . following damour and schfer , we introduce the rotation velocity vector @xmath126 , containing all involved angular velocities , and extract @xmath123 from it in the following way . we have already noted that there are two different ways to connect the observer - based triad to the pariastron - based one [ see eqs . and and related discussions there ] . the first way , given by eq . , involves the angular variables @xmath127 , @xmath43 , @xmath50 , and @xmath104 , appearing in our parametrization , and the second way , given by eq . , involves the angles @xmath128 , @xmath129 , and @xmath130 , where two of them are relevant for the observations , namely @xmath128 and @xmath129 . therefore , we can decompose the rotation velocity vector @xmath126 in the following two ways : [ eq : vec_omega_in_both_ways ] @xmath131 note , that the above two different decompositions are along four / three _ non - orthogonal _ directions , respectively . the elementary rotation velocities , appearing in eq . , can be defined with the help of our parametrization , whereas in eq . we have @xmath123 : the rate of change of the argument of the periastron . further note , that eq . is identical to eq . ( 5.11 ) of ref . @xcite . in order to extract @xmath123 and to relate it to the angular variables associated with our parametrization we proceed as follows . since all vectors , appearing in eqs . , can be expressed in terms of @xmath86 , it is clear that the rotation velocity vector @xmath126 may also be written as the sum of three elementary rotation velocities , along three _ orthogonal _ directions @xmath93 , @xmath95 , and @xmath35 , namely @xmath132 it is then straightforward to compute the components @xmath133 , @xmath134 , and @xmath135 in terms of the elementary rotation velocities and angles appearing in eqs . . the components of @xmath126 in the co - rotating periastron - based @xmath86 frame using the parametrization - based variables and the observation - based variables , respectively are given by [ eq : components_of_omega_in_abk ] @xmath136 these _ general _ set of equations provides the transformation from the parametrized angles to the angles of interest for the observation . we note , that the last component , @xmath135 , gives the so - called _ intrinsic _ contribution to the periastron advance @xmath137 , defined by @xmath138 . this is the component of @xmath126 acting _ within _ the orbital plane . however , in the timing of binary pulsars the parameter @xmath0 enters in the ( relativistic ) argument of the periastron @xmath111 , and so we have to extract @xmath123 from the above set of equations . a close inspection of fig . [ fig:3fullplanes ] indicates that @xmath104 is constant , since it connects two invariant reference frames . for the two distinct cases ( i ) and ( ii ) , where our parametrization is valid , we also have @xmath43 as a constant angle . this allows us to come from eqs . to the _ reduced _ set of equations for the components @xmath133 , @xmath134 , and @xmath135 , which reads [ eq : reduced_components_of_omega_in_abk ] @xmath139 solving the last equation , namely eq . , for @xmath123 leads to @xmath140 the first and the second contribution to @xmath123 are directly provided by our parametrization . we also figured out that they can be obtained via the time evolution of the laplace - runge - lenz vector without any parametrization ( see appendix [ appendix_lrl_vector ] for details ) . the third contribution to @xmath123 in eq . , namely @xmath141 , needs a closer inspection . to obtain @xmath142 , we multiply eq . with @xmath143 and add it to the product of eq . times @xmath144 , which results in @xmath145 using now the explicit expressions for @xmath143 and @xmath144 , given by eqs . in appendix [ appendix_binary_geometry ] , in the above equation , we get @xmath146 where @xmath147 , @xmath148 , @xmath149 , and @xmath150 are shorthand notations for the constant trigonometric functions @xmath151 , @xmath152 , @xmath153 , and @xmath154 , respectively . the time evolution for the orbital inclination @xmath112 , appearing in eqs . , is simply given by eqs . in appendix [ appendix_binary_geometry ] . let us now explicitly compute @xmath155 , @xmath156 and hence @xmath142 with the help of our parametric solution . using eqs . , , and for @xmath90 , and eq . for @xmath107 , we display below the relevant parametric expressions symbolically [ eq : display_the_separation ] @xmath157 the structure of eqs . indicates that we can handle the pn and the spin - orbit induced contributions to @xmath123 separately . to obtain the spin - orbit contributions to @xmath158 , we compute with the help of eqs . and the orbital averaged spin - orbit ( so ) contributions to the above required time derivatives [ eq : so_contr_to_pi_upsilon_omega_reduced ] @xmath159 the quantity @xmath160 gives the orbital average of @xmath161 , @xmath162 being the scaled radial separation , to the newtonian order [ see , e.g. , ref . @xcite ] @xmath163 with the help of eqs . , , and , it is straightforward to obtain the spin - orbit induced contribution to @xmath0 , defined by eqs . and , and it reads @xmath164 \,,\end{aligned}\ ] ] where @xmath165 is given in terms of our _ reduced _ variables by @xmath166 the proper nonspinning limit of eqs . and , can be easily deduced as when @xmath167 , @xmath34 is identical to @xmath40 , which implies @xmath168 and @xmath169 , @xmath112 being the angle between @xmath103 and @xmath40 , and therefore @xmath119 vanishes . the disappearance of @xmath119 as @xmath167 indicates that the presence of the 1pn accurate @xmath73 term in @xmath79 , given by eq . , does not have any observational consequence . our result for @xmath119 is in agreement with eqs . ( 5.17 ) and ( 5.17a ) of ref . @xcite , when we specialize these equations to the _ simple precession _ ( for the equal - mass case and for the single - spin case ) @xmath170 \,,\end{aligned}\ ] ] where @xmath171 , @xmath172 and the coefficient @xmath165 is given here in terms of the _ non - reduced _ variables , used by damour and schfer in ref . @xcite and in appendix [ appendix_lrl_vector ] , by @xmath173 the dimensionless mass dependent coupling constant @xmath174 is related to @xmath73 by @xmath175 ( see appendix [ appendix_lrl_vector ] for details ) . to check the above agreement between our result and the one of damour and schfer , we have used eqs . and eq . . the two expressions for the coefficient @xmath165 , namely eqs . and , are identical . let us now briefly consider the nonspinning 3pn accurate contributions to @xmath0 . these contributions , enter only via @xmath90 , orbital averaged analogue to eq . , our computation , and result in @xmath176 . however , it should be noted that the expression for @xmath118 , as given by eq . , is not yet ready for application . this is so as this expression for @xmath118 depends on the conserved _ theoretical _ quantities @xmath70 and @xmath71 rather than on _ observable _ quantities like mean motion @xmath54 and proper - time eccentricity @xmath177 . however , we have checked that up to 2pn order our expression for @xmath118 is in agreement with the one of damour and schfer , obtained in ref . @xcite . therefore , the 3pn accurate observable periastron advance parameter @xmath0 becomes @xmath178 where @xmath118 is the _ 3pn contribution , corresponding to the nonspinning case , given by the 3pn accurate contribution to eq . , namely eq . . the spin - orbit contribution @xmath119 is given by eqs . and or equivalently by eqs . and we recapitulate that @xmath176 and @xmath179 , where @xmath78 and @xmath79 are the corresponding contributions to @xmath56 , as given by eqs . and . this results from the construction of @xmath56 in ref . @xcite by combining the parameter @xmath180 of the nonspinning case with the internal parameter @xmath79 , given by eq . ( 4.45e ) in ref . @xcite , of the spin - orbit parametrization [ see sec . v in ref . @xcite for details ] . finally , let us present the following geometric explanation why @xmath181 . the difference between @xmath56 and @xmath0 results from the fact , that the angle @xmath90 associated with the parameter @xmath56 is measured from @xmath32 and in contrast to that , the angle @xmath111 associated with the observable parameter @xmath0 is measured with respect to @xmath114 . the difference between @xmath111 and @xmath90 is denoted by @xmath182 in fig . [ fig:3fullplanes ] ( see appendix [ appendix_binary_geometry ] for details ) . in addition , we want to point out , that the last rotation @xmath183 $ ] in eq . can be decomposed into two parts @xmath184 & = { \cal r}[\pi(t),{\boldsymbol{k}}(t ) ] \ ; { \cal r}[\delta(t),{\boldsymbol{k}}(t ) ] \,,\end{aligned}\ ] ] and so it is obvious that @xmath185 and correspondingly we have @xmath186 . the contribution of @xmath187 causes the difference between @xmath119 and @xmath79 . we have presented a new method to compute the pn accurate observable dimensionless relativistic periastron advance parameter @xmath0 of a spinning compact binary using our keplerian - type parametric solution to the pn accurate conservative dynamics of spinning compact binaries moving in eccentric orbits @xcite . the expression for @xmath0 is applicable , like our parametric solution , only for the following two distinct cases , namely , case ( i ) , the equal - mass case and case ( ii ) , the single - spin case . our result is consistent with a more general formula for @xmath0 , derived by damour and schfer , using a different procedure @xcite . the computation also helped us to clarify certain intriguing features that we observed in our parametrization . it is our pleasure to thank gerhard schfer for helpful discussions and encouragements . we gratefully acknowledge the financial support of the deutsche forschungsgemeinschaft ( dfg ) through sfb / tr7 `` gravitationswellenastronomie '' . the algebraic computations , appearing in this paper , were performed using maple and mathematica . in this section , we make contact with the variables and angles , commonly used in the literature @xcite . it is obvious that we have three planes in the binary geometry , namely the orbital plane , the invariable plane and the plane of the sky . the associated orthonormal triads are @xmath31 , @xmath105 , and @xmath100 . therefore , we may define three lines of nodes and associated inclination angles in the following way . we infer from fig . [ fig:3fullplanes ] that the plane of the sky meets the invariable plane along a line of nodes , characterized by @xmath188 , as we choose the line - of - sight unit vector @xmath103 to lie in the @xmath189-@xmath190 plane . the associated constant angle @xmath104 between @xmath103 and @xmath191 may be treated as the inclination angle , related to the above line of nodes . in the second case , we consider the line of nodes @xmath32 , created by the intersection of the orbital plane with the invariable plane . the corresponding inclination angle , namely the angle between the orbital angular momentum unit vector @xmath192 and @xmath191 , is denoted by @xmath43 . finally , we consider the intersection of the orbital plane with the plane of the sky . we associate @xmath114 with this line of nodes and the related inclination angle @xmath112 , the angle between @xmath35 and the line of sight unit vector @xmath103 , is the usually defined orbital inclination @xmath112 and hence the use of the subscript @xmath112 in @xmath114 . it is easy to figure out that the unit vectors , associated with the last two lines of nodes , @xmath32 and @xmath114 are defined by @xmath193 using @xmath35 , expressed in terms of @xmath105 and @xmath100 , given by @xmath194 we compute the following expressions for @xmath32 and @xmath114 as @xmath195 \,,\end{aligned}\ ] ] where @xmath196 . the time evolution for the ( observational ) orbital inclination @xmath112 is simply given by [ eq : time_evolution_of_i ] @xmath197^{1/2 } \nonumber \\ & = \left [ 1 - ( c_{i_{0 } } c_\theta - s_{i_{0 } } s_\theta \cos\upsilon ) ^2 \right]^{1/2 } \,.\end{aligned}\ ] ] the equation for @xmath198 is in agreement with eq . ( 6 ) in ref . we may also obtain a differential equation for the time evolution of @xmath112 , as given in ref . @xcite . to get that , we differentiate @xmath199 with respect to time and make use of @xmath200 with @xmath201 , @xmath196 and eq . . in this way we deduce the following differential equation for the orbital inclination @xmath112 : @xmath202 which agrees with eq . ( 5.15 ) of ref . @xcite , whereas we note that the quantity characterizing the _ ascending node _ in ref . @xcite , namely @xmath32 , is identical to our @xmath114 . following ref . @xcite , let @xmath113 be the longitude of the line of nodes @xmath114 , as measured from @xmath101 in the @xmath101-@xmath102 plane ( plane of the sky ) , which implies @xmath203 the time evolution for @xmath113 is defined by either of the following two equations @xmath204 finally , we note that it is also possible to obtain the time evolution for the longitude of periastron @xmath111 , measured from @xmath114 in the orbital plane , as defined in ref . @xcite . to do that , it is advantageous to define a periastron - based orthonormal triad @xmath205 . this is easily obtained by a rotation of the @xmath31 frame around the orbital angular momentum unit vector @xmath35 , by an angle @xmath90 . this gives us @xmath206 where @xmath90 is the angle between @xmath32 and the ( instantaneous ) periastron unit vector @xmath93 . the time evolution of @xmath111 is given by either of the following relations [ eq : cos_omega_sin_omega ] @xmath207 \ , , \\ \label{eq : sin_omega_in_appendix } \sin\omega & = |{\boldsymbol{i}}_{i } \times { \boldsymbol{a}}| = \frac{1}{\sin i } \left [ ( c_{i_{0 } } s_\theta + s_{i_{0 } } c_\theta \cos\upsilon ) \sin\pi \right . \nonumber \\ & \quad\left.+ s_{i_{0 } } \sin\upsilon \cos\pi \right ] \,.\end{aligned}\ ] ] the parametric solution to @xmath90 , which also gives its time evolution , is given by our parametrization via eq . . in addition , the parametric solution to @xmath90 can be also obtained by noting that @xmath208 , where @xmath125 is the laplace - runge - lenz vector given by @xmath209 \,.\end{aligned}\ ] ] further details are described in appendix [ appendix_lrl_vector ] . eq . for @xmath143 is in agreement with eq . ( 6 ) in ref . the additional check @xmath210 leads to @xmath211^{1/2 } \,,\end{aligned}\ ] ] which is equivalent to eq . ( using trigonometric relations ) . since @xmath114 , @xmath32 , and @xmath93 lie in the orbital plane , we may define @xmath182 such that it is the angle between @xmath114 and @xmath32 . this implies that @xmath212 ( see fig . [ fig:3fullplanes ] ) and the time evolution for @xmath182 is given by @xmath213 finally , let @xmath214 be the polar angle of @xmath87 , measured from @xmath114 , as defined by damour and schfer in ref . @xcite . as we define @xmath88 from our @xmath32 , this leads to @xmath215 . using these inputs , it is possible to write the relative separation vector @xmath87 in @xmath98 in terms of @xmath214 and @xmath111 as @xmath216 this is consistent with eqs . ( 5.5b ) , ( 5.6 ) , and ( 5.10 ) in ref . @xcite . in our angular variables , . reads @xmath217 in this section , we present an elegant way , starting from the famous laplace - runge - lenz vector , to obtain the time evolution of the apsidal motion without any parametrization . the quantities in this section are non - reduced for a simple comparism with the work of damour and schfer @xcite , and the work of wex and kopeikin @xcite . the first equation gives the orbital angular momentum vector @xmath124 and the second one the apsidal vector @xmath125 , which is the famous supplementary first integral of the kepler problem . the vector @xmath125 is usually named after runge and lenz or laplace , though he was first discovered by lagrange . where @xmath222 denotes the poisson brackets and @xmath223 is the full hamiltonian for the relative motion of our binary system . as @xmath124 and @xmath125 are first integrals of @xmath224 , only @xmath225 contribute to the right - hand sides of eqs . . the effects of @xmath226 are already studied in the literature , and so we just want to consider the contribution of @xmath227 where the effective spin @xmath228 is given by @xmath229 we recall that @xmath16 , @xmath17 , @xmath27 , and @xmath28 are canonical variables , such that the orbital variables commute with the spin variables , e.g. , see refs . @xcite . using the basic poisson brackets @xmath230 , @xmath231 , @xmath232 , @xmath233 , we find after orbital averaging these expressions , we can write down the equations for the precessional motion , induced by the spin - orbit interaction , in the vectorial form with the help of the rotational velocity vector @xmath235 in case of the so - called _ simple precession _ , where the orbital angular momentum @xmath124 and the total spin @xmath239 precess about the conserved total angular momentum vector @xmath240 at the same rate @xcite , the precession of the orbital plane and the longitude of periastron are best described by the angles @xmath107 and @xmath90 , defined with respect to the invariable plane , i.e. , the plane perpendicular to @xmath241 ( see fig . [ fig:3fullplanes ] for details ) : where @xmath243 denotes the unit vector in the direction of the orbital angular momentum vector @xmath124 . the conditions for the _ simple precession _ are fulfilled in our restricted cases , namely ( i ) , the equal - mass case and ( ii ) , the single - spin case . in these cases the effective spin @xmath228 and the total spin @xmath239 are related by @xmath244 , where the mass dependent coupling constant is given by @xmath245 note , that @xmath175 . using @xmath244 in eq . and comparing eqs . with an averaged version of eqs . , we obtain the precession in terms of the angles @xmath107 and @xmath90 following urls provide a wealth of information about the ground and space based gravitational wave interferometers : http://www.ligo.caltech.edu , http://www.virgo.infn.it , http://www.geo600.uni-hannover.de , http://tamago.mtk.nao.ac.jp , and http://lisa.nasa.gov .

our universe is filled of matter , and no macroscopic quantity of antimatter can be found . the asymmetry between the number of matters against that of antimatters is expressed in terms of the matter - photon ratio @xmath1 , where the number of photons is the synonym of that of the pairs of a particle and its antiparticle which have populated the early universe , and recent observation of planck is giving @xmath2 @xcite . to generate the baryon number asymmetry in our universe , several conditions have to be satisfied . those criteria were first given by sakharov , and the requirements are ( 1 ) the existence of baryon number violation , ( 2 ) the violation of the discrete @xmath3 and @xmath4 symmetries , and ( 3 ) the departure from thermal equilibrium @xcite . attempts to explain the baryon number asymmetry within the standard model ( sm ) have been made @xcite , but it is now known that the cp violation brought by the cabibbo - kobayashi - maskawa ( ckm ) matrix @xcite is in great deficit . the existence of matters around us is thus suggesting the existence of new physics beyond standard model ( bsm ) , and the search for new cp violating sources beyond it is now one of the most important subjects of particle physics . the search for cp violation , since its first observation in @xmath5 meson decays @xcite , was performed in many systems . one of the most attractive observables in this context is the _ electric dipole moment _ ( edm ) @xcite . the search for edms has several notable advantages . the first one is its accurate experimental measurability . it does not involve final state interaction effects , which are the main background mimicking the cp - odd observables in decays or reactions of particles . the ckm contribution to the edm is also known to be very small from several analyses . those two facts greatly reduce the source of systematics of edm experiments , and help us to improve the experimental accuracy beyond that of accelerator based experiments . the second advantage is that the edm can be measured in various systems , and that composite systems may enhance the fundamental level cp violation through the nontrivial many - body effects . the third strong point is that in many cases , the cost to prepare experiments are cheaper than other approaches , such as the accelerator based ones . owing to those advantages , the edm was extensively studied in experiments . the principle of the measurement of the edm is to observe the spin precession under an electric field @xcite . the edm was so far measured in charge neutral systems , such as the neutron @xcite , atoms and molecules@xcite , for which the spin precession frequency can be probed with high accuracy by applying simultaneously magnetic and electric fields . for the measurement of the edm of charged systems , the same method does not work due to the acceleration by the electric field , but it is possible to apply an effective electric field to them by rotating them under controlled magnetic and electric fields @xcite . recently , new techniques to measure the edm of charged particles using storage rings are becoming available @xcite , and the measurement of the edm of light nuclei is receiving much attention . the measurement of the nuclear edm has several notable advantages . first , schiff s screening phenomenon @xcite , encountered in the atomic systems , is not relevant for a bare nuclear system , and important suppression of the nuclear level cp violation is avoided . the second strong point is the prospective experimental sensitivity of @xmath6 cm , which is well below the current experimental lower limit of the neutron edm @xmath7 cm @xcite . finally , we have the possibility of a nuclear level enhancement of the cp violation due to the many - body effect @xcite . to discover new sources of cp violation , the realization of experiments sensitive to them are required . for that , finding nuclear systems where the cp violation is enhanced is esential . an important work is therefore to evaluate with a reasonable accuracy the nuclear many - body effect and theoretically find nuclei with large enhancement . we must also theoretically control the nuclear edm of various nuclei to disentangle the contribution of the new physics which lies in a very wide parameter space . in this review we present recent developments on the theoretical investigations of the edm of light nuclei in these directions . this review is organized as follows . we first define the nuclear edm and introduce the nucleon level cp violating interaction . we then show and analyze in section [ sec : intrinsicnucleonedm ] the results of the calculation of the intrinsic nucleon edm contribution to the nuclear edm . in section [ sec : abinitio ] , we present the result of ab initio investigations of the edm of the deuteron , @xmath8h , and @xmath8he . in section [ sec : cluster ] , we discuss the theoretical evaluation of the edm of light nuclei in the cluster model . we then present the strong cp problem caused by the @xmath0-term and its effect on the nuclear edm as the uncertainty of sm . to ensure that the sm contribution due to the ckm matrix is a negligible background , we discuss it in section [ sec : ckmedm ] . we then analyze in section [ sec : prospects ] the constraints on new physics bsm and future prospects on the study of the nuclear edm . finally , we summarize the discussion . there are three leading nucleon level cp violating processes which contribute to the nuclear edm : * the intrinsic nucleon edm . * the nuclear polarization due to the cp - odd nuclear force . * the exchange current effect . the schematic picture of the breakdown of the nuclear edm is shown in fig . [ fig : nuclear_edm_breakdown ] . diagrammatic representation of the breakdown of the nuclear edm : ( a ) intrinsic nucleon edm contribution , ( b ) nuclear polarization due to the cp - odd nuclear force , ( c ) exchange current effect . single and double lines denote the nucleon and nucleus , respectively . the wavy lines are the electromagnetic current insertion which probes the edm . , width=453 ] we first define the contribution of the intrinsic nucleon edm to the nuclear edm . the nucleon edm is proportional to the nucleon spin , and its effect is given by @xmath9 where @xmath10 is the polarized nuclear wave function ( in the @xmath11-axis ) . the proton and neutron edms are denoted by @xmath12 and @xmath13 , respectively . the nucleon spin matrix elements @xmath14 and @xmath15 can be calculated in nuclear physics , and they only depend on the nuclear structure . in this review , they are also called `` enhancement factors '' or `` spin quenching factors '' . the next effect is the nuclear polarization generated by the cp - odd nuclear force . this contribution is defined by @xmath16 where @xmath17 is the polarized nuclear wave function . the coordinate in the center of mass frame and the isospin pauli matrix of the @xmath18th nucleon are denoted by @xmath19 and @xmath20 , respectively . this nuclear polarization is induced only if there is a mixing of parity and cp in the nuclear wave function @xmath17 . finally , the exchange current contribution is given by @xmath21 where @xmath22 is the two - body exchange current operator acting on @xmath18th and @xmath23th nucleons . the three - body exchange current operator and so forth can also be defined in a similar manner . the exchange current is composed of the cp - even and cp - odd parts . this effect is suppressed by the power of the nucleon velocity @xmath24 , so we neglect it in this review . the nuclear edm is generated only if there is nucleon level cp violation . the leading order nucleon level cp violation receives contribution from the following p , cp - odd chiral lagrangian @xcite : @xmath25 where @xmath26 and @xmath27 . the pion decay constant is given by @xmath28 mev and the nucleon mass by @xmath29 mev . the intrinsic nucleon edm receives the leading contribution from the bare nucleon edm terms of the chiral effective lagrangian ( [ eq : cpv_chiral_lagrangian ] ) , and also from the pion loop diagram [ see fig . [ fig : cpv_pion - nucleon ] ( a ) ] . this radiative pion cloud effect is known to enhance the nucleon edm @xcite . the isoscalar and isovector nucleon edms are given by @xcite @xmath30 and @xmath31 respectively , where @xmath32 is the nucleon axial coupling @xcite and @xmath33 mev the pion mass . here we have neglected the isospin breaking terms of the pion and nucleon masses . the bare nucleon edms @xmath34 and @xmath35 act as counterterms and cancel the divergence of the pion loop diagram . the renormalization point @xmath36 is set to the hadronic scale , often between 1 gev and 500 mev . at @xmath37 mev , all mesons except pions are integrated out . we consider that the contributions from radiative processes involving @xmath5 mesons and hyperons @xcite are renormalized into @xmath34 and @xmath35 . leading contributions of the cp - odd pion - nucleon interaction to the nucleon level cp violation . ( a ) pion loop contribution to the nucleon edm . ( b ) pion exchange cp - odd nuclear force.,width=377 ] the most important nucleon level cp violating process is the pion - exchange cp - odd nuclear force [ see fig . [ fig : cpv_pion - nucleon ] ( b ) ] . at the leading order , it is generated by combining the cp - odd pion - nucleon interaction with the cp - even one . by taking the on - mass shell approximation of the nucleons , the momentum exchange representation of the leading cp - odd nuclear force in the chiral eft yields @xcite @xmath38\ , { \vc{\sigma}}_{- } \bigg\ } \cdot \hat { \vc{k } } \ , , \label{eq : cpveftpot}\end{aligned}\ ] ] where the subscripts label the nucleon . the unit vector of the exchanged momentum @xmath39 is given by @xmath40 . the spin and isospin pauli matrices are given by @xmath41 , @xmath42 , @xmath43 , and @xmath44 . here the contact couplings @xmath45 and @xmath46 contain the effects from short range physics , including those from heavy mesons with masses above the cutoff @xmath36 . a more phenomenological cp - odd nuclear force based on one meson exchange model can be conceived @xcite . in the coordinate representation , it is given as @xmath47 where the radial function @xmath48 is given by @xmath49 the shape of @xmath50 is shown in fig . [ fig : cpvnn_meson ] for the exchanges of @xmath51 , @xmath52 , @xmath53 , @xmath54 mesons . this cp - odd potential has been used in many previous ab initio analyses @xcite and has served as a benchmark of the ab initio nuclear edm calculations . the radial shape of the cp - odd nuclear force @xmath50 for @xmath55- , @xmath52- , @xmath53- and @xmath54-exchanges . , width=377 ] let us here mention several features of the cp - odd nuclear force ( [ eq : cpvhamiltonian ] ) and their expected impact on the nuclear polarization . the cp - odd nuclear force has the structure of a product between the nucleon density and the three dimensional divergence of the spin density . as it is a spin dependent interaction , nuclei with closed spin shell can not be polarized . the polarization is also dependent on the nucleon density , so that the increase of the nucleon number should enhance the nuclear edm . the cp - odd nuclear force is however an exchange of massive mesons , so its effect is limited in a finite range . as the nuclear density is saturated and the nuclear volume increases for large nuclei , the cp - odd interaction will know a limitation of the nuclear polarization when the nucleon number sufficiently grows . another important feature is the derivative , which makes the cp - odd interaction to be sensitive on the surface of the nucleus . this property suggests that nuclei with a well developed cluster structure may enhance the edm , due to the increase of the density gradient inside the system . the nuclear edm may also be enhanced due to octupole deformations , which facilitate the transition between opposite parity states by decreasing their level spacings in the spectrum @xcite . the p , cp - odd @xmath56 couplings @xmath57 @xmath58 are dimensionless . the isoscalar and isovector couplings can be given from the chiral eft analysis . to the leading order , the cp - odd pion - nucleon couplings and the cp - odd @xmath56 couplings are related by @xmath59 in the chiral analysis , the isotensor cp - odd nuclear force [ term with @xmath60 in eq . ( [ eq : cpvhamiltonian ] ) ] is suppressed by an additional factor of light quark mass , so it is negligible in the analysis of the edm of light nuclei . ( for heavy nuclei , it may be relevant due to the effective interaction generated by the isospin violation of the nuclear medium ) . the chiral eft analysis does not include mesons heavier than the pion , due to the cutoff placed near @xmath61 500 mev . instead , their effect is renormalized into the contact interaction [ terms with @xmath62 in eq . ( [ eq : cpv_chiral_lagrangian ] ) ] . in the hamiltonian ( [ eq : cpvhamiltonian ] ) , we can find terms with the same spin and isospin structures as the contact terms . as the cutoff is lower than the masses of @xmath52 , @xmath53 and @xmath54 mesons , their exchange processes can be approximated as contact interactions . we can therefore match the contact couplings @xmath45 and @xmath46 with those of @xmath52 and @xmath53 exchange , respectively , as @xmath63 the cp - odd chiral lagrangian ( [ eq : cpv_chiral_lagrangian ] ) contains a three - pion interaction which generates the cp - odd three - nucleon force and also additional contribution to the isovector cp - odd pion - nucleon interaction @xcite . the isovector cp - odd pion - nucleon interaction is radiatively generated by the diagram ( a ) of fig . [ fig : three - pion ] . this additional contribution , at the leading order , is given by the momentum dependent vertex function @xmath64 with the momentum dependent function @xcite @xmath65 \biggr\ } , \label{eq : f_g_1}\ ] ] where @xmath66 is the momentum of the outgoing pion . the momentum independent part of @xmath67 ( the first term in the curly bracket ) gives the leading contribution to the nuclear edm . the isovector cp - odd nuclear force is therefore corrected as @xmath68 , \ ] ] where we have neglected the momentum dependence . diagrammatic representation of the contribution of the three - pion interaction to the nucleon level cp violating processes . the solid and dashed lines denote the nucleon and the pion , respectively . ( a ) radiatively generated isovector cp - odd pion - nucleon interaction . ( b ) cp - odd three - nucleon force.,width=377 ] the three - pion interaction [ see fig . [ fig : three - pion ] ( b ) ] also generates the cp - odd three - nucleon force @xcite : @xmath69 \nonumber\\ & & \hspace{8em } \times \frac{(\vc{\sigma}_1 \cdot \vc{k}_1 ) ( \vc{\sigma}_2 \cdot \vc{k}_2 ) ( \vc{\sigma}_3 \cdot \vc{k}_3 ) } { \bigl [ \vc{k}_1 ^ 2 + m_\pi^2 \bigr ] \bigl [ \vc{k}_2 ^ 2 + m_\pi^2 \bigr ] \bigl [ \vc{k}_3 ^ 2 + m_\pi^2 \bigr ] } , \label{eq : cpv3n}\end{aligned}\ ] ] where the subscript of the momentum @xmath39 , isospin and spin matrices labels the nucleon . the above cp - odd three - nucleon force depends on the spin and isospin matrices of three nucleons at the same time . this means that it is active only for configurations where all interacting nucleons are unpaired . as the nucleons have tendency to be paired in singlet , the effect of the cp - odd three - nucleon force is expected to be small for stable nuclei . in this section , we present and analyze the results of the calculations of the intrinsic nucleon edm contribution to the nuclear edm ( [ eq : intrinsicnucleonedm ] ) . the results of the evaluations of the enhancement factors of the intrinsic nucleon edm for the deuteron , @xmath8he , @xmath8h ( ab initio using a@xmath70 @xcite , chiral eft @xcite ) , @xmath71li , @xmath72be , and @xmath73c ( cluster model @xcite ) are listed in table [ table : intrinsic_nucleon_edm_nuclearedm ] . let us now analyze them in detail . in many - body systems , the edm of the constituents may be enhanced if their spins are aligned . it is however known that the nucleons have strong pairing correlations , which tend to minimize the nuclear spin . stable even - even nuclei have an angular momentum @xmath74 without exception . moreover , there are no stable odd - numbered nuclei with aligned nucleon spin with the same isospin . we can not therefore expect enhancements of the nucleon edm effect from the alignment of the nucleon spin . rather , the nuclear spin matrix elements are suppressed by the mixing of angular momentum configurations . due to the spin dependent interactions between nucleons , orbital angular momentum configurations with flipped nucleon spin mix together , and their superposition suppresses the nuclear spin matrix elements . a good example is the intrinsic nucleon edm contribution to the deuteron edm , which is smaller than one for both the proton and the neutron ( see table [ table : intrinsic_nucleon_edm_nuclearedm ] ) . this suppression is due to the mixing of @xmath75-wave , which interferes destructively to the deuteron spin matrix elements @xcite . another possible mechanism of enhancement of the edm of composite systems is the relativistic effects @xcite . this process was especially important in the study of the atomic edm , where the electron edm is enhanced by the strong coulomb potential of heavy atoms @xcite . this effect is however negligible in nuclear systems , as the nucleons are nonrelativistic , at least for light nuclei which are the main target of this review . rather , the nucleon edm may be partially screened by a mechanism analogue to schiff s screening @xcite , encountered in atoms . this suppression effect is of the order of percentage for the deuteron , but may be more important for heavier nuclei . as a general feature , the nuclear spin matrix elements have tendency to become small when the neutron or proton numbers become far from their magic numbers @xcite . the enhancement factors of intrinsic nucleon edm contribution to the nuclear edm are therefore smaller than one for stable nuclei , and we can affirm that the nuclear edm generated by the intrinsic nucleon edm is generally not enhanced . as we can see in table [ table : intrinsic_nucleon_edm_nuclearedm ] , the linear coefficients of the nucleon edm contribution to the nuclear edm ( @xmath76 , @xmath77 ) are all less than one . in this sense , we should call them `` spin quenching factors '' , rather than `` enhancement factors '' . as a simple way to confirm the reliability of the results , we can estimate the enhancement factors of the nucleon edm in the simple shell model @xcite . in the nuclear shell model , the nucleus is considered as a system made of a core and valence nucleons . by assuming a spherical nucleus with a valence nucleon @xmath78 , the edm of odd - numbered nuclei is given by @xmath79 where @xmath23 and @xmath80 are the total and valence nucleon orbital angular momenta , respectively . for even - numbered nuclei with two valence nucleons such as the deuteron or @xmath81li , we just consider that there are two valence nucleons with constructively aligned spins . by comparing the results of the ab initio and cluster model calculations with the estimation in the simple shell model ( see table [ table : intrinsic_nucleon_edm_nuclearedm ] ) , we see that this valence nucleon picture works well in describing the nucleon edm contribution to the nuclear edm . the suppression of the formers against that of the simple shell model results is due to the mixing of angular momentum configurations , as already discussed in this section . the intrinsic nucleon edm contribution to the nuclear edm is important since it can provide effects of isoscalar cp - odd pion - nucleon interaction [ see eq . ( [ eq : d_1 ] ) ] on the nuclear edm , which tends to have stronger isovector dependence ( @xmath82 ) . this is particularly important when we want to probe the effect of the quark edm or the weinberg operator . for detail , see section [ sec : disentangling ] . in the _ ab initio _ evaluation of the nuclear edm , the schrdinger equation of a many - nucleon system is solved using the bare @xmath56 potential without any approximations of many - body physics , and the edms of the deuteron , @xmath8he , and @xmath8h nuclei were considered in previous works . as the phenomenological cp - even realistic nuclear forces , argonne @xmath70 @xcite , reid 93 , nijmegen ii @xcite , and inside non - local outside yukawa tail ( inoy ) @xcite potentials have been used . for the three - nucleon systems , the effect of the three - body force of urbana ix @xcite was also considered . finally , those edms were analyzed in the chiral effective field theory ( eft ) , which is based on the symmetry of qcd . in this case it is possible to analyze order - by - order the hadron level cp violation , and systematically estimate the error . the cp - odd interactions are small perturbations , so the nuclear edm should linearly depend on the them . here we parametrize the edm in terms of the phenomenological cp - odd @xmath56 couplings of eq . ( [ eq : cpvhamiltonian ] ) as @xmath83 , \label{eq : polarizationedmcoef}\ ] ] where we take up to isovector terms . in tables [ table : deuteronedm ] and [ table:3nucleonedm ] are summarized the results of the ab initio calculations of the coefficients @xmath84 @xmath85 of the edm of the deuteron , @xmath8he , and @xmath8h nuclei . the results of the chiral eft are given from the next - to - next - to - leading order analysis @xcite . the most important observation is that the contribution of the pion exchange cp - odd nuclear forces to the deuteron and three - nucleon systems agrees well for the realistic nuclear forces considered . this is not surprising , because the cp - odd polarization is generated by the long range pion exchange process , and the long range part of the nuclear wave functions is well described by the realistic nuclear forces as well as by the chiral eft , which respect the pion exchange . we also see some enhancement of the isoscalar cp - odd nuclear force contribution to the edm of @xmath8he and @xmath8h for the results of chiral eft and inoy potential . the shift due to the three - nucleon force is less than 10% ( compare a@xmath70 only and a@xmath70+uix in table [ table:3nucleonedm ] ) . the estimation of the theoretical uncertainty is available for the chiral eft analysis @xcite . the error bars of @xmath86 for the deuteron and @xmath87 ( @xmath88 ) for the three - nucleon systems are about 10% . they were determined from the variance of the results obtained with five different sets of cutoffs ( cutoff for the evaluation of the nuclear potential and that needed in the many - body calculation ) . for the calculations of the edm of light nuclei using realistic nuclear potentials , there are no firm way to define the error bar . the polarization contribution to the deuteron edm is insensitive to the isoscalar and isotensor cp - odd nuclear forces under the isospin symmetry . if the isospin breaking effect is considered ( e.g. chiral eft ) , the isoscalar cp - odd pion - nucleon coupling @xmath89 also contributes to the deuteron polarization , but this has a large theoretical uncertainty @xcite . this contribution is however smaller than the pion cloud effect of the single nucleon edm [ see eq . ( [ eq : d_0 ] ) ] , so we can neglect it . the isovector cp - odd nuclear force is relevant for the deuteron as well as for the three - nucleon systems . here it is important to note that the effect of the three - pion interaction also contributes through the isovector cp - odd pion - nucleon interaction [ see eqs . ( [ eq : g_1_3pi ] ) and ( [ eq : f_g_1 ] ) ] . another interesting feature is that the cp - odd three - nucleon interaction generated by the three - pion interaction was found to be subleading in the edm of @xmath8he and @xmath8h , compared to the radiative isovector cp - odd nuclear force @xcite . in regards to the phenomenological cp - odd potential , the contributions from the exchange of mesons heavier than the pion ( @xmath90 ) are suppressed . the polarization becomes smaller as the meson mass increases . if the meson is sufficiently heavy , the contact interaction approximation can be applied . for those contributions , the results become more dependent on the realistic nuclear force chosen . the coefficients vary from the first digit , and the theoretical uncertainty is @xmath91 ( see tables [ table : deuteronedm ] and [ table:3nucleonedm ] ) ) . this large uncertainty is due to the lack of knowledge of the short range part of the nuclear force . the effect of heavy meson exchange is integrated out in the chiral eft analysis , and this effect is renormalized into the contact interactions [ see eq . ( [ eq : cpv_chiral_lagrangian ] ) ] . in table [ table:3nucleonedm ] , we have displayed their effects by using the approximations of eqs . ( [ eq : contact1eta ] ) and ( [ eq : contact2rho ] ) . the calculation of the effect of contact interactions on the nuclear edm is affected by a large theoretical uncertainty for the same reason as that of the heavy meson exchange cp - odd nuclear force , and the results are very unstable @xcite . we should therefore consider their values as the error bar of the edm calculation . as we have seen in section [ sec : nucleonlevelcpv ] , the operator form of the cp - odd nuclear force ( [ eq : cpvhamiltonian ] ) is suggesting that the nuclear edm is enhanced for nuclei with well developed cluster structure . for that , we have to study the polarization effect of nuclei beyond three - nucleon systems . in this section , we introduce the cluster model and present the theoretical evaluation of the edm of @xmath81li , @xmath92be and @xmath73c . the computational cost of calculating the wave function of @xmath93-body systems is known to become exponentially difficult in growing @xmath93 . the evaluation of the nuclear wave functions of light nuclei is already difficult for @xmath81li or @xmath94li . in nuclear physics , however , the @xmath95 cluster is very stable , and has a strong correlation even inside nuclei . for light nuclei , the cluster structure is actually known to be well developed ( e.g. the @xmath81li nucleus is given as an @xmath96 system ) . the idea to treat light nuclei with @xmath95 clusters has been very successful in describing low lying spectra , and the cluster model has so far extensively been investigated @xcite . if we model light nuclei in terms of @xmath95 clusters , the computational cost greatly decreases , and may help us to evaluate their edm . to calculate the wave function of light nuclei in the cluster model , we have to determine the interactions involving clusters . the common concept is to construct an effective @xmath56 interaction which respect the model space of the nucleon inside the cluster , and apply the folding . for the @xmath97 or @xmath98 interactions , the effective interaction is determined from the experimental data of low energy @xmath99 or @xmath98 scattering @xcite . to reproduce the energy spectrum of light nuclei , the phenomenological three- and four - body interactions between @xmath95 clusters and/or nucleons are also introduced . another critically important feature of the cluster model is how to model the effect of pauli exclusion principle between nucleons of different clusters . the pauli exclusion is taken into account via the _ orthogonality condition model _ ( ocm ) , which consists of manually projecting out the forbidden states @xcite . in the few - body cluster model , it is convenient to remove pauli - forbidden states by including in the hamiltonian the pauli - blocking operator @xcite @xmath100 for example , the states @xmath101 and @xmath102 are removed for the @xmath103 and @xmath104 systems , respectively . in practice , the forbidden states are approximately given by the energy eigenfunctions of the harmonic oscillator potential . the coupling @xmath105 is typically set as @xmath106 mev . by excluding forbidden states , ocm can prevent nuclear clusters from collapsing , and can reproduce the saturation of the nuclear density . meanwhile , low lying energy eigenstates with well developed clusters , relevant near thresholds , can also well be described . the cluster model is therefore very adequate in analyzing the low energy dynamics of light nuclei , comprising shell and cluster correlations . it has been extensively applied in the study of the structure of multi-@xmath95 clustered nuclei , such as @xmath107be @xcite , @xmath108c @xcite , or @xmath109o @xcite , and their description was successful . the model is also well applicable to odd - numbered nuclei with a valence nucleon and several @xmath95-clusters , such as the @xmath73c nucleus . the analysis of the structure of @xmath73c in the @xmath110 cluster model with ocm yields bound states with a shell - like configuration and excited states with an enhanced @xmath92be + @xmath95 correlation , well reproducing the observed spectrum @xcite . in calculating the edm of light nuclei in the cluster model , the cp - odd intercluster force also has to be modeled . for light nuclei where the @xmath95 cluster is a relevant degree of freedom , we have to calculate the cp - odd @xmath103 interaction . here we present the derivation of the cp - odd @xmath103 interaction by folding the cp - odd @xmath56 interaction . we must note that only the isovector p , cp - odd nuclear force contributes to the nuclear edm , since the isoscalar and isotensor cp - odd interactions require the spin and isospin flips for both interacting nucleons , which excite the @xmath95 cluster . moreover , all cp - odd nuclear forces vanish for the @xmath98 interaction due to the closed shell . by only considering the direct contribution ( no exchange of nucleons between clusters ) to the @xmath103 system , the folding of the pion exchange cp - odd nuclear force yields the potential @xmath111 where @xmath112 is the coordinate of the nucleon with the origin located in the center of mass of the @xmath113he nucleus . the radial function of the cp - odd @xmath56 potential is given in eq . ( [ eq : barecpvnn ] ) . to fold the potential we must remove the unphysical effect of the center of mass motion of the @xmath114 system ( see fig . [ fig : cm_shift ] ) . the effective density of the @xmath95 cluster @xmath115 , approximated by a gaussian , must therefore be written as @xmath116 with the oscillator constant @xmath117 @xmath118 . the factor @xmath119 in front of @xmath120 accounts for the center of mass of the @xmath95 cluster . removal of the spurious center of mass effect in the folding of the cp - odd @xmath114 interaction . , width=453 ] by substituting @xmath121 , the folding potential is transformed as @xmath122 . \label{eq : cpvnn_folding}\end{aligned}\ ] ] the remaining radial integral can not be performed analytically . by numerically integrating the above integral , we obtain the curve of fig . [ fig : folding ] . the radial shape of the bare pion exchange cp - odd nuclear force @xmath123 and its folding potential @xmath124 . , width=302 ] we see that the cp - odd @xmath125 interaction has a maximal point around 1 fm , and decreases exponentially in growing @xmath126 . this bump is consistent with the property of the derivative interaction , for which the change of the nuclear density becomes important at the surface , as explained in section [ sec : nucleonlevelcpv ] . the interaction range is also extended due to the nuclear density distribution of the @xmath95 cluster . the cp - odd @xmath125 interaction can not be obtained by naively multiplying the bare cp - odd @xmath56 interaction by four due to the above properties . here we have to note that in eq . ( [ eq : cpvnn_folding ] ) , we have used the bare cp - odd @xmath56 interaction . to be exact , the folding of the bare @xmath56 potential is not respecting the model space of the @xmath114 system which has less degrees of freedom , and the use of the effective cp - odd @xmath56 interaction is actually required to control the systematics . it is expected that the long range part of the pion exchange cp - odd nuclear force is little affected by this change of model space , but the short range forces must be modified to a large extent , since they involve energies which may exceed the model space . it is therefore not possible to treat the cp - odd nuclear forces with heavy meson exchange , or the contact cp - odd interactions in the folding without deriving their effective interactions . in this review , we only consider the folding of the pion exchange cp - odd nuclear force . we also point that the isovector cp - odd @xmath114 interaction receives contribution from the three - pion interaction through eqs . ( [ eq : g_1_3pi ] ) and ( [ eq : f_g_1 ] ) . on the contrary , the cp - odd three - nucleon force ( [ eq : cpv3n ] ) is not relevant since the nucleons in the @xmath95 cluster are in closed shell . for the study of light nuclei in the cluster model , the latter can therefore be neglected . we now give the result of the evaluation of the edm of light nuclei in the cluster model . the nuclear edms of @xmath81li , @xmath92be , and @xmath73c have been calculated so far . the @xmath81li and @xmath92be were treated as @xmath127 and @xmath128 three - body systems , respectively @xcite . the @xmath73c was considered as an @xmath129 four - body system @xcite . the edm of those few - body systems were calculated in the gaussian expansion method , which can solve few - body schrdinger equations very accurately @xcite . the results are shown in table [ table : nuclearedm ] . let us see them in detail one by one . the @xmath81li is the lightest stable nuclear system with non - zero total angular momentum which can be found after the three - nucleon systems . the @xmath81li edm is well described by the @xmath130 cluster structure , and its edm inherits the property of that of the deuteron . the intrinsic nucleon edm contribution to @xmath81li is very similar to the deuteron . concerning the polarization by the cp - odd nuclear force , the @xmath81li is polarized by the deuteron subsystem , and also by the cp - odd @xmath97 interaction . those two effects are comparable , and their positive interference enhances the sensitivity of @xmath81li edm on the isovector cp - odd pion exchange nuclear force . this case is an example which shows the enhancement of the cp violation by the nuclear cluster structure . there are additional subdominant effects such as the isoscalar or isotensor cp - odd nuclear forces . those effects vanish in the cluster model setup with isospin symmetry . they may become relevant when the excitation of the @xmath95 cluster is considered , but this effect requires additional energy of about 20 mev , which is much larger than the @xmath81li binding energy 3.7 mev , and therefore negligible . as for the deuteron , the leading isoscalar effect should be brought by the intrinsic nucleon edm contribution . in ref . , the effect of exchanges of heavier mesons [ eq . ( [ eq : cpvhamiltonian ] ) ] was also considered . for the @xmath81li , the @xmath52 and @xmath53 exchange cp - odd nuclear forces contribute only to the polarization of the @xmath131 subsystem . this can be understood by the close values of the isovector coefficients of the @xmath81li nucleus with those of the deuteron . the @xmath54 exchange , however , also contributes to the polarization of the @xmath114 subsystem . as it was mentioned previously , a simple folding of heavy meson exchange processes yields results with unknown systematics due to the change of the model space , so we do not consider the @xmath54 exchange . we also mention the accuracy of this cluster model evaluation . in the cluster model used , the root mean square distance of the @xmath114 subsystem is about 4 fm . at this distance , the variation of the cp - odd @xmath114 interaction is mild , and the folding potential should well describe it . it is expected that the error bar of the isovector cp - odd nuclear force effect does not exceed 20 % . the @xmath92be edm is also analyzed in the same manner . in the cluster model , the edm of @xmath92be is polarized only by the cp - odd @xmath125 interaction , and its sensitivity to the isovector cp - odd nuclear force is close to that of the deuteron . as the distance between an @xmath95 cluster and the single neutron inside @xmath92be is about 4 fm , the cluster picture is expected to work well , so that the folding of the cp - odd potential describes well the cp - odd polarization . as the @xmath95 cluster is considered to be unbreakable , the isoscalar and isotensor cp - odd nuclear forces are not relevant for the polarization contribution to @xmath92be . for the @xmath92be nucleus , however , the intrinsic neutron edm also contributes to the nuclear edm through the spin quenching factor which is close to one ( see table [ table : nuclearedm ] ) . the leading isoscalar ( @xmath89 ) effect of @xmath92be is therefore generated by the pion cloud process of the neutron edm . the effect of heavier meson exchanges can not be considered in the case of @xmath92be , since the model space of those interactions does not match that of the folding potential . a similar problem is also encountered for the calculation of the contribution of the cp - odd contact interactions . in this case , they must be renormalized at the energy scale corresponding to the cutoff of the model space of the @xmath95 cluster model . the edm of @xmath73c was also evaluated within the cluster model @xcite . as for @xmath92be , the @xmath73c nucleus is sensitive to the isovector cp - odd pion - nucleon coupling due to the polarization by the cp - odd @xmath125 interaction , and its edm depends on the isoscalar cp - odd pion - nucleon coupling through the pion cloud effect of the intrinsic neutron edm . we do not consider the cp - odd nuclear force due to the exchange of heavier mesons or contact interactions , since the interaction is not renormalized to respect the model space . the edm of @xmath73c is less sensitive on the isovector cp - odd nuclear force than the other lighter nuclei by an order of magnitude . its analysis is however interesting in understanding how the cp - odd polarization effect is suppressed . the @xmath73c nucleus has a ground state ( @xmath132 ) which is well understood as a shell - like nucleus . if we describe it in terms of a system made of a valence nucleon and a @xmath108c core , the orbital angular momentum @xmath133 and the nucleon spin will be constructively combined due to the spin - orbit force . the angular momentum of @xmath108c core therefore has to be @xmath134 to obtain the @xmath132 state . the opposite parity excited state of @xmath73c ( @xmath135 ) is located 3.1 mev above the ground state , but this state is a neutron halo state with a @xmath74 @xmath108c core due to the damp of the spin - orbit force , and the transition between @xmath132 and @xmath135 states through the cp - odd nuclear force or the edm operator is suppressed . the next candidate of states which can couple to the @xmath132 state through the parity mixing is the @xmath108c + @xmath136 continuum state , which opens at 4.9 mev above the ground state . however , we also have here a dominant configuration with 0@xmath137 @xmath108c core , and the transition between them is small . the state with the best overlap is the continuum state with a neutron and an excited @xmath108c core ( 2@xmath137 , 4.4 mev above 0@xmath137 ) . its threshold is 9 mev above the ground state , and the transition is suppressed energetically . from the analysis of the edm of @xmath73c , we can learn that the existence of low lying opposite party states with the same angular momentum is not a sufficient condition to obtain large cp - odd nuclear polarization . the bad transition between low lying opposite parity states may also be relevant for @xmath138n , which have a @xmath139 ground state and a @xmath140 state at 5.3 mev . this analysis is also suggesting that the edm of nuclei with a shell - like structure is less advantageous than clustered nuclei in enhancing the cp - odd effect . let us add a brief comment on the effect of the cp - odd contact interaction of chiral eft . the isospin blind contact interaction [ term with @xmath45 of eq . ( [ eq : cpveftpot ] ) ] in principle does contribute to the cp - odd @xmath114 interaction . the contribution of @xmath45 to the edm of @xmath81li is forbidden by isospin selection rule . the @xmath92be and @xmath73c nuclei can probe the effect of @xmath45 . to evaluate it correctly , we have to derive the effective interaction by changing the model space , but it is not currently available . the isospin dependent isoscalar cp - odd contact interaction [ term with @xmath46 of eq . ( [ eq : cpveftpot ] ) ] can not contribute to the cp - odd @xmath114 folding potential , since it has to open the isospin shell of the @xmath95 cluster . therefore , @xmath46 can not be probed with the edms of @xmath81li , @xmath92be , @xmath73c . it is known that the strong interaction involves a cp violating interaction , the @xmath0-term , defined by @xmath141 where @xmath142 is the strong coupling and @xmath143 the gluon field strength . this term is not forbidden in sm , so it is of importance to quantify its effect on the hadronic cp violating observables . the @xmath0-term contribution to the nucleon edm was evaluated in many phenomenological approaches @xcite . the chiral eft analysis of the nucleon edm within the @xmath0-term involves several low energy constants which have to be fitted from phenomenology or lattice qcd data [ see eqs . ( [ eq : d_0 ] ) and ( [ eq : d_1 ] ) ] . using lattice qcd data with pion mass out of its physical one @xcite , the nucleon edms calculated in chiral eft are @xcite @xmath144 continuous efforts to calculate the nucleon edm in lattice qcd are also on - going , although simulations at the physical pion mass are still difficult @xcite . currently , results of the calculation of the nucleon edm at the pion mass @xmath145 mev are giving @xcite @xmath146 the direct calculation of the @xmath0-term contribution to the nucleon edm at the physical pion mass @xmath147 mev is one of the important goal in the study of hadronic cp violation . the most recent experimental data of the neutron edm is giving an upper limit @xcite @xmath148 the constraint on the @xmath0-parameter is then @xmath149 as this term is not suppressed or forbidden by some symmetries or mechanisms in sm , the natural size of the coupling should be of order one , like the cp - even qcd lagrangian . the constraint from the experimental data is suggesting a much smaller @xmath150 than its naturally expected size . this problem is known as the _ strong cp problem_. as a natural resolution of this problem , a mechanism making the @xmath0-term irrelevant was proposed by peccei and quinn @xcite . this consists of introducing a new scalar field , the axion , which has the following lagrangian @xmath151 with @xmath152 the decay constant of some symmetry which is broken at some energy scale much higher than the electroweak one . this field develops an expectation value at @xmath153 due to the nonzero topological susceptibility of qcd , and the effect of the @xmath0-term is dynamically canceled . this attractive mechanism was often considered in the study of hadronic cp violation @xcite . the axion mechanism cancels the effect of the bare @xmath0-term , but it may also induce dynamical one in the presence of additional cp - odd operators , through the correlation with the topological charge @xcite . as an example , the induced @xmath0-term due to the quark chromo - edm is @xmath154 where @xmath155 @xcite . this induced @xmath0-term has comparable contribution as other hadronic cp - odd effective interactions , and can not be neglected . this means that the evaluation of the @xmath0-term contribution to hadronic cp violation is important even if the axion mechanism is active . if the axion mechanism is inactive , eq . ( [ eq : thetalimit ] ) is a phenomenological constraint that corresponds to the theoretical uncertainty of the sm contribution . in the analysis of the new physics contribution to the nuclear edm , the effect of the @xmath0-term must therefore be investigated within the allowed region @xcite . as we have seen previously , the contribution of the cp - odd nuclear force is important in the study of the edm of light nuclei . the @xmath0-term contributes dominantly to the isoscalar cp - odd pion - nucleon interaction , and this contribution can be accurately evaluated in the chiral approach . the most recent prediction based on flavor @xmath156 interactions with @xmath156 splittings is giving @xcite @xmath157 for light nuclei , the effect of isovector cp - odd pion - nucleon interaction is often more important . the @xmath0-term , although being isoscalar , also generates sizable isovector coupling @xmath82 . the next - to - leading order analysis in chiral @xmath156 eft is giving @xcite @xmath158 the other hadron level cp violation such as the three - pion interaction or the contact cp - odd nucleon - nucleon interaction , generated by the @xmath0-term , has negligible effects . in table [ table : smnuclearedm ] , we show the uncertainty of the edm of several light nuclei when the @xmath0-term respects the experimental upper limit by the neutron edm ( [ eq : neutron_edm_exp_data ] ) . we see that the nuclear edm is typically of the same order as that of the neutron edm . the edm is attracting interest in part due to the small ckm contribution , which allows us to get rid of it in the background analysis . here we are considering the case where the @xmath0-term is not relevant , for instance when it is removed by the axion mechanism . in the case of the nuclear edm , simple estimation of ckm effect gives @xmath159 cm , with the jarlskog invariant @xmath160 @xcite , while earlier calculations are predicting nuclear edms in the range @xmath161 cm @xcite . at the quark level , there are several short distance processes which induce the cp violation due to the ckm phase , such as the quark chromo - edm @xcite , the weinberg operator @xcite , etc . those effects are however known to be small . here we discuss the nuclear edm induced by the ckm cp phase , which may be enhanced by the nuclear many - body effect . here we present current results of the calculations of the nuclear edm induced by the ckm phase through the long distance process @xcite . the leading ckm cp violation is generated by the tree level @xmath162 four - quark interaction ( with the product of ckm matrix elements @xmath163 ) and by the penguin diagram ( with @xmath164 ) @xcite . the combination of those elements forms the jarlskog invariant ( see fig . [ fig : deltas1 ] ) . those quark level processes are each matched with different hadron level cp violating interactions . the tree level @xmath162 interaction is matched with the hyperon - nucleon interaction , often using the quark model . the penguin diagram contribution is matched with the @xmath162 p - odd meson - baryon interaction . here the factorization model with the vacuum saturation approximation is often used . it is important to note that the penguin contribution is enhanced by an order of magnitude through the renormalization group equation when its wilson coefficient is evolved from the electroweak scale to the hadronic scale @xcite . leading ckm contributions to the quark level cp violation . ( a ) tree level diagram . ( b ) penguin diagram . , width=377 ] after obtaining the hyperon - nucleon transition and the @xmath165 p - odd meson - baryon interactions , we can combine them to calculate the nucleon level cp violation . the cp phase of the ckm matrix becomes relevant at the hadron level . the leading nucleon level cp violation is given by the nucleon edm [ see fig . [ fig : smcpv ] ( a ) ] and the cp - odd nuclear force [ see fig . [ fig : smcpv ] ( b ) ] . here we must note that hyperons and kaons have to be explicitly treated as dynamical degrees of freedom to evaluate the long distance effect ( we have to enlarge the cutoff ) . in table [ table : smnuclearedm ] , we show the estimated values of the edms of light nuclei generated by the ckm cp phase . in these calculations , the polarization effect due to the exchange of @xmath55 , @xmath5 , and @xmath52 mesons was considered @xcite . leading ckm contributions to the nucleon level cp violation . the inner solid lines are either a nucleon or a hyperon ( @xmath166 or @xmath167 ) . the dashed lines denote mesons ( @xmath55 , @xmath5 or @xmath52 ) . ( a ) pion loop contribution to the nucleon edm . ( b ) pion exchange cp - odd nuclear force . the grey blob denotes the @xmath162 interaction induced by the penguin diagram . the crosses denote the possible insertions of @xmath162 interactions generated by the tree level diagram . , width=377 ] the edm of light nuclei induced by the ckm cp violation are in the general case below the prospective sensitivity of the prepared measurement using storage rings [ @xmath6 cm ] . in this sense , we can safely neglect ckm backgrounds in the search for new physics bsm using nuclear edms . the evaluation of the nuclear edm involves however large systematics which are difficult to control . the largest theoretical uncertainty comes from the qcd calculation of the weak @xmath162 meson - baryon couplings . from the large @xmath168 analysis , this error is estimated to be @xmath169 . the accuracy of the contribution of the @xmath162 four - quark interactions to the @xmath162 interbaryon potential is expected to be improved through the phenomenological and eft analyses of the nonleptonic weak decays of hyperons @xcite and hypernuclei @xcite . the nucleon edm contribution to the nuclear edm is also an important source of uncertainty . the long distance effect of the nucleon edm from the ckm cp phase was estimated to be @xmath170 cm in many previous works @xcite , an order of magnitude smaller than the edm of light nuclei ( see table [ table : smnuclearedm ] ) . the contribution of the nucleon edm is not enhanced inside the nucleus , due to the spin quenching factor smaller than one ( see section [ sec : intrinsicnucleonedm ] ) . the uncertainty is however enlarged due to unknown relative sign with the nuclear edm @xcite . we should note that the tree level ckm contribution with higher dimension operators was also evaluated , yielding a nucleon edm of @xmath171 cm @xcite . there the baryon matrix elements were estimated using the naive dimensional analysis , with a suppression factor of @xmath172 due to the strange quark . however , from recent lattice qcd analyses of nucleon matrix elements such as the nucleon strange content or the axial charge , it is known that the strange quark effect is smaller by one or two orders of magnitude @xcite . this result for the nucleon edm should therefore be recognized as the upper limit of the theoretical uncertainty . for the deuteron , the dynamical effect between the nucleons and hyperons of the intermediate states was also considered in a recent work @xcite . this contribution can be calculated by taking into account the @xmath173 channel coupling , as shown in fig . [ fig : nn - yn ] . the shift of the edm due to this correction is about 10% , but there are no physical reasons to claim that this dynamical effect is not important for other nuclei . we therefore need to investigate this @xmath165 coupled channel effects for each nucleus to determine this systematics . diagrammatic representation of the deuteron polarization process with @xmath174 ( or @xmath175 ) coupled channel . the grey blobs denote the @xmath162 weak hadron level interactions . the grey square indicates the hyperon - nucleon interaction . , width=302 ] diagrammatic representation of several known important elementary level cp violating processes contributing to the nuclear edm . the dashed lines denote the boson of new physics bsm . ( a ) one - loop level quark chromo - edm , ( b ) barr - zee type two - loop level diagram , ( c ) cp - odd four - quark interaction , ( d ) weinberg operator . , width=453 ] let us now see the prospects for the discovery of new physics bsm . we first discuss the case of supersymmetric models @xcite . in supersymmetric models , the fermion edm and the chromo - edm receive contribution from the one - loop level [ see fig . [ fig : tev_cpv ] ( a ) ] . by assuming that the quark chromo - edm contributes to the leading order , the sensitivity of the edm of light nuclei of @xmath176 cm can probe @xmath177 and @xmath178 , the cp phase of the @xmath179 term and that of the trilinear supersymmetry breaking coupling , respectively , at the level of @xmath180 in supersymmetric models with the supersymmetry breaking scale @xmath181 tev @xcite . here we have assumed the axion mechanism @xcite is active and @xmath182 . even if the phases @xmath177 and @xmath178 are flavor independent , the current quark mass splitting generates both isoscalar and isovector contributions to the chromo - edm , and consequently the isoscalar and isovector cp - odd pion - nucleon interactions @xmath89 and @xmath82 . therefore , light nuclei which have high isovector sensitivity can probe the supersymmetric cp phases . the sensitivity to @xmath177 is increased with large @xmath183 @xcite . this prospective sensitivity may unveil the high scale supersymmetry breaking scenario , which is beyond the sensitivity of lhc experiment . we now model the effect of cp violation bsm by the cp - odd four - quark interaction [ see fig . [ fig : tev_cpv ] ( c ) ] with the exchange of a new boson with mass @xmath184 and with @xmath185 cp phase for the interactions among bosons and light quarks . a typical isovector cp - odd pion - nucleon interaction with the coupling @xmath186 is therefore generated in the factorization approach with vacuum saturation approximation . in this case @xmath82 is not suppressed by the light quark mass . this contribution is important when the boson is exchanged between two light quarks . here the qcd scale parameter is @xmath187 mev , and @xmath188 is the coupling between quarks and new particles . another important possibility is the generation of the three - pion interaction , which also contributes to @xmath82 and can be probed at the same level . if the edm of light nuclei can be measured at the level of @xmath176 cm , we can probe the new physics of the energy scale @xmath189 pev , with a typical coupling @xmath190 . this estimation works for models which generate cp - odd 4-quark interactions , such as the left - right symmetric model @xcite or the sfermion exchange processes in r - parity violating supersymmetry @xcite . another important class of models is that inducing the barr - zee type diagrams [ see fig . [ fig : tev_cpv ] ( b ) ] . this contribution is generated by the exchange of some scalar boson between a light quark and a heavy particle , and it is complementary to the cp - odd four - quark interaction . the combined analysis of the cp - odd four - quark interaction and the barr - zee type process is therefore important in the analysis of flavor physics . typical models contributing to the barr - zee type diagram are the higgs doublet models @xcite , generic supersymmetric models @xcite , and those with r - parity violation @xcite . the barr - zee type diagram with the gluon exchange gives the most important effect to the nuclear edm . using the simple formula @xmath191 , the sensitivity of the nuclear edm of @xmath176 cm can probe the energy scale of the new physics of @xmath192 tev . here @xmath193 is the mass of the inner loop particle @xmath194 , and @xmath195 is the coupling between the exchanged scalar boson and @xmath194 . we have assumed that the light quark yukawa coupling is @xmath196 . the barr - zee type diagram is suppressed by the yukawa coupling @xmath197 which reflects the light quark mass . in some models of new physics bsm , only the quark edm is relevant . this is the case of split supersymmetry @xcite , r - parity violation involving light quarks and heavy leptons @xcite , or models generating barr - zee type diagrams with charged higgs , @xmath198 or @xmath199 bosons @xcite . here we do not have to consider other cp - odd operators than the quark edm , since it does not mix with the others in the leading order renormalization group evolution @xcite . for those models , the only relevant hadron level cp violating process is the nucleon edm , for which the quark edm dominantly contributes . the quark edm effect to the nucleon edm is suppressed by the nucleon tensor charge due to the dynamical effect of qcd @xcite . recent lattice qcd data are giving @xmath200 @xcite . we must also note that the wilson coefficient of the quark edm operator is suppressed in the change of scale by the renormalization group evolution @xcite ( typically , about 80% when we run from @xmath201 tev to @xmath202 gev ) . moreover , the intrinsic nucleon edm contribution to the nuclear edm is smaller than one , due to the spin quenching of the nuclear spin matrix element ( see section [ sec : intrinsicnucleonedm ] ) . this last property is indicating that the nuclear edm is not an optimal probe of models bsm which only induce the quark edm . we finally have models contributing to the weinberg operator [ see fig . [ fig : tev_cpv ] ( d ) ] . this effect was first discussed for the case of the higgs doublet model @xcite . a recent important issue is the cp violation of models with vectorlike quarks and bosons @xcite . from the dimensional analysis , the weinberg operator contributes to the nucleon edm and to the isoscalar contact cp - odd nucleon - nucleon interaction without suppression by the light quark mass , whereas the cp - odd pion - nucleon interactions suffer from it . the analysis of the nucleon edm within qcd sum rules is estimating the weinberg operator contribution as @xmath203 mev @xcite , where @xmath204 is the weinberg operator coupling . we must note that the running of the weinberg operator reduces @xmath204 by an order of magnitude in the evolution of the scale from 1 tev to 1 gev , while the quark edm and the chromo - edm with wilson coefficients of the same order are generated due to the operator mixing @xcite . if we simply assume @xmath205 is the typical coupling , the sensitivity of the nuclear edm of @xmath176 cm can probe a model candidate of new physics with @xmath206 gev , with a typical coupling @xmath190 . we must note that the weinberg operator dominantly contributes to the nucleon edm , so the edm of light nuclei , being sensitive to the isovector cp - odd pion - nucleon interaction , is not the most appropriate in probing it . by summarizing the sensitivity of the nuclear edm on cp violating hadron level parameters , we see that the contribution from the nucleon edm becomes smaller than that of the isovector cp - odd pion - nucleon coupling as the nucleon number grows . as we have seen , the intrinsic nucleon edm effect can not be enhanced due to the pairing of nucleons , and it rather has tendency to decrease due to the mixing of angular momentum configuration . the cp - odd processes which can only be probed by the nucleon edm , such as the quark edm , therefore become subdominant . the isoscalar cp - odd nuclear force has also less effects on the nuclear edm than the isovector one . for light and stable nuclei , the number of protons and neutrons are almost equal , and the isoscalar cp - odd nuclear force becomes irrelevant in the nuclear polarization @xcite . as we have seen in this review , the sensitivity to @xmath89 is often brought by the pion cloud effect of the valence nucleon , which is not large . the isoscalar component of new physics contributing to the @xmath0-term ( see section [ sec : theta ] ) and to the quark chromo - edm @xcite suffers from this suppression . the large sensitivity of the nuclear edm on the isovector pion - nucleon coupling makes light nuclei to be good probes of new physics contributing to the isovector processes . up to mass dimension-6 operators , those are the isovector component of the quark chromo - edm and the isovector cp - odd four - quark interaction relevant in left - right symmetric models @xcite . it is useful to note that those two cp violating interactions do not mix with each other in the renormalization group evolution @xcite . this property is very remarkable because those contributions can be analyzed on a priority basis , without caring destructive interference with other cp - odd sources . this also means that the accurate hadron level calculation of @xmath82 from either the quark chromo - edm or the cp - odd four - quark interaction is not required as long as we are intending to discover the existence of cp violation bsm without going into investigations of the coupling constants beyond the second digit . this way of analyzing should work in a natural scenario where cp violating couplings are not fine - tuned . another very important role of measuring the edms of several nuclei is to fix the cp violating hadron level cp violating coupling constants using the linear independence of the edms on them . this analysis works for scenario with fine - tuned coupling constants . to fix unknown variables , we need the same number of equations , so to determine the seven unknown couplings of the leading cp violating chiral lagrangian ( [ eq : cpv_chiral_lagrangian ] ) , we need at least seven measurements of nucleon and nuclear edms . in this procedure , the individual data of the proton and neutron edms as well as those of atomic edms @xcite , which are more sensitive on @xmath89 , can also be used . the experimental measurement of t - odd neutron - nucleus scattering @xcite may also provide very useful informations on @xmath89 , since this process does not depend on the intrinsic nucleon edm . several investigations trying to constrain the multidimensional parameter space of the new physics with the experimental data of the edms of various systems ( neutron , atoms , and molecules ) exist in the literature @xcite . adding the experimental results of the nuclear edm measurement will additionally constrain the degrees of freedom of the parameter space of new physics . the dependence of the atomic and nuclear systems on the hadronic cp violation is complementary , since interesting atoms are generally heavy , and have sizable isoscalar sensitivity , whereas light nuclei are good in probing isovector cp - odd interactions . we have to note that this multidimensional analysis requires the hadronic cp violation to be accurately quantified . in the study of atomic systems , the nuclear level evaluations become also problematic , in addition to the hadron level difficulty already relevant in the study of the nuclear edm . to fix the elementary level parameters within reasonable accuracy , further improvements of the qcd and nuclear level calculations of the cp violating effects are required . finally , the determination of the weinberg operator has a critical difficulty . we have seen previously seen that the weinberg operator mixes with the quark edm and chromo - edm when the operator run from the tev scale to the hadronic scale . those three cp - odd operators are known to comparably contribute to the nucleon edm , and this fact requires tremendous efforts in the qcd calculation . the nuclear edm is providing an alternative approach in the determination of the weinberg operator , together with its own difficulty . the weinberg operator can be probed through the isoscalar contact cp - odd nucleon - nucleon interaction , but the calculation of this process suffers from a large uncertainty in the nuclear level calculation , due to the lack of informations of the nuclear force at short distance . this problem may be resolved if the short range behavior of the cp - even nuclear force is determined in high energy experiments @xcite , or by lattice qcd calculations @xcite . there are also no available results for the qcd calculation of the weinberg operator contribution to the contact cp - odd nucleon - nucleon interaction . moreover , the contribution of the quark chromo - edm appearing from the mixing due to the renormalization group evolution also has to be quantified . those issues can only be resolved using lattice qcd , if we wish to determine it beyond the order estimation . let us also see the prospects for finding nuclei sensitive on the nucleon level cp violation . in section [ sec : edm6li9be13c ] , we have seen that the bad transition between the ground state and low lying opposite parity states with the same angular momentum suppresses the polarization contribution to the nuclear edm . light nuclei with a dominantly shell - like structure have this tendency . we therefore have to find nuclei with a well developed cluster structure . restricting to stable nuclei , good candidates are @xmath94li and @xmath207f . the @xmath94li is known to have a @xmath208h cluster structure . the @xmath208h threshold is just 2.5 mev above the ground state , and the cp - odd transition to the continuum is expected to be large , as the @xmath95 core does not change its structure . the @xmath94li nucleus is manageable in the four - body cluster model ( @xmath209 ) , and it is thus a good target of future work . although being unstable , @xmath94be , the mirror nucleus of @xmath94li , may also be a good candidate since it has a life time of 53 days , which is experimentally manageable . in this case , the coulomb force decreases the energy needed to reach the @xmath210he threshold from the ground state to 1.6 mev , so that parity violation is enhanced , and the edm may be increased . the @xmath207f nucleus is also a good candidate , as it can be described by the mixing of @xmath138n@xmath211 and @xmath109o@xmath212h cluster states . the remarkable point is that its first opposite parity excited state ( @xmath139 ) is just 109 kev above the ground state ( @xmath140 ) , so that a sizable enhancement of the nuclear polarization by the isovector cp - odd nuclear force is expected . this nucleus can be calculated in the two - body cluster model with coupled channel ocm @xcite . in this review , we have summarized the current status of the theoretical development of the edm of light nuclei . the edm of @xmath213h , @xmath8he , @xmath8h , @xmath81li , @xmath92be , and @xmath73c nuclei have so far been evaluated . the study of the nuclear polarization of those nuclei by cp - odd effects is showing that the cluster structure may enhance the sensitivity on nucleon level cp violation , as for the case of @xmath81li . moreover , this extensive analysis is also suggesting the existence of nuclei furthermore sensitive on cp violation , such as @xmath94li , @xmath94be or @xmath207f , and their theoretical investigations are strongly wanted . the study of those nuclei will be an important target of our future work . the edm of light nuclei have dependences on various parameters of new physics bsm . among them , the most important nucleon level cp violation for light nuclei is the isovector cp - odd nuclear force . however , as the determination of the new physics bsm is not achievable by only measuring the edm of one very sensitive system , the combination of several experimental data of different observables is mandatory in constraining the large parameter space of new physics . to discern the cp violation bsm , the combined analysis with the experimental search of the single nucleon edm , the atomic edm , and t - odd neutron - nucleus scattering may be efficient , as these observables are sensitive to the isoscalar cp - odd pion - nucleon interaction . the experimental measurements of the edm of the proton and light nuclei are prepared at fermilab and at jlich , with a prospective experimental sensitivity of @xmath6 cm . the theoretical uncertainty due to the qcd calculation of low energy constants is still large , and further investigations are needed to fix at least the first digit , to be confident in observing signals bsm . the prospective experimental sensitivity has the potential to unveil typical models of new physics bsm between the energy scale of tev to pev . by experimentally observing the edm of the deuteron or three - nucleon systems , significant advance in the understanding of new physics is expected . we also recommend the development of the experimental and theoretical studies of the edm of other heavier light nuclei . the author is very grateful to jordy de vries who kindly helped him improving the review . he also thanks the organizers and participants of the workshop `` baryons over antibaryons : the nuclear physics of sakharov '' , held at ect * - european centre for theoretical studies in nuclear physics and related areas , italy , in july 2016 , for useful discussions and comments . this work was completed due to support of the rsf grant 15 - 12 - 20008 . it was also supported by riken ithes project . j. baron _ et al_. ( acme collaboration ) , science * 343 * , 269 ( 2014 ) . r. h. parker _ et al . _ , lett . * 114 * , 233002 ( 2015 ) . b. graner , y. chen , e. g. lindahl and b. r. heckel , phys . lett . * 116 * , 161601 ( 2016 ) . m. bishof _ et al . _ , phys . rev . c * 94 * , 025501 ( 2016 ) . g. w. bennett _ ( muon ( @xmath214 ) collaboration ) , phys . rev . d * 80 * , 052008 ( 2009 ) . b. plaster _ ( ucna collaboration ) , phys . c * 86 * , 055501 ( 2012 ) . k. fuyuto , j. hisano and n. nagata , phys . d * 87 * , 054018 ( 2013 ) . c. m. maekawa , e. mereghetti , j. de vries , and u. van kolck , nucl . a * 872 * , 117 ( 2011 ) . w. c. haxton and e. m. henley , _ phys . lett . _ * 51 * , 1937 ( 1983 ) . j. dobaczewski and j. engel , phys . lett . * 94 * , 232502 ( 2005 ) . j. de vries , e. mereghetti , r.g.e . timmermans and u. van kolck , annals phys . * 338 * , 50 ( 2013 ) . r. b. wiringa , v. g. j. stoks and r. schiavilla , phys . c * 51 * , 38 ( 1995 ) . j. bsaisou , j. de vries , c. hanhart , s. liebig , u .- g . meiner , d. minossi , a. nogga and a. wirzba , j. high energy phys . * 1503 * , 104 ( 2015 ) [ erratum ibid . * 1505 * , 083 ( 2015 ) ] . j. bsaisou , u .- g . meiner , a. nogga and a. wirzba , annals phys . * 359 * , 317 ( 2015 ) . m. pospelov and a. ritz , phys . lett . * 83 * , 2526 ( 1999 ) . m. pospelov and a. ritz , nucl . b * 573 * , 177 ( 2000 ) . d. k. hong , h .- c . kim , s. siwach and h .- u . yee , jhep * 0711 * , 036 ( 2007 ) . e. shintani , s. aoki and y. kuramashi , phys . d * 78 * , 014503 ( 2008 ) . guo , r. horsley , u .- g . meissner , y. nakamura , h. perlt , p. e. l. rakow , g. schierholz , a. schiller and j. m. zanotti , phys . * 115 * , 062001 ( 2015 ) . a. shindler , t. luu and j. de vries , phys . d * 92 * , 094518 ( 2015 ) . c. alexandrou , a. athenodorou , m. constantinou , k. hadjiyiannakou , k. jansen , g. koutsou , k. ottnad and m. petschlies , phys . d * 93 * , 074503 ( 2016 ) . e. shintani , t. blum , t. izubuchi and a. soni , phys . d * 93 * , 094503 ( 2016 ) . r. d. peccei and h. r. quinn , phys . * 38 * , 1440 ( 1977 ) . e. hiyama , k. suzuki , h. toki , and m. kamimura , prog . phys . * 112 * , 99 ( 2004 ) . t. inoue , s. takeuchi and m. oka , nucl . a * 597 * , 563 ( 1996 ) . t. inoue , m. oka , t. motoba and k. itonaga , nucl . a * 633 * , 312 ( 1998 ) . k. sasaki , t. inoue and m. oka , nucl . a * 669 * , 331 ( 2000 ) [ erratum ibid . a * 678 * , 455 ( 2000 ) ] . k. sasaki , t. inoue and m. oka , nucl . a * 707 * , 477 ( 2002 ) . a. prez - obiol , a. parreo and b. juli - daz , phys . rev . c * 84 * , 024606 ( 2011 ) . ( qcdsf collaboration ) , phys . d * 85 * , 054502 ( 2012 ) . m. engelhardt , phys . d * 86 * , 114510 ( 2012 ) . k. takeda , s. aoki , s. hashimoto , t. kaneko , j. noaki and t. onogi ( jlqcd collaboration ) , phys . d * 83 * , 114506 ( 2011 ) . h. ohki , k. takeda , s. aoki , s. hashimoto , t. kaneko , h. matsufuru , j. noaki and t. onogi ( jlqcd collaboration ) , phys . d * 87 * , 034509 ( 2013 ) . p. m. junnarkar and a. walker - loud , phys . d * 87 * , 114510 ( 2013 ) . m. gong _ et al . _ , d * 88 * , 014503 ( 2013 ) . c. alexandrou , m. constantinou , s. dinter , v. drach , k. hadjiyiannakou , k. jansen , g. koutsou and a. vaquero , phys . d * 91 * , 094503 ( 2015 ) . a. abdel - rehim _ et al . d * 89 * , 034501 ( 2014 ) . yang , a. alexandru , t. draper , j. liang and k .- f . liu ( @xmath215qcd collaboration ) , phys . d * 94 * , 054503 ( 2016 ) . g. s. bali _ et al . _ , phys . rev . d * 93 * , 094504 ( 2016 ) . g. degrassi and s. marchetti , e. franco and l. silvestrini , jhep * 0511 * , 044 ( 2005 ) . j. hisano , k. tsumura and m. j. s. yang , phys . b * 713 * , 473 ( 2012 ) . w. dekens and j. de vries , j. high energy phys . * 1305 * , 149 ( 2013 ) . m. radici , a. courtoy , a. bacchetta and m. guagnelli , j. high energy phys . * 1505 * , 123 ( 2015 ) . a. bacchetta , a. courtoy and m. radici , j. high energy phys . * 1303 * , 119 ( 2013 ) . a. courtoy , s. baeler , m. gonzlez - alonso and s. liuti , phys . lett . * 115 * , 162001 ( 2015 ) . kang , a. prokudin , p. sun and f. yuan , phys . d * 93 * , 014009 ( 2016 ) . k. choi , s. h. i m , h. kim and d. y. mo , phys . b * 760 * , 666 ( 2016 ) . d. demir , m. pospelov and a. ritz , phys . d * 67 * , 015007 ( 2003 ) . m. pospelov , phys . b * 530 * , 123 ( 2002 ) . v. p. gudkov , phys . rep . * 212 * , 77 ( 1992 ) . j. d. bowman and v. gudkov , phys . c * 90 * , 065503 ( 2015 ) . t. ibrahim and p. nath , phys . d * 58*. 111301 ( 1998 ) [ erratum ibid . d * 60 * , 099902 ( 1999 ) ] . y. li , s. profumo and m. ramsey - musolf , jhep * 1008 * , 062 ( 2010 ) . j. ellis , j. s. lee and a. pilaftsis , jhep * 1010 * , 049 ( 2010 ) . j. ellis , j. s. lee and a. pilaftsis , jhep * 1102 * , 045 ( 2011 ) . n. yamanaka , t. sato and t. kubota , jhep * 1412 * , 110 ( 2014 ) . m. m. sargsian , chapter to be published in `` @xmath56 and 3@xmath78 interactions '' ( nova science publishers , inc . ) [ arxiv:1403.0678 ] . n. ishii , s. aoki and t. hatsuda , phys . lett . * 99 * , 022001 ( 2007 ) . s. aoki , t. hatsuda and n. ishii , prog . . phys . * 123 * , 89 ( 2010 ) . t. sakuda and f. nemoto , prog . . phys . * 62 * , 1274 ( 1979 ) .

in relatively close primordial binaries which contain at least one star that can evolve substantially by the current epoch , mass transfer from the primary to the secondary is assured . the course of the evolution of each system depends on its initial orbital period , the two stellar masses , whether the binary is part of a triple system , spin and tidal interactions , and so forth . some of the possible evolutionary channels have been explored and even systematically mapped out by rappaport et al . ( 1994 ) , iben & tutukov ( 1995 ) , nelson & eggleton ( 2001 ) , han et al . ( 2002 ; 2003 ) , podsiadlowski et al . ( 2003a ) , and nelemans ( 2010 ) . depending largely upon the mass ratio and the initial orbital period , the mass transfer may be stable and mostly conservative , stable and mostly non - conservative , or dynamically unstable . in the case of a close binary that comprises a primordial primary star with mass @xmath5 and a secondary star that has @xmath6 , the mass transfer is most likely to be stable . however , depending on the ratio of thermal timescales of the two stars , the mass transfer may range from highly non - conservative for small @xmath7 ratios to mostly conservative for values of @xmath7 close to unity . whether the orbit ultimately evolves to be longer or shorter than the original orbital period depends on how much specific angular momentum is carried away with the matter ejected from the binary . in principle the best way to model such binary evolution is via a stellar evolution code that can simultaneously handle the nuclear and mass loss / gain evolution of _ both _ stars as well as the orbital evolution . the recently developed _ code ( paxton et al . 2011 ) is capable of doing such binary evolution studies , but , to our knowledge , no systematic studies have yet been carried out of ordinary close primordial binaries . we know empirically , and from some theoretical studies , that such evolution can lead to systems such as algol binaries ( see , e.g. , nelson & eggleton 2001 ) where the mass transfer is stable and the envelope of the primary has not yet been exhausted ; sdb binaries ( see , e.g. , han et al . 2002 ; 2003 ) which may have evolved from unstable mass transfer in a common envelope ; systems with hot , bloated , low - mass white dwarfs ( see , e.g. , van kerkwijk et al . 2010 ; carter et al . 2011 ) , where the envelope of the primary has been completely removed ; and even double - degenerate binaries ( marsh et al . 2004 ; iben , tutukov , & yungelson 1995 ; nelemans 2010 ) . in low - mass binaries where the mass transfer proceeds stably , regardless of whether it is or is not conservative , there is a fairly tight relation between the white dwarf mass , @xmath8 , and the final orbital period , @xmath9 ( see , e.g. , joss & rappaport 1983 ; savonije 1983 ; paczyski 1983 ; savonije 1987 ; rappaport et al . 1995 ; fig . 5 of lin et al . 2011 ; tauris & van den heuvel 2014 ) . for orbital periods @xmath10 years , and white dwarf masses , @xmath8 , up to 1.4 @xmath0 , in systems where the initial masses of the primaries , @xmath11 , were no larger than @xmath3@xmath12 , the following expression gives the final orbital period in terms of the white dwarf mass : @xmath13 for orbital periods up to @xmath320 days , which apply to systems with primaries of initial masses up to @xmath3@xmath12 , a more recently derived expression ( lin et al . 2011 ) is : @xmath14 these expressions have been checked against a number of binary radio pulsars ( fig . 3 of rappaport et al . 1995 ; fig . 2 of tauris & van den heuvel 2014 ) , and against a number of _ kepler _ binaries containing low - mass hot white dwarfs ( van kerkwijk et al . 2010 ; fig . 6 of carter et al . 2011 ) . prior to this work , only four low - mass white dwarf systems in the _ kepler _ field that can be used to study this relation between orbital period and white dwarf mass were identified . they are koi 74 , koi 81 ( rowe et al . 2010 ; van kerkwijk et al . 2010 ) , khwd3 ( koi 1375 ; carter et al . 2011 ) , and koi 1224 ( breton et al . 2012 ) . in this work we report the discovery of new low - mass , highly thermally bloated white dwarfs in the _ kepler _ binaries kic 9164561 and kic 10727668 . the orbital periods of these binaries are 1.267 and 2.305 days , the shortest periods among the six _ kepler _ binaries containing low - mass hot white dwarfs . the white dwarf in kic 9164561 is by far the most thermally bloated of any of the six white dwarfs . in section [ sec : data ] we discuss the search that led to these two systems among the _ kepler _ binaries , and indicate how the data were prepared for analysis . folded light curves are presented in section [ sec : lcs ] along with a simple geometric analysis , including a harmonic decomposition of the out - of - eclipse portions of the light curve . the masses of the primary a stars are inferred from information on their temperatures and mean densities in sect . [ sec : astars ] . in sect . [ sec : rv ] we present the radial velocity and spectral studies of kic 9164561 . the mass ratio of kic 10727668 is examined in the context of the doppler boosting effect in sect . [ sec : db ] . inferences about the mass ratios in the two systems are drawn from evaluating the ellipsoidal light variations in sect . [ sec : elv ] . in sect . [ sec : wds ] the system parameters for these two binaries are evaluated with emphasis on the white dwarf properties . evidence for a third star in kic 9164561 is presented in sect . [ sec : third ] . finally , in sect . [ sec : discuss ] we discuss kic 9164561 and kic 10727668 in the context of different binary evolution scenarios and white dwarf cooling models . in the process of searching for third - body companions to the binary stars in the _ kepler _ sample ( see , e.g. , rappaport et al . 2013 ; see also conroy et al . 2014 ) , we produced both an @xmath15 curve for the eclipses ( `` observed minus computed '' eclipse times ) as well as a folded light curve for each of the @xmath32600 cataloged _ kepler _ binaries ( matijevic et al . 2012 ; kirk et al . we used the _ kepler _ long - cadence time - series photometric data ( 29.4 min samples ) obtained from quarters 0 through 17 . for all the known binary systems for which at least one quarter of photometry is available , the version 5.0 fits files , which are available for all quarters , were downloaded from the stsci mast website . we used the data that had been processed with the pdc - map algorithm ( stumpe et al . 2012 ; smith et al . 2012 ) , which is designed to remove many instrumental artifacts from the time series while preserving any astrophysical variability . we also made use of the information in the file headers to obtain estimates of the combined differential photometric precision ( cdpp ) . we then examined by eye all 2600 @xmath15 curves for interesting variations . some of the results have been reported elsewhere ( rappaport et al . 2013 ; tran et al . the @xmath15 curve for kic 9164561 is discussed in sect . [ sec : third ] . we also searched for folded light curves that ( i ) exhibit a flat - bottomed eclipse that is deeper than the other eclipse , and ( ii ) are associated with hot _ kepler _ targets , i.e. , targets that have a reported @xmath16 k. the deep flat - bottomed eclipse - high temperature combination is a signature of a hot white dwarf in orbit with a normal , f , a , or b star . in addition to the four known _ kepler _ low - mass bloated hot white dwarfs , we found two others , previously unreported , in the binaries kic 9164561 and kic 10727668 . the nondegenerate components of kic 9164561 and kic 10727668 are , according to the kepler input catalog , hot with effective temperatures of @xmath38000 k and 10,000 k , respectively , and thus are not susceptible to the starspot modulations that are exhibited by cooler stars . an examination of the fourier transform of the kic 10727668 flux time series shows no evidence for any periodicities , including those from stellar pulsations , other than that of the orbit . in the case of kic 9164561 , the fourier transform of the flux time series shows a set of pulsations at frequencies @xmath17 cycles / day , which does not significantly affect our light curve or timing analysis . we therefore made no corrections to the data except for normalizing each quarter to its median value . the orbital periods given in the _ kepler _ binary catalog ( matijevic et al . 2012 ; kirk et al . 2015 ) were used to produce initial folds of the data . from these folds , we established a trial template for the orbital light curve . this profile was then used to fit each individual eclipse in the time series , thereby determining a set of 893 and 108 eclipse times for kic 9164561 and 10727668 , respectively . we then fit a linear function to the sequence of eclipse times of the form : @xmath18 , where @xmath19 is the nth eclipse . the two orbital periods thus determined are in good agreement with those given in the _ kepler _ binary catalog , but have somewhat smaller uncertainties . the periods and their uncertainties for these two systems are listed in table [ tab : sys ] . we also fit the eclipse times for each system including a quadratic term to allow for possible changes in the orbital period or light travel time effects that might occur if the binary is part of a triple system . these are discussed in sect . [ sec : third ] below . the data for both kic 9164561 and kic 10727668 were folded at the orbital periods we determined from fitting the individual eclipses , as described above . the results were averaged into 3-minute bins , and are shown in figures [ fig:916 ] and [ fig:107 ] . due to the small curvature observed in the eclipse timing residuals for kic 9164561 ( see sect . [ sec : third ] ) we removed a quadratic term from the eclipse phases before producing the folded light curve in fig . [ fig:916 ] . the phase corrections amount to less than 0.002 days ( or 0.0016 orbital cycles ) . in each case , the light curves are normalized to unity at the bottom of the deeper eclipse when the hot white dwarf has been totally eclipsed . this is then the light level of the primary a star at a minimum of its ellipsoidal variations . the relatively small separation of the two components of kic 9164561 is manifest by the short orbital period , the large amplitude of the ellipsoidal light variations ( `` elv '' ) , and the duration of the eclipse relative to the orbital period . the light curves of these two binaries were analyzed with a relatively simple geometrical model which involves 13 free parameters . in brief , there are 7 free parameters associated with the eclipses : @xmath20 , @xmath21 , @xmath22 , @xmath23 , two limb - darkening coefficients , and @xmath24 , where subscript `` 1 '' refers to the primary a star , @xmath25 is the orbital separation , @xmath22 the orbital inclination angle , and @xmath24 , the time of the middle of a primary eclipse . the out - of - eclipse phases are fitted with a set of 6 terms , i.e. , a sine and a cosine term for each of the orbital frequency and its next two higher harmonics . each of these can be associated with different contributions to the elvs , the illumination term , and the doppler boosting effect , as described in some detail in carter et al . ( 2011 ) . once the best - fitting parameters were found with a levenberg - marquardt fitting algorithm , the allowed region in parameter space was defined with a markov chain monte carlo ( mcmc ) routine , which used the gibbs sampler and the metropolis - hastings algorithm , and a likelihood proportional to @xmath26 ( see , e.g. , holman et al . 2006 ; sanchis - ojeda et al . 2013 ) . the results for the fitted model are plotted as red curves in figs . [ fig:916 ] and [ fig:107 ] . the high quality of the fit in each case is indicated by the white - noise appearance of the residuals . the results for the fitted parameters and their uncertainties are summarized in table [ tab : sys ] . the mass and radius of the primary star in each system are essential to quantitative models of the two binaries . there are two accessible parameters that can be utilized to estimate the primary mass : the effective temperature , @xmath27 , and the mean stellar density , @xmath28 . the _ kepler _ input catalog ( kic ) gives @xmath298059 k for the primary star of kic 9164561 and 9696 k for the primary of kic 10727668 . these temperatures are based on 5-color photometry . for kic 9164561 we measured @xmath30 k from the same spectra used for the radial velocity measurements ( see sect . [ sec : spec ] ) . in the case of kic 10727668 , where we have no additional spectral information , we estimate the uncertainty in the kic value of @xmath27 at approximately @xmath31 300 k ( huber et al . 2104 ) . the second readily , and more accurately , determined parameter associated with each primary star is its mean density , @xmath32 . this mean density is a function only of the light curve - determined parameters @xmath33 and @xmath34 , with a weak dependence on @xmath35 ( the mass ratio @xmath36 ) . the dependence of @xmath28 on these parameters may be found by dividing kepler s third law by @xmath37 to obtain : @xmath38 this relation was pointed out by seager & malln - ornelas ( 2003 ) in the context of transiting exoplanets . for exoplanets , the value of @xmath35 is much smaller than the error in @xmath34 so that @xmath28 can be determined independently of any mass information . while the same is not quite true with our two binaries , the value of @xmath35 is still sufficiently small , i.e. , @xmath39 that a rather accurate value of @xmath28 may be determined nearly independently of @xmath35 . in particular , by using eqn . ( 3 ) with the results from our photometric analysis of @xmath21 and with , as a first approximation , @xmath40 , we find that @xmath41 g @xmath42 for kic 9164561 , and @xmath43 g @xmath42 for kic 10727668 . these values are refined below using estimates for @xmath35 that are based on estimates of the masses of the two component stars . figure [ fig : yy ] shows stellar evolution tracks for stars of a range of initial masses in the @xmath44@xmath45 plane . the locations of the two target primary stars are indicated by the rectangular boxes . we conclude from this that the mass of the primary star in kic 9164561 must be @xmath46 ( kic 9164561 ) and that of the primary in kic 10727668 must be @xmath47 @xmath0 . the spectroscopic follow - up of kic9164561 was carried out with the fibre - fed chelle spectrograph ( fies ; * ? ? ? * ; * ? ? ? * ) mounted at the 2.56-m nordic optical telescope ( not ) of roque de los muchachos observatory ( la palma , spain ) . spectra were taken under clear and stable sky conditions at 11 epochs between may and september 2014 . we used the @xmath48 _ med - res _ fibre , which provides a resolving power of about 47000 in the spectral range 36007400 . the observations taken at each epoch were split into two or three consecutive sub - exposures of 1200 seconds each to remove cosmic ray hits . following the observing strategy described in @xcite , we monitored the radial velocity drift of the instrument by acquiring long - exposed ( t@xmath49=15 - 30 sec ) thar spectra in sandwich mode " , i.e. , right before and after each epoch observation . the data were reduced using a customized idl software suite , which includes bias subtraction , flat fielding , order tracing and extraction , and wavelength calibration . radial velocities ( rvs ) were determined by cross - correlating the target spectra with the spectrum of vega , for which we adopted a heliocentric radial velocity of @xmath50@xcite . table [ rv - table ] lists the fies rv measurements , the signal - to - noise ( s / n ) ratio per pixel at 4650 , and the full - width at half maximum ( fwhm ) of the fies cross - correlation function ( ccf ) . full - widths at half maximum were extracted by fitting a gaussian to the ccfs between the observed spectra and the atlas9 model spectrum having the same photospheric parameters as the target star ( see sect.[sec : spec ] ) and smoothed to the fies spectral resolution . figure [ fig : rvs ] shows the phase - folded rv curve of kic9164561 , based on the phase established from the _ kepler _ photometry , and the keplerian circular - orbit fit to the results . the k velocity of the primary a star is 21.64 @xmath31 0.35 km s@xmath51 , and the corresponding @xmath52-velocity is -28.43 @xmath31 0.25 km s@xmath51 . the corresponding mass function is @xmath53 for an orbital inclinational angle of @xmath54 ( see table [ tab : sys ] ) , this implies : @xmath55 the mass for the a star of @xmath56m@xmath57 obtained above may be combined with the mass function to obtain a white dwarf mass of @xmath58 . the mass ratio for the kic 9164561 system is therefore @xmath59 . because kic 10727668 is faint , i.e. , 2.6 _ kepler _ magnitudes fainter than kic 9164561 , we were unable to obtain any useful rv measurements for that system . [ rv - table ] ccccccc bjd@xmath60 & rv & @xmath61 & ccf@xmath62 fwhm & @xmath63 & s / n / pixel@xmath64 & date of obser.@xmath65 + ( @xmath66 2450000 ) & & & & & @4650 & ( at start ) + 6783.67056 & -9.397 & 0.673 & 103.6 & 1.4 & 23 & 06 may 2014 + 6833.58987 & -38.015 & 0.714 & 106.7 & 1.2 & 19 & 25 jun 2014 + 6833.67106 & -43.987 & 0.768 & 102.9 & 1.2 & 17 & 25 jun 2014 + 6844.66674 & -7.539 & 0.968 & 112.1 & 1.2 & 11 & 06 jul 2014 + 6845.43519 & -38.741 & 0.794 & 109.1 & 1.0 & 15 & 06 jul 2014 + 6845.71215 & -11.787 & 0.690 & 102.8 & 1.2 & 23 & 07 jul 2014 + 6901.37427 & -22.004 & 0.847 & 107.6 & 1.2 & 15 & 31 aug 2014 + 6902.40137 & -43.375 & 0.743 & 104.6 & 1.3 & 17 & 01 sep 2014 + 6915.38485 & -12.215 & 0.784 & 106.5 & 1.3 & 15 & 14 sep 2014 + 6915.55901 & -7.337 & 0.820 & 107.2 & 1.1 & 14 & 15 sep 2014 + 6920.43515 & -14.633 & 0.791 & 105.6 & 1.3 & 13 & 19 sep 2014 + \(a ) width of the cross correlation function . ( b ) s / n ratio per pixel at 4650 . ( c ) calendar date in ut at the beginning of the target observation . we started the spectral analysis by coadding all the fies spectra , after shifting them to a zero - velocity reference frame , which yielded a s / n of @xmath67 55 per pixel at 4650 . the spectral parameters of the primary component in kic 9164561 were then derived by fitting this spectrum to a grid of synthetic spectra calculated with the spectrum code @xcite using atlas9 models @xcite . we assumed a secondary - to - primary flux ratio of 0.0285 , as estimated from the eclipse light curve modeling ( table [ tab : sys ] ) . we adopted the calibration equations from @xcite and @xcite for the purpose of estimating the microturbulent @xmath68 and macroturbulent @xmath69 velocities . these equations were extrapolated to the effective temperature of the star . the spectral type was determined following the method described in @xcite . the estimated stellar parameters are listed in table [ star - parm ] . figure [ fies_spectrum ] shows the co - added fies spectrum in the spectral region encompassing the h@xmath70 , h@xmath71 , and mgi b lines , along with the best fitting synthetic model . lc parameter & + effective temperature @xmath72 ( k ) & @xmath73 + spectroscopic surface gravity log@xmath74 ( log@xmath75 cms@xmath76 ) & @xmath77 + metallicity [ m / h ] ( dex ) & @xmath78 + microturbulent velocity @xmath79 ( kms@xmath51 ) & @xmath80 + macroturbulent velocity @xmath81 ( kms@xmath51 ) & @xmath82 + projected rotational velocity @xmath83sin@xmath22 ( kms@xmath51 ) & @xmath84 + spectral type & 7 + we looked for the spectroscopic signature of a third ( bound ) component to kic 9164561 by cross - correlating the fies spectra with a set of a- through m - type observed templates . the templates were gathered between 2011 and 2014 , as part of the on - going rv follow - up of _ corot _ and _ kepler _ stars carried out with fies@not ( e.g. , gandolfi et al . 2013 ) . assuming that the third star is on or near the zams , its mass can not exceed @xmath32 @xmath0 corresponding to an a7v spectral type without significantly violating a number of observational constraints ( see below ) . the eclipse timing variations detected in the _ kepler _ light curve ( see sect . [ sec : third ] ) constrain the orbital period of a putative third star in the system to lie between 8 and 19 years , implying that its rv is within @xmath312 km s@xmath51 of the @xmath52 velocity of the binary system . the fies rv follow - up covers a time span of about 4.5 months , which corresponds to less than 4.6% of the orbital period of the third component , constraining its rv variation to be less than @xmath34 km s@xmath51 within the observing window . because of the relatively high projected rotational velocity of the primary star ( see table [ star - parm ] ) , the almost `` stationary '' ccf peak of the third source is expected to be blended with the broad peak of the a7v component . the fies ccfs show neither a peak with constant rv superimposed on top of the broad peak of the a7v star , nor a skewed profile that would suggest the presence of a another rapidly rotating star ) . ] . in order to assess the significance of the non - detection and constrain the parameters of a possible third body , we built up a set of synthetic double - lined spectroscopic binary ( ` sb2 ' ) spectra using atlas9 model atmospheres . we adopted gray s ( 2005 ) calibration for dwarf stars to convert spectral types to stellar parameters . we combined the rotationally broadened model spectrum of an a7v star resembling the primary component with that of dwarf stars with different spectral types ( from a7v to m5v ) and values of @xmath83sin@xmath22 . we added random noise to account for the average s / n ratio of the fies spectra and the contaminant light of the second source to account for the white dwarf . we assumed a zero velocity for the third light and radial velocities in the range @xmath85rv@xmath86 kms@xmath51 for the a7v primary star . finally , we cross correlated the combined sb2 spectra with narrow - lined synthetic spectra covering different spectra types , in the same fashion as for the observed spectra . since the contrast of a ccf peak decreases with increasing @xmath83sin@xmath22 , we considered the following two cases . _ spectral type later than f5v_. we used gyro - chronology to constrain the @xmath83sin@xmath22 of the third star . given the age of the system ( @xmath3925 myr ; see sect . [ sec : cool ] ) , the rotation period of the third component is expected to be longer than about 4 days for spectral later than f5v ( barnes et al . this places an upper limit of 18 km s@xmath51 on @xmath83sin@xmath22 . although this value likely overestimates the true @xmath83sin@xmath22 for the late - type stars , we conservatively assumed 18 km s@xmath51 to be the projected rotational velocity associated with the third light . we found that any stars with spectral type between k2v and f5v would be detectable in the fies ccf ( for an example see fig . [ fig : simccfs ] ) . for later spectral types the peak of the third light is embedded in the noise of the ccf , owing to the relatively low s / n ratio of the spectra and the low flux of the third star . sin@xmath87km / s and a k0v star with @xmath83sin@xmath88km / s ( see text for details ) . the dashed red and blue lines mark the rv of the k0v ( 0 km / s ) and the a7v star ( 21.6 km / s ) , respectively . the black curve , shown for reference , is the same ccf , but without the presence of the k0v star . ] _ spectral type between f5v and a7v_. for @xmath83sin@xmath89 km s@xmath51 the ccf peak of the third component is clearly visible on top of the peak of the primary star , as described in the previous case . for @xmath83sin@xmath90 km / s the ccf would assume a variable skewed profile , whose shape changes with the changing rv of the primary star , which we do not observe . finally , for @xmath91sin@xmath92 km / s the two ccf peaks are merged together and larger actual rv variations of the primary component would be needed to account for the measured rvs ( see below ) . in the case of comparable amounts of light contributed by the a7v star and a putative third star , we can say two things . first , the fwhm of the ccf would vary in a correlated way with the measured rvs . the extra widening can be estimated analytically , and the broadening factor is given by @xmath93 $ ] , where @xmath94 is the observed ( i.e. , ` diluted ' ) instantaneous value of the a7v star rv , and @xmath95 is the gaussian width of the ccf curve . since the ccf fwhm remains constant at 106.5 km s@xmath51 to within an rms scatter of @xmath33 km s@xmath51 ( see table [ rv - table ] ) , and a 42 km s@xmath51 shift ( i.e. , @xmath96 km s@xmath51 ) is expected to widen the ccf by 12 km s@xmath51 , which is distinctly not observed , we conclude that there is no other a- to early f - type star of comparable brightness in the system . second , as mentioned above , a stationary line would ` dilute ' the actual doppler shift by a factor of @xmath3@xmath97 , where @xmath98 and @xmath99 are the fluxes of the a7v star and the third body , respectively . in this case , the properly inferred mass of the white dwarf ( based on the rv amplitude ) would be a factor of approximately @xmath100 higher . this , in turn , would yield an implausibly high mass for the white dwarf when @xmath99 exceeds @xmath98 by even a factor of 2 . therefore , the mass of an early - type third body is independently constrained to be @xmath101 . thus , based on our simulations , we can exclude nearly any plausible near - ms third star that has a spectral type earlier than k2v . while there is still some small ` wiggle room ' left in parameter space , these constraints likely imply an upper limit for the mass of any third star to be about 0.8 @xmath0 . in principle , the doppler boosting amplitude ( loeb & gaudi 2003 ; van kerkwijk et al . 2010 ) can serve as a surrogate in place of radial velocity ( ` rv ' ) measurements for measuring the binary k velocity . as formulated by carter et al . ( 2011 ) the doppler boosting ( hereafter ` db ' ) amplitude in a binary is given by : @xmath102\ ] ] where the @xmath103 s are the fluxes of the two stars in the _ kepler _ band . the @xmath104 s are doppler boosting pre - factors calculated from : @xmath105 where @xmath106 is the spectral flux density at frequency @xmath107 , and @xmath104 must be evaluated using stellar atmosphere models averaged over the _ kepler _ band . the second term in eqn . ( [ eqn : db ] ) represents the contribution to the db amplitude from the white dwarf in the system ( note the minus sign ) . this method for evaluating the k velocities has been verified by ehrenreich et al . ( 2011 ) and bloemen et al . ( 2012 ) for the case of koi 74 , a system quite analogous to the ones discussed in this paper . _ kic 9164561 _ : the measured doppler boosting amplitude in this system , i.e. , the amplitude of the @xmath108 term , where @xmath109 is the orbital frequency , and @xmath110 is time , is -10 @xmath31 26 parts per million ( ` ppm ' ) . given the @xmath111 velocity obtained from the rv measurements ( see sect . [ sec : rv ] ) , for nominal values of @xmath112 , @xmath113 for this system should be @xmath67160 ppm . one possible explanation for this discrepancy is based on the large illumination semi - amplitude , i.e. , the coefficient of the @xmath114 term , that is measured to be 2385 @xmath31 36 ppm . in this case , if the phase of the illumination is shifted in the proper sense from its nominal alignment with the eclipses by as little as 4@xmath115 , the entire db effect would be cancelled . we note that in koi 74 the illumination term is 600 times smaller because of the wider orbital separation and the lower luminosity of the white dwarf , and is therefore very unlikely to materially affect the observed doppler boosting amplitude . in order for the phase of the illumination effect in kic 9164561 to be shifted by the amount required to cancel most of the doppler boosting signal , the a - star primary would have to be rotating faster than the rate that is synchronous with the orbit , such that some of the heated surface material is carried into the night side of the star . the measured rotation velocity of this star is @xmath67 75 @xmath31 5 km s@xmath51 , a value that is to be compared with the equatorial rotation velocity of a star of radius 2.5 @xmath116 ( see table [ tab : sys ] ) rotating with a period of 1.267 days , i.e. , @xmath67 100 km s@xmath51 . thus , if anything , the primary is rotating subsynchronously and the db signal is therefore expected to be enhanced relative to its nominal amplitude . this is a potentially important discrepancy which merits follow - up study . _ kic 10727668 _ : in this system , the doppler boosting signal is measured to have a significantly non - negative value , 143 @xmath31 20 km s@xmath51 . the illumination amplitude , 557 ppm , is 4 times smaller than in the case of kic 9164561 ; this system may therefore be less susceptible to an illumination - induced perturbation of the db amplitude . if eqn . ( [ eqn : db ] ) is used to evaluate @xmath111 we find : @xmath117 where we have taken @xmath118 , @xmath119 ( van kerkwijk , private communication ) , and @xmath120 ( see fig . [ fig:107 ] ) . this expression for @xmath111 may be entered into the mass function , and the cube root of the result taken to obtain the following equation involving the masses of the two stars : @xmath121 where we have taken the orbital inclination angle to be 86.4@xmath115 ( see table [ tab : sys ] ) . this equation can be solved numerically for @xmath122 given any specific value for @xmath123 . for a primary star mass of 2.2 @xmath0 ( see fig . [ fig : yy ] ) , the mass of the secondary is @xmath124 where the error bar takes into account the uncertainty in @xmath123 as well as the uncertainty in @xmath111 . again we caution that this estimate for @xmath122 does not take into account any systematic error due to an `` amplitude leak '' in the db term . the ellipsoidal light variations ( ` elvs ' ) seen in the out - of - eclipse region of the light curves also contain information on the mass ratios , @xmath125 . the basic expression for the elvs ( kopal 1959 ; morris & naftilan 1993 ; carter et al . 2010 ) is : @xmath126 and @xmath127 where @xmath128 is the amplitude of the cosine term at twice the orbital frequency , @xmath129 is the linear limb - darkening coefficient , and @xmath130 is the gravity darkening coefficient . for the hot primary stars in kic 9164561 and kic 10727668 we take @xmath131 and @xmath132 . this leads to a numerical value for @xmath133 of 1.68 . we can then utilize the measured values of @xmath128 of @xmath134 ppm and @xmath135 ppm for the two systems , respectively ( see table [ tab : sys ] ) . similarly , we take the values for @xmath21 and @xmath22 for each system from table [ tab : sys ] . finally , we solve for @xmath35 to find : @xmath136 where the error bars are statistical only and do not include any systematic uncertainties in the use of eqn . ( [ eqn : elv ] ) to evaluate @xmath35 from the elv amplitude . the value of @xmath35 found for kic 9164561 from the rvs is @xmath137 ( sect . [ sec : spec ] ) which is in quite satisfactory agreement with @xmath138 . on the other hand , the value of @xmath35 for kic 10727668 found from the elv analysis ( eqn . ( [ eqn : qelv ] ) ) , while still in marginal agreement with the value obtained from the doppler boosting analysis ( @xmath139 ) , would imply a white dwarf mass as high as @xmath140 . this is quite on the high side of what we expect for this system ( see sect . [ sec : discuss ] ) . in order to determine the sizes of the binary systems we need the total masses . in the case of kic 9164561 , where a direct rv measurement has been made , the total system mass is 2.22 @xmath31 0.07 @xmath0 . with the application of kepler s third law , this leads immediately to an orbital separation of @xmath141 . for kic 10727668 the doppler boosting analysis yielded a mass ratio @xmath142 . if we tentatively accept this mass ratio , and apply kepler s third law , we find the orbital separation is @xmath143 . the reason that the uncertainty is quite small , in spite of the larger uncertainty in @xmath35 for this system , is that @xmath25 depends only on the cube root of the total system mass , and @xmath35 is a small number . these values for the orbital separation are summarized in table [ tab : sys ] . the radii of the primary and secondary stars follow immediately from the @xmath21 and @xmath144 values derived from the light curve analyses and the values of the orbital separations . these radii are summarized in table [ tab : sys ] . of particular note are the large radii of the two white dwarfs , 0.277 @xmath116 and 0.151 @xmath116 for kic 9164561 and kic 10727668 , respectively , which indicate that these stars are quite thermally bloated . the degenerate radii for cold white dwarfs of either he or hybrid heco compositions , are given by eqn . ( 4 ) in nelson & rappaport ( 2003 ) . the degenerate radii for the white dwarfs in kic 9164561 and kic 10727668 are 0.020 @xmath116 and 0.018 @xmath116 , respectively , implying thermal bloating factors of 14 and 8 . we carried out an eclipse timing analysis for both binaries discussed in this paper . we used a code to produce the observed minus calculated curve ( ` @xmath15 ' curve ) that was tested in our search for triple star systems ( rappaport et al . only the @xmath15 curve for kic 9164561 showed an interesting non - linear behavior . the results are shown in fig . [ fig : omc ] with black points for the average of the @xmath15 curves for the primary and secondary eclipses . the @xmath15 curve shows two notable features : ( i ) a clear curvature , and ( ii ) small - amplitude oscillations with a period of @xmath3170 days . the small amplitude oscillations are due to an artifact that results from a beat between the orbital period and _ kepler _ long - cadence sampling ( see eqns . ( 4 ) and ( 5 ) in rappaport et al . 2013 ) . the quadratic behavior in the @xmath15 curve is highly significant ( 35 @xmath95 ) , and can be interpreted as either a change in the orbital period , or as a roemer delay due to the motion of the binary induced by a third body . the first option requires changes in the binary orbital period on a characteristic timescale of @xmath145 years . since there is no significant mass transfer taking place in the system at the current time , this can not be the cause of the apparent orbital period changes . one would have to invoke another mechanism , such as the applegate effect ( applegate & patterson 1987 ; applegate 1992 ; kotze & charles 2010 ) to explain the short timescale changes in the orbit . we focus here on a more interesting and , in our view , more likely scenario , i.e. , that there is a third body in the system . it is then possible to constrain the orbital radius of such a third body as well as its mass , if we restrict our attention to coplanar and circular orbits . otherwise , the possibilities are nearly limitless . the results of such constraints are shown in fig . [ fig : constraints ] . we find that the orbital period must lie in the range of @xmath146 years in order to provide at least a somewhat satisfactory fit to the @xmath15 curve ( @xmath147 limits ) . the blue curve shows the best fit ( projected ) orbital radius as a function of @xmath148 . the red curve is the corresponding mass function , while the green curve is the mass of the third body under the assumptions that the mass of the inner binary is @xmath149 and that the orbital inclination of the third body orbit is @xmath150 . since the mass of the third body , if it is assumed to be on or near the zams , can not exceed @xmath32 @xmath0 without affecting the spectrum , rv analysis , and the third light of the system in ways that are not observed ( see sects . [ sec : spec ] and [ sec : ccf ] ) , the orbital period is further restricted to the range of @xmath151 years . if we adopt the formal constraint on the mass of a main - sequence third star in the system from sec . [ sec : ccf ] of @xmath152 we find that : @xmath153 and @xmath154 years . to this point we have not yet considered the possibility that the putative third body might be a white dwarf . the orbital period would then be restricted to the range @xmath155 years since the mass would have to be @xmath156 ( see fig . [ fig : constraints ] ) . if there had been a prior episode of mass transfer from the progenitor of the outer body white dwarf to the inner binary ( kic 9164561 ; see , e.g. , tauris & van den heuvel 2014 ) then , according to eqn . ( [ eqn : mwd_p ] ) , the mass would lie in the range of @xmath157 . such white dwarfs would have cooled to the point where they would contribute very little light to the system ( @xmath158 ) . based on the inferred luminosity of kic 9164561 , and the appropriate bolometric corrections , we infer that its distance from the earth is @xmath159 pc . typical orbital separations to the putative third body , for periods in the range of @xmath160 years , are expected to be @xmath161 au . at a distance of 390 pc , the corresponding angular separations are then @xmath162 mas . this may be detectable with _ gaia_. c c c parameter & kic 9164561 & kic 10727668 + [ 0.5ex ] ra ( j2000 ) & 19h 42 m 27.64s & 19h 21 m 45.58s + dec ( j2000 ) & 45@xmath115 30@xmath163 17.0@xmath164 & 48@xmath115 02@xmath163 42.4@xmath164 + @xmath33 [ days ] & 1.2670400 @xmath165 & 2.305897 @xmath166 + @xmath24 [ mjd]@xmath167 & 64.61956 @xmath168 & 1207.3677 @xmath169 + @xmath170 [ mag . ] & 13.71 & 16.32 + @xmath171 & 0.1091 @xmath172 & 0.0893 @xmath173 + @xmath21 & 0.397 @xmath174 & 0.170 @xmath175 + @xmath25 [ @xmath116 ] & 6.40 @xmath177 & 9.95 @xmath31 0.15 + @xmath178@xmath179 & 0.0285 @xmath180 & 0.0245 @xmath180 + inclin . [ deg ] & 71.59 @xmath181 & 86.45 @xmath182 + + @xmath123 @xmath183 $ ] & @xmath184 & @xmath185 + @xmath186 [ @xmath116 ] & 2.54 @xmath31 0.03 & 1.69 @xmath31 0.04 + @xmath27 [ k ] & 7870 @xmath31 150 & 9696 @xmath31 300 + @xmath187 [ @xmath188 & 22.3 @xmath31 1.5 & 22.9 @xmath31 3.1 + @xmath189 [ g / cc ] & 0.171 @xmath310.04 & 0.63 @xmath31 0.04 + @xmath190 [ cgs ] & 3.93 @xmath31 0.02 & 4.32 @xmath31 0.04 + + @xmath122 [ @xmath0 ] & 0.197 @xmath31 0.005 & 0.266 @xmath31 0.035 + @xmath192 [ @xmath116 ] & 0.277 @xmath31 0.005 & 0.151 @xmath31 0.004 + @xmath193 [ k ] & 10410 @xmath31 200 & 14110 @xmath31 440 + @xmath194 [ @xmath188 & 0.82 @xmath31 0.07 & 0.82 @xmath31 0.11 + @xmath195 [ g / cc ] & 12.8 @xmath31 1.2 & 120 @xmath3119 + @xmath196 [ cgs ] & 4.84 @xmath31 0.04 & 5.55 @xmath31 0.06 + + @xmath197 [ @xmath198@xmath199 & @xmath200 @xmath31 27 & 143 @xmath31 20 + @xmath201 [ @xmath202@xmath203 & 2385 @xmath31 36 & 557 @xmath31 22 + @xmath204 [ @xmath205 & @xmath206 @xmath31 27 & @xmath207 @xmath31 20 + @xmath208 [ @xmath209@xmath210 & @xmath134 @xmath31 35 & @xmath135 @xmath31 21 + @xmath211 [ @xmath212 & @xmath213 @xmath31 28 & @xmath214 @xmath31 19 + @xmath215 [ @xmath216 & @xmath217 @xmath31 36 & @xmath218 @xmath31 23 + [ tab : sys ] here we discuss the properties of the kic 9164561 and kic 10727668 systems and some of the evolutionary paths that may have led to their formation . we can then check whether the expected system parameters match those determined empirically . there are two important evolutionary pathways that could lead to the formation of low - mass helium white dwarfs in close binaries ( see , e.g. , van kerkwijk et al . 2010 ; carter et al . 2011 ; breton et al . the first channel includes one episode of common envelope ( ce ) evolution wherein the more massive star of the primordial binary overflows its roche lobe after having become a giant with a degenerate core , and then undergoes dynamically unstable mass transfer onto its binary companion ( see , e.g. , taam et al . 1978 ; webbink 1984 ; taam and bodenheimer 1992 ) . the rapid rate of mass transfer causes the companion to spiral inward towards the degenerate core of the giant while concomitantly expelling the envelope of the giant into the ism . the resulting binary is composed of a degenerate white dwarf in a tight circular orbit with a largely unchanged companion star . the second channel involves stable mass transfer between the components of the progenitor binary . unfortunately the degree to which mass transfer is conservative and the effectiveness of orbital angular momentum loss mechanisms ( e.g. , magnetic braking ; see , for example , verbunt & zwaan 1981 ) are uncertain . for reasonable physical assumptions in regard to the mass transfer and angular momentum loss processes , this channel can explain the formation and observed properties of these two systems ( see nangreaves et al . 2015 , and references therein ) . _ ( i ) the common envelope channel _ although a physically realistic 3-d hydrodynamic simulation of ce evolution has not yet been carried out , a reasonably good approximation of the effects of the process can presumably be obtained from first principles using energy conservation . as the companion star moves at high velocity through the envelope of the giant , a fraction @xmath219 of the orbital energy is transferred to the tenuous envelope material . the companion star then spirals inward and more and more energy and angular momentum are transferred to the giant envelope . the gravitational potential energy that is released may be sufficient to eject the envelope . it is assumed that this process occurs on a timescale much shorter than the thermal timescale of the giant star . one of the more commonly used ce energy constraints was analyzed by dewi & tauris ( 2000 ) . it is given by @xmath220 \label{eqn : ce}\ ] ] where @xmath221 , @xmath222 , and @xmath223 are the core and envelope masses of the wd progenitor , and the mass of primordial secondary , respectively ( see , e.g. , webbink 1984 ; pfahl et al . note that @xmath219 is a dimensionless measure of the efficiency of the conversion of orbital energy that goes toward unbinding the envelope of the giant ( the energy required to eject the envelope , @xmath224 , must satisfy @xmath225 ) and is usually taken to be @xmath226 . the dimensionless factor @xmath227 is intended to take into account the actual binding energy of the giant at the onset of the ce phase . it largely depends on the ratio of the core mass to the pre - ce - phase mass of the giant . values of @xmath227 have been tabulated for many different evolutionary states and masses of giants by tauris & dewi ( 2001 ) and podsiadlowski et al . ( 2003b ) . for the types of giants that could produce helium white dwarfs , @xmath227 is of order unity , normally having a value between 0.5 and 1.0 , which is possible in certain circumstances , this only makes the argument presented here more robust . ] . also note in equation ( [ eqn : ce ] ) that @xmath221 , the core mass of the giant , is approximately equal to the final white dwarf mass , @xmath8 . thus the original mass of the primary can be written as @xmath228 and its radius is given by @xmath229 . finally , @xmath230 is the separation of the white dwarf and secondary star in their mutual circular orbit after the ce phase while @xmath231 is the separation of the components of the progenitor binary . using kepler s third law , equation ( [ eqn : ce ] ) can be rewritten as @xmath232\left(\frac{r_{1,0}}{a_0}\right)=1 \label{eqn : ce2}\ ] ] where @xmath233 is a dimensionless parameter that is @xmath226 . also @xmath234 ) and @xmath235 ) are the mass ratios of the primordial and present day binaries , respectively . at the onset of ( unstable ) mass transfer , the ratio , @xmath236 , is approximately equal to the roche lobe radius of the progenitor primary divided by the primordial separation ; its value is constrained by roche geometry and can be expressed as a simple function of @xmath237 ( see , eggleton 1983 ) . given that the radius of the red giant will have a very strong dependence on @xmath238 and a very weak dependence on @xmath11 , the constraint of roche geometry allows the approximation of @xmath239 by a function of @xmath221 ( see , e.g. , rappaport et al . 1995 ) . assuming a specific value of @xmath240 , and using the observed orbital period of the binary ( @xmath241 ) and the inferred mass of the present - day companion @xmath123 ( @xmath242 ) , strong limits can be imposed on the possible values of the white dwarf mass ( we take @xmath243 ) , by solving eqn . ( [ eqn : ce2 ] ) implicitly for @xmath8 and applying the constraint that @xmath244 ( i.e. , @xmath245 ) . it is reasonable to suppose that @xmath240 will lie in the range 0.5 to 1.0 . taking the upper limit of this range for @xmath240 , we find that the only acceptable values of the core mass for kic 9164561 are @xmath246 while for kic 10727668 they are @xmath247 . assuming the lower limit of this range of @xmath240 , we find that the only acceptable values of the core mass for kic 9164561 are @xmath248 , while for kic 10727668 they are @xmath249 . these ranges of values , by definition , while the upper limits correspond to @xmath250 to take into account the fact that we are considering only primaries of masses @xmath251 so that only he wds are produced . ] of @xmath221 are clearly incompatible with the estimated values of @xmath8 for both kic 9164561 and kic 10727668 based on the cooling evolution of the white dwarfs ( see the next section ) , as well as with the empirically determined value for kic 9164561 from rvs . we conclude that for a reasonable range of common envelope efficiencies , the formation of either or both of these systems via the ce channel is rather unlikely . _ ( ii ) the algol channel _ the other channel involves stable mass transfer during the evolution of the progenitor binary . in this scenario , the star that is more massive ( the donor ) evolves more quickly and fills its roche lobe on its nuclear timescale before its less massive companion ( the accretor ) can do so . even if the mass ratio is as large as @xmath252 , it is entirely possible for the mass transfer to proceed stably , if the donor has a substantial radiative envelope ( nelson & goliasch 2014 ) . as an example of evolution that could explain the formation of the two observed systems , consider a 1.5 @xmath0 donor that has evolved beyond the tams and lies in the hertzsprung gap when it fills its roche lobe . this star can stably transfer mass to a 1.2 @xmath0 companion on approximately a nuclear and/or thermal timescale ( case ab evolution ) . at this mass transfer rate ( of @xmath3@xmath253 yr@xmath51 ) , the accretor is not likely to expand and fill its own roche lobe . during the mass transfer process , the donor can evolve up the red giant branch and form a substantial degenerate helium core . since the radius of the giant has a very strong dependence on @xmath221 but a very weak dependence on its total mass , the giant will continuously expand with a concomitant increase in the core mass until the envelope surrounding the core is so depleted ( to a mass of the order of 0.01 @xmath0 ; the remainder having been transferred to the companion star ) that it can not be sustained and ultimately collapses onto the remnant degenerate helium core . if approximately 1@xmath0 is transferred conservatively to the accretor during this process , the resulting system may then resemble either of the observed ones . details concerning the possible progenitor evolutions will be discussed in a future paper ( nangreaves et al . 2015 ) . assuming that stable mass transfer from a red giant led to the formation of the bloated helium wds in both kic 9164561 and kic 10727668 , it is possible to check whether the observed properties of the wds are consistent with cooling models of degenerate helium dwarfs . moreover , if the properties of either wd are found to be consistent , it is then possible to use it as a chronometer in order to determine the age of the binary after the termination of mass transfer , i.e. , just after the collapse of the giant to form a helium wd ( hewd ) . the study of the formation and cooling of hewds from progenitor red giants has been carried out by a number of groups ( see , e.g. , podsiadlowski et al . 2003 ; nelson et al . 2004 ; althaus et al . 2005 , 2013 ; istrate et al . we use the code described by nelson et al . ( 2004 ) to analyze the cooling evolution of very low - mass helium wds . it is assumed that the giant ( i.e. , the wd progenitor ) had a solar metallicity ( @xmath254 ) , and that its initial mass was between 1.0 and 1.5 @xmath0 . the initial hydrogen abundance ( by mass ) of the progenitor star while on the zams is taken to be equal to 0.71 and the ratio of the mixing length to the pressure scale - height ratio ( @xmath255 ) is set equal to 1.5 . the initial orbital separation was carefully selected so that only very low mass helium wds were created . diffusion ( gravitational settling and radiative levitation ) was not included in the computations because diffusion is a very fragile physical process that can be negated by turbulence in the outer layers of these rapidly rotating ( proto-)white dwarf stars . moreover , the inclusion of diffusion has little effect on the cooling times for the models of interest ( i.e. , bloated helium wds ) because these systems are very young and have not even evolved to reach their maximum surface temperatures before joining the usual wd cooling track in the hr diagram . it is only after the wds have passed through their maximum @xmath45 that the compression of the outer hydrogen - rich layers , due to contraction , can cause the residual hydrogen to start to burn , leading to a rapid expansion of the outer layers as a result of a thermonuclear runaway ( tnr ) . in figure [ fig : cooling ] , we show the results of the cooling evolution of several models that have masses near or smaller than the tnr lower mass limit , i.e. , @xmath256 . the evolution of the radius of the wd is shown as a function of its effective temperature , @xmath45 . when the envelope of the progenitor giant collapses , the radius of the proto - wd continually decreases , as long as no tnrs occur , until it asymptotically reaches its zero - temperature radius . the surface temperature of the object initially increases ( this increasing temperature defines the proto - wd phase ) until it reaches a maximum value of @xmath45 where it then follows a cooling track and asymptotically approaches zero temperature . assuming that the white dwarfs in both kic 9164561 and kic 10727668 formed according to the scenario discussed above , we can infer reasonably precise values of the mass based on the observed radii and @xmath45 values superposed on the cooling tracks ( see fig . [ fig : cooling ] ) . we find that the mass of the wd in kic 9164561 is @xmath257 , and that of the wd in kic 10727668 is @xmath258 . these masses agree with those inferred from the cooling evolutions of althaus et al . ( 2013 ) to within 5 to 10% ( their values of the mass being smaller ) , and to within @xmath259% for the values inferred from the work of istrate et al . we also carried out this same type of wd cooling calculations with the mesa code ( paxton et al . 2011 ) and found very similar results . the ages of these objects as measured from the cessation of mass transfer are approximately 925 myr and 950 myr , respectively . if the companion star in kic 9164561 has a mass of @xmath260 , then the time required to reduce its average density to the inferred value of 0.17 g @xmath42 is in good agreement with the theoretical predictions for the evolution of a solar metallicity star . the orbital period predicted by our models is approximately 1.7 days , but it should be noted that our assumptions about the degree of mass conservation during the mass - transfer phase and the magnitude of angular momentum losses leading to the formation of the giant are not precisely known , and this uncertainty could have a significant impact on the final value of the orbital period . as for kic 10727668 , the predicted orbital period from our models almost matches the observed one . however , the predicted age and somewhat higher mass of the companion star are very problematic because the observationally inferred average density implies that the companion star has not significantly evolved from its initial position on the zams . based on the implied cooling age of the system and a mass of 2.2 @xmath0 , we would expect the star to be much closer to the tams with a correspondingly lower density ( contrary to what is implied in figure 3 ) . if the mass of the companion were substantially less ( say , @xmath671.7 @xmath0 ) , then a wd composed primarily of helium and with a mass of @xmath261 would be entirely compatible with the observations . based on the cooling evolution , we conclude that the theoretical models for the evolution of kic 9164561 lead to properties that agree well with the observationally inferred properties . the predictions for kic 10727668 are problematic , especially given the implied cooling age of the white dwarf in the system . for an age in the range of 100200 myr , a white dwarf mass of @xmath262 ( inferred from the doppler boosting effect ; see sect . [ sec : db ] ) , and @xmath263 , it is very difficult to model the properties with any type of solar metallicity hewd . to understand why this is true , consider the evolutionarily limiting case for which all of the hydrogen has been stripped from the atmosphere and envelope of the hewd . based on the wd cooling tracks of nelson & davis ( 2000 ) , a 0.27 @xmath0 wd of this type can reach a temperature of 14,000k at either of two ages : ( i ) considerably less than 10 myr ; or , ( ii ) @xmath264 myr . the luminosity of the wd for the latter case is two orders of magnitude smaller than what is inferred ( see table [ tab : sys ] ) and can be automatically ruled out . for the other solution ( corresponding to the early - wd phase before the maximum @xmath45 is reached ) , the luminosity is approximately correct but a cooling age of only several million years after the birth of the wd seems to be unacceptably small based on probability arguments . if a thick layer of hydrogen were to be added to the envelope , the cooling would be slowed but incompatibilities with the observations would persist . for example , the luminosity of the hewd while it is still young would be considerably higher and it would likely undergo a tnr . perhaps we are observing the hewd while it is undergoing one of its tnr cycles ; the probability of observing it at this epoch may be unacceptably small . another possible explanation might be a difference in the actual metallicity from that which we assumed ; however , making the metallicity smaller generally makes the incompatibilities larger . thus if the mass of the wd in kic 10727668 is actually about 0.27 @xmath0 , then it might have been derived from a higher mass progenitor ( @xmath265 ) or perhaps via a different evolutionary pathway than we have considered here . we plan to investigate all of these possibilities in a future paper . for the sake of completeness , we have included the robust measurements of effective temperatures and radii of eight other eclipsing or tidally distorted , extremely - low mass wds ( found in double degenerate systems ; see hermes et al . the observations are plotted in fig . [ fig : cooling ] as open blue circles . it should be noted that these additional wds have properties that are reasonably similar to the six hot wds found by kepler except that they are typically less bloated . based on the cooling models , we expect the masses of all of these systems to reside in the range of 0.15 - 0.25 @xmath0 , which is generally consistent with the values reported in hermes et al . ( 2014 ) . finally , in this regard , we note that these white dwarfs have a very different evolutionary scenario from the systems examined in our work in that they are likely the result of two common - envelope episodes . here we examine very briefly what the future holds for systems such as koi 74 , koi 81 , koi 1224 , koi 1375 , kic 9164561 , and kic 10727668 . when these systems evolve to the point where the primary a star fills its roche lobe and commences mass transfer onto the white dwarf , their he core masses will range from 0.19 to 0.27@xmath266 for the shortest ( kic 9164561 ) and longest ( koi 81 ) orbital periods of these systems , respectively ( as estimated using eqn . ( [ eqn : mwd_p2 ] ) ) . once the mass transfer onto the white dwarf begins , it will be dynamically unstable due to the very large mass ratio ( typically 10:1 ) , and a common envelope will ensue . using eqn . ( [ eqn : ce ] ) , we can compute an approximate expression for the ratio of the final orbital separation to the initial orbital separation , and we find : @xmath267 where @xmath8 is the mass of the hot white dwarfs being studied in these systems , @xmath222 and @xmath221 will be the envelope and core mass of the a star when it evolves to fill its roche lobe , @xmath268 , @xmath240 is the common envelope parameter of order unity discussed above , and @xmath269 is the roche - lobe radius of the a star in units of the orbital separation ( with an expected numerical value of @xmath270 ) . if we apply eqn . ( [ eqn : aoai ] ) to estimate the final orbital separation that will result from this upcoming common envelope phase , we find that @xmath271 . the current orbital separations of the six bloated hot white dwarfs in the _ kepler _ field range from 6.4 to 48@xmath272 . from this we can infer that the final orbital separations will be @xmath273 , for kic 9164561 to koi 81 , respectively . all of these will be binaries comprised of two roughly equal mass white dwarfs each of @xmath274 , if the core of the a star does not merge with the initial low - mass white dwarf in the system . the corresponding orbital periods and roche - lobe radii would range from 3/4 to 14 min , and @xmath275 , respectively . all but the largest of these roche - lobe radii are too small to harbor the white dwarf cores that will be unveiled by the common envelope . thus , we conclude that the orbital periods of the group of double degenerate systems recently studied by brown et al . ( 2014 ) and hermes et al . ( 2014 ) result from much wider systems . this is emphasized by the fact that the masses of the primary in the double degenerate binaries in the latter studies are typically much more massive than would result from the six binaries containing bloated hot white dwarfs in the _ kepler _ field . in this study we report two new low - mass , thermally bloated , hot white dwarfs in binary systems with companion a stars . the systems are similar in their basic nature to those in koi 74 , koi 81 , koi 1224 , and koi 1375 . this brings to six the total number of such systems in the _ kepler _ field . of the six systems , the two new systems have the shortest orbital periods at 1.27 d and 2.3 d , while kic 9164561 has the largest , i.e. , the most thermally bloated , of the six white dwarfs its white dwarf has a radius that is 14 times larger than its fully degenerate radius it is more than 1/2 @xmath116 in _ diameter _ ! the mass of the white dwarf in kic 9164561 is 0.197 @xmath31 0.005 @xmath0 and is determined from direct rv measurements of the primary a star as well as some inferences about the mass of the a star ( i.e. , the mean density and @xmath45 lead to an estimate for @xmath123 ) . by contrast , the mass of the white dwarf in kic 10727668 is more uncertain at 0.27 @xmath31 0.03 @xmath0 . since there are no rv measurements of this fainter system , we relied on measurements of the doppler boosting effect to determine the orbital velocity of the a star . the accuracy of this latter mass is not quite sufficient for detailed comparisons with binary evolution or white - dwarf cooling models . we strongly advocate that a larger telescope be used to determine an accurate rv for this system . binary evolution scenarios involving two low - mass , e.g. , @xmath67 1.1 and 1.5 @xmath0 , primordial stars that undergo stable , conservative mass transfer can nicely explain the current mass and evolutionary state of both the companion star and the hot white dwarf in kic 9164561 . for this system , we estimate an age of @xmath3925 myr since the envelope of the white dwarf progenitor was transferred to the current a star . by contrast , it is more difficult to explain the current state of both the a star and its white dwarf companion in kic 10727668 . if the white dwarf actually has a mass as high as is indicated by the doppler boosting technique ( i.e. , 0.27 @xmath0 ) then it is difficult to understand how the white dwarf can still be so thermally bloated . if , on the other hand , we take the uncertainty seriously , and allow for a mass as low as , say , 0.22 @xmath0 , then there is still a discrepancy between the inferred cooling age of @xmath67 950 myr , and the apparent location of the a star in the hr diagram which is still near the zams . the growing statistical ensemble of these systems should make it worthwhile to ( 1 ) make renewed attempts to measure the rvs in four of the remaining bloated hot white dwarf systems in order to better determine @xmath8 , and ( 2 ) directly model the binary evolution of such systems with a code that can evolve two stars _ simultaneously _ , e.g. , mesa ( paxton et al . 2011 ) . _ note added in manuscript _ : after this work was completed we became aware of an impressive collection of 18 similar bloated hot white dwarf systems found in the wasp photometric database ( maxted et al . 2013 ; 2014 ; and references therein ) . two of these have accurate measurements of the mass , radius , and @xmath45 for the bloated hot white dwarf and show multi - periodic pulsations . these are now also shown in fig . [ fig : cooling ] . we are grateful to the entire _ kepler _ team for making this study possible . we thank jon swift for taking exploratory spectra of kic 10727668 to see if rv measurements can be acquired in the future , and dave latham for his early contribution to this project . the rv observations were made with the nordic optical telescope , operated by the nordic optical telescope scientific association at the observatorio del roque de los muchachos , la palma , spain , of the instituto de astrofisica de canarias . we gratefully acknowledge that two of the rv points were acquired by pilar montaes - rodrguez , and an additional one was obtained during time allocated to hans j. deeg . we thank antonio frasca of catania astrophysical observatory for some important tips in combining the model spectra . mattia gandolfi was quite inspirational for this work . we are grateful for the cat and not service time awarded to this project , and to the staff members at the not for their valuable and unique support during the observations . this research was supported in part by the mineco esp2013 - 48391-c4 - 2-r grant . thanks the natural sciences and engineering research council ( nserc ) of canada for financial support , calcul qubec for providing computing facilities , and ernie dubeau and heather nangreaves for technical assistance . r.s.o . acknowledges nasa support through the kepler participating scientist program . this research has made use of the nasa exoplanet archive , which is operated by the california institute of technology , under contract with the national aeronautics and space administration under the exoplanet exploration program . the data presented in this article were obtained from the mikulski archive for space telescopes ( mast ) . stsci is operated by the association of universities for research in astronomy , inc . , under nasa contract nas5 - 26555 . support for mast for non - hst data is provided by the nasa office of space science via grant nnx09af08 g and by other grants and contracts . we made use of j - band images that were obtained with the united kingdom infrared telescope ( ukirt ) which is operated by the joint astronomy centre on behalf of the science and technology facilities council of the u.k . brown , w.r . , kilic , m. , allende prieto , c. , gianninas , a. , & kenyon , s. 2013 , apj , 769 , 66 allende prieto c. , barklem p. s. , lambert d. l. , cunha k. , 2004 , a&a , 420 , 183 althaus , l.g . , serenelli , a.m. , panei , j.a . , crsico , a.h . , garca - berro , e. , & scccola , c.g . 2005 a&a , 435 , 631 althaus , l.g . , miller bertolami , m.m . , crsico , a.h . 2013 , a&a , 435 , 631 applegate , j.h . 1992 , apj , 385 , 621 applegate , j.h . , & patterson , j. 1987 , apj , 322 , l99 barnes , s.a . 2007 , apj , 669 , 1167 batalha , n. , et al . 2010 , apj , 713 , l109 bloemen , s. , marsh , t.r . , degroote , p. , et al . 2012 , mnras , 422 , 2600 breton , r. , rappaport , s. , van kerkwijk , m.h . , & carter , j.a . 2012 , apj , 748 , 115 bruntt , h. , bedding , t.r . , quirion , p .- o . , et al . 2010 , , 405 , 1907 buchhave , l.a . , bakos , g.a . , hartman , j.d . , et al . 2010 , apj , 720 , 1118 carter , j.a . , rappaport , s. , & fabrycky , d. 2011 , apj , 728 , 139 castelli , f. & kurucz , r.l . 2004 , eprint arxiv : astro - ph/0405087 conroy , k. e. , pra , a. , stassun , k. g. , et al . 2014 , , 147 , 45 eggleton , p.p . 1983 , apj , 268 , 368 ehrenreich , d. , lagrange , a .- m . , bouchy , f. et al . 2011 , a&a , 525 , a85 . frandsen , s. & lindberg , b. 1999 , in `` astrophysics with the not '' , proceedings eds : karttunen , h. & piirola , v. , anot . conf , 71 gandolfi , d. , alcal , j.m . , leccia , s. , et al . 2008 , , 687 , 1303 gandolfi , d. , parviainen , h. , fridlund , m. , et al . 2013 , a&a , 557 , 74 gray , r.o . & corbally , c.j . 1994 , aj , 107 , 742 gray , d.e . 2005 , ` the observation and analysis of stellar photospheres ' , ( uk : cambridge university press ) . han , z. , podsiadlowski , ph . , maxted , p.f.l . , marsh , t.r . , & ivanova , n. 2002 , mnras , 336 , 449 han , z. , podsiadlowski , ph . , maxted , p.f.l . , & marsh , t.r . 2003 , mnras , 341 , 669 hermes , j.j . , brown , w.r . , kilic , m. , et al . 2014 , apj , 792 , 39 holman , m.j . , winn , j.n . , latham , d.w . 2006 , apj , 652 , 1715 huber , d. , silva aguirre , v. , matthews , j. m. , et al . 2014 , apjs , 211 , 2 iben , i. , tutukov , a.v . , & yungelson , l.r . 1995 , apjs , 100 , 233 istrate , a.g . , tauris , t.m . , langer , m. , & antoniadis , j. 2014 , a&a , in press joss , p.c . , & rappaport , s. 1983 , nature , 304 , 419 kotze , m.m . , & charles , p.a . 2010 , mnras , 402 , l16 kirk , b. , pra , a. et al . 2015 , in preparation . lin , j. , rappaport , s. , podsiadlowski , ph . , nelson , l. , paxton , b. , & todorov , p. 2011 , apj , 732 , 70l loeb , a. , & gaudi , b.s . 2003 , apj , 588 , l117 matijevic , g. , prsa , a. , orosz , j.a . , welsh , w. , bloemen , s. , & barclay , t. 2012 , aj , 143 , 123 marsh , t.r . , nelemans , g. , & steeghs , d. 2004 , mnras , 350 , 113 maxted , p.f.l . , serenelli , a.m. , miglio , a. , et al . 2013 , nature , 498 , 463 . maxted , p.f.l . , serenelli , a.m. , marsh , t.r . , cataln , s. , mahtani , d.p . , & dhillon , v.s . 2014 , mnras , 444 , 208 morris , s.l . , & naftilan , s.a . 1993 , apj , 419 , 344 . nangreaves , h. , nelson , l. et al . 2015 , in preparation nelemans , g. 2010 , ap&ss , 329 , 25 nelson , l. & rappaport , s. 2003 , apj , 598 , 431 nelson , c. & eggleton , p.p . 2001 , apj , 552 , 664 nelson , l.a . , dubeau , e. , & maccannell , k.a . 2004 , apj , 616 , 1124 nelson , l.a . , & goliasch , j. 2014 , in preparation paxton , b. , bildsten , l. , dotter , a. , herwig , f. , lesaffre , p. , & timmes , f. 2011 , apjs , 192 , 3 paczyski , b. 1983 , nature , 304 , 421 pfahl , e. , rappaport , s. , & podsiadlowski , ph . 2003 , 597 , 1036 podsiadlowski , ph . , han , z. , & rappaport , s. 2003a , mnras , 340 , 1214 podsiadlowski , ph . , rappaport , s. , & han , z. 2003b , mnras , 341 , 385 rappaport , s. , di stefano , r. , & smith , j.d . 1994 , apj , 426 , 692 rappaport , s. , podsiadlowski , ph . , joss , p.c . , de stefano , r. , & han , z. 1995 , mnras , 273 , 731 rappaport , s. , deck , k. , levine , a. , borkovits , t. , carter , j. , el mellah , i. , sanchis - ojeda , r. , & kalomeni , b. 2013 , apj , 768 , 33 rowe , j. , borucki , w. , koch , d. , et al . 2010 , apj , 713 , 150 sanchis - ojeda , r. , rappaport , s. , winn , j.n . , levine , a. , kotson , m.c . , latham , d.w . , & buchhave , l.a . 2013 , apj , 774 , 54 savonije , g.j . 1983 , nature , 304 , 422 savonije , g.j . 1987 , nature , 325 , 416 seager , s. , & malln - ornelas , g. 2003 , apj , 585 , 1038 smith , j.c . , et al . 2012 , pasp , stumpe , m.c . , van cleve , j.e . , et al . 124 , 1000 stumpe , m.c . , smith , j.c . , van cleve , j.e . , et al . 2012 , pasp , 124 , 985 taam , r.e . , bodenheimer , p. , & ostriker , j.p . 1978 , apj , 222 , 269 taam , r.e . & bodenheimer , p. 1992 , in `` nato advanced research workshop on x - ray binaries and the formation of binary and millisecond radio pulsars '' , p. 281 tauris , t.m . & dewi , j.d.m . 2001 , a&a , 369 , 170 tauris , t.m . & van den heuvel , e.p.j . 2014 , apj , 781 , 13 telting , j.h . , avila , g. , buchhave , l. , et al . 2014 , an , 335 , 41 tran , k. , levine , a. , rappaport , s. , borkovits , t. , csizmadia , sz . , & kalomeni , b. 2013 , apj , 774 , 81 . van kerkwijk , m.h . , rappaport , s. , breton , r. , justham , s. , podsiadlowski , ph . , & han , z. 2010 , apj , 715 , 51 verbunt , f. , & zwaan , k. 1981 , a&a , 100 , l7 webbink , r.f . 1984 , apj , 277 , 355 yi , s. , demarque , p. , kim , y .- c . , lee , y .- w . , ree , c. h. , lejeune , t. , & barnes , s. 2001 , apjs , 136 , 417

the accurate localisation of gamma - ray bursts ( grbs ) through their optical and x - ray afterglows has enabled detailed studies of their environments . selection by the unobscured gamma - ray flash alone has allowed the discovery of a unique sample of galaxies which span an enormously wide range of redshifts from @xmath0 @xmath1 0.009 ( grb980425 , e.g. tinney et al . 1998 ; galama et al . 1998a ) to 6.3 ( grb050904 , kawai et al . 2005 , 2006 ) . the subset of long - duration ( @xmath2 2 s ) grbs are almost certainly caused by the collapse of certain massive stars to black holes , confirmed by observations of supernova components in the late - time afterglows of a number of long - grbs ( woosley & bloom 2006 ; kaneko et al . 2006 ) and by the observed location of grbs in uv - bright regions within their host galaxies ( bloom et al . 2002 ; fruchter et al . grbs are located in host galaxies which are generally small , faint , blue and highly star - forming ( e.g. chary et al . 2002 ; fruchter et al . 1999 ; le floch et al . hence , detailed and extensive host galaxy observations provide a wealth of information on the gas and dust properties of star - forming galaxies throughout cosmological history . accurately measuring the dust content of these galaxies is of great importance in , to name one example , the determination of their unobscured star formation rates where uncertainty in the correction for dust can easily dominate the errors on the measured star formation rates for high redshift galaxies ( e.g. pettini et al . 1998 ; meurer et al . absorption within our own galaxy along a particular line of sight can be estimated and removed , but absorption which is intrinsic to the grb host galaxy as a function of wavelength is unknown , and is especially difficult to determine given its dependence on metallicity and the possible existence of dusty intervening systems whose extinction curves can not be disentagled from those of the host galaxy . afterglow spectroscopy and/or photometry can be used to provide an estimate of the total extinction along the line - of - sight to the grb . if the host galaxy itself is bright and extended enough to be observed once the afterglow has faded , different lines - of - sight may be probed besides the off - centre uv - bright regions within which grbs are generally situated . extinction in the optical / uv regime due to dust grains is typically modelled using either milky way ( mw or galactic ) , large magellanic cloud ( lmc ) or small magellanic cloud ( smc ) extinction curves ( e.g. pei 1992 ) because these curves can be measured and so are well known , or with the calzetti extinction law derived empirically from uv observations of starburst and blue compact galaxies ( calzetti et al . it appears that dust content in most grb hosts produces an smc - like extinction law ( e.g. galama & wijers 2001 ; vreeswijk et al . 2004 ; stratta et al . 2004 ; kann et al . 2006 ; schady et al . in preparation ) , owing to an observed lack of the 2175 feature thought to be caused by carbonaceous dust grains ( draine & lee 1984 ) . this feature has , however , been clearly observed in grb afterglow spectra where the line of sight between us and the grb is intercepted by intervening systems : the best example to date , in which the extinction curve of an intervening system could for the first time be disentangled from that of the host galaxy , is grb060418 ( ellison et al . 2006 ) . in general , low amounts of optical extinction are found towards grbs , unexpected if grbs are located in dusty star - forming regions , whilst the x - ray spectra reveal a different picture ( first noted by galama & wijers 2001 ) . at x - ray wavelengths absorption is caused by metals in both gas and solid phase , predominantly oxygen and carbon ( see e.g. wilms et al . 2000 ) , and we often measure high values for the absorbtion columns . however , these absorption edges are shifted out of the x - ray observing window for high redshifts ( @xmath3 ) beyond which only large columns can be measured and there is a degeneracy between redshift and x - ray column density ( e.g. watson et al . the grb host metallicity is observed to be low ( compared with the milky way ) in measurements via optical spectroscopy of a dozen or so afterglows . host metallicities can reach values as low as 1/100 solar ( grb050730 , starling et al . 2005 ; chen et al . 2005 , but see also prochaska 2006 for potential caveats ) - even lower than found for the smc ( see figure 3 of fynbo et al . 2006 for an overview ) . this only increases any measured x - ray column , which is expressed as an equivalent hydrogen column density , @xmath4 . but here we note that metallicities are not generally obtainable for lower redshift grbs ( @xmath5 ) due to the hydrogen lyman-@xmath6 line lying in the far uv outside typical observing windows . the apparent discrepancy between optical and x - ray extinction resulting in high gas - to - dust ratios in grb host galaxies ( often far higher than for the mw , lmc or smc , e.g. grb020124 , hjorth et al . 2003 , but see schady et al . in preparation ) is not satisfactorily explained , though the suggestion that dust destruction can occur via the high energy radiation of the grb ( e.g. waxman & draine 2000 ) could possibly account for the discrepancy . it is thought that circumburst dust may be destroyed by sublimation of dust grains due to uv emission ( waxman & draine 2000 ) , sputtering ( draine & salpeter 1979 ) or dust grain heating and charging ( fruchter et al . alternative models for the extinction by dust grains , including skewing the dust grain size distribution towards larger grains , have been investigated , and in fact such a grain size distribution may result from exposure of the dust to the grb radiation field since destruction of small grains is more efficient than for larger grains ( perna et al . attempts to model the process of dust destruction have been made by e.g. perna & lazzatti ( 2002 ) . however , such models have not replaced smc - like ( low metallicity ) extinction as the best description of most grb environments ( e.g. stratta et al . 2004 ; kann et al . 2006 ) . traditionally the optical and x - ray spectra have been treated seperately in extinction studies . since the underlying spectrum is likely a synchrotron spectrum ( power law ( pl ) or broken power law ( bknpl ) , e.g. galama & wijers 1998 ) extending through both wavelength regions , it would be most accurate to perform simultaneous fits . more recently such fits have been made , either by fitting the x - ray spectrum individually and thereby transforming the model counts to flux to create a spectral energy distribution ( sed ) with the optical data in flux space ( e.g. stratta et al . 2005 ; watson et al . 2006 ) , or by using the _ swift _ xrt x - ray and uvot ultraviolet ( uv ) and @xmath7,@xmath8,@xmath9 band data together in a fit to the count spectrum ( e.g. blustin et al . 2005 ; schady et al . in preparation ) . the _ swift _ uvot data can be loaded directly into the xspec spectral fitting package and treated in count space owing to its calibration , which is generally not true for ground - based data . we present here an alternative method , which makes use of simultaneous fits in count space extending from near - infrared ( nir , in this case @xmath10 band ) to x - ray ( 10 kev ) to obtain the most accurate possible measurements of both the underlying continuum spectrum and the extinction . in section 4 we fit the broad - band seds ( from nir through x - ray ) of a subsample of the _ bepposax _ sample of grb afterglows to better measure the extinction properties of their host galaxies - a sample chosen for its availability of suitable data well studied in the seperate optical and x - ray band passes . section 2 outlines the data sample and reduction techniques . section 3 describes the method we use to model the broad - band seds and section 4 presents the results of our fitting through discussion of individual bursts and comparison with previous studies . in section 5 we discuss the implications of our findings for galaxy extinction curves , and summarise and compare the various methods now available to measure extinction in the hosts . we conclude by summarising our method and findings in section 6 . an analysis of the blastwave parameters and density profiles for the circumburst media obtained from these fits will be presented in a forthcoming paper ( starling et al . , paper ii ) . this sample of 10 long grbs observed with the _ bepposax _ narrow field instruments ( approx . 0.110 kev ) is chosen for the good availability ( 3 bands or more ) of optical / nir photometry ( tables 1 and 2 ) . the optical / nir bands available for each source and their references are listed in table [ tbl-3 ] . as these grbs are all previously studied , overlapping with the samples studied by galama & wijers ( 2001 ) and stratta et al . ( 2004 ) , this constitutes a good sample on which to first adopt this method of simultaneous sed fitting . all x - ray observations are taken from the _ bepposax _ data archives , using the lecs and mecs instruments raw data within the energy ranges 0.14 and 1.010 kev respectively . data have been reduced using the saxdas routines . we combined data from the mecs2 and mecs3 instruments ( except in the case of grb970228 , where we use the mecs3 instrument only , see stratta et al . 2004 ) , including a gain equalisation . we then combined multiple observations for each source and instrument type among the narrow field instruments , omitting the last observation if it was @xmath11 3 days later than the previous one , before extracting spectra . background x - ray spectra were taken from blank fields and count rates checked against the local background finding no adjustments necessary : the net count rates of the two types of field agree on average to within 0.0001 counts s@xmath12 . the latest canned arf and rmf files were used with mecs data , but were created for lecs observations at their off - axis source positions ( listed in table 3 of stratta et al . we group all spectra such that a minimum of 20 counts are in each bin in order to use the @xmath13 statistic . however , in the case of grb000926 there were very few counts in the x - ray spectrum so we have required only 10 counts per bin . there is a known offset between the normalisations of the lecs and mecs instruments . we fit for this offset in the x - ray spectra only , adding a constant - value free parameter to the model and adopting mecs as the reference for the lecs instrument . we fix the offset values in the sed of each grb to these values ( table [ tbl-1 ] ) . all temporal decay slopes , both for x - ray and optical lightcurves , have been taken from the literature and are listed together with galactic extinction corrections in table [ tbl-2 ] . optical and nir photometry was taken from the literature and from our own nir observations of grb990510 ( described in curran et al . in preparation ) . [ cols="^,^,^,^,^,^,^,^ " , ] the seds and the results of fits to the seds for all grbs in the sample are listed in table [ tbl-4 ] , and best fitting models are shown overlaid on the data in figure [ fig2 ] . figure [ fig3 ] shows a comparison of the absorption measurements with galactic , lmc and smc gas - to - dust ratios , which we discuss in the following section . this plot has been constructed in a number of previous works ( e.g. galama & wijers 2001 ; stratta et al . 2004 ; kann et al . 2006 ; schady et al . in preparation ) but here we show the observed distribution of @xmath14 and @xmath4 for the first time derived simultaneously from a fit to x - ray , optical and nir data . we find an excess in absorption above the galactic values particularly significant in two sources : grbs 000926 ( @xmath14 only ) and 010222 ( figure [ fig4 ] ) , whilst no significant intrinsic absorption is necessary in grbs 970228 and 990510 . the cooling break can be located in three of the afterglows : grbs 990123 , 990510 and 010222 and to all other seds a single power law is an adequate fit . details are given below for each individual afterglow . no significant absorption is measured for grb970228 . we find only a single power law is required and it is possible to pin down the power law slope relatively well ( @xmath13/dof @xmath15 11.6/14 ) . a well - defined temporal decay slope starting 1.9 days after trigger ( zeh et al . 2006 ) . preceding this time there is an increase in flux followed by an apparent flattening . the time of the x - ray observation log midpoint occurs a little before the 1.9-day break at 1.679 days , so we have extrapolated the optical data post 1.9 days back to 1.9 days and then assume that the evolution of the lightcurve is flat back to 1.679 days after trigger . to allow for a different behaviour before 1.9 days we include a constant value offset in the model between the optical and x - ray data , which we both fix at 1.0 and leave as a free parameter . the improvement in the fits when the offset is a free parameter is somewhat marginal . an f - test indicates that the free - parameter model is better at a 98% level , with the offset increasing to 3 . given the uncertain extrapolation , however , we use the fits with the offset as a free parameter in our further analysis . the best fitting model for grb970508 is a single power law with relatively low intrinsic x - ray absorption at the level of @xmath16 10@xmath17 @xmath18 . kann et al . ( 2006 ) found a best fit with mw - like dust , but as we measure no significant optical extinction we can not distinguish between different extinction laws . grb971214 is the highest redshift source in the sample at @xmath0 = 3.418 , and we note that the source was faint , particularly as seen by lecs , crucial for the low x - ray energies . we measure an optical - to - x - ray spectral index @xmath19 of 0.6 and find that a single power law is an acceptable fit to these data . the intrinsic x - ray absorption appears to be extremely large whilst the optical extinction is moderately large , but we note that the high redshift of the grb makes measurement of the x - ray extinction more difficult . the curvature in the optical part of the spectrum has been previously interpreted as a cooling break ( wijers & galama 1999 ) and as extinction by either smc - like extinction ( stratta et al . 2004 ) or by a presently unknown , more complex extinction law ( halpern et al . 1998 ; ramaprakash et al . whilst smc extinction is the best fitting law of the 3 used here , it is not sufficient to accurately reproduce the shape of the optical sed , reflected in the large errors on @xmath4 and @xmath14 . in the absence of an accurate redshift determination for grb980329 we adopt the photometric redshift of @xmath0 = 3.6 ( jaunsen et al . 2003 ) , hence all results must be taken with caution . the most striking feature of this sed is the apparent flux deficit in the @xmath20 band , which is present even after correction for galactic absorption and the probable high redshift . we tested for the possibility that the r - band flux deficit is due to the 2175 feature in the milky way extinction curve by fitting the pl+mw model whilst leaving the redshift as a free parameter lying between @xmath0 = 1.2 and @xmath0 = 4.2 ( jaunsen et al . 2003 ) ( we note that at @xmath0 = 3.6 the 2175 bump would lie between the @xmath21 and @xmath22 bands if extinction is mw - like , which is not observed ) . we obtain a fit with @xmath13/dof = 32.7/26 and @xmath0 = 1.95@xmath23 . the other best - fitting parameters , as listed in table 5 for standard fits , do not differ greatly from those found in the @xmath0 = 3.6 fit and the optical curvature is no better fitted . we also tested for the possibility that the r - band deficit is caused by a break in the power law , by allowing the break energy to reside within the optical regime ( and adopting @xmath0 = 3.6 with smc extinction ) . in this fit the break energy could not be well constrained , and the power law slopes are not well described by the difference of 0.5 as expected for a cooling break ( the second slope is steeper ) , hence we rule out this possibility . however , it is also possible that we are seeing an @xmath21 band excess rather than an @xmath20 band deficit . two late time @xmath21 band points taken at 110 days since burst ( yost et al . 2002 ) appear not to agree with all of the @xmath21 band data used here ( reichart et al . 1999 ) , which could be the result of overestimation of the early @xmath21 magnitudes , underestimation of the later @xmath21 magnitudes or the occurrence of color evolution . among the models applied to the whole grb sample here , a single power law unabsorbed in the x - rays is the best fit , with a moderate @xmath14 of 0.178 and a slight preference for smc extinction . we caution that all the results are based on a redshift estimate equal to the mean of the sample spectroscopic redshifts of @xmath0 = 1.58 . a few attempts to constrain the redshift have not been very precise , amounting to 0.5@xmath243.6 or 1.5@xmath243.6 ( lower limits from jaunsen et al . 2003 and upper limit from the fact that we detect a @xmath7 band counterpart ) . we find that a single power law with a small @xmath14 and x - ray absorption consistent with zero but with a large error is sufficient to model this afterglow . we note that the galactic x - ray extinction towards grb980519 is the highest for this sample which , together with the lack of known redshift hampers a good measurement of @xmath4 for this source . a single power law provides an acceptable fit to the spectrum , when absorbed by a large amount at both x - ray and optical / uv wavelengths . in this afterglow we measure the largest @xmath14 value in the sample of 0.29@xmath250.05 or @xmath26@xmath10.85 at 1.33 days after burst assuming smc extinction ( which is marginally preferred ) . this is consistent with the value found by bloom et al . ( 1998b ) of @xmath27 , 5.3 days after the burst . a great deal of work has been done on the host galaxy properties of this burst because the host is bright , with optical extinction measurements by 5 groups . there is a hint that the optical extinction may be decreasing with time ( see e.g. holland et al . 2001 ) since measurements of @xmath26 at different times are inconsistent : @xmath26@xmath12.2 at 0.9 days , castro - tirado et al . 1999 ; @xmath26@xmath11.5 at 1.2 days , vreeswijk et al . 1999 ; @xmath26@xmath10.3 at 4.4 days , djorgovski et al . 1998 ; @xmath10.9 at at 5.3 days , bloom et al . 1998b , and there appears to be a discrepancy between the measured optical spectral and temporal slopes when assuming @xmath26 is constant . however , we note that the optical spectral slope was taken to be @xmath28 = -2.71@xmath250.12 from vreeswijk et al . ( 1999 ) , and in this study we obtain a lower value of @xmath28 = -0.92@xmath250.03 which would be completely consistent with @xmath29 = ( 1 + 2@xmath6)/3 = 0.9 using @xmath6 = 0.85 from zeh et al . ( 2006 , noting their @xmath6 value has a very large associated error - table [ tbl-1 ] ) . we use the vreeswijk et al . ( 1999 ) optical data here and scale it from 1.2 days to 1.33 days after trigger . combining the optical and x - ray data when fitting provides us with a different estimate for the extinction than was obtained by vreeswijk et al . for the optical data alone . we note that we have made only a minimal extrapolation of the original optical data used in this analysis , from @xmath11.2 days to 1.3 days after trigger . grb990123 does not have significant x - ray absorption above the galactic value , and the optical extinction is consistent with zero . we set an upper limit to the latter which is ten times lower than the value found by savaglio et al . ( 2003 ) from fits to zn ii and h i in the optical spectrum . a single power law fit to both optical and x - rays results in a spectral slope of @xmath30=0.61@xmath31 at 1.24 days since burst , comparable to the @xmath29=0.67@xmath250.02 at about the same time since the burst found by galama et al . the latter authors note that the cooling break must lie at or above x - ray frequencies at that time . however , we find an improved fit with a broken power law , constraining the cooling break to 0.15 @xmath32 2.4 kev , within the x - ray spectrum , also found by stratta et al . ( 2004 ) , corsi et al . ( 2005 ) and maiorano et al . we note , however , that the x - ray spectrum comprises flux accumulated between 0.2 and 2.6 days since burst , including the proposed jet break time of 2 days ( table [ tbl-2 ] ) . we tested for the possibility that the offset between optical and x - ray data was incorrect , but this made very little difference to our overall goodness of fit . it is possible that the cooling break has entered the x - ray band during these observations , since this break is expected to decrease in frequency with time . if this is the cooling break , the spectral slopes above and below the break differ by the expected factor of 0.5 when left free . we therefore took only the data from the first x - ray observation , before the jet break ( zeh et al . 2006 ) , together with scaled optical / nir data , which has a log midpoint of @xmath10.7 days since trigger . we repeated all our fits . we find acceptable fits again only with the broken power law model , with the break energy lying somewhere between the optical and @xmath12 kev ( preferring central values just below x - ray frequencies , in line with our x - ray only analysis in which a single power law is a good fit to all spectra ) . we note that an excess of flux at high energies ( as seen by the _ bepposax _ pds instrument not used here ) is reported by corsi et al . they attribute this to an inverse compton component ( though we note that this remains inconsistent with the radio data , kulkarni et al . since the x - ray spectrum is adequately fit with a single power law we assume that the tail of any such component is not significant below 10 kev . fits with galactic , lmc- and smc - like extinctions show that @xmath14 is very low in this source and we can only determine upper limits . the low amount of extinction makes it impossible to pin down the extinction curve shape , hence a fit using galactic - like extinction is sufficient here and lmc- and smc - like extinctions give similar results . there is considerable improvement in the @xmath13 when allowing for a break in the power law , noted by previous authors , which we find is located at 0.0160.020 kev at @xmath11.06 days since burst ( of the order of the value of @xmath10.029 kev at @xmath11 day since burst given by pian et al . the slope change is as expected for a cooling break in the slow cooling regime when leaving both power law slopes free . our measured spectral slope in the optical regime agrees well with that measured in the very large telescope ( vlt ) spectra of @xmath30 = 0.6@xmath250.1 by vreeswijk et al . ( 2001 ) . we note that kuulkers et al . ( 2000 ) analysed these x - ray data in several time bins and found no spectral evolution , hence the cooling break remains outside the x - ray frequencies during the observations . we constrain the optical extinction to be @xmath14 @xmath33 0.01 , and the x - ray equivalent hydrogen column to be @xmath33 0.15@xmath3410@xmath35 @xmath18 for smc extinction and @xmath33 0.87@xmath3410@xmath35 @xmath18 for mw extinction . from the x - ray spectrum alone a higher column of @xmath4 = 2.1 @xmath25 0.6@xmath3410@xmath17 @xmath18 was measured by kuulkers et al . optical spectra have provided a lower limit on the amount of neutral hydrogen towards grb990510 of log @xmath36(h i ) @xmath11 19.7 @xmath18 ( vreeswijk et al . 2001 ) . these authors obtain an approximate estimate for the metallicity from the optical spectra using fe / h , and find 12 + log [ fe / h ] = -1.5 @xmath25 0.5 or 0.010.1 times the solar value . this range approximately covers the metallicity of both the lmc ( 0.33 , pei 1992 ) and the smc ( 0.125 , pei 1992 ) so we also fitted the data with these two metallicities . as there is no substantial absorption observed , the fits do not change significantly . the optical and ir sed of this burst is very well sampled , but the x - ray afterglow was very faint at the time of observation . for this reason these x - ray data have been grouped to have a maximum of only 10 counts per bin , and strictly speaking this means gaussian statistics should be treated with caution . however , we do use the @xmath13 statistic as a goodness of fit for comparison with the rest of the sample . given the poor quality of the x - ray data , and the lack of sufficient counts in low x - ray energy bins , we have fixed the x - ray column density at zero and fit only for the optical extinction . we find a large amount of intrinsic extinction is necessary to describe the flux deficit with respect to a single power law , consistent with the value found by most other studies ( stratta et al . 2004 ; fynbo et al . 2001 ; harrison et al . 2001 ; price et al . a large extinction was also derived from an optical spectrum by savaglio et al . ( 2003 ) , and the @xmath26 found in that study is approximately twice the value found here . however , savaglio et al . use the spectral line measurements to first fit for the depletion pattern and then infer an extinction . harrison et al . ( 2001 ) interpreted the optical flux deficit as indicating that a significant fraction of the x - ray flux was in fact inverse compton emission - later also suggested for grb990123 ( see above ) . this is the only afterglow for which an lmc extinction law is ( marginally ) preferred . fynbo et al . ( 2001 ) report a tentative h i column density measurement of @xmath36(h i ) @xmath12@xmath3410@xmath17 @xmath18 which leads to a relatively high metallicity with [ zn / h ] = -0.13 . this metallicity is between the lmc and the mw values , which may explain the preference here for lmc - like extinction if no 2175 bump is present . a good dataset for grb010222 allows the spectral properties to be well constrained . we find that whilst a single power law is a reasonable fit to these data , a broken power law significantly improves the fit . the break energy lies around 0.01 kev , above the frequency of the last optical band in our sed . optical extinction is clearly non - zero , with @xmath14 = 0.06@xmath250.02 , consistent with that found by lee et al . ( 2001 ) for an smc extinction law but about three times lower than that inferred from the spectral lines by savaglio et al . x - ray absorption is also required with an effective hydrogen column of @xmath4 = 1.15@xmath37 @xmath34 10@xmath35 @xmath18 . panaitescu & kumar ( 2002 ) in their fits to the multiwavelength afterglow of this source find significant reddening of the optical spectrum of @xmath38 with an smc extinction curve , which is consistent with our value . but they find a large fitting error and attribute this to 8 outlying points suggesting that either some reported observations have underestimated uncertainties or there are short timescale fluctuations in the afterglow emission ( cowsik et al . their jet model requires the cooling break to pass through the x - ray band at about 1 day , which they find incompatible with their observations . our analysis places the cooling break at optical / uv wavelengths at 1.51 days since burst . we can compare our results directly with those of previous studies of samples overlapping with this _ bepposax _ subsample . in general , we are finding similar central values for extinction as all previous studies , and are improving upon the uncertainties , fitting all afterglows in the same consistent manner allowing for direct comparison . galama & wijers ( 2001 ) performed the first systematic study of line - of - sight @xmath4 and @xmath14 with a sample of 8 afterglows , consisting of all but the two most recent bursts in our sample . from fits to the x - ray spectra they found intrinsic @xmath4 amounting to 10@xmath35 - 10@xmath39 @xmath18 , ruling out the possibility that some hosts have no x - ray column at all . they noted that these values lie in the range of galactic giant molecular clouds ( estimating cloud sizes of 1030 pc ) - a conclusion also recently arrived at when including swift bursts ( campana et al . 2006a ) and when measuring h i from damped lyman alpha absorption in grb optical afterglow spectra ( jakobsson et al . they used a simple extinction law @xmath40 with a smoothly broken power law . comparing their dust to gas ratios with that of the milky way they obtain an optical extinction 10100 times smaller than expected . their finding of generally low @xmath26 was attributed to previously predicted dust destruction by the grb . of the bursts common to both their and our samples and having a known redshift we obtain consistent extinction values within the 2@xmath41 uncertainties in all cases except for grb971214 for which the two optical extinctions are only consistent at the 3@xmath41 level , our central value lying four times lower than the galama & wijers measurement . we obtain smaller uncertainties in our extinction measurements in all cases except for the x - ray column of grb980703 . stratta et al . ( 2004 ) have also measured @xmath4 and a@xmath42 in the optical and x - ray data seperately , and later plotted the combined data in flux space ( after assuming the x - ray model to be correct ) in order to judge the position of the cooling break . to derive optical spectral slopes , stratta et al . first adopt the @xmath43-value derived from the x - ray fits ( the input electron energy index , see paper ii for our fits for this parameter ) , use this to fix the optical spectral index and fit for @xmath26 . our results for @xmath14 with smc - like extinction are consistent in all cases to within the 90 % confidence limits , adopting the cooling break positions found from this work if known ( the values for grb970508 are only consistent if @xmath44 ) . even for the two grbs for which we have assumed different redshifts we find consistency in extinction estimates ( stratta et al . adopt @xmath0 = 1 for both 980329 and 980519 ) . in 7 of the 10 cases we derive a better constrained value or upper limit to the extinction . the x - ray absorbing columns we measure are also generally consistent with those found by stratta et al . in a fit to the x - ray data only ( these are of course the same data used in our study , except for differences in number of observations combined for some sources , and we include grb000926 as well ) . our method obtains improved estimates for @xmath4 for 7 of the 9 sources common to both studies . for grb990510 stratta et al . find a value higher than our derived @xmath4 using solar metallicity , and for grb980329 our result for @xmath4 is less accurate at the 90 % confidence level , though we note the different redshifts assumed hence direct comparison is not possible . kann et al . ( 2006 ) fitted only the optical seds for a sample of pre - swift bursts including 8 analysed here ( not including grbs 970228 and 980329 and using @xmath0 = 1.5 for grb980519 ) . we find similar central values of @xmath45 and improve upon their mean error by a reduction of 510 % . values disagree at the 90 % confidence level only for grb970508 , where we find an upper limit to the extinction which is 2.4 times lower than the lower limit of kann et al . but would be consistent with their estimate at the 3@xmath41 level . for half the afterglows the best - fitting model to the sed includes smc - like extinction ( as opposed to lmc or mw ) and in one case lmc - like extinction . in no cases is there a preference for mw - like extinction . we are sensitive to the 2175 bump ( mw ) in the redshift range @xmath0 = 0.469.9 , covering all our selected grbs , but clearly we do not detect any such feature . we find a wide spread in central values for the gas - to - dust ratios , and for 4 grbs the gas - to - dust ratios are formally inconsistent with mw , lmc and smc values at the 90 % confidence limit assuming the smc metallicity ( figure [ fig5 ] ) . in these 4 cases the ratio is several orders of magnitude higher than the smc value of 4.4@xmath250.7@xmath3410@xmath35 @xmath18 mag@xmath12 ( koornneef 1982 ; bouchet et al . 1985 ) and must mean that either gas - to - dust ratios in galaxies can span a far larger range than thought from the study of local galaxies , or the ratios are disproportionate in grb hosts because the dust is destroyed by some mechanisms ( likely the grb jet ) , or that the lines of sight we probe through grbs tend to be very gas - rich or dust - poor compared with random lines of sight through galaxies . finally , a dust grain size distribution which is markedly different than considered here may also affect these ratios . in fact , a recent study has shown that for the lmc the former is true . a recent observation of four core - collapse supernova remnants ( snrs ) in the lmc with spitzer has shown ir emission associated with the supernova blastwave ( williams et al . this is interpreted as dust with an lmc - like grain size distribution which has been collisionally heated by the x - ray emitting plasma . the observations require that some fraction of the small dust grains has been destroyed by sputtering by high energy ions in fast shocks . dust destruction is known to occur in snr shock fronts ( jones 2004 ) , and we will return to this issue later in the section . the derived gas - to - dust ratios are several times higher than the lmc ratio , as we see in the line - of - sight measurements of grbs and has been observed in other types of supernova ( e.g. borkowski et al . 2006 ) , the cause of which is not known . we measure a large amount of intrinsic absorption in some of the sample ( figure [ fig4 ] ) , and can state that absorption is insignificant in others , as seen for example in the contrast between grbs 010222 and 990510 . we have tested for the possibility that the galactic column density in the grb direction is affecting our intrinsic column derivations by plotting @xmath46 vs. @xmath47 , and found no correlation . we assumed therefore that all the measured absorption lies within the host galaxy . we note that a few of these afterglows have spectra : for grb990510 no intervening systems were clearly identified , an intervening system is measured at 168 km s@xmath12 from the host redshift for 000926 ( assumed to lie within the host galaxy , castro et al . 2003 ) , 2 intervening systems are found for grb 010222 ( jha et al . 2001 ; mirabal et al . 2002 ) and at the host redshift there are two components separated by 119 km s@xmath12 ( mirabal et al . none of these intervening systems are close enough to us as the observer to significantly affect measurement of the intrinsic host extinction . in this study we provided a thorough , uniform study of both optical and x - ray extinction along the lines of sight towards a sample of 10 grb afterglows . the well known spectral shape and relative brightness of the afterglow emission make grb afterglows a powerful line - of - sight probe of high redshift extinction . this is one of several approaches to measuring absorption in grb host galaxies . one can globally divide the studies of extinction in the field of grbs in two categories : line - of - sight extinction studies and studies of extinction of the integrated host galaxy or parts thereof . line - of - sight studies generally involve fitting the afterglow spectral energy distributions in optical and/or x - rays with template extinction models ( i.e. mw , lmc , smc or more parametrized models ) as we have done here . the standard conversions between x - ray extinction and optical extinction for the milky way and the two magellanic clouds are generally in disagreement with the column densities measured through this method , but these skewed gas - to - dust ratios are also being found in other astrophysical situations as discussed above for snrs in the lmc , and the destruction of dust can go some way to alleviating the mismatch . a further way to probe line - of - sight extinction properties is through optical spectroscopy of the afterglow . in this case the careful measurement of column densities of heavy elements can be used to study the dust depletion pattern along the line of sight ( e.g. savaglio & fall 2004 ) . the measured metal column densities in combination with the best - fitting depletion pattern and the empirically determined conversion between @xmath45 and the dust column , can provide a prediction of the dust extinction along the line of sight to a grb ( for a detailed explanation see e.g. savaglio & fall 2004 ) . savaglio & fall ( 2004 ) show that the extinction derived from the dust depletion method is significantly higher than the value derived from direct fitting to the continuum of the afterglow spectrum , e.g. for grb020813 they find an overestimation by at least five times . we show that this conclusion holds when fitting the afterglow continuum emission over a much larger wavelength range and can quantify that overestimation factor for grbs 990123 , 000926 and 010222 as approximately @xmath48 , @xmath49 and @xmath50 times overestimated respectively . the reasons for this apparent discrepancy may be two - fold : firstly the fitted extinction profiles to the afterglow sed are likely poor approximations to the true extinction profile , and secondly the host dust depletion chemistry may well differ from the milky way chemistry . in addition , the grb or afterglow may preferentially destroy small dust grains , skewing the extinction profile towards larger grains , resulting in a `` grey '' extinction curve . this would alter the derived extinction from sed fitting , and possibly bring estimates from dust depletion methods and sed fitting closer together . whist a grey extinction curve was the best fitting extinction curve to grb020405 ( stratta et al . 2005 ) , we note that grey extinction curves have been fitted to samples of afterglow seds with no conclusive improvement in fit ( e.g. stratta et al . 2004 ) . one can also study the extinction properties of host galaxy as a whole , and there are again several methods to do this . whilst it has been shown that grbs occur in starforming regions in the host ( bloom et al . 2002 ; fruchter et al . 2006 ) , many host galaxies are small and mixing timescales may be short , enabling global properties to be measured . the host and afterglow have similar , moderate reddening in grb000418 ( gorosabel et al . 2003a ) which is taken as evidence that the ism is well mixed . but more extreme values of reddening are also seen , such as the extremely red afterglow and host of grb030115 , in which the host is an extremely red object ( ero , levan et al . 2006 ) . one of the most common methods of integrated host galaxy studies is fitting of the broadband optical and near - infrared seds of the hosts themselves ( e.g. christensen et al . 2004 ) . galaxy templates can be fit to the data , using photometric redshift programmes such as hyperz ( bolzonella et al . 2000 ) providing values for the photometric redshift , the age of the dominant stellar population and the extinction , by fitting a series of galaxy templates . the extinction measured this way is the extinction by the ism of the galaxy on the stellar light , @xmath51 , in which the geometry of the dust in the galaxy can play an important role ( e.g. dustlanes as opposed to an homogeneous dust distribution ) . a study of a large sample of hosts has been performed through hyperz template fitting by christensen et al . ( 2004 ) , who find that grb hosts generally exhibit little extinction , and have young stellar populations . the dependence on metallicity and assumed initial mass function is small ( gorosabel et al . 2003a , b ; christensen et al . one of the difficulties faced here is that the galaxies are often very faint . in low redshift ( @xmath52 ) cases , an optical spectrum of the host galaxy can be taken , typically showing several nebular and balmer emission lines . the balmer lines can be used to derive values for the reddening by calculating their deviation from case b recombination values expected in typical starforming region conditions ( osterbrock 1989 ) . the derived reddening @xmath53 is the reddening of the ionized gas in the source , i.e. the dominant starforming region(s ) producing the balmer emission lines . in general the reddening found from the balmer decrement is low ( see e.g. prochaska et al . 2004 ) , to very low ( wiersema et al . 2006 ) . the two host galaxy extinction estimates @xmath53 and @xmath51 may be correlated for most galaxies ( see calzetti 2001 for a review ) . the increasing data volume on nearby grb host galaxies will allow a test of these correlations , providing further insight into effective grb host galaxy extinction curves . on rare occasions it is possible to obtain high - resolution spectra of an afterglow that also shows host galaxy emission lines , allowing one to obtain a simultaneous view of the extinction along the line of sight and of the balmer decrement . in the case of grb060218 , both absorption lines and emission lines are detected at high resolution using the uves spectrograph on the vlt ( wiersema et al . the spectrum shows asymmetric emission lines that are well fitted with two gaussians separated by 22 km s@xmath12 . the same two velocity components can be seen in absorption in ca ii and na i , and have different chemical properties . these two systems can be interpreted as two seperate starforming regions , through which the light of the afterglow shines . a broadband measure of the extinction either from the afterglow or from template fitting of the host would not have been able to seperate out the contributions of the two individual systems . yet another way to detect the presence of dust , is the detection of grb host galaxies in the far - infrared or sub - mm ( see e.g. barnard et al . 2003 ; tanvir et al . 2004 ) , where the uv radiation from massive stars is reprocessed by dust and reradiated in the far infrared . detection of hosts in the far infrared as well as optical can severely constrain their sed ( le floch et al . 2006 ) , providing estimates for the unobscured star formation rate of grb hosts . in a few cases values up to hundreds of solar masses per year have been reported , while optical indicators give much lower values , indicating a lot of dust - obscured star formation . a different probe of unobscured star formation is the radio continuum flux , which is thought to be formed by synchrotron emission from accelerated electrons in supernova remnants and by free - free emission from h ii regions ( condon 1992 ) . berger et al . ( 2003 ) performed a survey of host galaxies at radio and sub - mm wavelengths , finding that a significant fraction of grb host galaxies have a much higher radio - derived star formation rate than optical methods indicate , pointing again to significant dust extincted star formation but based on only a few detections . all methods considered , sed fitting of the afterglow as described and carried out in this paper is the most broadband view of line - of - sight properties . given the very well known underlying continuum shape , the extinction curve can in principle be extremely well modelled using this method . as discussed in section 1 , the effects of dust extinction in the optical are better measured at higher redshifts whereas the x - ray absorption is best measured at lower redshifts , so this method is most reliable for some middle range of redshifts centred around @xmath54 which approximately corresponds to the current _ swift _ median redshift ( e.g. jakobsson et al . 2006b ) . as with any line - of - sight method , the measured columns may not be representative of the host galaxy as a whole , so comparison with the integrated host galaxy methods is important . the general drawbacks of this method lie in the inability to disentangle the locations of absorbers , both intervening ( between us and the host ) and within the host itself . intervening absorbers may be identified using afterglow spectroscopy ( ellison et al . distribution of absorbers within the host may be tackled from the point of view of searches for variability in the absorption which may indicate destruction of dust or ionisation by the grb and/or afterglow , and comparison of the measured columns with known absorbers such as molecular clouds . campana et al . ( 2006a ) find that the x - ray absorptions of 17 bursts ( including most of this sample ) are consistent with the grbs lying along random sightlines towards galactic - like molecular clouds . our sample is too small to make such a statistical comparison , but we note that the campana et al . sample includes the results of the stratta et al . ( 2004 ) study of the grbs presented here for which we obtain largely consistent extinction values . dust destruction is of course impossible to measure in a single sed , but may be seen with multiple epochs of data . destruction of grains by for example the afterglow uv / x - ray emission would in principle be observable , because depletion indicative line ratios would vary on observable timescales as metals are released from grains into the ism . variable line emission has been searched for in afterglows with multiple optical spectra and has been seen in only two cases , grb020813 , in which the fe ii @xmath552396 transition equivalent widths decreases by at least a factor of five over 16 hours ( dessauges - zavadsky et al . 2006 ) and grb060418 in which several transitions of fe ii and ni ii are seen to vary in observations covering 11 to 71 minutes post - burst ( vreeswijk et al . this line emission is thought to arise at 50 - 100 pc from the grb site , possibly within range of the grb ionising flux . a variable @xmath45 has only been suggested for grb980703 ( described in section 4.1.6 ) . there is evidence for time variable x - ray absorption in a small number of grbs ranging from tentative to moderately strong , in particular for grbs 011121 ( piro et al . 2005 ) , 050730 ( starling et al . 2005 ) and 050904 ( bor et al . 2006 ; campana et al . 2006b ; gendre et al . 2006 ) and 060729 ( grupe et al . this implies ionisation of the line - of - sight gas , probably by the high energy grb jet ( though we note that column density changes and early - time spectral evolution can have similar effects on the spectrum ) . dust destruction can , however , only affect the immediate environment around a burst , ( perhaps out to 10 pc , waxman & draine 2000 ) , and not all the dust in that region need be destroyed since this depends on the efficiency of the mechanism and on the dust grain size distribution . kann et al . ( 2006 ) noted a possible tentative correlation between @xmath26 and star formation rate as measured via host galaxy emission in the sub - mm ( for a sample of 7 grbs ) . if this turns out to hold when tested against larger samples it would demonstrate that a substantial amount of the dust we see comes from star forming regions located throughout the whole host galaxy rather than very close to the burst where it could be destroyed . however , this correlation is unlikely to be real given the very low resolution of the sub - mm observations which make it difficult to differentiate between host galaxy emission and field galaxy emission . this differentiation can be made with spitzer and lower star formation rates in comparison with the sub - mm derived values have been found for some hosts including those of grbs 980703 and 010222 ( le floch et al . 2006 ) . /@xmath14 , derived from best fits to the seds assuming smc metallicity , plotted for each sample grb in date order from left to right ( grb970228 first ) , excluding grb000926 where we did not fit for x - ray column density . we include the ratios for 3 grbs taken from the literature for comparison ( square data points ) : grbs 000301c ( jensen et al . 2001 ) , 000926 ( fynbo et al . 2001 ) and the lower limit on 020124 ( hjorth et al . 2003 ) . the solid , dotted and dot - dashed lines show the measured gas - to - dust ratios and their errors for the milky way ( diplas & savage 1994 ) , lmc and smc ( koornneef 1982 ; bouchet et al . error bars are at the 90 % confidence level.[fig5],width=264 ] here we have demonstrated the advantages of simultaneous fitting of broadband data using a subsample of _ bepposax _ grb afterglows . in no cases is a mw - like extinction preferred , when testing mw , lmc and smc extinction laws . the 2175 bump would in principle be detectable in all these afterglows , but is not present in these data . an smc - like gas - to - dust ratio ( or lower value ) can be ruled out for 4 of the hosts analysed here ( assuming smc metallicity and extinction law ) whilst the remainder of the sample have too large an error to discriminate . we discuss the various methods employed to derive host galaxy extinctions , and compare our results with previous works . we find that this method provides similar central values of @xmath14 and @xmath4 to previous works in which extinction or absorption is determined through afterglow continuum fitting , and in the majority of cases we obtain tighter constraints . we confirm that , with respect to continuum fitting methods such as this , optical extinction is overestimated with the depletion pattern method , and quantify this for a small number of cases . swift , robotic telescopes and rapid response mode on large telescopes such as the william herschel telescope on la palma and the very large telescopes in chile now allow earlier and higher quality data to be obtained , which will help immensely in discriminating between the different extinction laws at work in the host galaxies . we are grateful to mike nowak for his assistance with isis . we thank wim hermsen , jean in t zand , erik kuulkers , tim oosterbroek and nanda rea for advice on the _ bepposax _ data reduction , and the referee for useful comments . this research has made use of saxdas linearized and cleaned event files produced at the _ bepposax _ science data center . the authors acknowledge benefits from collaboration within the research training network ` gamma - ray bursts : an enigma and a tool ' , funded by the eu under contract hprn - ct-2002 - 00294 . rlcs and er acknowledge funding from pparc . references arnaud k. et al . , 1996 , aspc 101 , 17 barnard v. e. et al . , 2003 , mnras , 338,1 berger e. et al . 2003 , apj , 588 , 99 bloom j. s. et al . 1998b , apj , 508 , l21 bloom j. s. , djorgovski s. g. , kulkarni s. r. , 2001 , apj , 554 , 678 bloom j. s. , djorgovski s. g. , kulkarni s. r. , frail d. a. , 1998a , apj , 507 , l25 bloom j. s. , kulkarni s. r. , djorgovski s. g. , 2002 , aj , 123 , 1111 blustin a. j. et al . , 2006 , apj , 637 , 901 bor m. , atteia j. l. , damerdji y. , gendre b. , klotz a. , stratta g. , 2006 , apj , 638 , l71 bolzonella m. , miralles j , pello r. , 2000 , a&a , 363 , 476 borkowski k. j. et al . , 2006 , apj , 642 , l141 calzetti d. , 2001 , new astron . , 45 , 601 calzetti d. , kinney a.l . , storchi - bergmann t. , 1994 , apj , 429 , 582 campana s. et al . , 2006 , a&a , 449 , 61 campana s. et al . , 2006 , apj letters , in press cardelli j. a. , clayton g. c. , mathis j. s. , 1989 , apj , 345 , 245 castro s. , galama t. j. , harrison f. a. , holtzman j. a. , bloom j. s. , djorgovski s. g. , kulkarni s. r. , 2003 , apj , 586 , 128 castro - tirado a. j. et al . , 1999 , apj , 511 , l85 chary r. et al . , 1998 , apj , 498 , l9 chary r. , becklin e. e. , armus l. 2002 , apj , 566 , 229 chen h .- w . , prochaska j. x. , bloom j. s. , thompson i. b. , 2005 , apj , 634 , l25 christensen l. , hjorth j. , gorosabel j. , 2004 , a&a , 425 , 913 condon j. j. , 1992 , ara&a , 30 , 575 corsi a. et al . , 2005 , a&a , 438 , 829 cowsik r. , prabhu t. p. , anupama g. c. , bhatt b. c. , sahu d. k. , ambika s. , padmakar , bhargavi s. g. , 2001 , basi , 29 , 157 dessauges - zavadsky m. , chen h .- w . , prochaska j. x. , bloom j. s. , barth a. j. , 2006 , apjl , submitted dickey j. m. & lockman f. j. , 1990 , ara&a , 28 , 215 diercks a. et al . , 1997 , iau circular 6791 diplas , a. , savage , b. d. , 1994 , apj , 427 , 274 draine b. t. , lee h. m. , 1984 , apj , 285 , 89 draine b. t. , salpeter e. e. , 1979 , apj , 231 , 438 djorgovski s. g. , kulkarni s. r. , bloom j. s. , goodrich r. , frail d. a. , piro l. , palazzi e. , 1998 , apj , 508 , l17 ellison s. l. et al . , 2006 , mnras , 372 , l38 fitzpatrick e. l. , 1985 , apj , 299 , 219 fruchter a. s. et al . , 1999 , apj , 519 , l13 fruchter a. s. et al . , 2006 , nature , 441 , 463 fruchter a. s. , krolik j. h. , rhoads j. e. , 2001 , apj , 563 , 597 fukugita m. , shimasaku k. , ichikawa t. , 1995 , pasp 107 , 945 fynbo j. p. u. et al . , 2001 , a&a , 373 , 796 fynbo j. p. u. et al . , 2006 , a&a , 451 , l47 galama t. j. et al . , 1998a , nature , 395 , 670 galama t. j. et al . , 1998b , apj , 497 , l13 galama t. j. et al . , 1999 , nature , 398 , 394 galama t. j. et al . , 2000 , apj , 536 , 185 galama t. j. , wijers r. a. m. j. , 2001 , apj , 549 , l209 gendre b. , bor m. , 2005 , a&a , 430 , 465 gendre b. , galli a. , corsi a. , klotz a. , piro l. , stratta g. , bor m. , damerdji y. , 2006 , a&a , in press gorosabel j. et al . , 2003a , a&a , 409 , 123 gorosabel j. et al . , 2003b , a&a , 400 , 127 grupe d. et al . , 2006 , apj , submitted , astro - ph/0611240 halpern j. et al . , 1998 , nature , 393 , 41 harrison f. a. et al . , 2001 , apj , 559 , 123 hjorth j. et al . , 2003 , apj , 597 , 699 holland s. et al . , 2001 , a&a , 371 , 52 houck j. c. , denicola l. a. , 2000 , aspc , 216 , 591 in t zand j. j. m. et al . , 1998 , apj , 505 , l119 jakobsson p. et al . , 2006 , a&a , 447 , 897 jakobsson p. et al . , 2006 , a&aletters , in press jaunsen a. o. et al . , 2001 , apj , 546 , 127 jensen , b. l. et al . , 2001 , a&a , 370 , 909 jha s. et al . , 2001 , apj , 554 , l155 jones a. p. , 2004 , astrophysics of dust , eds . witt , c.c . clayton and b.t . draine , asp conference series , 309 , 347 kalberla p. m. w. , burton w. b. , hartmann d. , arnal e. m. , bajaja e. , morras r. , poppel w. g. l. , 2005 , a&a , 440 , 775 kaneko y. et al . , 2006 , apj , in press kann d. a. , klose s. , zeh a. , 2006 , apj , 641 , 993 kawai n. et al . , 2006 , nature , 440 , 184 kawai n. , yamada t. , kosugi g. , hattori t. , aoki k. , 2005 , gcn circular 3937 koornneef j. , 1982 , a&a , 107 , 247 kulkarni s. r. et al . , 1999a , nature , 398 , 389 kulkarni s. r. et al . , 1999b , apj , 522 , l97 kuulkers e. et al . , 2000 , apj , 538 , 638 le floch , e. , charmandaris , v. , forrest , w.j . , 2006 , apj , 642 , 636 le floch , e. , duc , p .- a mirabel , i .- f . , 2003 , a&a , 400 , 499 lee b. c. et al . , 2001 , apj , 561 , 183 levan a. et al . , 2006 , apj , 647 , 471 madau p. , 1995 , apj , 441 madau p. et al . , 1996 , mnras , 283 maiorano e. et al . , 2005 , a&a , 438 , 821 martin n. , maurice e. , lequeux j. , 1989 , a&a , 215 , 219 masetti n. et al . , 2001 , a&a , 374 , 382 meurer g. r. , heckman t. m. , calzetti d. , 1999 , apj , 521 , 64 mirabal n. et al . , 2002 , apj , 578 , 818 nicastro l. et al . , 1999 , a&as , 138 , 437 nousek j. a. et al . , 2006 , apj , 642 , 389 osterbrock d. e. , 1989 , astrophysics of gaseous nebulae and active galactic nuclei ( mill valley : university science books ) panaitescu a. , kumar p. , 2002 , apj , 571 , 779 pei y. c. , 1992 , apj , 395 , 130 perna r. , lazzatti d. , 2002 , apj , 580 , 261 perna r. , lazzatti d. , fiore f. , 2003 , apj , 585 , 775 pettini m. , kellogg m. , steidel c. , dickinson m. , adelberger k. l. , giavalisco m. , 1998 , apj , 508 , 539 pian e. et al . , 2001 , a&a , 372 , 456 piro l. et al . , 2005 , apj , 623 , 314 predehl p. , schmitt j. h. m. m. , 1995 , a&a , 293 , 889 price p. a. , harrison f. a. , galama t. j. , 2001 , apj , 549 , l7 prochaska j. x. , 2006 , apj , in press , astro - ph/0606500 prochaska j. x. et al . , 2004 , apj , 611 , 200 ramaprakash a. n. et al . , 1998 , nature , 393 , 43 reichart d. e. et al . , 1999 , apj , 517 , 692 savaglio s. , fall s. m. , 2004 , apj , 614 , 293 savaglio s. , fall s. m. , fiore f. , 2003 , apj , 585 , 638 schlegel d. j. , finkbeiner d. p. , davis m. , 1998 , apj , 500 , 525 stanek k. z. , garnavich p. m. , kaluzny j. , pych w. , thomson i. , 1999 , apj , 522 , l39 starling r. l. c. et al . , 2005 , a&a , 442 , l21 stratta g. , fiore f. , antonelli l. a. , piro l. , de pasquale m. , 2004 , apj , 608 , 846 stratta g. , perna r. , lazzatti d. , fiore f. , antonelli l. a. , conciatore m. l. , 2005 , a&a , 441 , 83 tanvir n. et al . , 1997 , iau circular 6796 tanvir n. r. et al . , 2004 , mnras , 352 , 1073 tinney c. et al . , 1998 , iau circular 6896.1 vreeswijk p. m. , 1999 , apj , 523 , 171 vreeswijk p. m. , 2001 , apj , 546 , 672 vreeswijk p. m. , 2004 , a&a , 419 , 927 vreeswijk p. m. , 2006 , a&a , submitted , astro - ph/0611690 watson d. et al . , 2006 , apj , in press watson d. , reeves j. n. , osborne j. p. , tedds j. a. , obrien p. t. , tomas l. , ehle m. , 2002 , a&a , 395 , l41 waxman e. , draine b. t. , 2000 , apj , 537 , 796 wiersema k. et al . , 2006 , a&a , submitted wijers r. a. m. j. , galama t. j. , 1999 , apj , 523 , 177 williams b. j. et al . , 2006 , apj , submitted ( astro - ph/0610166 ) wilms j. , allen a. , mccray r. , 2000 , apj , 542 , 914 woosley s. e. , bloom j. s. , 2006 , ara&a , 44 , 507 yost s. a. et al . , 2002 , apj , 577 , 155 zeh a. , klose s. , kann d. a. , 2006 , apj , 637 , 889

the global dynamical properties of galaxies can be summarised by scaling relations between their mass or luminosity , their size , and the velocity of their constituent stars and gas . for spirals , the tully fisher relation ( tfr ) describes a tight correlation between galaxy mass and rotation velocity , with a scatter of around 0.08 dex in velocity when stellar mass is used as the independent variable @xcite , and 0.03 dex for baryonic mass @xcite . although a further correlation exists between mass and radius , with a scatter of around 0.17 dex @xcite , it appears that no relation tighter than the tfr can be constructed by including radius information . for ellipticals , however , the analogue of the tfr , the faber jackson relation ( fjr ; @xcite ) , _ does _ exhibit a significant correlation of its residuals with galaxy size , in the sense that smaller than average galaxies have larger than average velocity dispersion . in other words , there exists a relation tighter than the fjr in the form of a fundamental plane ( fp ) : @xmath1 @xcite . the best - fitting relation has @xmath2 , @xmath3 , and a scatter of around 0.09 dex in @xmath4 @xcite . since the internal motions of galaxies are set largely by their dark matter mass , these relationships provide key insight into the connection between galaxies and their host haloes . in this work , we construct a semi - empirical framework in @xmath0cdm to predict the velocity dispersions of early - type galaxies , and hence the fjr and fp . in particular , our modelling approach is based on subhalo abundance matching ( am ) , which posits a nearly one - to - one relation between stellar mass and a halo proxy , typically a mass or velocity at a chosen epoch . models based on this technique have been shown to provide an excellent fit to a wide range of spatial statistics of the galaxy population , including the galaxy - galaxy two and three - point correlation functions , galaxy - galaxy lensing , and group statistics @xcite . however , am as it is usually implemented does not distinguish between galaxies of different morphology , and therefore requires testing against the scaling relations of both late and early - type galaxies . here we investigate whether the properties of the fjr and fp are consistent with the galaxy halo connection implied by am . we intend in this way to evaluate agreement between galaxy - scale and large - scale constraints on this connection , and bring out the additional information contained in the fjr and fp on the processes governing galaxy formation . our approach may be usefully situated at the intersection of two broad classes of methodology for understanding the dynamical relations of elliptical galaxies . a bottom - up approach focuses on detailed kinematic analysis and mass modelling of a small number of individual galaxies ( e.g. @xcite ) . the jeans equation is solved in full , the anisotropy profile may be determined simultaneously with the velocity dispersion profile , and dark matter fractions in the inner regions are estimated by fitting one or more halo profiles to the data . this approach has the advantage of affording tight control over the model baryonic and dark matter masses of individual systems and hence enables a high precision comparison of theory to data . however , the samples analysed in this way tend to be small ( up to a few hundred galaxies ) , and may therefore be unrepresentative in relevant galaxy properties . furthermore , dark matter modelling on a galaxy - by - galaxy basis prevents comparison with cosmological expectations for the halo population obtained from @xmath5-body simulations and a complete galaxy formation model . except in the most extreme cases , therefore , it is difficult to assess statistically the extent to which the results are consistent with @xmath0cdm predictions . the alternative top - down approach uses a fully cosmological framework to derive model values for galaxy observables . this may be attempted in several different ways depending on the theoretical assumptions employed . hydrodynamical simulations implement physically motivated prescriptions for baryonic physics in order to simulate galaxy formation in situ and concurrently with the evolution of dark matter haloes ( e.g. @xcite ) . although in principle applying basic physics in the most straightforward way , such simulations are too computationally expensive to generate theoretical galaxy populations of sufficient size to allow robust statistical comparison with observations . this method may therefore be usefully complemented by semi - analytic and empirical models ( e.g. @xcite ) , which forgo complete treatment of baryonic physics in favour of simple mappings between galaxy and halo properties in order to predict galaxy observables from the outputs of @xmath5-body simulations . although potentially dependent on uncertain parametrizations of key properties , these methods in principle allow statistically rigorous comparison of theoretical and observed galaxy populations . hydrodynamical , semi - analytical , and empirical methods have the advantage that they permit a direct test of concrete , cosmologically - motivated galaxy formation scenarios using a representative halo population . on the other hand , theoretical results are typically compared with large and often heterogeneous observational data sets without regard to selection effects , the assumptions and approximations used in the galaxy modelling , or the peculiar nature of particular galaxy subsets . the systematic error budget is therefore uncertain and potentially large . furthermore , it is rare for such studies to simultaneously investigate the dependence of velocity dispersion on galaxy size in the fundamental plane , the scaling relations scatter , or the effect of radial orbit anisotropy . by combining the precision of the bottom - up approach with the scope and power of the top - down approach , we intend our methodology to permit detailed inferences of cosmological scope . we will use a complete galaxy formation model based on halo abundance matching to create predictions for the fjr and fp , and establish a set of key diagnostics with which to compare the output of the model to a single homogeneous data set with well - defined selection criteria . we prune the observational data set to those galaxies whose structure is compatible with our mass model , use an identical velocity dispersion measure in the theoretical and observed fjr , ensure the distributions and covariances of size , stellar mass and srsic index are the same between theory and data , and explicitly consider the possible impact of radial orbit anisotropy . furthermore , we pay particular attention to the tilt and scatter of the fp . we note that an alternative approach which has been used in the literature to model elliptical galaxies is based on the modified newtonian dynamics ( mond ) as opposed to @xmath0cdm paradigm . these models seek to explain the dynamics of galaxies by positing a breakdown of newtonian gravity at low acceleration , instead of invoking dark matter haloes . work done along these lines includes @xcite , @xcite , @xcite , and @xcite , and suggests that this framework may be able to offer a convincing explanation of the tilt and small scatter of the fundamental plane with realistic assumptions about galaxy structure and radial orbit anisotropy . in mond , an elliptical lies on a fundamental plane near the newtonian virial relation when the acceleration of stars in its inner region is high and hence their dynamics are quasi - newtonian ; the faber jackson relation ( of the specific form @xmath6 ) is the effective virial relation in the low - acceleration deep - mond regime that obtains in low surface brightness galaxies and the outer regions of high surface brightness ones . however , none of these models have been shown to match the spatial statistics of galaxies on large scales . the structure of this paper is as follows . section [ sec : data ] describes the observational data against which we test our model , and documents the @xmath5-body simulation from which halo properties are drawn . section [ sec : method ] details our methodology . in section [ sec : results ] we present our results for the faber jackson relation and fundamental plane . we will develop a series of models of increasing complexity , comparing in each case the predicted and observed values of five key properties of the fjr and fp . the main results are summarised in table [ tab : table ] and figure [ fig : fjr_cont ] . in section [ sec : discussion ] we discuss the general significance and broader ramifications of our results , as well as their relation to results in the literature . section [ sec : conclusion ] concludes . this work may be seen as a companion to @xcite ( hereafter dw15 ) , to which it is similar in approach and methodology . in dw15 we explored the aptitude of our am framework to reproduce the spiral galaxy tully fisher and mass size relations ; here we redeploy and augment these models in order to investigate the elliptical galaxies that we previously excluded . in several places we refer to dw15 for more detail . we compare our model to data from the nasa - sloan atlas ( nsa ) . this is a compilation of galaxy properties within the footprint of sdss dr8 out to redshift @xmath7 , cross - matched with a number of other catalogues . sky - subtraction and photometric determinations in the nsa are believed to be among the most reliable available @xcite . we will cast all scaling relations in terms of stellar mass rather than luminosity , as the fundamental plane . this is sometimes called the `` mass plane '' in the literature to distinguish it from the relations @xmath8 and @xmath9 . ] and take mass - to - light ratios from the kcorrect algorithm @xcite with a universal chabrier imf ( but see section [ sec : contraction ] for a discussion of the effect of imf choice ) . we select from the nsa a subsample of galaxies to compare to our fjr and fp model predictions . first , we clean the data by eliminating galaxies with unreliable size or redshift measurements @xcite , and restrict to @xmath10 . we next deduce the angular diameter distance @xmath11 , effective radius @xmath12 , and srsic index @xmath13 of each galaxy . the angular diameter distance is calculated from the redshift assuming a flat cosmology with @xmath14 and @xmath15 . @xmath12 is derived by first multiplying the ` sersic_th50 ` field of the nsa ( the angular half - light radius along the major axis ) by the angular diameter distance , and then circularising the light profile by multiplying by the square root of the minor - to - major axis ratio . the spherically - symmetric srsic index @xmath13 is set equal to the 2d srsic index recorded in the nsa , from which we find it is not systematically offset . we thus determine the relative frequency distribution \{@xmath16 , @xmath12 , @xmath13 , @xmath11 } , which we will use in section [ sec : model ] to assign @xmath12 , @xmath13 and @xmath11 values to our model galaxies as a function of @xmath16 . , we use units of @xmath17 throughout . ] next , we impose the following selection criteria to ensure that the galaxies in our final sample have mass distributions that adhere as closely as possible to the spherically - symmetric srsic profile that we will give our model galaxies . these are typical cuts for eliminating rotation - dominated galaxies ( e.g. @xcite ) . 1 . srsic index @xmath18 ( low-@xmath13 galaxies are more likely to be flattened and discy , and the sdss analysis code has a hard upper limit at @xmath19 ) . 2 . concentration ratio @xmath20 ( pressure - supported systems tend to be less compact ) . apparent axis ratio @xmath21 ( low projected ellipticity favours more spherical early - type galaxies ) . we will refer to the galaxies passing these cuts as `` early - type '' or `` elliptical '' and the remainder as `` late - type '' or `` spiral . '' we caution , however , that these definitions may not conform to others in the literature , which may include a cut on colour ( e.g. @xcite ) , or be based on visual inspection ( e.g. @xcite ) . since we will find that the precision of our analysis is not limited by statistical uncertainties due to sample size , we choose the threshold values for our cuts to be fairly stringent to maximise agreement with our model assumptions . taken together , these factors result in `` elliptical fractions '' in our analysis significantly lower than in many literature studies ( e.g. @xcite ) . we next impose a stellar mass cut to generate a volume - limited sample with the same @xmath16 distribution as our am mocks . assuming a conservative apparent magnitude limit of @xmath22 , the sdss and hence nsa data are complete to @xmath23 out to @xmath24 @xcite . we find that the stellar mass value above which all galaxies in both the early and late - type subsample have @xmath25 is @xmath26 , which we impose as the lower @xmath16 limit in our analysis . these cuts leave 6,088 elliptical galaxies out of 36,520 total . since we will be interested in the possible impact of the elliptical selection criteria on our theoretical galaxy population ( section [ sec : model ] ) , we record also the fraction of galaxies retained as a function of stellar mass . this rises from @xmath27 at @xmath28 to @xmath29 at @xmath30 , the highest stellar mass we consider . finally , we generate an early - type catalogue of \{@xmath16 , @xmath12 , @xmath31 } values , to which we will compare our model predictions . we include at this stage a final requirement , @xmath32 , which removes galaxies whose velocities are untrustworthy @xcite . we will apply this cut in the same way to the mock data sets . the final early - type subsample contains 5,682 galaxies . our theoretical galaxy population is based on the haloes generated in the darksky-400 simulation @xcite , a @xmath33 box with @xmath34 particles , run with the 2hot code @xcite . the darksky suite assumes a flat @xmath0cdm cosmology with @xmath35 , @xmath36 , @xmath37 , and @xmath38 . the darksky-400 box is large enough to contain sufficient haloes at the high end of our stellar mass range to permit robust statistical testing , and at the same time has a small enough particle mass that haloes at the low mass end are well resolved ( according , for example , to the criteria of @xcite ) . we identify haloes using the rockstar halo finder @xcite . the simulation contains around 1,000,000 haloes hosting galaxies in the stellar mass range @xmath39 . the overall aim of our method is to create a map between a set of theoretical parameters describing galaxy formation processes and five key properties of the corresponding theoretical fjr and fp . we do this by generating for particular input parameter values a population of galaxies with stellar masses ( @xmath16 ) , effective radii ( @xmath12 ) , and velocity dispersions averaged over the 1.5 sdss aperture ( @xmath31 ) . we then draw mock data sets from these populations and determine their fjr and fp . by comparing to the observed relations , we assess the aptitude of the models to account for elliptical galaxies velocity dispersions . the scope of our parametrizations encompasses many of the galaxy formation models under consideration today . we begin by assigning a stellar mass to each halo in the darksky-400 simulation by abundance matching to the @xmath40 snapshot @xcite . specifically , we adopt the am model of @xcite , which uses a continuous parameter @xmath41 to interpolate proxy between virial mass ( @xmath42 ) and maximum circular velocity @xmath43 ( @xmath44 ) at the time of peak mass . a universal gaussian scatter ( which we refer to as the `` am scatter '' ) describes the deviation of the stellar mass proxy relation from exact monotonicity . @xcite show that a good fit to the correlation function and satellite fraction of sdss can be obtained in the case @xmath45 , am scatter @xmath46 , and we take these as our fiducial values . we use the stellar mass function of @xcite based on single - srsic fits , which is consistent with the stellar mass determinations of the nsa . each model galaxy is assumed to have a @xcite mass density profile - band light , so that the luminosity and mass profiles are directly related . strong trends in stellar mass - to - light ratio are not observed in elliptical galaxies , and there are only small differences in nsa measurements between bands . ] @xmath47\right\},\ ] ] where @xmath48 is chosen so that the luminosity within @xmath12 is half the total luminosity @xcite @xmath49 a good approximation to the de - projected volume density is @xcite @xmath50,\ ] ] with @xmath51}\ ] ] and @xmath52 we next assign an effective radius @xmath12 , srsic index @xmath13 and angular diameter distance @xmath53 to each model galaxy by randomly drawing from the relative frequency distribution calculated in section [ sec : obs ] at the stellar mass of the galaxy in question . this purely empirical method guarantees that the model galaxies have precisely the same \{@xmath16 , @xmath12 , @xmath13 , @xmath11 } distribution as the data to which we will compare them .. ] for reference , the best - fitting power - law mass size relation ( msr ) is @xmath54 , where @xmath55 is the median mass of the nsa sample . the scatter in size is 0.136 dex . the angular diameter distance is then used to calculate the physical size of the sdss aperture at the location of each galaxy , within which the weighted velocity dispersion @xmath31 is defined . combined with present day halo virial mass ( @xmath56 ) and concentration , \{@xmath16 , @xmath12 , @xmath13 , @xmath11 } form the inputs to the calculation of @xmath31 , which we now perform . for a given cumulative spherical mass distribution @xmath57 , density distribution @xmath58 , and radial orbit anisotropy profile @xmath59 , the radial velocity dispersion @xmath60 obeys the spherical jeans equation @xmath61}{dr } + \frac{2 \beta(r)}{r } \rho(r ) \sigma_r^2(r ) = -\frac{g \rho(r ) m(r)}{r^2}.\ ] ] the line - of - sight velocity dispersion @xmath62},\ ] ] is then given by @xcite @xmath63 where @xmath64\end{aligned}\ ] ] and @xmath65 is the incomplete beta function . finally , we calculate the physical radius @xmath66 of the sdss aperture at the position of the galaxy by multiplying the angular radius of 1.5 by @xmath11 . @xmath67 is then averaged ( weighted by brightness ) over the area of the aperture to yield the aperture velocity dispersion @xmath68 it now remains to specify 1 ) the dark matter mass distribution @xmath69 , 2 ) the radial orbit anisotropy @xmath70 , and 3 ) which haloes in the simulation should be assumed to harbour the elliptical galaxies in question , and hence included in the analysis . knowing the virial mass and concentration of each halo , we can reconstruct the nfw density profile ; however , it is likely that this profile is altered in some way by the formation of galaxies at the halo centres . one possibility is for the dark matter to contract adiabatically @xcite , but it is also possible for baryons to transfer energy to the halo and cause it to expand instead ( e.g. @xcite ) . we employ here the same parametrization of these effects as in dw15 : a free , continuous parameter @xmath71 interpolates between the adiabatic contraction prescription of @xcite ( @xmath72 ) , and an expansion of the same magnitude ( @xmath73 ) . @xmath74 corresponds to no effect of disc formation , in which case haloes retain their nfw form . the anisotropy parameter @xmath75 characterises the extent to which stars follow predominantly tangential or radial orbits . @xmath76 describes motion that is purely tangential , @xmath77 isotropic and @xmath78 purely radial . in principle , @xmath75 is not only a function of position within each galaxy , but also varies between galaxies . canonical studies of anisotropy include @xcite , @xcite and @xcite , and indicate that @xmath75 typically rises in the central regions of ellipticals , and then flattens out or falls slowly towards larger @xmath79 . variations among galaxies appear to be relatively small and not significantly correlated with other galaxy properties , and in relatively few cases is @xmath75 observed to fall outside the range [ @xmath80 , @xmath81 . we will therefore assume @xmath75 to be independent of @xmath79 in our models ( equivalently , we use a single effective value averaged over the aperture ) . in the most general case we allow @xmath75 to be a function of stellar mass but uncorrelated with all other galaxy and halo variables , and give it a universal gaussian scatter @xmath82 among galaxies : @xmath83 . in section [ sec : obs ] we calculated the fraction of galaxies in the nsa that are ellipticals according to our selection criteria , as a function of stellar mass . this must equal the fraction of haloes hosting elliptical galaxies , and only these haloes ought to be included in our mock population . , the elliptical fraction is a function of our early - type cuts . if less stringent cuts were used , leading to an increase in the elliptical fraction , selection effects would have a weaker impact on the theoretical population . our model therefore provides an upper limit for the possible effect of selection on the fjr and fp . ] given that there is still significant theoretical uncertainty in which haloes should be included , we adopt the flexible selection model of dw15 ( in which the motivation and method is spelled out in detail ) . in short , we assume early - type galaxies to inhabit haloes that are above average in mass or concentration at a given stellar mass , so that their characteristic velocities are higher . the strength of this correlation is governed by a free parameter we call the `` selection factor '' : when this parameter takes its largest allowed value ( @xmath84 ) ellipticals are put in the haloes which give the _ largest _ values for @xmath31 at given @xmath16 , and when it is minimised ( @xmath85 ) the correlation between galaxy morphology and halo type is switched off . we expect from the results of dw15 ( in addition to recent indications in data ; see e.g. @xcite , @xcite and @xcite ) that the selection factor ought to take a value in the upper half of its allowed range . we begin with the value 1.4 , corresponding to a moderate increase in predicted velocity for ellipticals over the case of random selection . this completes the description of the inputs to the model : \{@xmath41 , am scatter , @xmath71 , selection factor } along with a function @xmath86 for setting @xmath87 . given values for each of these , we are now in a position to calculate @xmath31 for the galaxy in each halo of the elliptical subsample . since there are @xmath88 haloes , and 16.7 per cent of the nsa galaxies in this mass range were determined to be ellipticals ( section [ sec : obs ] ) , our theoretical parent population contains around 167,000 objects . we now describe our procedure for comparing this population with the observational sample in order to evaluate goodness - of - fit and investigate the significance of the model parameters . we begin by using the model to generate a mock data set with as many galaxies as the observational sample and at the same stellar masses , but with radii and velocity dispersions drawn randomly from the theoretical population . for consistency with our observational sample , we impose at this stage the requirement @xmath32 . the stellar masses used in am as well as the sizes and srsic indices we give our mock galaxies all derive from the observed distributions , and hence no correction need be applied to convert between true and observed values of these quantities when comparing the mock and real data . measurement uncertainties do however play a role in determining @xmath89 for a given mock galaxy , which is a function of the true rather than observed values of @xmath16 , @xmath12 and @xmath13 , and furthermore is different to the @xmath90 value naturally output by the model . in a rigorously bayesian analysis , the true positions of the galaxies in \{@xmath16 , @xmath12 , @xmath13 , @xmath31 } space would be treated as free parameters with posterior distributions deduced simultaneously with the others . at the maximum likelihood point , the true positions would lie somewhat closer to the model prediction than the observed positions ( within the measurement uncertainties ) , and hence we would find slightly greater agreement between theory and data . however , since this analysis is too expensive to perform here we take the true and observed positions to be coincident . this is justified by the relatively small uncertainties in the measurements of the local nsa galaxies , and is no worse an estimate for the maximum - likelihood values of the true parameters than those derived by randomly scattering the observed values . we fit to each mock data set a power - law fjr cut . we are not concerned here with the determination of the fitting parameters themselves , but only their comparison between real and mock data sets . this requires merely that each data set be treated in the same way . ] of the form @xmath91 with scatter @xmath92 in @xmath93 , and a planar fp @xmath94 with scatter @xmath95 in @xmath96 . we then reduce the set \{@xmath97 , @xmath98 , @xmath92 , @xmath99 , @xmath100 , @xmath101 , @xmath95 } to 5 _ independent _ parameters by removing duplicate information : the normalisation @xmath99 of the fp contains no information absent from the combination of the normalisation of the msr ( matched by construction ) and that of the fjr ( @xmath97 ) , and one linear combination of @xmath100 and @xmath101 replicates the information in the msr and fjr slopes . the other degree of freedom , @xmath102 , which we denote by @xmath103 , describes the correlation of the msr and fjr residuals and is a measure of the fp tilt . @xmath104 will be the diagnostic statistic that we use to test the theory . by fitting a plane to the \{@xmath16 , @xmath12 , @xmath31 } data we neglect the possibility of curvature in the fp and a dependence of @xmath95 or @xmath103 on @xmath16 . although we have verified by eye that the predicted and observed fps are not far from planar , we can not preclude the possibility of a larger measured @xmath95 being due in part to greater curvature . the fp tilt is known to vary with stellar mass @xcite , and we have qualitatively identified such a trend in our data sets . in principle this dependence contains further information . we will find , however , that the parameters of our framework are well constrained by fitting an averaged tilt and hence we leave an investigation of the significance of this effect within the am framework to further work . we evaluate goodness - of - fit as follows . for each set of parameter values we generate 1000 monte carlo \{@xmath16 , @xmath12 , @xmath31 } realisations , and calculate for each one the best - fitting value of @xmath105 . we then condense this information into a set of median values @xmath106 over the realisations ( the values one would be most likely to find in the data if the model were true ) and the standard deviation @xmath107 ( an indication of the expected spread between different observational samples ) . we shall find the skews of the monte carlo @xmath105 distributions to be negligible , making them fully characterisable by these two statistics . a measure of goodness - of - fit is then @xmath108 , where @xmath109 is the vector @xmath105 for the nsa data . the more standard deviations the real data from the model median , the worse the model is performing . we will investigate a total of seven different sets of values for the input model parameters , arranged in order of increasing complexity . to begin with , and for reference , we examine the scenario ( which we label `` no - dm '' ) in which velocity dispersions are determined entirely by the baryonic mass , and stellar motions are isotropic . in this case , none of the other model parameters apply . we then introduce dark matter haloes using the prescription of section [ sec : model ] , first with a fiducial model in which the parameters are given values which might be considered most likely a priori ( although still with @xmath77 ) , then with individual perturbations to the parameters , and finally with parameter values chosen to optimise the fit to the fjr and fp . we then lift the @xmath77 restriction and consider a model with small radial anisotropies ( favoured by existing data ) , and one in which the effective anisotropy increases with stellar mass . for each set of parameter values we show a contour plot of the predicted fjr ( fig . [ fig : fjr_cont ] ) , and list the median @xmath106 of the five diagnostics ( table [ tab : table ] ) . in certain cases we show also a contour plot of the correlation of the velocity and radius residuals ( fig . [ fig : rescorr_cont ] ) , and the standard deviations @xmath107 of the diagnostics among the monte carlo realisations ( table [ tab : table2 ] ) . for reference , the free parameters of our framework are listed and described in table [ tab : params1 ] , and the statistics of the fjr and fp to which we compare our model are given in table [ tab : params2 ] . [ cols= " < , > " , ] we now consider a fiducial am model with @xmath110 , am scatter = 0.16 , @xmath111 and selection factor = 1.4 ( fig . [ fig : fjr_fid ] and table [ tab : table ] , third row ) . these might be reasonable estimates a priori : the am parameters are those preferred by @xcite , the @xmath71 choice corresponds to moderate halo contraction , and the selection factor corresponds to a moderate preference for early - type galaxies to inhabit more massive or more concentrated haloes at fixed mass . the added dark matter , distributed in this way , has four main effects on the fjr and fp : 1 . the normalisation of the fjr is increased due to the extra dynamical mass . the velocity dispersions of low - mass galaxies are boosted more than those of high - mass galaxies , making the fjr shallower . this effect is exacerbated by the strong selection in this model : since the elliptical fraction falls towards lower stellar mass , the effect of selecting only the highest @xmath31 objects to be ellipticals is greatest there . the scatter of the fjr and ( particularly ) the fp are increased by the addition of halo variables ( @xmath56 and concentration ) into the effective formula for @xmath31 . these variables themselves have a spread of values at fixed @xmath16 . the precision with which one can estimate @xmath96 from @xmath12 and @xmath31 is only 1.5 times greater than from @xmath31 alone due to the relative weakening of the correlation between the three observables . @xmath103 is increased , indicating that @xmath31 varies less rapidly with @xmath12 at fixed @xmath16 than in the no - dm case . now that @xmath31 is set largely by the halo ( and under our assumption that @xmath12 is independent of halo properties ) , the dependence of @xmath31 on surface brightness is reduced . we now consider the effect of individual changes to the model parameters . we note first that varying @xmath41 within reasonable bounds ( @xmath112 ; @xcite ) has very little effect on any of our diagnostics . when @xmath41 is larger , the rank of a halo in am is conditioned more by its concentration than its virial mass , making the haloes of our high-@xmath16 galaxies more concentrated and less massive on average . however , the total dark matter mass within the aperture , which is responsible for setting @xmath31 , is only slightly larger at high @xmath41 . we note that the halo proxy has a significantly weaker effect on the fjr here than on the tfr in dw15 . this is because in dw15 velocities were measured at larger radius , so the ratio of enclosed dark matter to baryonic mass was larger and the choice of halo proxy had a larger effect . in addition , while the squared tfr circular velocity is directly proportional to the enclosed dark matter mass , the fjr aperture velocity dispersion is determined by a set of three differential equations that depend non - trivially on density and cumulative mass . the result is a washing out of the subtle effect of the am proxy . we therefore fix @xmath110 for the remainder of this work . the effects of variations in the other three model parameters are shown in rows four to six of table [ tab : table ] and figs . [ fig : fjr_scatt][fig : fjr_sel ] , and may be accounted for as follows . * the most significant effect of the am scatter is on the scatter of the fjr and fp . increasing the am scatter increases the spread of halo properties assigned to galaxies of a given @xmath16 , and hence the spread of the resulting velocity dispersions . a higher scatter also flattens the fjr ; since the halo mass function is a falling function of mass , high am scatter tends to associate lower mass haloes with galaxies at a given high @xmath16 . finally , increasing the am scatter reduces @xmath103 . * @xmath71 largely sets the normalisation of the fjr : the lower it is , the more dark matter is expelled from the inner regions of a galaxy , and hence the lower @xmath31 . the normalisation of the fjr can be reconciled with that of the data by setting @xmath113 . in addition , lowering @xmath71 tends to reduce the rate at which @xmath31 falls as @xmath12 rises , as manifest by an increase in @xmath103 . when @xmath114 , more compact galaxies ( i.e. lower @xmath12 at fixed @xmath16 ) pull more dm into the inner regions , boosting the rate at which @xmath31 rises as @xmath12 falls . conversely , when @xmath115 more compact galaxies _ expel _ more dm from their inner regions , mitigating the rise in @xmath31 due to the baryonic mass as @xmath12 falls . * a larger selection factor preferentially puts elliptical galaxies in more massive or more concentrated haloes at given @xmath16 , increasing the normalisation of the fjr and reducing its scatter . since the elliptical fraction falls towards low @xmath16 , selection effects have largest impact in the low - mass regime , and hence increasing the selection factor lowers the fjr slope . table [ tab : table ] shows that the predicted @xmath98 and @xmath97 are brought into better agreement with the data by adopting @xmath116 and selection factor = @xmath117 . we display the results of such a model in figs . [ fig : fjr_opt ] and [ fig : rescorr_opt ] , and row seven of table [ tab : table ] . we focus hereafter on the case \{@xmath118 , am scatter = 0.16 , @xmath116 , selection factor = @xmath117}. all models examined so far have assumed @xmath120 . we now consider two models where this constraint is relaxed , the first in which a @xmath75 value for each galaxy is drawn from a normal distribution with fixed mean @xmath87 and standard deviation @xmath82 , and the second in which @xmath87 is a function of @xmath16 . detailed kinematic studies of ellipticals suggest that @xmath75 is typically fairly independent of radius and takes a value around 0.3 with an inter - galaxy scatter of roughly @xmath121 @xcite . the results of a model with @xmath122 , separately for each mock galaxy , are shown in fig . [ fig : fjr_3beta ] , fig . [ fig : rescorr_3beta ] and row eight of table [ tab : table ] . this model agrees fairly well with the data in terms of both @xmath92 and @xmath103 , and the large effect of @xmath75 demonstrates its crucial importance in any study aimed at uncovering the theoretical significance of elliptical galaxy scaling relations . we distinguish and explain the effects of @xmath87 and @xmath82 as follows . 1 . all else being equal , the larger a galaxy s @xmath75 , the greater its @xmath31 . this is because larger @xmath75 corresponds to more radial stellar motions , which provide a larger contribution to @xmath60 and hence @xmath67 ( eqs . thus @xmath97 rises with @xmath87 . a larger @xmath75 increases @xmath31 for smaller galaxies more than it does for larger ones . increasing @xmath87 therefore increases the average @xmath31 of galaxies at fixed @xmath16 and low @xmath12 more than those at the same @xmath16 but high @xmath12 , reducing @xmath103 . 3 . at small @xmath12 , a large @xmath75 increases @xmath31 more than a small @xmath75 decreases it , and vice versa at large @xmath12 . hence increasing @xmath82 at fixed @xmath87 increases the average @xmath31 of galaxies at fixed @xmath16 and low @xmath12 , while decreasing it at high @xmath12 . the result is a decrease in @xmath103 . 4 . since @xmath75 is assumed to be uncorrelated with galaxy or halo properties , the effect of @xmath82 roughly sums in quadrature with the other sources of scatter to increase @xmath92 and @xmath95 . we now consider a simple toy model in which @xmath87 depends on @xmath16 . this is motivated by the observation from fig . [ fig : fjr_cont ] and table [ tab : table ] that none of the model refinements made so far have had a significant effect on the _ slope _ of the fjr , which has remained in all cases too low . we therefore show in figs . [ fig : fjr_vbeta ] and [ fig : rescorr_vbeta ] , and row nine of table [ tab : table ] , the effect of the alternative model @xmath123 , in which @xmath87 rises from @xmath124 at @xmath28 to 1.0 at @xmath30 . , and @xmath75 tends to fall slowly with @xmath125 beyond a certain point ( e.g. @xcite ) . ] the fjr slope in this model is increased to that in the data , and furthermore @xmath95 is reduced . however , the model fjr has acquired a curvature not present in the data , and @xmath103 is significantly increased . within our parameter space , the fjr curvature could be reduced only by setting @xmath87 even lower at @xmath28 ( @xmath75 can never exceed 1 ) , which would further increase @xmath103 and necessitate an increase in @xmath71 to mitigate the fall in @xmath97 . we therefore conclude that introducing a dependence of @xmath87 on @xmath16 does not offer a particularly attractive prospect for further improving fjr and fp predictions . note that a positive @xmath87 and @xmath82 would increase the predicted fjr intercept and scatter also in the no - dm case , bringing it into better agreement with the data , and including a positive @xmath87 dependence on @xmath16 would increase the slope . within the broader parameter space opened up by the anisotropy , it is not clear therefore that the no - dm model can be excluded by the fjr alone . now that all the parameters of our model have been introduced and their effects described , we discuss their degeneracies . the parameters @xmath71 , @xmath87 and the selection factor have a degenerate effect on the normalisation of the fjr . there is also a strong degeneracy between the am scatter and @xmath82 , both of which predominantly affect @xmath92 and @xmath95 . however , strong prior constraints exist on the am scatter ( e.g. * ? ? ? * and references therein ) , so we have chosen to fix its value here . we note however , that although most current measurements are consistent with a constant value or weak trend of am scatter with mass , the constraints weaken significantly at low mass . finally , the selection factor , am scatter and effective dependence of @xmath75 on @xmath16 are degenerate in their effect on @xmath98 , and @xmath71 , @xmath82 and the am scatter are degenerate in determining @xmath103 . alternative sets of parameter values capable of adequately reproducing the observed @xmath97 , @xmath92 and @xmath103 are therefore likely to exist . we have investigated a series of am - based models , which have previously been shown to match several spatial statistics of galaxies , in order to assess the extent to which important characteristics of the faber jackson relation and fundamental plane are expected in contemporary galaxy formation scenarios . while some aspects of the relations ( the normalisation and scatter of the fjr , and the rate of change of velocity with size at fixed mass ) may be satisfactorily accounted for , others ( the fjr slope and fp scatter ) are challenging to match with our approach . in section [ sec : contraction ] we discuss our preference for halo expansion in the context of previous findings in the literature , in section [ sec : selection ] we comment on the possibility that early - type galaxies tend to inhabit more massive or more concentrated haloes at given stellar mass , and in section [ sec : discrepancies ] we elaborate on the discrepancy in the fp scatter and consider potential solutions . in order to obtain the correct magnitude of the aperture velocity dispersions , we require moderate expansion of haloes in response to disc formation ( @xmath113 ) . this roughly agrees with the tfr calculations in dw15 , which favour @xmath126 for mass - based am and @xmath113 for the more realistic case of velocity - based matching . these results may be considered an extension of the cusp core @xcite and too big to fail @xcite problems to higher mass : the nfw profile with a canonical mass - concentration relation puts too much dark matter in the central regions of @xmath127 galaxies . the necessity of halo expansion has already been noted by several observational and theoretical studies of early - type galaxies . @xcite find that reproducing the low normalisation , scatter and curvature of the fp requires either that the haloes surrounding ellipticals have significantly lower mass or concentration than predicted by @xmath5-body simulations , or that they are substantially cored . @xcite observe keplerian fall - off of the velocity dispersion profiles of three bright ellipticals , indicating a low dark matter fraction out to a significant radius . indeed , many studies indicate that dark matter constitutes only a small fraction of the total mass in the central regions of ellipticals ( e.g. @xcite ) . finally , since the majority of ellipticals under investigation here are high surface brightness systems , our findings add to the growing body of evidence that higher stellar surface mass density implies lower enclosed dark matter mass @xcite . indeed , it is found from the slope of the nsa msr that the surface mass density at @xmath12 falls on average as @xmath16 increases , while fig . [ fig : fjr_nodm ] demonstrates that the inferred dark matter mass rises . while `` bottom up '' studies prioritising detailed treatment of the data over cosmological predictions typically find evidence for low or negligible dark matter fractions in the central regions , several `` top down '' studies that start from a cosmological model argue that halo _ contraction _ is required in order to fit the fjr normalisation . we discuss two . @xcite compile a set of dynamical mass measurements at 10 kpc from the centres of elliptical galaxies , and combine this with tfr observations to create a single luminosity velocity data set . they find that elliptical galaxies tend to have larger circular velocities than spirals at fixed luminosity . taking the theoretical prediction to be the halo plus baryon velocities at 10 kpc after am , they argue that halo contraction is necessary to reproduce the observations . comparing the measured nsa fjr presented here to standard tully fisher relations ( e.g. that of @xcite ) , we on the other hand find the fjr to have a lower normalisation and higher slope . only at the very largest stellar masses ( @xmath128 ) do the elliptical galaxy aperture velocity dispersions exceed the spiral galaxy @xmath129 values . @xcite combine a number of empirical power - law correlations relating galaxy size , gas content , halo virial mass , and halo concentration with stellar mass to generate a mock tfr and fjr . they do not in fact solve the jeans equation to model @xmath31 , but instead adopt an average empirical relation between the circular velocity at @xmath12 ( which is output by their model ) and the velocity dispersion there , and an additional relation between that and the integrated velocity dispersion within the sdss aperture . their fiducial model with halo contraction generates an fjr with slope and intercept in almost perfect agreement with their data , i.e. with a considerably lower normalisation and higher slope than ours . the differences between our results and those of @xcite may be largely attributable to the use of different stellar - to - halo mass ratios : where @xcite employs a compilation of satellite kinematics and weak lensing observations , we use abundance matching . by direct comparison , we find that our halo masses are around @xmath130 dex higher at fixed @xmath16 , resulting in an increased need to move dark matter out of the central regions in order to match the fjr normalisation . these differences are within the uncertainties of direct measurements of the stellar - to - halo mass ratio ( e.g. @xcite ) , but have an important effect of mass modelling . the choice of imf is another important systematic . the chabrier imf that we use is among the lightest of those considered routinely in the literature for early - type galaxies ; adopting the heavier kroupa or salpeter imf would increase stellar masses and hence reduce the inferred dark matter mass . imf studies in the literature may be settling on a non - universal imf for early - type galaxies which increases roughly linearly from chabrier at @xmath131 to salpeter at @xmath132 ( @xcite ; @xcite and references therein ) . significantly , it is found not only that this imf favours low central dark matter fractions ( @xmath133 per cent ) , but also that the stellar mass fundamental plane is then fully consistent with the virial plane @xmath134 , with zero scatter . this suggest that were we to adopt this imf in our analysis we would find the data to be in agreement with our no - dm prediction ; in particular , the @xmath98 and @xmath95 values measured from the data would be significantly reduced , and @xmath103 increased . changing the imf would also affect the fjr and fp output by our model , since greater stellar masses at the bright end correspond to larger stellar mass fractions with the halo mass function fixed . however , predicted velocities would increase along with the stellar masses , causing a shift roughly along the fjr and hence impacting the predicted relations less than the observed ones . the non - universal imf of @xcite would therefore still require @xmath135 . more generally , while the choice of imf may affect any of the fjr or fp diagnostics in table [ tab : table ] , the true imf is very unlikely to generate a higher fp scatter in the data . hence the @xmath95 discrepancy could not be eliminated in this way . additional sources of discrepancy between our results and those of @xcite and @xcite may include a potential systematic error associated with combining incommensurate @xmath31 and @xmath136 measurements ( measured at different radii and governed by qualitatively different equations ) , and an offset between data and model velocity dispersions when the latter are not derived from jeans analysis . we stress the importance of measuring velocities in precisely the same way in real and mock galaxies . finally , we note that @xcite assume galaxy and halo variables to be correlated with @xmath16 only , while we include also their expected and observed correlations with one another . there is no guarantee that these covariances are insignificant . @xcite argue that simultaneously matching the normalisation of the tfr and fjr requires that early - type galaxies either experience stronger halo contraction in response to disc formation than late - types , or have a heavier imf . we note that the relative amount of halo contraction or expansion required by spirals and ellipticals depends on whether the two types of galaxies are placed in haloes of systematically different mass or concentration . this effect is included in our framework via the selection factor . the higher the selection factor , the stronger the expansion required for ellipticals relative to spirals . in particular , we find that a high selection factor ( i.e. the haloes of more elliptical galaxies have more mass in the central regions at fixed @xmath16 ) favours @xmath137 for spirals ( dw15 ) , and @xmath138 for ellipticals ( see fig . [ fig : fjr_nu ] and row five of table [ tab : table ] ) , while random selection ( selection factor = arctan(0.5 ) ) favours @xmath138 for spirals and @xmath137 for ellipticals . selection effects are important in any attempt to model the dynamics of spiral and elliptical galaxies within the same framework . our model assumes that the dependence of the halo expansion factor on @xmath16 is the same as that of adiabatic contraction , viz . , maximised at @xmath139 where @xmath140 is largest . while this dependence is motivated in the case of contraction ( @xmath141 ) , there is no guarantee that it would be preserved by the baryonic feedback processes presumed to cause halo expansion in the case @xmath135 . we see from figs . [ fig : fjr_opt ] and [ fig : fjr_3beta ] that to bring the predicted fjr slope into better agreement with the observations would require even more expansion at @xmath139 relative to @xmath142 where the baryon fraction is very low . expansion schemes which act to flatten cusps at intermediate @xmath140 ( e.g. @xcite ) would not help here . the sensitivity of our halo response to @xmath140 makes our constraint on @xmath71 dependent on our being able to probe lower stellar masses ; for the very highest mass galaxies some halo contraction may be permitted . two potential alternatives to halo expansion would require only slight deviation from our assumptions . one possibility is that @xmath143 or @xmath144 is too large , resulting in haloes of too high mass or concentration being generated by the @xmath5-body simulation . however , we showed in appendix b of dw15 that a significant effect on galaxy velocities would require these variables to change by more than their cosmological uncertainties ( a conclusion also reached by @xcite ) . alternatively , the am model could be changed to modify the relation between stellar mass and halo mass and concentration . in particular , lowering the normalisation of the fjr would require galaxies in the range @xmath145 to lie in less massive or less concentrated haloes , and steepening the relation would require this effect to be larger at @xmath146 than @xmath147 . however , this effect would have to be large to affect @xmath97 and @xmath98 significantly : it can be seen from fig . [ fig : fjr_nodm ] there is little need for dark matter at @xmath139 , and hence @xmath56 would have to be lowered to near @xmath16 to obtain agreement here . at @xmath142 , we estimate that @xmath56 would have to be increased by a factor of @xmath148 , or halo concentrations by @xmath149 per cent , to increase @xmath31 by 0.1 dex . @xcite construct a framework in which to constrain the stellar mass dependence of the imf and halo response to disk formation using the slope and normalisation of the fjr and the tilt of the fp ( our @xmath98 , @xmath97 and @xmath103 statistics ) . in agreement with our analysis , they find that a universal value of the imf and @xmath71 underpredicts the fjr slope , and that a baryon - only ( mass follows light ) model with imf slightly heavier than chabrier best accounts for @xmath31 at low @xmath16 . halo contraction is disfavoured for @xmath150 , but may be allowed at higher mass provided the imf is sufficiently light . the value they derive for the tilt of the observed fp is dissimilar to ours , and in particular lies on the other side of the virial value @xmath151 . this may be due to different sample selection ( our cuts on srsic index and light concentration are replaced by one on colour , and they go out to a higher redshift @xmath152 ) , different baryonic mass modelling ( they use a de vaucouleurs plus exponential fit where we take a general srsic index @xmath13 ) , or different stellar mass determinations . nevertheless we agree that the strong dependence of @xmath31 on halo variables induced by contracted or nfw haloes washes out the fp tilt to an unacceptable level , and hence overpredicts @xmath103 . we note that while indeed the average radial orbit anisotropy @xmath87 has only a small effect on @xmath97 and @xmath103 within the range of its observed values , the variation among galaxies , @xmath153 , is less well constrained a priori and may have a larger effect . we note finally that further insight into the shape of dark matter density profiles in this stellar mass range may come from combining precise probes of the inner regions ( e.g. aperture velocity dispersion ) with measurements of the large - scale mass distribution from weak lensing . while some analyses of this type have already been performed ( e.g. @xcite ) , these studies can be significantly expanded and improved with the current generation of large photometric surveys such as des and hsc . we have found that lowering the selection factor that is , reducing the relative preference for ellipticals to live in more massive or more concentrated haloes at fixed @xmath16 improves the model prediction of the fjr slope , intercept and scatter , as well as the average rate at which @xmath31 varies with @xmath12 at fixed @xmath16 . the opposite was argued for by dw15 from the normalisation of the tfr . taken in combination , these results suggest a scenario in which the overall dark matter mass in _ all _ galaxies is reduced ( e.g. by halo expansion ) , rather than one in which the two types of galaxy populate haloes with markedly different dynamical properties . it does however remain possible to situate early - types in more massive haloes by increasing the degree of their halo expansion . a natural next step would be to model these two populations and their selection simultaneously in a fully self - consistent way . we have found two of our five fjr and fp diagnostics intractable within our framework : the fjr slope @xmath98 and the fp scatter @xmath95 . while the slope prediction may be improved by a more sophisticated model for radial orbit anisotropy @xmath75 , a more realistic imf , or a change to the mass - dependence of halo expansion , the discrepancy in the fp scatter likely points to a deeper problem with the model . we discuss this issue here . a number of authors have suggested previously that the scatter in the fp predicted by @xmath0cdm is too large ( e.g. @xcite ) ; our contribution is to quantify these claims precisely within the am framework . when galaxies are situated in haloes , scatter in halo mass and concentration at fixed @xmath31 and @xmath12 provide additional scatter in @xmath16 : roughly , @xmath154 is increased by @xmath155 ( see table [ tab : table ] ) . further , in all of our models that include dark matter ( rows three to nine of table [ tab : table ] ) , the reduction in scatter in progressing from the fjr to the fp is much lower than in the data . in the presence of a halo , the relative dependence of @xmath31 on @xmath12 is reduced , and hence so too is the increase in the precision with which velocity may be determined when its mean dependence on size is accounted for in the optimum way . we have assumed that galaxy size is correlated only with stellar mass and srsic index , and not separately with halo properties at a given stellar mass and profile . thus varying @xmath12 at fixed @xmath16 ( and @xmath66 ) affects only the baryonic mass within the aperture and leaves the dark matter mass unchanged . the failure of our models to match the observed fp scatter while at the same time performing well in terms of @xmath92 and @xmath103 suggests a correlation between galaxy size and halo properties that we have not included . we note for example that @xcite have shown that galaxy half - mass radius is tightly correlated with halo mass , though it is unclear from that work whether this is true at fixed stellar mass . one such possibility for an underlying correlation between size and halo properties was explored in dw15 , where size was determined by the assumption that galaxy and halo specific angular momentum are proportional . this correlation acts to increase the rate at which @xmath156 falls as @xmath4 rises , since larger galaxies end up in less concentrated haloes at fixed @xmath16 and halo spin . a model along these lines would therefore likely reduce @xmath95 ( since the dependence of @xmath31 on @xmath12 would be strengthened relative to its dependence on halo properties that do not appear in the fp ) , and decrease @xmath103 . since predicted and observed @xmath103 values agree under the empirical size model implemented here , this would run the risk of making @xmath103 too negative . this could perhaps be remedied by reducing @xmath82 , but at the price of predicting too low an fjr scatter . furthermore , this type of correlation between @xmath12 and halo spin was shown in dw15 to overpredict the scatter in size at fixed @xmath16 . an alternative approach would be to start with the observed values of @xmath12 , as we have done here , but correlate @xmath12 explicitly with a halo property such as spin or concentration at fixed @xmath16 . this is a form of conditional abundance matching @xcite , and would retain the feature of matching the msr by construction . although it may be possible to find such a correlation that reduces @xmath95 , this method lacks both the theoretical motivation of the dw15 approach and the simplicity of the empirical method presented here . this is worth exploring but we leave it for further work . we note that previous semi - empirical models of elliptical galaxy dynamics such as those of @xcite , @xcite and @xcite do not incorporate significant correlations between galaxy and halo properties beyond those investigated here . if we are right that matching the fp scatter requires such a correlation , it follows that these models too would disagree with the fp in detail . ultimately , hydrodynamical simulations may be necessary to identify and understand all dynamically - relevant correlations of galaxy and halo variables . the fundamental plane ( fp ) and faber jackson relation ( fjr ) of elliptical galaxies provide an important theoretical testing ground , and contain information on galaxy formation processes unlikely to be found elsewhere . we have isolated five key characteristics of these relations and investigated the extent to which they are expected on the basis of a generalised abundance matching model in concordance @xmath0cdm . comparing to a large and homogeneously - selected sample of elliptical galaxies from the nasa - sloan atlas , our main findings are the following . 1 . the magnitude of observed velocity dispersions indicates on average a small amount of dark matter within the central regions of ellipticals with @xmath157 . the discrepancy with the prediction of a baryon - only model rises at higher stellar mass , reaching @xmath158 dex at @xmath159 . 2 . situating galaxies within the haloes produced by an @xmath5-body simulation using currently - favoured cosmological parameter values causes an overprediction of their velocity dispersions , unless dark matter is removed from the central regions . in the context of our framework , successfully matching the normalisation of the fjr requires a degree of halo expansion in response to disc formation similar to that required by @xcite for agreement with the tully fisher relation . abundance matching naturally predicts an fjr slope shallower than that observed . this discrepancy may potentially be accounted for by a variation in either effective radial orbit anisotropy or halo expansion factor with stellar mass , or may indicate non - universality in the imf . 4 . across the parameter space spanned by our framework , the scatter predicted around the fp is significantly too high compared to the observations . this may indicate the presence of correlations between galaxy and halo variables not included in our models . . the rate at which velocity dispersion varies with size at fixed @xmath16 is a measure of the tilt of the fp . we demonstrate that the observed non - homology of elliptical galaxies ( in this case , the variation of srsic index with @xmath16 and @xmath12 ) , coupled with the variations in dynamical mass - to - light ratio generated by abundance matching , are sufficient to account for this key statistic . this is true in detail , however , only when the anisotropy of stellar motions is accounted for . there are several directions in which this work could be taken . first , it is important to understand the correlation between galaxy size and halo properties implied by the small scatter in the fundamental plane . this could be achieved either by exploring the output of hydrodynamical simulations which produce galaxies of realistic size , or by creating toy models in which correlations can be varied and constrained manually . second , the parameters of our framework have implications for various statistics of the large - scale galaxy distribution , including galaxy clustering and galaxy - galaxy lensing , and combining such analyses with dynamical modelling in a single self - consistent framework may be expected to considerably reduce the remaining uncertainties . finally , in this work and @xcite we have treated early and late - type galaxies separately ; future tests of galaxy formation models should do both together . we thank aaron dutton , simon foreman and yao - yuan mao for comments on the manuscript . we further thank the anonymous referee for comments which improved the paper . this work received partial support from the u.s . department of energy under contract number de - ac02 - 76sf00515 . this research made use of the dark sky simulations , which were produced using an incite 2014 allocation on the oak ridge leadership computing facility at oak ridge national laboratory . we thank sam skillman , mike warren , matt turk and our other dark sky collaborators for their efforts in creating these simulations and for providing access to them . additional computation was performed at slac national accelerator laboratory . we thank the slac computational team for their consistent support . this paper made use of data from the nasa - sloan atlas ( www.nsatlas.org ) . we thank michael blanton for making this resource available and for his help in using it . this paper also made use of data from the sloan digital sky survey . funding for sdss - iii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , and the u.s . department of energy . the sdss - iii web site is www.sdss3.org . sdss - iii is managed by the astrophysical research consortium for the participating institutions of the sdss - iii collaboration including the university of arizona , the brazilian participation group , brookhaven national laboratory , university of cambridge , university of florida , the french participation group , the german participation group , the instituto de astrofisica de canarias , the michigan state / notre dame / jina participation group , johns hopkins university , lawrence berkeley national laboratory , max planck institute for astrophysics , new mexico state university , new york university , ohio state university , pennsylvania state university , university of portsmouth , princeton university , the spanish participation group , university of tokyo , university of utah , vanderbilt university , university of virginia , university of washington , and yale university .

surveys of stars with large proper motions are very effective in identifying low - luminosity objects in the vicinity of the sun . most of the white dwarfs and red dwarfs in the solar neighborhood have first been discovered in proper motion surveys . much of the current census of nearby stars is based on follow - up analysis of objects listed in large catalogs of high proper motion stars , most notably the nltt catalog of stars with @xmath9 yr@xmath1 @xcite and ( especially ) its subset , the lhs catalog of stars with @xmath3 yr@xmath1 @xcite . objects with annual proper motions in excess of @xmath10 are now often referred to as `` lhs stars '' . the current census of nearby stars is largely based on follow - up observations of those luyten objects @xcite . the luyten catalogs have long been known @xcite to suffer from significant incompleteness both at low galactic latitudes ( @xmath11 ) , and in southernmost declinations ( @xmath12 ) . the desire to build a complete census of stars in the neighborhood of the sun has motivated a series of new proper motion surveys which , over the years , have gradually filled up these gaps . recently , improved data - mining techniques have been used to conduct a massive proper motion survey of the northern sky , using data from the digitized sky survey ( dss ) . in a systematic analysis of dss scans with the superblink software , an automated blink comparator , several hundred new @xmath3 yr@xmath1 stars have been discovered in the northern sky , effectively filling up the low galactic latitude gap @xcite , and also significantly increasing the completeness of the proper motion census at high galactic latitude @xcite . in fact , the new _ lspm - north catalog _ of stars with proper motions @xmath13 yr@xmath1 @xcite , which compiles the superblink results , is now superseding the luyten catalogs for all areas north of the celestial equator . the southern sky has been very attractive for proper motion hunters in recent years , because of the low completeness of the lhs catalog at southern declinations . several small proper motion surveys have been carried , including the caln - eso ( ce ) proper motion survey @xcite , the wt survey @xcite , and the apmpm survey @xcite , which have all made significant , additions to the zoo of @xmath3 yr@xmath1 proper motion stars , although limited in area . two larger proper motion surveys have recently been conducted based on data from the supercosmos sky survey ( sss ) . the liverpool - edinburgh high proper motion ( lehpm ) survey @xcite has covered @xmath14 square degrees of the south galactic cap , including a large fraction of the largely incomplete @xmath12 cap . the supercosmos - recons survey ( scr ) now aims to obtain an improved proper motion survey of the entire southern sky , although results have so far been presented for @xmath15 sq.deg . south of @xmath16 @xcite . encouraged by the impressive results already obtained with superblink in the northern sky , i have now expanded the analysis of dss images to the southern declinations . in this paper , i report the discovery of a first set of 182 new stars with @xmath17 yr@xmath1 , found after completing the analysis of @xmath18 of the sky south of @xmath19 ( 8980 square degrees ) . the search methodology is outlined in 2 . results are described in 3 . the completeness of our survey is estimated in 4 . finder charts for all new objects are provided in the appendix . our method for identifying high proper motion stars from the digitized sky survey ( dss ) is described in much detail in @xcite . scans from the dss are analyzed using superblink . for each detection , superblink generates two - epoch finding charts which are blinked on the computer screen and examined by eye , in order to filter out any false detection . the whole northern sky has already been analyzed with superblink , and the first major data release can be found in @xcite . analysis of the southern declinations of the dss by superblink presents an additional complication . in the northern hemisphere , the dss incorporates scans from photographic plates in the red passband at both the first ( poss - i red ; 103ae emulsion ) and second epoch ( poss - ii red ; iiiaf emulsion ) . in the south , however , the dss only contains first epoch blue plate ( serc - j blue ; iiiaj emulsion ) while the second epoch images are available only from red ( ses red ; iiiaf emulsion ) and far red plates ( ivn emulsion ) . this is problematic , because superblink is based on image subtraction algorithms , and thus works best on pairs of images that are as similar as possible . nevertheless , tests have shown the procedure to work reasonably well on pairs of images in different passbands , as long as there is a sufficient number of stars with similar plate intensities on both images . tests have shown that superblink can recover high proper motion stars from pairs of blue / red plates , although it fails to detect objects with extreme colors . encouraged by these results , it has been decided to expand the superblink analysis of the dss to southern declinations . the analysis of areas south of decl.=-30@xmath2 has now already been completed . not all areas of the dss are suitable for analysis by superblink , however . there are several zones where first epoch serc - j plates are unavailable , most notably at low galactic latitudes . proper motion searches also require that there be a sufficient temporal baseline between the first and second epoch . this was rather apparent in the northern sky , where the poss - i plates preceed the poss - ii plates by @xmath20 years . but in the southern sky , the baseline between the serc and ses surveys is significantly shorter ( @xmath21 years ) . in some cases , the baseline is too short for accurate proper motion detection / measurement . pairs of plates with baselines under 5 years were thus not considered for our analysis . the superblink algorithms also fail in subfields ( @xmath22 ) where there is an extended , saturated object , such as very bright @xmath23 star . in the end , 8980 square degrees , or @xmath24 of the sky south of decl.=-30@xmath2 , were successfully analyzed ( figure 1 ) . dss scans were provided as input to the superblink software . lists of candidate high proper motion stars were generated on output , along with the usual two - epoch @xmath25 finder charts . the charts were visually examined by blinking on a computer screen , and bogus detections were filtered out . counterparts for all confirmed objects were searched for in the _ usno - b1.0 catalog _ @xcite and in the _ 2mass all - sky catalog of point sources _ @xcite . the 2mass positions were used as a reference to calculate the 2000.0 epoch position . optical magnitudes were obtained from usno - b1.0 , infrared magnitudes from 2mass . a total of 1,147 stars with proper motions in the range @xmath26 yr@xmath1 were located by superblink . a majority of these have been previously reported in the literature as high proper motion stars . the list of superblink detections was cross - correlated with the nltt catalog , and with lists of detections from the apmpm survey @xcite , the ce survey @xcite , the lehpm catalog @xcite , and the recent scr proper motion survey @xcite . a search was also made with simbad , to identify known high proper motion stars from other sources , especially the wt stars @xcite . in all , 965 of the superblink detections were found to be known high proper motion stars . a total of 182 new high proper motion stars were discovered with superblink . a complete list is presented in table 1 . names for the stars are based on the convention introduced by @xcite . we use the letters `` pm '' for _ proper motion _ , followed by a space , followed by the letter `` j '' which denotes that the following digits refer to the j2000 coordinates . the numerals then correspond to the right ascension ( to 0.1 minutes of time ) and declination ( to 1 minute of arc ) . in case of confusion between two objects , most notably close proper motion doubles , we add the directional suffix `` n '' , `` s '' , `` w '' , or `` e '' to distinguish between the two components . finder charts for all the stars can be found in the appendix . stars range in magnitude from @xmath27 to @xmath28 . most notably , the list includes 5 new stars with proper motions exceeding @xmath29 yr@xmath1 . these are pm j00522 - 6201 , lpm j12277 - 4541 , lpm j13384 - 3752 , lpm j19167 - 3638 , and lpm j19261 - 4310 . the list also includes two common proper motion doubles : lpm j19105 - 4132 / lpm j19105 - 4133 , and lpm j22403 - 4931e / lpm j22403 - 4931w . figure 2 plots the distribution on the sky of all the stars found with superblink , with new discoveries denoted by larger symbols . new stars are found all over the survey area , although relatively fewer ones are found near the south galactic cap ( an area covered by the lehpm catalog ) , and in areas south of decl.=-47@xmath2 ( where the scr survey has already made an efficient pass ) . table 1 provides j2000 coordinates for the stars at the 2000.0 epoch . the temporal baseline between the first and second epoch of the discovery plates is given . the measured proper motions are _ relative _ , calculated with respect to background galactic stars . proper motion errors ( @xmath30 ) are estimated based on the detailed analysis of superblink errors from @xcite , with normalization to the temporal baseline ( a shorter baseline makes the proper motion less accurate ) . optical magnitudes b@xmath31 , r@xmath32 , and i@xmath33 ( respectively photographic iiia - j , iiia - f , and iv - n ) are listed ; these are obtained from counterparts of the stars found in the usno - b1.0 catalog @xcite . infrared j , h , and k@xmath34 magnitudes are also given for stars that have counterparts in the 2mass all - sky catalog of point sources @xcite . a simple reduced proper motion analysis provides an estimate of the spectral class of the stars . optical to infrared proper motion diagrams @xcite can be used to separate faint high proper motion stars into three general classes : white dwarfs ( wd ) , subdwarfs ( sd ) , and disk dwarfs ( d ) . here we use a @xmath35 $ ] reduced proper motion diagram and relationships defined in @xcite , where @xmath36 . figure 3 shows the distribution , in the reduced proper motion diagram , of all 1,147 stars found by superblink , with the new discoveries again labeled with larger symbols . among the new discoveries , there are 14 candidate white dwarfs , 52 candidate halo subdwarfs , and 106 candidate disk dwarfs . the status of nine other stars remains ambiguous because of incomplete photometric information . to estimate the completeness of our survey , we have compared our list of superblink detections to lists of high proper motion stars from five previous southern surveys which overlap with our search areas . within the 8980 square degrees area analyzed with superblink , we compiled a total 769 stars with magnitude @xmath7 and proper motion with @xmath8 from the nltt catalog @xcite , the apmpm survey @xcite , the ce survey @xcite , the lehpm catalog @xcite , and the scr survey . note that stars brighter than magnitude @xmath37 are generally not recovered by superblink because they form extended saturated patches on the serc - j and ses plates ; these are thus excluded from the completeness analysis . proper motions for all those bright stars can be obtained from the tycho-2 catalog @xcite . from the reference sample of fainter ( @xmath38 ) stars superblink recovered 612 objects , missing 157 , suggesting recovery rate of @xmath39 . it must be noted , however , that nltt catalog entries are sometimes affected by large positional errors @xcite . it is thus very possible that a number of nltt objects have been `` missed '' by superblink simply because they do not exist at the position recorded in the nltt catalog . likewise , @xcite report problems recovering some the stars listed in the lehpm catalog ; it appears that a number of lehpm entries are either spurious , or have large errors in their reported position . further investigations would be required to determine which of the nltt and/or lehpm objects are not real . in any case , the superblink recovery rate quoted above is probably an underestimate . in fact , if we exclude from the analysis stars all the stars that are listed only in the nltt or lehpm , we reduce the reference sample to 259 previously known stars in our survey area , of which only 26 have been missed by superblink . this suggests a higher recovery rate of @xmath40 . the actual recovery rate of superblink most probably lies somewhere between these two estimates . in any case , it appears that the superblink recovery rate in the south is significantly lower than the much superior detection rate ( @xmath41 ) achieved at northern declinations @xcite . the lower efficiency of superblink in the southern sky is most likely due to the fact that plates of different colors ( blue / red ) are used for the first and second epoch . because superblink is an image - subtraction analysis software , more confusion arises when plates of different passbands are compared . in particular , one of our algorithms for the identification of moving objects requires that they display similar intensities ( to within a factor of 2 ) on both the first and second epoch plates . this filter is required to minimize the number of false detections from the often noisy / grainy photographic plates . the software does renormalize the intensities so that the total plate flux from all stars in the field are equal in the first and second epoch image . in principle , if all the stars in the field have the same color index , the renormalization will correct for it , and the stars will show the same ( renormalized ) flux on both images . however , if there are objects on the field that are bluer , or redder than the typical background star , their renormalized flux will be different on both images . the median value for the color of a field star is @xmath42 . in a typical field , where the red and blue plates have been renormalized by such that a @xmath43 star yields similar instrumental flux values on both images , stars with @xmath44 or with @xmath45 will be missed by superblink . in practice , this effect is mitigated by the fact that the photographic plate response is non - linear , but it remains true that unusually blue or red stars are likely to be missed by the code . in fact , close examination of some apmpm and scr stars missed by superblink shows that they are indeed quite red . preliminary tests reveal that some tweaking of the superblink software parameters , and relaxing of the automated rejection filters would increase the efficiency of the code , albeit at the expense of a larger number of false detections . these would then have to be filtered out by eye , requiring a larger investment of person - hours . a search for high proper motion stars in the digitized sky survey with the superblink software has been completed for an area of 8980 square degrees covering @xmath18 of the sky south of @xmath46 . a total of 1,147 stars with proper motion @xmath17 yr@xmath1 were detected by superblink , of which 182 are found to be new discoveries . of these , there are 123 new stars with @xmath3 yr@xmath1 ( the canonical limit of the lhs catalog ) , including 5 stars with extremely large proper motions ( @xmath47 yr@xmath1 ) . these make significant additions to the census of high proper motion stars at southern declinations . a standard designation system is used to name the new high proper motion stars . it is based on the naming convention first introduced by @xcite . at the time , the standard system was used by eggen as a means to simplify the designation of some high proper motion stars which were then known under a variety of names . the original convention used pm_hhmmm@xmath48ddmm _ , for a star with r1950 coordinates r.a.=_hh mm.m _ , decl.=_@xmath48dd mm_. nowadays , there has been several new surveys and catalogs , each using its own prefix system , with much confusion resulting . we are again reaching a point where it is becoming desirable to uniformize the naming convention of the high proper motion stars . i propose a slight modification to the eggen convention , in which the j2000 coordinates are used instead . high proper motion stars are thus to be designated according to the model pm j_hhmmm@xmath48ddmm _ , for a star at the j2000 coordinates r.a.=_hh mm.m _ i strongly encourage all researchers in the field to start following this convention as well , at least when naming a new discovery . a reduced proper motion analysis suggests that 14 of the new high proper motion stars are white dwarfs , while 52 more are candidate halo subdwarfs . most of the others are probably nearby red dwarfs . all these objects should be high priority targets for follow - up spectroscopy and astrometry ( parallax measurements ) . one notes that in the area covered by our survey , there are now 940 known @xmath38 stars with @xmath8 yr@xmath1 , where luyten s nltt catalog listed 564 . much of the improvement comes from the fact that the ses blue and aao red plates used in the superblink survey are significantly deeper ( 20th magnitude ) than the bruce proper motion survey plates used by luyten in the south , which had a limiting photographic magnitude @xmath49 . at southern declinations , the superblink software has a detection rate of @xmath50 for stars with @xmath7 , which falls short of matching the very high detection rates @xmath41 achieved in the northern sky survey . the main reason is that in the south , the dss has only first epoch scans of blue ( iiiaj ) plates , but second epoch scans of red ( iiiaf ) plates , while in the north , red plates were available for both the first and second epochs . the image subtraction techniques used by superblink makes handling of blue / red pairs more problematic . strategies to increase the superblink detection rate of high proper motion stars in pairs of blue / red images are currently being considered and tested . in the meantime , the superblink analysis of the remaining half of the southern sky ( @xmath51 ) is now being performed , and new superblink discoveries will be available soon . * acknowledgments * the author gratefully acknowledges support from the cordelia corporation , from hilary lipsitz , and from the american museum of natural history . this work has been made possible through the use of the digitized sky surveys . the digitized sky surveys were produced at the space telescope science institute under u.s . government grant nag w-2166 . the images of these surveys are based on photographic data obtained using the oschin schmidt telescope on palomar mountain and the uk schmidt telescope . the plates were processed into the present compressed digital form with the permission of these institutions . the national geographic society - palomar observatory sky atlas ( poss - i ) was made by the california institute of technology with grants from the national geographic society . the second palomar observatory sky survey ( poss - ii ) was made by the california institute of technology with funds from the national science foundation , the national geographic society , the sloan foundation , the samuel oschin foundation , and the eastman kodak corporation . the oschin schmidt telescope is operated by the california institute of technology and palomar observatory . the uk schmidt telescope was operated by the royal observatory edinburgh , with funding from the uk science and engineering research council ( later the uk particle physics and astronomy research council ) , until 1988 june , and thereafter by the anglo - australian observatory . the blue plates of the southern sky atlas and its equatorial extension ( together known as the serc - j ) , as well as the equatorial red ( er ) , and the second epoch [ red ] survey ( ses ) were all taken with the uk schmidt . this publication makes use of data products from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center / california institute of technology , funded by the national aeronautics and space administration and the national science foundation . the data mining required for this work has been made possible with the use of the simbad astronomical database and vizier astronomical catalogs service , both maintained and operated by the centre de donnes astronomiques de strasbourg ( http://cdsweb.u-strasbg.fr/ ) . cutri , r. m. , _ et al . _ 2003 , the 2mass all - sky catalog of point sources university of massachusetts and infrared processing and analysis center ( ipac / california institute of technology _ cds - vizier catalog number ii/246 _ ) dawson , p. c. 1986 , , 311 , 984 eggen , o. j. 1979 , , 39 , 89 eggen , o. j. 1980 , , 43 , 457 hog e. , fabricius c. , makarov v.v . , urban s. , corbin t. , wycoff g. , bastian u. , schwekendiek p. , & wicenec a. 2000 , the tycho-2 catalogue of the 2.5 million brightest stars , , 355 , 27 ( _ cds - vizier catalog number i/259 _ ) gizis , j. e. , & reid , i. n. 1997 , , 109 , 849 gliese , w. , & jahreiss , h. 1991 , preliminary version of the third catalogue of nearby stars , cds - vizier on - line data catalog : v/70a . lpine , s. , shara , m. m. , & rich , r. m. 2002 , , 124 , 1190 lpine , s. , rich , r. m. , & shara , m. m. 2003 , , 125 , 1598 lpine , s. , shara , m. m. , & rich , r. m. 2002 , , 126 , 921 lpine , s. , & shara , m. m. 2005 , , in press luyten w. j. 1979 , lhs catalogue : a catalogue of stars with proper motions exceeding 0.5 " annually , university of minnesota , minneapolis ( _ cds - vizier catalog number i/87b _ ) luyten w. j. 1979 , new luyten catalogue of stars with proper motions larger than two tenths of an arcsecond ( nltt ) , university of minnesota , minneapolis ( _ cds - vizier catalog number i/98a _ ) monet , d. g. , _ et al . _ 2003 , , 125 , 984 , ( the usno - b1 catalog _ cds - vizier catalog number i/284 _ ) pokorny , r. s. , jones , h. r. a. , & hambly , n. c. 2003 , , 397 , 584 reid , i. n. , _ et al . _ 2003 , , 126 , 3007 ruiz , m. t. , wischnjewsky , m. , rojo , p. m. , & gonzalez , l. e. 2001 , , 133 , 119 salim , s. , & gould , a. 2002 , , 575 , l83 salim , s. , & gould , a. 2003 , , 582 , 1011 scholz , r .- d . , irwin , m. , ibata , r. , jahreiss , h. , & malkov , o. yu . 2000 , , 353 , 958 subasavage , j. p. , henry , t. j. , hambly , n. c. , brown , m. a. , & jao , w .- c . 2005 , , in press wroblewski , h. , & torres , a. 1997 , , 122 , 447 wroblewski , h. , & costa , e. 2001 , , 367 , 725 finder charts of high proper motion stars are generated as a by - product of the superblink software . we present here finder charts for all the new , high proper motion stars presented in this paper and listed in table 1 . all the charts consist of pairs of @xmath25 images showing the local field at two different epochs . the name of the star is indicated in the center just below the chart , and corresponds to the name given in table 1 . to the left is the poss - i field , with the epoch of the field noted in the lower left corner . to the right is the modified poss - ii field which has been shifted , rotated , and degraded in such a way that it matches the quality and aspect of the poss - i image . the epoch of the poss - ii field is noted on the lower right corner . high proper motion stars are identified with circles centered on their positions at the epoch on the plate . the charts are oriented in the local x - y coordinate system of the poss - i image ; the poss - ii image has been remapped on the poss - i grid . this means that north is generally up and east left , but the fields might appear rotated by a small angle for high declination objects . sometimes a part of the field is missing : this is an artifact of the code . superblink works on @xmath22 dss subfields . if a high proper motion star is identified near the edge of that subfield , the output chart appears truncated . lrrrrrrrrrrrrrr pm j00151 - 5919 & 0 15 10.13 & -59 19 56.7 & 0.483 & 0.473 & 0.097 & 16.02 & 0.020 & 19.6 & 17.0 & 13.5 & 12.14 & 11.56 & 11.21 & d + pm j00168 - 7233 & 0 16 51.97 & -72 33 46.4 & 0.507 & -0.383 & -0.332 & 9.92 & 0.032 & 19.0 & 17.0 & 14.8 & 13.33 & 12.84 & 12.57 & d + pm j00294 - 6900 & 0 29 26.87 & -69 00 13.3 & 0.486 & 0.465 & -0.139 & 13.89 & 0.023 & 18.0 & 16.7 & 14.3 & 13.22 & 12.74 & 12.47 & d + pm j00327 - 7641 & 0 32 43.62 & -76 41 28.7 & 0.593 & 0.592 & 0.032 & 9.92 & 0.032 & 19.3 & 17.4 & 15.5 & 14.58 & 14.03 & 13.81 & sd + pm j00344 - 6849 & 0 34 28.77 & -68 49 29.7 & 0.508 & -0.016 & -0.508 & 13.89 & 0.023 & 19.5 & 18.8 & 17.8 & & & & + pm j00522 - 6201 & 0 52 15.28 & -62 01 54.6 & 1.073 & 1.065 & 0.122 & 15.09 & 0.021 & 19.7 & 16.7 & 13.3 & 12.15 & 11.74 & 11.37 & d + pm j00568 - 7603 & 0 56 53.20 & -76 03 43.8 & 0.457 & 0.194 & -0.414 & 13.96 & 0.023 & & 17.5 & 15.1 & 12.57 & 11.99 & 11.62 & d + pm j01009 - 7904 & 1 00 56.07 & -79 04 25.2 & 0.450 & -0.270 & -0.360 & 14.97 & 0.021 & & 11.9 & 10.0 & 8.80 & 8.17 & 7.88 & d + pm j01082 - 3714w & 1 08 13.21 & -37 14 39.2 & 0.509 & -0.128 & -0.492 & 17.39 & 0.018 & & 15.2 & 13.2 & 12.84 & 12.36 & 12.12 & sd + pm j01082 - 3714e & 1 08 13.69 & -37 14 39.1 & 0.509 & -0.128 & -0.492 & 17.39 & 0.018 & & 15.2 & 14.1 & 13.41 & 12.97 & 12.70 & sd + pm j01184 - 7504 & 1 18 29.66 & -75 04 55.0 & 0.454 & 0.446 & 0.081 & 13.96 & 0.023 & 19.7 & 15.5 & & 10.85 & 10.29 & 10.02 & d + pm j01253 - 4545 & 1 25 18.02 & -45 45 31.1 & 0.748 & 0.517 & -0.541 & 13.82 & 0.023 & 18.0 & 16.9 & 18.7 & 15.11 & 14.84 & 14.91 & wd + pm j01598 - 4736 & 1 59 48.21 & -47 36 12.3 & 0.466 & 0.272 & -0.378 & 13.05 & 0.025 & & 16.2 & 15.1 & 13.80 & 13.30 & 13.04 & sd + pm j03007 - 4653 & 3 00 45.15 & -46 53 50.6 & 0.745 & 0.676 & 0.312 & 16.95 & 0.019 & 18.3 & 15.4 & 12.7 & 11.79 & 11.31 & 11.02 & d + pm j03064 - 3325 & 3 06 26.46 & -33 25 42.9 & 0.972 & -0.164 & -0.957 & 16.01 & 0.020 & 19.5 & 16.7 & 15.8 & 14.35 & 13.85 & 13.58 & sd + pm j03116 - 3113 & 3 11 37.99 & -31 13 21.2 & 0.544 & 0.162 & -0.520 & 13.99 & 0.023 & 20.8 & 18.3 & 16.3 & 14.37 & 13.88 & 13.63 & d + pm j03229 - 8159 & 3 22 59.54 & -81 59 54.4 & 0.504 & 0.383 & 0.327 & 12.11 & 0.026 & & 18.2 & 14.8 & 13.02 & 12.48 & 12.08 & d + pm j03245 - 5220 & 3 24 32.99 & -52 20 39.3 & 0.823 & -0.444 & -0.692 & 11.92 & 0.027 & 17.2 & 14.6 & 11.6 & 11.10 & 10.51 & 10.17 & d + pm j03286 - 6756 & 3 28 38.30 & -67 56 26.5 & 0.469 & 0.451 & 0.129 & 5.28 & 0.061 & & 17.2 & 14.7 & 13.09 & 12.66 & 12.33 & d + pm j03536 - 5006 & 3 53 36.57 & -50 06 00.9 & 0.523 & 0.255 & 0.457 & 17.21 & 0.019 & & 20.5 & 18.5 & & & & + pm j04356 - 6105 & 4 35 37.39 & -61 05 40.4 & 0.502 & 0.432 & 0.256 & 14.71 & 0.022 & 17.5 & 16.5 & 16.0 & 15.70 & 15.21 & 15.10 & wd + pm j05175 - 4227 & 5 17 35.96 & -42 27 54.4 & 0.452 & 0.191 & -0.410 & 15.06 & 0.021 & & 12.8 & 11.7 & 11.80 & 11.31 & 11.06 & sd + pm j05472 - 6009 & 5 47 14.62 & -60 09 50.5 & 0.475 & 0.164 & -0.445 & 17.28 & 0.019 & 20.9 & 16.4 & & 13.39 & 12.89 & 12.68 & sd + pm j05597 - 4519 & 5 59 44.05 & -45 19 10.5 & 0.819 & 0.449 & 0.685 & 12.87 & 0.025 & 18.7 & 16.8 & 15.4 & 14.15 & 13.61 & 13.41 & sd + pm j06491 - 7603 & 6 49 06.65 & -76 03 50.1 & 0.512 & 0.428 & 0.280 & 12.88 & 0.025 & & 16.7 & 16.0 & 14.88 & 14.43 & 14.33 & sd + pm j07096 - 3941 & 7 09 37.05 & -39 41 52.0 & 0.472 & -0.124 & -0.455 & 14.86 & 0.022 & 16.7 & 14.7 & 12.5 & 11.77 & 11.21 & 10.99 & d + pm j07096 - 4648 & 7 09 37.27 & -46 48 59.1 & 0.456 & 0.137 & 0.434 & 14.81 & 0.022 & & 14.0 & 13.1 & 12.20 & 11.70 & 11.49 & sd + pm j07110 - 3824 & 7 11 00.96 & -38 24 46.1 & 0.893 & 0.516 & -0.728 & 14.06 & 0.023 & 20.6 & 15.6 & 13.0 & 11.09 & 10.55 & 10.23 & d + pm j07185 - 3351 & 7 18 30.75 & -33 51 24.5 & 0.461 & -0.024 & 0.461 & 14.93 & 0.021 & 19.0 & 18.4 & 17.8 & & & & + pm j07228 - 3042 & 7 22 51.34 & -30 42 34.4 & 0.494 & 0.323 & 0.374 & 15.11 & 0.021 & 18.3 & 17.4 & 17.2 & 16.42 & 15.69 & 15.52 & wd + pm j07253 - 3742 & 7 25 21.57 & -37 42 59.1 & 0.587 & 0.211 & -0.548 & 14.06 & 0.023 & 13.3 & 12.1 & 10.1 & 9.06 & 8.42 & 8.22 & d + pm j07283 - 5812 & 7 28 23.20 & -58 12 31.1 & 0.482 & 0.367 & -0.312 & 14.83 & 0.022 & & 12.9 & 12.4 & 11.60 & 10.97 & 10.84 & sd + pm j07293 - 4804 & 7 29 21.66 & -48 04 27.3 & 0.450 & 0.361 & 0.266 & 14.16 & 0.023 & 14.5 & 12.7 & 10.9 & 10.02 & 9.44 & 9.19 & d + pm j07401 - 4257 & 7 40 11.83 & -42 57 40.7 & 0.703 & -0.436 & 0.551 & 14.00 & 0.023 & 15.5 & 12.5 & 10.3 & 8.68 & 8.09 & 7.77 & d + pm j07433 - 5457 & 7 43 18.57 & -54 57 30.0 & 0.484 & -0.098 & 0.474 & 14.89 & 0.021 & & 18.1 & 15.1 & 13.19 & 12.61 & 12.27 & d + pm j07571 - 4021 & 7 57 07.87 & -40 21 33.5 & 0.551 & -0.532 & 0.143 & 12.20 & 0.026 & 17.3 & 14.5 & 15.7 & 11.41 & 10.92 & 10.65 & d + pm j08152 - 3600 & 8 15 15.99 & -36 00 59.5 & 0.639 & -0.105 & 0.630 & 15.17 & 0.021 & 16.2 & 15.0 & 12.6 & 10.74 & 10.17 & 9.88 & d + pm j08172 - 6808 & 8 17 16.66 & -68 08 36.7 & 0.551 & -0.301 & 0.461 & 14.06 & 0.023 & 18.9 & 16.8 & 15.5 & & & & + pm j08186 - 3110 & 8 18 40.24 & -31 10 19.6 & 0.817 & 0.224 & -0.785 & 14.05 & 0.023 & 16.1 & 15.1 & 14.9 & 14.92 & 14.73 & 14.83 & wd + pm j08276 - 3003 & 8 27 40.89 & -30 03 00.0 & 0.594 & -0.326 & 0.497 & 14.05 & 0.023 & 16.3 & 14.1 & 12.2 & 10.67 & 10.17 & 9.92 & d + pm j08300 - 5039 & 8 30 00.78 & -50 39 46.2 & 0.989 & -0.568 & 0.810 & 14.11 & 0.023 & 15.7 & 13.8 & 12.5 & 10.70 & 10.15 & 9.90 & d + pm j08355 - 3400 & 8 35 31.73 & -34 00 36.9 & 0.546 & -0.138 & -0.529 & 12.26 & 0.026 & 16.0 & 14.3 & 11.2 & 9.90 & 9.37 & 9.08 & d + pm j08446 - 4805 & 8 44 38.92 & -48 05 21.8 & 0.748 & -0.224 & 0.713 & 14.11 & 0.023 & 14.4 & 12.7 & 10.7 & 9.36 & 8.81 & 8.56 & d + pm j08458 - 3051 & 8 45 51.96 & -30 51 31.3 & 0.518 & -0.477 & -0.201 & 13.97 & 0.023 & 16.4 & 14.5 & 12.7 & 10.82 & 10.30 & 10.04 & d + pm j08463 - 4407 & 8 46 20.99 & -44 07 22.0 & 0.477 & 0.462 & -0.117 & 15.17 & 0.021 & 17.5 & 16.3 & 14.8 & 12.80 & 12.26 & 12.04 & d + pm j08471 - 3046 & 8 47 09.70 & -30 46 12.0 & 0.528 & 0.062 & -0.524 & 13.97 & 0.023 & 15.5 & 13.4 & 12.1 & 10.39 & 9.91 & 9.60 & d + pm j08496 - 3138 & 8 49 38.94 & -31 38 22.9 & 0.506 & -0.174 & 0.475 & 13.97 & 0.023 & 16.4 & 14.8 & 13.2 & 11.69 & 11.16 & 10.91 & d + pm j08503 - 5848 & 8 50 21.34 & -58 48 06.7 & 0.612 & -0.418 & 0.447 & 14.84 & 0.022 & 18.0 & 17.7 & 17.1 & 16.74 & 16.90 & 16.83 & wd + pm j09047 - 3804 & 9 04 46.49 & -38 04 07.0 & 0.614 & 0.408 & -0.458 & 14.80 & 0.022 & 17.3 & 14.8 & 13.7 & 12.03 & 11.57 & 11.36 & sd + pm j09172 - 3849 & 9 17 13.61 & -38 49 36.0 & 0.494 & -0.086 & 0.486 & 14.80 & 0.022 & 21.4 & 14.8 & 13.5 & 11.56 & 11.12 & 10.80 & d + pm j09181 - 4437 & 9 18 08.62 & -44 37 24.5 & 0.501 & -0.405 & 0.295 & 13.98 & 0.023 & 17.3 & 16.8 & 16.5 & 15.59 & 15.37 & 14.89 & wd + pm j09204 - 4922 & 9 20 26.02 & -49 22 34.7 & 0.561 & -0.406 & 0.387 & 14.81 & 0.022 & 17.8 & 15.9 & 12.7 & 11.33 & 10.82 & 10.53 & d + pm j09233 - 3537 & 9 23 23.68 & -35 37 58.3 & 0.485 & -0.437 & -0.210 & 14.03 & 0.023 & 16.8 & 13.8 & 12.9 & 12.24 & 11.72 & 11.53 & sd + pm j09237 - 7248 & 9 23 42.50 & -72 48 36.7 & 0.472 & -0.360 & 0.305 & 14.88 & 0.022 & & 15.1 & 13.6 & 12.63 & 12.07 & 11.82 & sd + pm j09271 - 4137 & 9 27 07.19 & -41 37 12.0 & 0.476 & 0.389 & -0.274 & 14.28 & 0.022 & & 11.3 & 11.0 & 10.32 & 9.89 & 9.80 & sd + pm j09376 - 3852 & 9 37 36.23 & -38 52 23.0 & 0.586 & -0.446 & 0.380 & 14.28 & 0.022 & & 16.6 & 16.5 & 16.03 & 15.60 & 15.97 & wd + pm j09396 - 6135 & 9 39 38.82 & -61 35 11.3 & 0.481 & 0.467 & 0.112 & 14.81 & 0.022 & 19.1 & 13.2 & & 10.80 & 10.30 & 9.98 & d + pm j09566 - 4234 & 9 56 36.99 & -42 34 27.3 & 0.658 & 0.382 & -0.536 & 14.99 & 0.021 & & 14.9 & 12.5 & 10.99 & 10.47 & 10.21 & d + pm j09598 - 5027 & 9 59 53.94 & -50 27 17.7 & 0.677 & -0.665 & 0.128 & 15.83 & 0.020 & 16.4 & 15.6 & 15.6 & 15.46 & 15.20 & 14.82 & wd + pm j10026 - 7519 & 10 02 37.66 & -75 19 22.7 & 0.523 & -0.371 & 0.368 & 14.95 & 0.021 & & 16.9 & 14.6 & 12.60 & 11.98 & 11.67 & d + pm j10050 - 4322 & 10 05 03.11 & -43 22 28.2 & 0.632 & -0.589 & 0.228 & 14.07 & 0.023 & 15.1 & 13.3 & 11.0 & 9.85 & 9.32 & 9.06 & d + pm j10185 - 5254 & 10 18 30.01 & -52 54 44.3 & 0.514 & -0.297 & 0.419 & 19.10 & 0.017 & 16.4 & 13.8 & 13.3 & 13.74 & 13.29 & 13.04 & wd + pm j10256 - 4703 & 10 25 39.40 & -47 03 51.1 & 0.477 & -0.445 & 0.170 & 14.07 & 0.023 & 19.9 & 17.5 & 16.2 & 15.05 & 14.63 & 14.23 & sd + pm j10326 - 5255 & 10 32 40.69 & -52 55 05.8 & 0.558 & -0.512 & 0.221 & 19.10 & 0.017 & & 15.8 & & 11.98 & 11.38 & 11.01 & d + pm j10346 - 4259 & 10 34 37.86 & -42 59 57.0 & 0.471 & -0.422 & 0.208 & 14.06 & 0.023 & 16.7 & 14.9 & 13.4 & 11.65 & 11.15 & 10.89 & d + pm j10352 - 7424 & 10 35 12.32 & -74 24 57.0 & 0.461 & -0.397 & 0.233 & 14.95 & 0.021 & & 15.8 & & 13.40 & 12.84 & 12.58 & sd + pm j10382 - 4933 & 10 38 15.18 & -49 33 44.6 & 0.682 & -0.333 & -0.595 & 15.03 & 0.021 & & 13.9 & 11.7 & 10.31 & 9.72 & 9.43 & d + pm j10395 - 3820 & 10 39 32.40 & -38 20 02.0 & 0.702 & -0.196 & -0.674 & 13.40 & 0.024 & 16.9 & 15.1 & & 10.85 & 10.30 & 10.00 & d + pm j10538 - 3858 & 10 53 49.42 & -38 58 59.0 & 0.558 & -0.347 & 0.436 & 18.10 & 0.018 & 15.7 & 14.1 & 12.4 & 10.91 & 10.43 & 10.13 & d + pm j10587 - 3854 & 10 58 47.12 & -38 54 14.8 & 0.622 & -0.603 & 0.151 & 18.10 & 0.018 & 16.4 & 14.5 & 12.4 & 11.01 & 10.52 & 10.21 & d + pm j11076 - 4105 & 11 07 40.05 & -41 05 44.2 & 0.492 & -0.475 & 0.129 & 14.90 & 0.021 & 22.3 & 18.1 & 16.0 & 14.28 & 13.80 & 13.51 & d + pm j11094 - 4631 & 11 09 28.34 & -46 31 10.0 & 0.452 & -0.447 & 0.066 & 18.77 & 0.017 & & 15.6 & 14.0 & 12.35 & 11.88 & 11.55 & d + pm j11104 - 3608 & 11 10 29.04 & -36 08 24.5 & 0.535 & -0.531 & -0.071 & 14.13 & 0.023 & & 14.6 & 18.7 & 10.93 & 10.34 & 10.00 & d + pm j11256 - 3834 & 11 25 37.31 & -38 34 42.9 & 0.541 & -0.528 & -0.119 & 14.90 & 0.021 & & 13.8 & & 10.09 & 9.51 & 9.19 & d + pm j11308 - 4120 & 11 30 48.27 & -41 20 20.9 & 0.454 & -0.441 & -0.107 & 14.90 & 0.021 & 16.8 & 15.0 & 13.2 & 12.09 & 11.52 & 11.32 & d + pm j11329 - 4039 & 11 32 57.95 & -40 39 21.6 & 0.786 & -0.685 & 0.385 & 19.14 & 0.017 & 15.7 & 13.5 & 11.5 & 10.38 & 9.89 & 9.65 & d + pm j11413 - 3624 & 11 41 21.53 & -36 24 34.7 & 0.590 & 0.532 & 0.255 & 15.93 & 0.020 & & 12.4 & 10.1 & 8.49 & 7.97 & 7.70 & d + pm j11480 - 4523 & 11 48 03.32 & -45 23 01.8 & 0.635 & -0.536 & -0.340 & 14.12 & 0.023 & 16.9 & 14.2 & 13.8 & 14.89 & 14.76 & 14.60 & wd + pm j11495 - 4248 & 11 49 31.69 & -42 48 10.4 & 0.989 & -0.979 & -0.136 & 19.14 & 0.017 & 16.0 & 13.3 & 12.3 & 11.67 & 11.11 & 10.90 & sd + pm j11511 - 4142 & 11 51 07.82 & -41 42 17.3 & 0.600 & -0.569 & -0.189 & 19.14 & 0.017 & & 15.3 & 12.8 & 11.51 & 10.99 & 10.68 & d + pm j11578 - 5022 & 11 57 50.47 & -50 22 32.1 & 0.478 & -0.397 & -0.265 & 14.85 & 0.022 & & 12.4 & 10.8 & 9.27 & 8.65 & 8.37 & d + pm j11596 - 4256 & 11 59 37.65 & -42 56 39.1 & 0.603 & -0.446 & -0.405 & 17.65 & 0.018 & 15.4 & 12.3 & 10.6 & 9.54 & 8.98 & 8.72 & d + pm j12042 - 4037 & 12 04 15.50 & -40 37 52.1 & 0.603 & 0.230 & -0.557 & 17.06 & 0.019 & 14.9 & 12.9 & 10.8 & 9.57 & 9.02 & 8.75 & d + pm j12088 - 3723 & 12 08 51.03 & -37 23 27.3 & 0.477 & 0.346 & -0.329 & 15.86 & 0.020 & 16.0 & 14.3 & 12.0 & 10.62 & 10.08 & 9.78 & d + pm j12146 - 4603 & 12 14 40.02 & -46 03 14.2 & 0.797 & -0.747 & -0.276 & 17.65 & 0.018 & 16.9 & 14.6 & 12.3 & 10.32 & 9.75 & 9.44 & d + pm j12159 - 5014 & 12 15 56.96 & -50 14 19.3 & 0.665 & -0.665 & 0.001 & 14.85 & 0.022 & 19.0 & 17.2 & 15.8 & 14.64 & 14.21 & 14.13 & sd + pm j12201 - 4546 & 12 20 08.01 & -45 46 18.3 & 0.750 & -0.702 & 0.262 & 17.65 & 0.018 & & 14.8 & 13.6 & 12.70 & 12.16 & 11.95 & sd + pm j12225 - 4002 & 12 22 32.28 & -40 02 12.2 & 0.602 & -0.577 & -0.170 & 17.06 & 0.019 & 19.8 & 17.5 & 14.5 & 12.60 & 11.98 & 11.62 & d + pm j12277 - 4541 & 12 27 46.91 & -45 41 17.1 & 1.321 & -1.285 & 0.301 & 15.11 & 0.021 & 16.4 & 14.5 & 13.2 & 12.75 & 12.40 & 12.27 & sd + pm j12300 - 3411 & 12 30 01.77 & -34 11 23.9 & 0.563 & -0.465 & -0.317 & 15.92 & 0.020 & 15.3 & 13.6 & 11.2 & 9.34 & 8.77 & 8.44 & d + pm j12313 - 4018 & 12 31 21.19 & -40 18 36.8 & 0.569 & -0.568 & -0.023 & 13.02 & 0.025 & 16.2 & 14.0 & 12.1 & 10.44 & 9.94 & 9.64 & d + pm j12346 - 5640 & 12 34 40.35 & -56 40 12.4 & 0.559 & -0.527 & -0.183 & 14.65 & 0.022 & 19.8 & 16.8 & & & & & + pm j12355 - 4527 & 12 35 35.00 & -45 27 03.8 & 0.450 & -0.302 & 0.333 & 15.11 & 0.021 & 15.3 & 13.4 & 11.9 & 10.57 & 10.04 & 9.76 & d + pm j12415 - 4717 & 12 41 33.18 & -47 17 05.7 & 0.471 & -0.454 & -0.125 & 15.11 & 0.021 & 16.2 & 14.5 & 14.2 & 12.77 & 12.21 & 12.06 & sd + pm j12472 - 4441 & 12 47 15.88 & -44 41 50.0 & 0.688 & -0.683 & 0.080 & 15.11 & 0.021 & 19.0 & 17.7 & 16.2 & 14.65 & 14.26 & 14.11 & sd + pm j12515 - 3846 & 12 51 31.82 & -38 46 12.5 & 0.625 & -0.616 & 0.102 & 15.10 & 0.021 & 18.6 & 16.9 & 13.9 & 12.09 & 11.56 & 11.25 & d + pm j13002 - 6024 & 13 00 15.78 & -60 24 19.3 & 0.490 & -0.437 & 0.221 & 15.02 & 0.021 & 16.9 & 15.4 & 12.9 & 11.76 & 11.21 & 10.84 & d + pm j13077 - 7925 & 13 07 44.45 & -79 25 10.3 & 0.514 & -0.486 & -0.165 & 17.10 & 0.019 & & 17.8 & 17.3 & & & & + pm j13084 - 3903 & 13 08 27.01 & -39 03 32.0 & 0.456 & -0.443 & -0.110 & 15.10 & 0.021 & & 16.9 & 14.1 & 12.60 & 12.07 & 11.78 & d + pm j13095 - 3848 & 13 09 31.06 & -38 48 04.4 & 0.572 & -0.554 & -0.141 & 15.10 & 0.021 & & 12.8 & 10.9 & 9.86 & 9.30 & 9.05 & d + pm j13139 - 5011 & 13 13 54.14 & -50 11 09.6 & 0.468 & -0.414 & -0.216 & 15.04 & 0.021 & 18.9 & 16.3 & 13.7 & 12.38 & 11.83 & 11.50 & d + pm j13276 - 3551 & 13 27 39.59 & -35 51 01.0 & 0.518 & -0.404 & -0.325 & 15.94 & 0.020 & 16.8 & 14.7 & 12.3 & 11.13 & 10.60 & 10.33 & d + pm j13338 - 5738 & 13 33 53.03 & -57 38 42.5 & 0.710 & -0.695 & -0.144 & 17.93 & 0.018 & 19.1 & 17.2 & 15.0 & 12.84 & 12.32 & 12.03 & d + pm j13379 - 4311 & 13 37 56.06 & -43 11 30.1 & 0.522 & -0.408 & -0.325 & 16.03 & 0.020 & 17.1 & 15.3 & 13.5 & 11.70 & 11.11 & 10.79 & d + pm j13384 - 3752 & 13 38 26.53 & -37 52 50.4 & 1.265 & -1.264 & -0.031 & 15.94 & 0.020 & & 15.5 & 13.2 & 11.75 & 11.28 & 10.97 & d + pm j13423 - 3534 & 13 42 21.29 & -35 34 50.8 & 0.903 & -0.878 & -0.211 & 15.94 & 0.020 & 19.0 & 16.5 & & 13.76 & 13.22 & 12.94 & sd + pm j13420 - 3544 & 13 42 00.21 & -35 44 51.5 & 0.468 & -0.454 & 0.113 & 15.94 & 0.020 & 18.8 & 16.3 & & 13.31 & 12.80 & 12.52 & sd + pm j13438 - 3447 & 13 43 48.93 & -34 47 49.5 & 0.487 & -0.479 & -0.086 & 15.94 & 0.020 & 18.5 & 17.0 & & 16.25 & 15.90 & 15.59 & wd + pm j13548 - 4129 & 13 54 48.47 & -41 29 04.8 & 0.690 & -0.592 & -0.354 & 14.70 & 0.022 & 19.4 & 18.1 & 16.8 & 15.52 & 15.06 & 14.63 & sd + pm j14005 - 3935 & 14 00 32.32 & -39 35 29.3 & 0.508 & -0.496 & -0.109 & 14.70 & 0.022 & 17.0 & 15.8 & 14.1 & 13.47 & 12.90 & 12.66 & sd + pm j14104 - 5001 & 14 10 29.10 & -50 01 59.5 & 0.538 & -0.158 & -0.514 & 14.19 & 0.023 & & 17.1 & 14.7 & 12.72 & 12.31 & 11.97 & d + pm j14123 - 3941 & 14 12 21.16 & -39 41 33.4 & 0.613 & -0.538 & -0.293 & 16.14 & 0.020 & 16.3 & 14.3 & 12.3 & 10.99 & 10.43 & 10.18 & d + pm j14129 - 4001 & 14 12 58.26 & -40 01 21.7 & 0.450 & -0.414 & -0.176 & 16.14 & 0.020 & 20.8 & 18.1 & 17.8 & & & & + pm j14330 - 3846 & 14 33 03.36 & -38 46 59.5 & 0.475 & -0.465 & -0.095 & 16.14 & 0.020 & 18.6 & 16.0 & 15.3 & 14.37 & 13.78 & 13.59 & sd + pm j14344 - 4700 & 14 34 27.25 & -47 00 15.6 & 0.517 & -0.517 & -0.002 & 18.09 & 0.018 & 21.1 & 15.2 & & 11.94 & 11.47 & 11.20 & d + pm j14373 - 4002 & 14 37 21.44 & -40 02 49.1 & 0.509 & -0.408 & -0.304 & 13.80 & 0.023 & 15.9 & 14.3 & 12.4 & 10.79 & 10.21 & 9.90 & d + pm j14382 - 6231 & 14 38 13.58 & -62 31 39.5 & 0.695 & -0.546 & -0.429 & 17.34 & 0.018 & & 13.4 & 12.8 & 10.52 & 9.98 & 9.72 & d + pm j14435 - 5430 & 14 43 35.13 & -54 30 44.6 & 0.645 & -0.631 & -0.131 & 15.19 & 0.021 & 18.0 & 15.8 & 13.9 & 11.96 & 11.41 & 11.14 & d + pm j14441 - 3426 & 14 44 06.57 & -34 26 46.5 & 0.493 & -0.030 & -0.492 & 13.98 & 0.023 & & 13.7 & 11.0 & 9.74 & 9.18 & 8.88 & d + pm j14440 - 4414 & 14 44 01.97 & -44 14 40.8 & 0.509 & -0.441 & -0.254 & 18.09 & 0.018 & 20.5 & 17.0 & 15.1 & 13.33 & 12.81 & 12.50 & d + pm j14500 - 3742 & 14 50 02.87 & -37 42 09.6 & 0.511 & -0.297 & -0.416 & 13.80 & 0.023 & 15.0 & 13.3 & 11.0 & 9.95 & 9.37 & 9.07 & d + pm j14558 - 3914 & 14 55 51.61 & -39 14 33.1 & 0.786 & -0.780 & -0.095 & 13.80 & 0.023 & 20.7 & 14.7 & & 12.50 & 11.98 & 11.79 & sd + pm j14570 - 4705 & 14 57 05.32 & -47 05 25.8 & 0.548 & -0.411 & -0.362 & 10.34 & 0.031 & 16.7 & 15.2 & 14.5 & 13.53 & 12.96 & 12.82 & sd + pm j14570 - 5943 & 14 57 02.97 & -59 43 46.8 & 0.617 & -0.407 & -0.463 & 17.34 & 0.018 & 17.9 & 15.7 & 13.9 & 12.48 & 12.01 & 11.78 & d + pm j14585 - 5916 & 14 58 31.22 & -59 16 42.3 & 0.472 & -0.423 & -0.210 & 18.02 & 0.018 & & 13.8 & & 10.48 & 9.85 & 9.58 & d + pm j14596 - 4043 & 14 59 40.72 & -40 43 19.6 & 0.754 & -0.687 & -0.308 & 13.87 & 0.023 & 17.9 & 16.6 & 14.5 & 14.29 & 13.80 & 13.47 & sd + pm j15044 - 4353 & 15 04 29.26 & -43 53 43.8 & 0.478 & -0.414 & -0.240 & 10.34 & 0.031 & 16.8 & 15.4 & 13.1 & 11.79 & 11.18 & 10.88 & d + pm j15054 - 4620 & 15 05 27.37 & -46 20 16.1 & 0.530 & -0.462 & -0.259 & 10.34 & 0.031 & 16.0 & 14.5 & & 11.07 & 10.51 & 10.28 & d + pm j15116 - 3403 & 15 11 38.66 & -34 03 15.8 & 0.567 & -0.165 & -0.543 & 16.25 & 0.020 & 15.7 & 13.7 & 11.3 & 10.05 & 9.42 & 9.13 & d + pm j15128 - 4354 & 15 12 52.34 & -43 54 12.4 & 0.474 & -0.244 & -0.407 & 10.34 & 0.031 & 16.2 & 13.6 & 11.5 & 10.57 & 9.96 & 9.75 & d + pm j15145 - 4625 & 15 14 31.88 & -46 25 55.3 & 0.539 & -0.538 & -0.005 & 17.38 & 0.018 & 15.6 & 14.9 & 14.4 & 14.71 & 14.61 & 14.65 & wd + pm j15184 - 3544 & 15 18 29.68 & -35 44 11.2 & 0.481 & -0.067 & -0.477 & 16.25 & 0.020 & & 15.8 & 14.5 & 13.73 & 13.28 & 13.10 & sd + pm j15206 - 7137 & 15 20 41.70 & -71 37 08.3 & 0.478 & -0.396 & -0.267 & 16.33 & 0.020 & 18.0 & 16.8 & & 13.25 & 12.75 & 12.46 & d + pm j15231 - 7711 & 15 23 11.17 & -77 11 23.5 & 0.831 & -0.772 & -0.305 & 17.98 & 0.018 & & 15.8 & 14.7 & 13.76 & 13.27 & 13.03 & sd + pm j15246 - 6239 & 15 24 40.53 & -62 39 27.1 & 0.497 & -0.394 & -0.303 & 17.06 & 0.019 & & 13.9 & 12.4 & 10.93 & 10.48 & 10.18 & d + pm j15252 - 4549 & 15 25 15.12 & -45 49 36.5 & 0.457 & -0.202 & -0.410 & 17.38 & 0.018 & 18.4 & 16.4 & 15.3 & 14.49 & 14.02 & 13.76 & sd + pm j15292 - 5620 & 15 29 15.30 & -56 20 40.6 & 0.598 & -0.576 & -0.158 & 17.04 & 0.019 & 17.8 & 15.6 & 14.5 & 13.23 & 12.73 & 12.48 & sd + pm j15322 - 3622 & 15 32 13.92 & -36 22 30.7 & 0.462 & -0.372 & -0.274 & 15.85 & 0.020 & & 13.0 & & 10.10 & 9.54 & 9.28 & d + pm j15334 - 3634 & 15 33 27.72 & -36 34 02.4 & 0.552 & -0.453 & -0.316 & 15.85 & 0.020 & 16.4 & 14.2 & 12.5 & 11.54 & 10.99 & 10.76 & d + pm j15345 - 6452 & 15 34 31.45 & -64 52 54.8 & 0.455 & -0.139 & -0.434 & 17.06 & 0.019 & 15.9 & 14.1 & 12.4 & 11.19 & 10.65 & 10.39 & d + pm j15413 - 3609 & 15 41 19.05 & -36 09 10.8 & 0.529 & -0.520 & -0.094 & 15.85 & 0.020 & 17.6 & 16.1 & 13.5 & 11.97 & 11.45 & 11.11 & d + pm j15547 - 3856 & 15 54 46.12 & -38 56 17.4 & 0.454 & -0.203 & -0.406 & 17.82 & 0.018 & 16.2 & 13.6 & 11.8 & 11.19 & 10.69 & 10.46 & d + pm j16006 - 3234 & 16 00 39.59 & -32 34 11.4 & 0.534 & -0.443 & -0.298 & 14.22 & 0.023 & 20.2 & 17.3 & 15.9 & 14.43 & 14.04 & 13.69 & sd + pm j16019 - 3421 & 16 01 55.61 & -34 21 56.4 & 0.672 & 0.586 & -0.328 & 14.22 & 0.023 & & 15.0 & 12.5 & 10.96 & 10.33 & 9.98 & d + pm j16087 - 4442 & 16 08 43.95 & -44 42 28.3 & 0.596 & -0.142 & -0.579 & 17.89 & 0.018 & 20.2 & 14.3 & & 10.88 & 10.35 & 10.10 & d + pm j16138 - 3040 & 16 13 53.61 & -30 40 57.9 & 0.457 & -0.304 & -0.341 & 14.72 & 0.022 & 16.5 & 15.1 & 13.5 & 13.15 & 12.58 & 12.38 & sd + pm j16141 - 4044 & 16 14 07.03 & -40 44 15.0 & 0.523 & -0.433 & -0.292 & 17.82 & 0.018 & 16.8 & 14.5 & 13.1 & 12.33 & 11.87 & 11.61 & sd + pm j16169 - 4352 & 16 16 58.36 & -43 52 17.1 & 0.496 & -0.486 & -0.099 & 16.84 & 0.019 & 17.5 & 15.6 & 13.9 & 12.28 & 11.78 & 11.48 & d + pm j16256 - 6712 & 16 25 38.95 & -67 12 21.4 & 0.698 & -0.561 & -0.415 & 14.08 & 0.023 & 18.1 & 16.8 & 14.3 & 12.45 & 11.95 & 11.60 & d + pm j16364 - 8737 & 16 36 26.10 & -87 37 06.2 & 0.451 & -0.309 & -0.328 & 18.67 & 0.017 & 17.8 & 17.2 & 17.4 & 16.59 & 16.26 & 16.75 & wd + pm j16492 - 5142 & 16 49 13.81 & -51 42 52.2 & 0.452 & -0.262 & -0.369 & 16.31 & 0.020 & 15.4 & 14.4 & 11.9 & 12.12 & 11.63 & 11.44 & sd + pm j17037 - 6823 & 17 03 46.81 & -68 23 29.3 & 0.503 & -0.347 & -0.363 & 15.02 & 0.021 & 16.8 & 14.6 & 13.1 & 13.13 & 12.65 & 12.47 & sd + pm j17056 - 8608 & 17 05 38.35 & -86 08 44.1 & 0.512 & -0.305 & -0.411 & 18.67 & 0.017 & 19.2 & 16.9 & 13.6 & 12.82 & 12.28 & 11.96 & d + pm j17397 - 6322 & 17 39 46.10 & -63 22 07.1 & 0.547 & -0.098 & -0.538 & 17.07 & 0.019 & & 13.3 & 11.8 & 11.08 & 10.50 & 10.25 & d + pm j18004 - 6358 & 18 00 29.82 & -63 58 21.4 & 0.472 & -0.183 & -0.435 & 17.07 & 0.019 & 18.4 & 16.1 & 14.8 & 14.40 & 13.94 & 13.72 & sd + pm j18036 - 6340 & 18 03 37.17 & -63 40 47.9 & 0.484 & -0.256 & -0.411 & 17.27 & 0.019 & 17.2 & 15.5 & 14.3 & 13.67 & 13.13 & 12.92 & sd + pm j18217 - 6101 & 18 21 46.25 & -61 01 54.5 & 0.559 & 0.074 & -0.554 & 19.99 & 0.016 & 20.3 & 18.1 & 14.6 & 12.82 & 12.29 & 11.96 & d + pm j18286 - 4531 & 18 28 38.23 & -45 31 08.1 & 0.501 & -0.331 & -0.376 & 15.92 & 0.020 & 19.0 & 14.7 & 15.3 & 12.24 & 11.72 & 11.43 & sd + pm j18298 - 5131 & 18 29 51.30 & -51 31 31.7 & 0.577 & -0.241 & -0.524 & 15.04 & 0.021 & 19.5 & 18.0 & & 15.63 & 14.95 & 14.79 & sd + pm j19105 - 4132 & 19 10 33.60 & -41 32 50.6 & 0.676 & 0.125 & -0.664 & 14.95 & 0.021 & & & & 11.14 & 10.55 & 10.24 & d + pm j19105 - 4133 & 19 10 34.60 & -41 33 44.4 & 0.676 & 0.125 & -0.664 & 14.95 & 0.021 & & 13.0 & 11.0 & 9.85 & 9.24 & 9.03 & d + pm j19103 - 4338 & 19 10 23.58 & -43 38 37.2 & 0.450 & -0.004 & -0.450 & 14.74 & 0.022 & & 15.7 & 13.7 & 11.86 & 11.28 & 10.99 & d + pm j19110 - 3820 & 19 11 00.24 & -38 20 31.9 & 0.585 & 0.547 & -0.205 & 14.95 & 0.021 & & 16.9 & & 15.99 & 15.79 & 15.62 & wd + pm j19167 - 3638 & 19 16 46.58 & -36 38 04.1 & 1.340 & -0.162 & -1.329 & 15.04 & 0.021 & 18.1 & 15.8 & 14.9 & 13.66 & 13.12 & 12.95 & sd + pm j19184 - 4554 & 19 18 29.44 & -45 54 30.8 & 0.616 & -0.414 & -0.456 & 14.74 & 0.022 & & 15.1 & 13.1 & 11.21 & 10.65 & 10.30 & d + pm j19248 - 3356 & 19 24 48.28 & -33 56 09.7 & 0.574 & 0.319 & -0.477 & 15.04 & 0.021 & 16.0 & 13.7 & 12.8 & 12.45 & 11.99 & 11.77 & sd + pm j19261 - 4310 & 19 26 08.60 & -43 10 56.4 & 1.142 & -0.142 & -1.133 & 16.96 & 0.019 & & 15.8 & 13.7 & 11.94 & 11.42 & 11.12 & d + pm j19285 - 3634 & 19 28 33.61 & -36 34 30.3 & 0.503 & 0.024 & -0.503 & 14.72 & 0.022 & & 14.2 & 12.4 & 10.61 & 10.06 & 9.81 & d + pm j19280 - 6028 & 19 28 00.48 & -60 28 52.4 & 0.460 & 0.068 & -0.455 & 13.77 & 0.023 & & 16.6 & 14.5 & 13.97 & 13.41 & 13.23 & sd + pm j19403 - 3944 & 19 40 21.30 & -39 44 10.0 & 0.529 & 0.156 & -0.505 & 14.20 & 0.023 & & 13.8 & 17.4 & 10.38 & 9.84 & 9.57 & d + pm j20444 - 4123 & 20 44 27.89 & -41 23 51.6 & 0.505 & 0.357 & -0.356 & 15.25 & 0.021 & & 14.3 & 12.4 & 11.75 & 11.16 & 10.99 & d + pm j20460 - 5458 & 20 46 01.48 & -54 58 06.7 & 0.461 & 0.134 & -0.441 & 15.04 & 0.021 & & 18.5 & 16.7 & 15.29 & 14.80 & 14.41 & sd + pm j20530 - 5409 & 20 53 03.90 & -54 09 37.5 & 0.633 & 0.177 & -0.608 & 15.04 & 0.021 & 18.7 & 16.6 & 13.8 & 12.17 & 11.66 & 11.35 & d + pm j21324 - 3922 & 21 32 29.67 & -39 22 50.1 & 0.513 & 0.447 & -0.252 & 16.08 & 0.020 & & 15.8 & 13.4 & 12.21 & 11.70 & 11.35 & d + pm j22040 - 3347 & 22 04 02.27 & -33 47 38.3 & 0.948 & 0.472 & -0.822 & 14.14 & 0.023 & & 14.5 & & 12.32 & 11.81 & 11.60 & sd + pm j22178 - 7753 & 22 17 53.89 & -77 53 36.0 & 0.584 & 0.315 & -0.492 & 15.92 & 0.020 & 18.9 & 17.0 & 15.5 & 14.53 & 14.07 & 13.91 & sd + pm j22209 - 3346 & 22 20 57.91 & -33 46 57.9 & 0.559 & -0.102 & -0.550 & 15.12 & 0.021 & 19.8 & 17.8 & 18.1 & & & & + pm j22359 - 7722 & 22 35 57.74 & -77 22 16.8 & 0.601 & -0.224 & -0.558 & 15.92 & 0.020 & 18.4 & 16.3 & 15.3 & 14.17 & 13.67 & 13.46 & sd + pm j22387 - 6232 & 22 38 47.28 & -62 32 22.7 & 0.504 & 0.305 & -0.401 & 18.11 & 0.018 & 20.3 & 17.4 & 18.2 & & & & + pm j22403 - 4931e & 22 40 18.96 & -49 31 01.4 & 0.504 & 0.457 & 0.212 & 14.87 & 0.022 & 15.1 & 13.3 & 10.7 & 9.93 & 9.38 & 9.03 & d + pm j22403 - 4931w & 22 40 18.67 & -49 31 04.6 & 0.504 & 0.457 & 0.212 & 14.87 & 0.022 & & & & 9.84 & 9.26 & 9.01 & d + pm j23350 - 4904 & 23 35 04.31 & -49 04 54.3 & 0.611 & 0.334 & -0.511 & 17.33 & 0.018 & & 15.5 & 13.9 & 13.29 & 12.75 & 12.52 sd

in particle physics , the neutrino plays a remarkable role of a `` tiny '' particle . the scale of neutrino mass @xmath0 is much lower than that of the charged fermions ( @xmath1 , @xmath2 ) . interaction of neutrinos with matter is extremely weak as compared to that in the case of other known elementary fermions . according to the standard model ( sm ) , it can be mediated only via exchange of the @xmath3 and @xmath4 bosons . however , the recent development of our knowledge of neutrino mixing and oscillations , supported by the discovery of flavor conversions of neutrinos from different sources ( see @xcite ) , substantiates the assumption that neutrinos can possess electromagnetic properties and , hence , take part in electromagentic interactions ( see , for instance , the review articles @xcite ) . these properties include , in particular , the electric charge , the charge radius , the anapole moment , and the dipole electric and magnetic moments . such nontypical neutrino features are of particular interest , because they open a door to `` new physics '' beyond the sm ( bsm ) . in spite of appreciable efforts in searches for neutrino electromagnetic characteristics , up to now there is no experimental evidence favoring their nonvanishing values . among the electromagnetic properties of neutrinos , the most studied and well understood theoretically are neutrino magnetic moments ( nmm ) , along with electric dipole moments . for the most recent and complete review on theoretical and experimental aspects of nmm , as well as for the corresponding references , see @xcite . the effective lagrangian , which describes the coupling of nmm to the electromagnetic field @xmath5 , can be written in the form @xmath6 where the magnetic moments @xmath7 , in the presence of mixing between different neutrino states , are associated with the neutrino mass eigenstates @xmath8 . an interplay between the magnetic moment and neutrino mixing effects is important . note that the electric ( transition ) moments @xmath9 do also contribute to the coupling . a dirac neutrino may have non - zero diagonal electric moments in models where cp invariance is violated . for a majorana neutrino the diagonal magnetic and electric moments are zero . therefore , nmm can be used to distinguish dirac and majorana neutrinos @xcite . in the standard model the magnetic moment of a massless neutrino is zero . in the minimal extension of the sm , the explicit evaluation of the one - loop contributions to the dirac nmm in the leading approximation over small parameters @xmath10 ( @xmath11 are the neutrino masses , @xmath12 ) , that however exactly accounts for the parameters @xmath13 ( @xmath14 ) , yields the following result @xcite : @xmath15 where @xmath16 is the neutrino mixing matrix , and @xmath17.\ ] ] a majorana neutrino can also have nondiagonal ( transition ) magnetic moment @xmath18 ( @xmath19 ) . the obtained value for nmm is proportional to the neutrino mass and is , in general , of the order @xmath20 . much larger values for nmm can be obtained in different other sm extensions ( see @xcite for the detailed discussion ) . however , there is a problem @xcite for any bsm theory how to get a large nmm value and simultaneously to avoid an unacceptable large contribution to the neutrino mass . recently this problem has been reconsidered for a class of bsm theories and it has been shown in a model - independent way that in principle it is possible to avoid the above mentioned contradiction in the case of dirac @xcite and majorana @xcite neutrinos . it has been shown that in this kind of theoretical models the nmm can naturally reach values of @xmath21 . these values are at least two orders of magnitude smaller than the present laboratory experimental limits ( see bellow ) . there is also a huge gap of many orders of magnitude between these values and the prediction of the minimal extension of the sm . therefore , if any direct experimental confirmation of non - zero nmm is obtained in the laboratory experiments , it will open a window to `` new physics '' . the neutrino magnetic moments are being searched in reactor @xcite , accelerator @xcite and solar @xcite experiments on low - energy elastic neutrino - electron scattering ( for more details see the review articles @xcite and references therein ) . the current best upper limit on the nmm value obtained in such direct laboratory measurements is @xmath22 where @xmath23 is a bohr magneton . this bound , which is due to the gemma experiment @xcite with a hpge detector at kalinin nuclear power station , is by an order of magnitude larger than the tightest constraint obtained in astrophysics @xcite : @xmath24 and it by many orders of magnitude exceeds the value derived in the minimally extended sm that includes right - handed neutrinos @xcite : @xmath25 where @xmath0 is a neutrino mass . at the same time , there are different theoretical bsm scenarios that predict much higher @xmath26 values . for example , the effective nmm value in a class of extra - dimension models can be as large as about @xmath27 @xcite . future higher precision reactor experiments can therefore be used to provide new constraints on large extra - dimensions . the paper is organized as follows . [ nmm_searches ] outlines the current status of searches for nmm and the problem of atomic - ionization effects in reactor experiments . [ gen_theory ] is devoted to the theoretical background for neutrino scattering on atomic electrons . in sec . [ 1e ] , we discuss the case of neutrino scattering on one bound electron . hydrogen - like states and a semiclassical limit are considered . [ mea ] focuses on ionization of manyelectron atoms by neutrino impact . the case of a helium atomic target and the thomas - fermi and _ ab initio _ approaches are discussed . finally , sec . [ summary ] summarizes this review . the strategy of experiments searching for nmm is as follows . one studies an inclusive cross section for elastic ( anti)neutrino - electron scattering which is differential in the energy transfer @xmath28 . in the ultrarelativistic limit @xmath29 , it is given by an incoherent sum of the sm contribution @xmath30 , which is due to weak interaction that conserves the neutrino helicity , and the helicity - flipping contribution @xmath31 , which is due to @xmath26 , @xmath32 the sm term is well - documented and is given by @xcite @xmath33,\end{aligned}\ ] ] where @xmath34 is the incident antineutrino energy , @xmath35 and @xmath36 for @xmath37 and @xmath38 and @xmath39 for @xmath40 and @xmath41 , with @xmath42 being the weinberg angle . for antineutrinos one must substitute @xmath43 . the possibility for neutrino - electron elastic scattering due to nmm was first considered in ref . @xcite , and the cross section of this process was calculated in ref . @xcite ( the related brief historical notes can be found in ref . @xcite ) . here we would like to recall the paper by domogatsky and nadezhin @xcite , where the cross section of ref . @xcite was corrected and the antineutrino - electron cross section was considered in the context of the earlier experiments with reactor antineutrinos of cowan and reines @xcite and cowan et al . @xcite , which were aimed to reveal the nmm effects . discussions on the derivation of the cross section and on the optimal conditions for bounding the nmm value , as well as a collection of the cross section formulas for elastic scattering of neutrinos ( antineutrinos ) on electrons , nucleons , and nuclei , can be found in refs . @xcite and @xcite . the result relevant to the @xmath26 component in eq . ( [ cr_sec ] ) reads @xcite @xmath44 where @xmath45 is the fine - structure constant . thus , the two components of the cross section ( [ cr_sec ] ) exhibit qualitatively different dependencies on the recoil - electron kinetic energy @xmath28 . namely , at low @xmath28 values the sm cross section is practically constant in @xmath28 , while that due to @xmath26 behaves as @xmath46 . this means that the experimental sensitivity to nmm value critically depends on lowering the energy threshold of the detector employed for measurement of the recoil - electron spectrum . the current reactor experiments with germanium detectors @xcite have reached threshold values of @xmath28 as low as few kev and are to further improve the sensitivity to low energy deposition in the detector @xcite . at low energies , however , one can expect a modification of the free - electron formulas due to the binding of electrons in the germanium atoms , where e.g. the energy of the @xmath47 line , 9.89kev , indicates that at least some of the atomic binding energies are comparable to the already relevant to the experiment values of @xmath28 . thus a proper treatment of the atomic effects in neutrino scattering is necessary and important for the analysis of the current and , even more , of the future data with a still lower threshold . furthermore , there is no known means of independently calibrating experimentally the response of atomic systems , such as the germanium , to the scattering due to the interactions relevant for the neutrino experiments . therefore one has to rely on a pure theoretical analysis in interpreting the neutrino data . for the first time this problem was addressed in ref . @xcite , where a 2 - 3 times enhancement of the electroweak cross section in the case of ionization from a 1@xmath48 state of a hydrogen - like atom with nuclear charge @xmath49 had been numerically determined at neutrino energies @xmath50 . subsequent numerical calculations within the relativistic hartree - fock method for ionization from inner shells of various atoms showed much lower enhancement ( @xmath51 ) of the electroweak contribution @xcite . it was found that in the scattering on realistic atoms , such as germanium , the so - called stepping approximation works with a very good accuracy . the stepping approach , introduced in ref . @xcite from an interpretation of numerical data , treats the process as scattering on individual independent electrons occupying atomic orbitals and suggests that the cross section follows the free - electron behavior in eqs . ( [ cr_sec_sm ] ) and ( [ cr_sec_mu ] ) down to @xmath28 equal to the ionization threshold for the orbital , and that below that energy the electron on the corresponding orbital is ` deactivated ' thus producing a sharp ` step ' in the dependence of the cross section on @xmath28 . the interest to the role of atomic effects was renewed in several more recent papers . the early claim @xcite of a significant enhancement of the nmm contribution in the case of germanium due to the atomic effects has been later disproved @xcite and it was argued @xcite that the modification of the free - electron formulas ( [ cr_sec_sm ] ) and ( [ cr_sec_mu ] ) by the atomic - binding effects is insignificant down to very low values of @xmath28 . this conclusion appeared to be also in contradiction with the results of ref . @xcite , where it was deduced by means of numerical calculations that the @xmath26 contribution to ionization of the helium atomic target by impact of electron antineutrinos strongly enhances relative to the free - electron approximation . however , from calculations performed in ref . @xcite it follows that the stepping approximation is well applicable practically down to the ionization threshold for helium . as indicated in the introduction , the most sensitive and widely used method for the experimental investigation of the neutrino electromagnetic properties is provided by direct laboratory measurements of low - energy elastic scattering of neutrinos and antineutrinos with electrons in reactor , accelerator and solar experiments . in this section we deliver a theoretical background for such studies . let us consider the elastic - scattering process @xmath52 where an incident neutrino with energy @xmath34 transfers to a free electron , which is initially at rest in the laboratory frame , the energy - momentum @xmath53 . there are two recoil - electron observables : the kinetic energy @xmath28 , which amounts to the energy transfer , and the outgoing angle @xmath54 measured with respect to the incident neutrino direction . in the ultrarelativistic limit @xmath55 , these kinematical variables are related by @xmath56 the maximal value of the kinetic electron energy is thus realized when @xmath57 and is given by @xmath58 within the sm , the scattering process ( [ nu - e_process ] ) takes place due to exchange of the weak bosons , as shown in fig . [ fig : e - nu_sm ] . elastic neutrino - electron scattering due to the weak interaction . exchange by the @xmath59 ( left ) and @xmath4 ( right ) bosons is shown.,title="fig:",scaledwidth=45.0% ] elastic neutrino - electron scattering due to the weak interaction . exchange by the @xmath59 ( left ) and @xmath4 ( right ) bosons is shown.,title="fig:",scaledwidth=45.0% ] the @xmath59-boson channel corresponds to the charged current interaction and is absent in the cases of the muon and tau neutrinos . if @xmath60 , where @xmath61 is the @xmath59-boson mass , the scattering amplitude is given by @xcite @xmath62 where @xmath63 ( @xmath64 ) and @xmath65 ( @xmath66 ) are initial and final neutrino ( electron ) spinors . the @xmath4 boson mediates the neutral current interaction . the corresponding scattering amplitude in the case @xmath67 , where @xmath68 is the @xmath4-boson mass , reads @xcite @xmath69 using the matrix elements ( [ m_w ] ) and ( [ m_z ] ) , one arrives , after averaging over the initial and summing over the final electron spins , at the sm single - differential cross section ( [ cr_sec_sm ] ) . fig . [ fig : e - nu_em ] shows the electromagnetic channel of the scattering process ( [ nu - e_process ] ) . contribution of the neutrino electromagnetic vertex function to neutrino elastic scattering on a charged lepton @xcite.,scaledwidth=45.0% ] in general the matrix element of the neutrino electromagnetic current @xmath70 can be considered between different neutrino initial @xmath71 and final @xmath72 states of different masses , @xmath73 and @xmath74 : @xmath75 in the most general case consistent with lorentz and electromagnetic gauge invariance , the vertex function is defined as ( see refs . @xcite and references therein ) @xmath76(q^2\gamma_\mu-\gamma_\mu\!\not{\!q})+f_m(q^2)_{ij}i\sigma_{\mu\nu}q^\nu+f_e(q^2)_{ij}i\sigma_{\mu\nu}q^\nu\gamma_5,\ ] ] where @xmath77 , @xmath78 , @xmath79 , and @xmath80 are respectively the charge , anapole , dipole magnetic , and dipole electric neutrino form factors , which are matrices in the space of neutrino mass eigenstates @xcite . consider the diagonal case @xmath81 . the hermiticity of the electromagnetic current and the assumption of its invariance under discrete symmetries transformations put certain constraints on the form factors , which are in general different for the dirac and majorana neutrinos . in the case of dirac neutrinos , the assumption of cp invariance combined with the hermiticity of the electromagnetic current @xmath70 implies that the electric dipole form factor vanishes , @xmath82 . at zero momentum transfer only @xmath83 and @xmath84 , which are called the electric charge and the magnetic moment , respectively , contribute to the hamiltonian @xmath85 that describes the neutrino interaction with the external electromagnetic field @xmath86 . the hermiticity also implies that @xmath87 , @xmath88 , and @xmath89 are real . in contrast , in the case of majorana neutrinos , regardless of whether cp invariance is violated or not , the charge , dipole magnetic and electric moments vanish , @xmath90 , so that only the anapole moment can be non - vanishing among the electromagnetic moments . note that it is possible to prove @xcite that the existence of a non - vanishing magnetic moment for a majorana neutrino would bring about a clear evidence for cpt violation . in the off - diagonal case @xmath19 , the hermiticity by itself does not imply restrictions on the form factors of dirac neutrinos . it is possible to show @xcite that , if the assumption of the cp invariance is added , the form factors @xmath87 , @xmath89 , @xmath91 , and @xmath88 should have the same complex phase . for the majorana neutrino , if cp invariance holds , there could be either a transition magnetic or a transition electric moment . finally , as in the diagonal case , the anapole form factor of a majorana neutrino can be nonzero . the neutrino dipole magnetic and electric form factors ( and the corresponding magnetic and electric dipole moments ) are theoretically the most well understood among the form factors . the value of the magnetic form factor @xmath79 at @xmath92 defines the nmm , @xmath93 . in the low - energy limit , the nmm contribution to the effective electromagnetic vertex can be expressed in the following form : @xmath94 thus , the corresponding scattering amplitude is @xmath95 this leads to the nmm single - differential cross section given by ( [ cr_sec_mu ] ) . consider the process where a neutrino with energy - momentum @xmath96 scatters on an atom at energy - momentum transfer @xmath97 . in what follows the recoil of the atomic nucleus is neglected because of the typical for current experiments situation @xmath98 , where @xmath99 is the nuclear mass . the atomic target is supposed to be unpolarized and in its ground state @xmath100 with the corresponding energy @xmath101 . it is also supposed that @xmath102 and @xmath103 , where @xmath49 is the nuclear charge and @xmath45 is the fine - structure constant , so that the initial and final electronic systems can be treated nonrelativistically . the neutrino states are described by the dirac spinors assuming @xmath104 . thus , the magnetic moment interaction of the neutrino field @xmath105 with the atomic electrons is described by the lagrangian @xmath106 where @xmath107 is the final neutrino four - momentum . the electromagnetic field @xmath108 of the atomic electrons is . ] @xmath109 , @xmath110 , where @xmath111 and @xmath112 are the fourier transforms of the electron number density and current density operators , respectively , @xmath113 @xmath114 , \label{ce}\ ] ] and the sums run over the positions @xmath115 of all the @xmath49 electrons in the atom . the double - differential cross section can be presented as @xmath116 where @xmath117 and @xmath118 where @xmath119 , also known as the dynamical structure factor @xcite , and @xmath120 are @xmath121 @xmath122 with @xmath123 being the @xmath124 component perpendicular to @xmath125 and parallel to the scattering plane , which is formed by the incident and final neutrino momenta . the sums in eqs . ( [ dsf ] ) and ( [ dsfc ] ) run over all the atomic states @xmath126 with energies @xmath127 of the electron system , with @xmath128 being the initial state . the longitudinal term ( [ dcs_par ] ) is associated with atomic excitations induced by the force that the neutrino magnetic moment exerts on electrons in the direction parallel to @xmath125 . the transverse term ( [ dcs_perp ] ) corresponds to the exchange of a virtual photon which is polarized as a real one , that is , perpendicular to @xmath125 . it resembles a photoabsorption process when @xmath129 and the virtual - photon four - momentum thus approaches a real - photon value . due to selections rules , the longitudinal and transverse excitations do not interfere ( see @xcite for detail ) . the factors @xmath119 and @xmath120 are related to respectively the density - density ( or polarization ) and current - current green s functions @xmath130 @xmath131 where @xmath132 @xmath133@xmath134 being the hamiltonian for the system of electrons . from these relations it follows that , due to the parity selection rule , the functions @xmath119 and @xmath120 are even with respect to @xmath53 . for small @xmath135 values , in particular , such that @xmath136 , only the lowest - order non - zero terms of the expansion of eqs . ( [ sfrel ] ) and ( [ sfrelc ] ) in powers of @xmath137 are of relevance ( the so - called dipole approximation ) . in this case , one has @xcite @xmath138 taking into account eq . ( [ dipole_approx ] ) , the experimentally measured singe - differential inclusive cross section is , to a good approximation , given by ( see e.g. in @xcite ) @xmath139 the standard electroweak contribution to the cross section can be similarly expressed in terms of the same factor @xmath119 @xcite as @xmath140 where the factor @xmath119 is integrated over @xmath137 with a unit weight , rather than @xmath141 as in eq . ( [ d1s ] ) . the kinematical limits for @xmath137 in an actual neutrino scattering are explicitly indicated in eqs . ( [ d1s ] ) and ( [ d1sw ] ) . at large @xmath142 , typical for the reactor neutrinos , the upper limit can in fact be extended to infinity , since in the discussed here nonrelativistic limit the range of momenta @xmath143 is indistinguishable from infinity on an atomic scale . the lower limit can be shifted to @xmath144 , since the contribution of the region of @xmath145 can be expressed in terms of the photoelectric cross section @xcite and is negligibly small ( at the level of below one percent in the considered range of @xmath28 ) . for this reason one can discuss the momentum - transfer integrals in eqs . ( [ d1s ] ) and ( [ d1sw ] ) running from @xmath144 to @xmath146 : @xmath147 and @xmath148 for a free electron , which is initially at rest , the density - density correlator is the free particle green s function @xmath149 so that the dynamical structure factor is given by @xmath150 and the discussed here integrals are in the free - electron limit as follows : @xmath151 @xmath152 clearly , these expressions , when used in the formulas ( [ d1s ] ) and ( [ d1sw ] ) , result in the free - electron cross sections for the case @xmath153 , @xmath154 and @xmath155 , \label{sew}\end{aligned}\ ] ] correspondingly . in this section we consider neutrino scattering on an electron bound in an atom following consideration of @xcite . the binding effects generally deform the density - density green s function , so that both the integrals ( [ defi_1 ] ) and ( [ defi_2 ] ) are somewhat modified . namely , the binding effects spread the free - electron @xmath156-peak in the dynamical structure function ( [ struc_factor_fe ] ) at @xmath157 and also shift it by the scale of characteristic electron momenta in the bound state . consider the situation when the initial electron occupies the discrete @xmath158 orbital in a coulomb potential @xmath159 . the dynamical structure factor for this hydrogen - like system is given by @xmath160 where @xmath161 is the bound - state wave function , @xmath162 is the outgoing coulomb wave for the ejected electron with momentum @xmath163 , and @xmath164 , with @xmath165 being the electron momentum in the @xmath166th bohr orbit . the closed - form expressions for the bound - free transition matrix elements in eq . ( [ sfnl ] ) can be found , for instance , in ref . @xcite . in principle , they allow for performing angular integrations in eq . ( [ sfnl ] ) analytically . this task , however , turns out to be formidable for large values of @xmath166 . therefore , below we restrict our consideration to the @xmath167 states only , which nevertheless is enough for demonstrating the validity of the semiclassical approach developed in sec . [ 1e ] . using results of ref . @xcite , we can present the function ( [ sfnl ] ) when @xmath167 as @xmath168}\frac{q^2f_{nl}(q^2)}{[(q^2-k^2+p_n^2)^2 + 4p_n^2k^2]^{2n+1}}\nonumber\\ & { } & \times \exp\left[-2\eta\arctan\left(\frac{2p_nk}{q^2-k^2+p_n^2}\right)\right ] , \end{aligned}\ ] ] where the branch of the arctangent function should be used that lies between 0 and @xmath169 , @xmath170 is the sommerfeld parameter , and @xmath171 @xmath172,\nonumber\\\end{aligned}\ ] ] @xmath173.\nonumber\\\end{aligned}\ ] ] fig . [ fig1 ] shows the magnetic single - differential cross section ( [ d1s ] ) for ionization from the @xmath174 orbital , which is normalized to the free - electron value ( [ cr_sec_mu ] ) , as a function of @xmath175 , with the electron binding energy given by @xmath176 . as can be seen , the numerical results for @xmath177 are close to the free - electron ones in magnitude . this can be qualitatively explained by noticing the following facts . first , in an attractive coulomb potential there is an infinite set of bound states , with the discrete spectrum smoothly transforming into the continuum at the ionization threshold . second , the average value of the @xmath174-electron momentum is @xmath178 and the average change in the electron momentum after ejection , @xmath179 , is such that @xmath180 , which is analogous to the free - electron case . the ratio of single - differential cross sections for magnetic neutrino scattering from the @xmath174 hydrogen - like and free - electron states , respectively , versus @xmath175 at different values of @xmath34 . the @xmath181 and @xmath182 curves are practically indistinguishable.,scaledwidth=90.0% ] thus , taking into account the results in fig . [ fig1 ] , one might expect the atomic - binding effects to play a subsidiary role when @xmath177 . the authors of @xcite , however , came to the contrary conclusion that the single - differential cross section dramatically enhances due to atomic ionization when @xmath183 . the enhancement mechanism proposed in @xcite is based on an analogy with the photoionization process . as mentioned above , when @xmath184 the virtual - photon momentum approaches the physical regime @xmath185 . in this limit , we have for the transverse component of the double - differential cross section ( [ dcs_perp ] ) @xmath186 where @xmath187 is the photoionization cross section at the photon energy @xmath28 @xcite . the limiting form ( [ photoeffect ] ) was used in @xcite in the whole integration interval , when deriving the single - differential cross section . such a procedure is obviously incorrect , for the integrand rapidly falls down as @xmath137 ranges from @xmath188 up to @xmath189 , especially when @xmath190 , where @xmath191 is a characteristic atomic size ( within the thomas - fermi model @xmath192 @xcite ) . this fact reflects a strong departure from the real - photon regime . for this reason we can classify the enhancement of the differential cross section determined in @xcite as spurious . insertion of eq . ( [ sfnl1 ] ) into the integrals ( [ defi_1 ] ) and ( [ defi_2 ] ) and integration over @xmath193 , using the change of variable @xmath194 and the standard integrals involving the products of the exponential function and the powers of sine and cosine functions , yields @xcite @xmath195\nonumber\\&{}&\times\left . \left(1-\frac{4}{y_1}+\frac{16}{3y_1 ^ 2}\right)\right \},%\nonumber\\\end{aligned}\ ] ] @xmath196\nonumber\\&{}&\left.\times \left(1-\frac{8}{y_2}+\frac{80}{3y_2 ^ 2}-\frac{448}{15y_2 ^ 3}+\frac{1792}{15y_2 ^ 4}\right)\right \},%\nonumber\\\end{aligned}\ ] ] @xmath197\nonumber\\&{}&\left.\times \left(1-\frac{8}{y_2}+\frac{80}{3y_2 ^ 2}-\frac{448}{15y_2 ^ 3}+\frac{1024}{15y_2 ^ 4}\right)\right \},%\nonumber\\\end{aligned}\ ] ] @xmath198\nonumber\\&{}&\left.\times \left(1-\frac{8}{y_2}+\frac{80}{3y_2 ^ 2}-\frac{704}{15y_2 ^ 3}+\frac{3328}{45y_2 ^ 4}\right)\right \},%\nonumber\\\end{aligned}\ ] ] @xmath199\nonumber\\&{}&\left.\times \left(1-\frac{8}{y_2}+\frac{80}{3y_2 ^ 2}-\frac{704}{15y_2 ^ 3}+\frac{512}{15y_2 ^ 4}\right)\right \ } , % \nonumber\\\end{aligned}\ ] ] where @xmath200 . the largest deviations of these integrals from the free - electron analogs ( [ intf_1 ] ) and ( [ intf_2 ] ) occur at the ionization threshold @xmath201 . the corresponding relative values in this specific case are @xcite @xmath202 @xmath203 @xmath204 the above results indicate a clear tendency : the larger @xmath166 and @xmath205 , the closer @xmath206 and @xmath207 are to the free - electron values . the departure from the free - electron behavior does not exceed several percent at most . these observations provide a solid base for the semiclassical approach described below . in the one - electron approximation , the hamiltonian has the form @xmath208 , and the density - density green s function from eq.([fdef ] ) can be written as @xmath209^{-1 } \ , e^{i { \bf q } \cdot { \bf r}}\ , \right | 0 \right \rangle \nonumber \\ & = & \left \langle 0 \left | \ , \left [ t -h({\bf p } + { \bf q } , { \bf r } ) + e_0 \right ] ^{-1 } \ , \right | 0 \right \rangle \nonumber\\ & = & \left \langle 0 \left | \ , \left [ t -{{q}^{2 } \over 2 m_e}- { { \bf p } \cdot { \bf q } \over m_e } - h({\bf p } , { \bf r } ) + e_0 \right ] ^{-1 } \ , \right | 0 \right \rangle \nonumber \end{aligned}\ ] ] where the infinitesimal shift @xmath210 is implied . clearly , a nontrivial behavior of the latter expression in eq.([f1 ] ) is generated by the presence of the operator @xmath211 in the denominator , and the fact that it does not commute with the hamiltonian @xmath134 . thus an analytical calculation of the green s function as well as the dynamical structure factor is feasible in only few specific problems . in sec . [ h ] such a calculation has been presented for ionization from the @xmath174 , @xmath212 , and @xmath213 hydrogen - like states . in particular , we have seen that the deviation of the discussed integrals ( [ defi_1 ] ) and ( [ defi_2 ] ) from their free values are very small : the largest deviation is exactly at the ionization threshold , where , for instance , each of the @xmath174 integrals is equal to the free - electron value multiplied by the factor @xmath214 ( see eq . ( [ mw1s_num ] ) ) . it can be also noted from ( [ mw1s ] ) that both integrals are modified in exactly the same proportion , so that their ratio is not affected at any @xmath28 : @xmath215 . we find however , that this exact proportionality is specific for the ionization from the ground state in the coulomb potential . the problem of calculating the integrals ( [ defi_1 ] ) and ( [ defi_2 ] ) , however , can be solved in the semiclassical limit , where one can neglect the noncommutativity of the momentum @xmath216 with the hamiltonian , and rather treat this operator as a number vector . taking also into account that @xmath217 , one can then readily average the latter expression in eq.([f1 ] ) over the directions of @xmath125 and find the formula for the dynamical structure factor : @xmath218 , \label{scls}\ ] ] where @xmath219 is the standard heaviside step function . the expression in eq . ( [ scls ] ) is nonzero only in the range of @xmath220 satisfying the condition @xmath221 , i.e. between the ( positive ) roots of the binomials in the arguments of the step functions : @xmath222 and @xmath223 . one can notice that the previously mentioned ` spread and shift ' of the peak in the dynamical structure function in this limit corresponds to a flat pedestal between @xmath224 and @xmath225 . the calculation of the integrals ( [ defi_1 ] ) and ( [ defi_2 ] ) with the expression ( [ scls ] ) is straightforward , and yields the free - electron expressions ( [ intf_1 ] ) and ( [ intf_2 ] ) for the discussed here integrals in the semiclassical ( wkb ) limit : @xmath226 the appearance of the free - electron expressions here is not surprising , since the equation ( [ scls ] ) can be also viewed as the one for scattering on an electron boosted to the momentum @xmath227 . the difference from the pure free - electron case however is in the range of the energy transfer @xmath28 . namely , the expressions ( [ wkbi ] ) are applicable in this case only above the ionization threshold , i.e. at @xmath228 . below the threshold the electron becomes ` inactive ' . we believe that the latter conclusion explains the so - called stepping behavior observed empirically @xcite in the results of numerical calculations . namely the calculated cross section @xmath229 for ionization of an electron from an atomic orbital follows the free - electron dependence on @xmath28 all the way down to the threshold for the corresponding orbital with a very small , at most a few percent , deviation . this observation led the authors of ref . @xcite to suggest the stepping approximation for the ratio of the atomic cross section ( per target electron ) to the free - electron one : @xmath230 where the sum runs over the atomic orbitals with the binding energies @xmath231 and the filling numbers @xmath232 . clearly , the factor @xmath233 simply counts the fraction of ` active ' atomic electrons at the energy @xmath28 , i.e. those for which the ionization is kinematically possible . for this reason we refer to @xmath233 as an _ atomic factor_. we conclude here that the stepping approximation is indeed justified with a high accuracy in the approximation of the scattering on independent electrons , i.e. if one neglects the two - electron correlations induced by the interference of terms in the operator @xmath111 in eq . ( [ ne ] ) corresponding to different electrons . the effects of such an interference will be discussed in the next section . in considering the neutrino scattering on actual manyelectron atoms one needs to evaluate the dependence of the number of active electrons on @xmath28 and generally also evaluate the effect of the two - electron correlations . the latter can be studied , for example , in the case of a helium atom , where the electron - electron correlations are known to play a very significant role . recently , the authors of @xcite deduced by means of numerical calculations that the @xmath26 contribution to ionization of the he atomic target by impact of electron antineutrinos from reactor and tritium sources strongly departures from the stepping approximation , exhibiting large enhancement relative to the free - electron case . according to @xcite , the effect is maximal when the @xmath28 value approaches the ionization threshold in helium , @xmath234ev , where the relative enhancement is as large as almost eight orders of magnitude . it was thus suggested that this finding might have an impact on searches for @xmath26 , provided that its value falls within the range @xmath235 . in this section we show that ( i ) the result of @xcite is erroneous and ( ii ) the stepping approximation for helium is well applicable , except the energy region @xmath236 where the differential cross section substantially decreases relative to the free - electron case . we consider the process where an electron antineutrino with energy @xmath34 scatters on a he atom at energy and spatial - momentum transfers @xmath28 and @xmath125 , respectively . in what follows we focus on the ionization channel of this process in the kinematical regime @xmath153 , which mimics a typical situation with reactor ( @xmath237mev ) and tritium ( @xmath238kev ) antineutrinos when the case @xmath239 is concerned . the he target is assumed to be in its ground state @xmath240 with the corresponding energy @xmath231 . since for helium one has @xmath241 , where @xmath242 is the nuclear charge , the state @xmath240 can be treated nonrelativistically . as we are interested in the energy region @xmath236 , the final he state @xmath243 ( with one electron in continuum ) can also be treated in the nonrelativistic approximation . under the above assumptions , the dynamical structure factor ( [ dsf ] ) is given by @xmath244 here the @xmath245 sum runs over all final he states having one electron ejected in continuum , with @xmath246 being their energies . for evaluation of the dynamical structure factor ( [ dyn_str_fact ] ) we employ the same models of the initial and final he states as in @xcite . the initial state is given by a product of two @xmath174 hydrogenlike wave functions with an effective charge @xmath247 , @xmath248 where @xmath249 is the bohr radius . the final state has the form @xmath250,\ ] ] where @xmath251 is an outgoing coulomb wave for the ejected electron with spatial momentum @xmath163 . @xmath252 is the effective charge experienced by the ejected electron in the field of the final he@xmath253 ion . contributions to the dynamical structure factor from excited he@xmath253 states are neglected due to their very small overlap with the @xmath254-electron state in the he atom . to avoid nonphysical effects connected with nonorthogonality of states ( [ init_state ] ) and ( [ fin_state ] ) , we use the gram - schmidt orthogonalization @xmath255 substitution of ( [ init_state ] ) and ( [ fin_state ] ) into ( [ dyn_str_fact ] ) thus yields @xmath256 where @xmath257 , and @xmath258 is the inelastic form factor , with @xmath259 it is straightforward to perform the further calculation of the dynamical structure factor analytically ( see , for instance , the textbook @xcite ) . the resulting expression is @xmath260,\ ] ] where @xmath261 and ( introducing @xmath262 ) @xmath263 ^ 3 } \bigg\{(p_{i}+\eta k)^2(p_{i}^2+q^2+k^2)^2\nonumber \\&{}&+4kq^2\left[\frac{1}{3}kp_{i}^2-\frac{2}{3}\eta^2kp_{i}^2-\eta^2k^3-\eta p_{i}(p_{i}^2+q^2+k^2)\right ] \bigg\},\end{aligned}\ ] ] @xmath264^{i\eta}}{[(p_{eff}-ik)^2+q^2 ] } % \left\{\frac{(k - q+ip_{eff})^2}{[(k - q)^2+p_{eff}^2]^{1+i\eta}}-\frac{(k+q+ip_{eff})^2}{[(k+q)^2+p_{eff}^2]^{1+i\eta}}\right\}. a_2(k , q)&=&\frac{2\exp\left(\eta\,{\rm arccos}\frac{p_{i}^2+q^2-k^2}{\sqrt{(p_{i}^2+q^2+k^2)^2 - 4k^2q^2}}\right)}{(p_{i}^2+q^2+k^2)^2 - 4k^2q^2 } \left[p_{i}\cos\left(\frac{\eta}{2}\ln\frac{(k+q)^2+p_{i}^2}{(k - q)^2+p_{i}^2}\right)\right.\nonumber\\ & { } & \left.+\frac{p_{i}^2-q^2+k^2}{2q } \sin\left(\frac{\eta}{2}\ln\frac{(k+q)^2+p_{eff}^2}{(k - q)^2+p_{i}^2}\right)\right],\end{aligned}\ ] ] @xmath265 ^ 2}-\frac{32p_{i}^4}{(4p_{i}^2+q^2)^2}\right\}.\nonumber\\\end{aligned}\ ] ] finally , the usual choice of the effective charges is @xmath266 and @xmath267 ( see , for instance , @xcite and references therein ) . the value @xmath268 follows from the variational procedure that minimizes the ground - state energy @xmath231 , while the value @xmath267 ensures the correct asymptotic behavior of the final state . however , the authors of @xcite utilized in their calculations the values @xmath269 and @xmath270 derived from fitting the photoionization cross - section data on helium with the present model of the he states . the departures of the differential cross sections ( [ d1s ] ) and ( [ d1sw ] ) from the free - electron approximation are characterized by the respective atomic factors @xmath271 where @xmath272 and @xmath273 are the sm and @xmath26 contributions to the differential cross section for scattering of an electron antineutrino on two free electrons . let us recall that following @xcite one should expect the @xmath274 value to be of about @xmath275 at @xmath239 . atomic factors ( [ at_factor ] ) as functions of the energy transfer @xcite.,scaledwidth=90.0% ] numerical results for atomic factors ( [ at_factor ] ) are shown in fig . they correspond to the kinematical regime @xmath276 , which is typically realized both for reactor and for tritium antineutrinos when @xmath277ev . note that in such a case one can safely set the upper limit of integrals in ( [ d1s ] ) and ( [ d1sw ] ) to infinity , as the dynamical structure factor @xmath278 rapidly falls down when @xmath279 and practically vanishes in the region @xmath280 . it can be seen from fig . [ fig ] that atomic factors exhibit similar behaviors for both sets of the @xmath281 and @xmath252 parameters discussed in the previous section . namely , their values are minimal ( @xmath282 ) at the ionization threshold , @xmath283 , and tend to unity with increasing @xmath28 . the latter tendency is readily explained by approaching the free - electron limit . it can be also seen that a more or less serious deviation ( @xmath284 ) of the present results from the stepping approximation is observed only in the low - energy region @xmath285ev . this deviation can be attributed to the effect of the electron - electron correlation in a helium atom . indeed , if the electrons do not interact with each other , then they occupy two @xmath174 hydrogen - like states ( with opposite spins ) , in which case the departure of the atomic - factor values from unity is , according to the results of sec . [ h ] , less than 5% . thus , the calculations presented in fig . [ fig ] do not confirm the huge enhancement of the @xmath26 contribution with respect to the free - electron approximation . moreover , in accord with various calculations for other atomic targets @xcite , we find that at small energy - transfer values the electron binding in helium leads to the appreciable reduction of the differential cross section relative to the free - electron case . we attribute the erroneous prediction of @xcite to the incorrect dynamical model that draws an analogy between the nmm - induced ionization and photoionization . indeed , as discussed in sec . [ nu - at_el ] , the virtual photon in the nmm - induced ionization process can be treated as real only when @xmath184 . however , the integration in ( [ d1s ] ) involves the @xmath53 values ranging from @xmath28 up to @xmath286 . since @xmath287 , the real - photon picture appears to be applicable only in the vicinity of the lower integration limit . when moving away from that momentum region , one encounters a strong departure from the real - photon approximation which treats the integrand as a constant in the whole integration range , assuming it to be equal to its value at @xmath288 , that is , @xmath289 such an approach is manifestly unjustified , and it gives rise to the spurious enhancement of the @xmath26 contribution to the differential cross section . in the thomas - fermi model ( see e.g. ref . @xcite ) the atomic electrons are described as a degenerate free electron gas in a master potential @xmath290 filling the momentum space up to the zero fermi energy , i.e. up to the momentum @xmath291 such that @xmath292 . the electron density @xmath293 then determines the potential @xmath290 from the usual poisson s equation . in the discussed picture at an energy transfer @xmath28 the ionization is possible only for the electrons whose energies in the potential are above @xmath294 , i.e. with momenta above @xmath295 with @xmath296 . the electrons with lower energy are inactive . calculating the density of the inactive electrons as @xmath297 and subtracting their total number from z , one readily arrives at the formula for the atomic factor , i.e. the effective fraction of the active electrons @xmath298 as a function of @xmath28 : @xmath299^{3/2 } \ , x^2 \ , dx~ , \label{zeff}\ ] ] where @xmath300 is the thomas - fermi function , well known and tabulated , of the scaling variable @xmath301 , the energy scale @xmath302 is given by @xmath303 and , finally , @xmath304 is the point where the integrand becomes zero , i.e. corresponding to the radius beyond which all the electrons are active at the given energy @xmath28 . the energy scale @xmath302 in germanium ( z=32 ) evaluates to @xmath305kev . the thomas - fermi atomic factor for germanium calculated from the formula ( [ zeff ] ) is shown by the dashed line in the plot of fig . the discussed statistical model is known to approximate the average bulk properties of the atomic electrons with a relative accuracy @xmath306 and as long as the essential distances @xmath307 satisfy the condition @xmath308 , which condition in terms of the scaling variable @xmath309 reads as @xmath310 . in terms of the formula ( [ zeff ] ) for the number of active electrons , the lower bound on the applicability of the model is formally broken at @xmath311 , i.e. at the energy scale of the inner atomic shells . however the effect of the deactivation of the inner electrons is small , of order @xmath312 in comparison with the total number @xmath49 of the electrons . on the other hand , at low @xmath28 , including the most interesting region of @xmath313 , the integral in eq . ( [ zeff ] ) is determined by the range of @xmath309 of order one , where the model treatment is reasonably justified . the atomic factor @xmath245 for germanium in the stepping approximation with the actual energies of the orbitals ( solid line ) and its interpolation in the thomas - fermi model ( dashed ) @xcite.,scaledwidth=90.0% ] the energies of the inner @xmath314 and @xmath315 orbitals in the germanium atom are well known ( see e.g. ref . @xcite ) and provide the necessary data for a description of the neutrino scattering by the stepping formula ( [ step ] ) down to the values of the energy transfer @xmath28 in the range of the binding of the @xmath315 electrons , i.e. at @xmath316kev . the corresponding steps in the atomic factor are shown in fig . one can see that the stepping atomic factor ( [ step ] ) mimics upon average over the energy intervals between the electron shells in germanium the thomas - fermi result . thus , it can be considered as refinement of the latter due to account for the quantization of the electron binding energies . it can be mentioned that if one applies formulas of sec . [ h ] to the onset of the @xmath254 shell step , i.e. just above 10.9kev , the difference from the shown in the plot step function would be practically invisible in the scale of fig . [ fig2 ] . while the treatments based on a generic model of manyelectron atomic targets allow to determine characteristic features and behaviors of the differential cross sections ( [ d1s ] ) and ( [ d1sw ] ) , to obtain accurate numerical results one needs to resort to _ ab initio _ calculations . such calculations can be realized using the hartree - fock ( hf ) method ( see , for instance , @xcite ) and its modifications . in the hf approximation , atomic electrons occupy one - electron states in a spherically - symmetric mean - field potential which is derived self - consistently from the solution of the hf equations . accordingly , each one - electron state independently contributes to the atomic ionization process . for the first time this approach was formulated in refs . @xcite , where it was illustrated with numerical calculations of neutrino - impact ionization of the f and mo atomic targets . the wave functions and energies of atomic bound states were calculated within the relativistic hf method @xcite with local exchange - correlation potential @xcite . the wave functions of outgoing electrons were obtained by a numerical solution of the dirac equation in the same mean - field potential as for the wave functions of discrete states . performed in ref . @xcite numerical calculations for ionization of the iodine atoms by impact of reactor antineutrinos led the authors to suggest the stepping approximation ( [ step ] ) . in a very recent theoretical study @xcite , the authors adopted the multiconfiguration relativistic random - phase approximation ( mcrrpa ) @xcite to evaluate the germanium atomic factors . this particular method is based on the time - dependent hf approximation @xcite , however , several important features make it a better tool beyond the usual hf approximation to describe transitions of open - shell atoms of high atomic number @xmath49 . first , for open - shell atoms , typically there are more than one configuration which have the desired ground - state properties . therefore , a proper hf reference state should be formed by a linear combination of these allowed configurations , i.e. , a multiconfiguration reference state . second , for atoms of high @xmath49 , the relativistic corrections can no longer be ignored . by using a dirac equation , instead of a schrdinger one , the leading relativistic terms in the atomic hamiltonian are treated nonperturbatively from the onset . third , two - body correlations in addition to the hf approximation are generally important for excited states and transition matrix elements . the random - phase approximation ( rpa ) is devised to account for part of the additional two - body correlations ( particles can be in the valence or core states ) not only for the excited but also for the reference state , and in a lot of cases , it gives good agreement with experiment @xcite . furthermore , it has been shown that rpa equations preserve gauge invariance @xcite ; this provides a measure of stability of their solutions . the mcrrpa has been applied successfully to photoexcitation and photoionization of divalent atoms such as be , mg , zn , and etc . ( some of the results are summarized in @xcite ) . following similar treatments , the authors of @xcite treated the electronic configuration of germanium as a core filled up to the 4@xmath48 orbits , with two valence electrons in the @xmath317 orbits . as the ge ground state is a @xmath318 state , it is a linear combination of two configurations , namely [ zn]@xmath319 and [ zn]@xmath320 . the wave function was calculated using the multiconfiguration dirac - fock package @xcite . the atomic excitations due to weak and magnetic scattering were solved by the mcrrpa equation , and consequently transition matrix elements were yielded . in that calculation , all the current operators were expanded by spherical multipoles , and the resulting final scattering states were represented in the spherical wave basis and subject to the incoming - wave boundary condition . compared with the previous works on the same subject @xcite which are also in the similar spirit of the relativistic hf method , the mcrrpa approach differs in several respects . first , due to the near degeneracy of the @xmath321 and @xmath322 levels , using a multiconfiguration reference state is necessary . second , the nonlocal fock term is treated exactly , without resorting to the local exchange potentials . third , the excited states are calculated with two - body correlation built in by mcrrpa , not simply by solving a coulomb wave function with a static one - hole mean field . the sm ( weak ) and nmm contributions to the differential cross section of @xmath323-ge ionization at @xmath324mev @xcite in comparison with the corresponding stepping - approximation results ( fea ) . the nmm value is set to be the current upper limit @xmath325 @xcite.,scaledwidth=90.0% ] fig . [ fig : ge_ab_initio ] shows numerical results from ref . @xcite for ionization of germanium by impact of an electron antineutrino . as can be seen , in the energy region @xmath326kev the results are very well explained by the stepping - approximation formula ( [ step ] ) . at the same time , in the sub - kev region , i.e. , where the electrons from the @xmath254 and @xmath327 shells in germanium stay ` inactive ' , both the sm and nmm contributions appear to be significantly suppressed as compared to the stepping approximation . the latter finding seems to disagree with the semiclassical approach discussed in sec . [ wkb ] , according to which the ionization involving more loosely - bound electron states , such as those belonging to the @xmath315 and @xmath328 shells , is expected to follow more closely the free - electron scenario . notice that a similar suppression of the atomic - factor values close to the ionization threshold was observed in the case of helium ( see fig . [ fig ] ) , and it was attributed to the two - electron correlation effect . thus , we can suggest that the correlation effects beyond the approximation of independent electrons lead to the suppression of atomic factors in the low - energy region . this feature will be important for the next - generation experiments with ge detectors having energy thresholds in the sub - kev region @xcite . in this review , we have considered the neutrino - atom ionizing collisions with focus on the most important theoretical issues related to the problem . the main results discussed in the paper can be summed up as follows . the differential over the energy transfer cross section given by the free - electron formulas ( [ cr_sec_sm ] ) and ( [ cr_sec_mu ] ) and the stepping behavior of the atomic factor given by eq . ( [ step ] ) provides a reasonable description of the neutrino - impact ionization of a complex atom , such as germanium , down to quite low energy transfer . the deviation from this approximation due to the onset of the ionization near the threshold is less than 5% ( of the height of the step ) for the @xmath254 electrons , if one applies the analytical behavior of this onset that one finds for the ground state of a hydrogen - like ion . it is also found that the free - electron expressions for the differential cross section are not affected by the atomic binding effects in the semiclassical limit and for independent electrons . these analytical results can support the numerically determined behaviors of the electroweak and magnetic contributions to the neutrino - impact ionization of various atomic targets within the mean - field model @xcite . at the same time , very recent numerical calculations of the @xmath323-impact ionization processes of helium @xcite and germanium @xcite exhibit suppression of the sm and nmm differential cross sections relative to the stepping approximation with lowering the energy - transfer value . this suppression can be assigned to the electron - electron correlation effects . a theoretical analysis @xcite of ionization of helium by electron - antineutrino impact shows no evidence of the recently predicted enhancement @xcite of the electromagnetic contribution as compared with the free - electron case . in contrast , in line with previous studies on other atomic targets , it is found that the magnitudes of the differential cross sections decrease relative to the free - electron approximation when the energy transfer is close to the ionization threshold . thus , no sensitivity enhancement can be expected when using the he atomic target in searches for nmm . and the stepping approximation appears to be valid , within a few - percent accuracy , down to the energy - transfer values as low as almost 100ev . we thus conclude that for practical applications , i.e. for the analysis of data of the searches for nmm , one can safely apply the free - electron formulas and the stepping approximation at the energy transfer down to this range . when analyzing the low-@xmath28 data of the current high - sensitivity experiments searching for neutrino electromagnetic properties , one must go beyond the free - electron approximation for the elastic neutrino - electron scattering and take into account the atomic - ionization effects , at least , in the case of @xmath254 electrons . at the present time , the experiment gemma - ii with reactor antineutrinos is in preparation @xcite . its sensitivity to the nmm value is expected to be at the level of @xmath329 . to achieve such a sensitivity level , which is the region of astrophysical interest @xcite , it is planed to reduce the effective energy threshold of a ge detector from 2.8 to 1.5kev . this threshold value will be very close to the binding energies of the @xmath327 electrons in germanium ( 1.2 - 1.4kev @xcite ) . recently , a @xmath227-type point - contact ge detector @xcite has been implemented in the texono experiment with reactor antineutrinos . the energy threshold of this detector is about 0.3kev , which value is comparable to the binding energies of the @xmath330 electrons in germanium ( 0.12 - 0.18kev @xcite ) . this means that an accurate analysis of the corresponding data will require numerical calculations based on the _ ab initio _ methods . with lowering the @xmath28 value down to @xmath331 , an additional collision channel apart from ionization opens up , namely the coherent elastic neutrino - nucleus scattering @xcite , which has not been observed experimentally so far . the early treatments of the atomic effects in the coherent elastic neutrino - nucleus scattering within sm can be found in @xcite . it should be noted that any deviation of the measured cross section of the coherent elastic neutrino - nucleus scattering from the very precisely known sm value @xcite will provide a signature of the bsm physics ( see @xcite ) . in this context , the accurate calculations of the nmm - induced contribution to the cross section of the coherent elastic neutrino - nucleus scattering appear to be of particular importance . the nmm - induced coherent neutrino scattering by single atoms as well as by crystals was discussed theoretically only in @xcite . however , further studies are necessary for the correct interpretation of future measurements at low @xmath28 values @xcite . this work was partially supported by the russian foundation for basic research ( grants no . 14 - 01 - 00420-a and no . 14 - 22 - 03043-ofi - m ) . 99 c. giunti and c.w . kim , _ fundamentals of neutrino physics and astrophysics _ , oxford university press , oxford , uk , 2007 . m. c. gonzalez - garcia and m. maltoni , `` phenomenology with massive neutrinos , '' _ physics reports _ , vol . 460 , no . 1 - 3 , pp . 1 - 129 , 2008 . s. bilenky , _ introduction to the physics of massive and mixed neutrinos _ , 817 of lecture notes in physics , springer , new york , ny , usa , 2010 . z. z. xing and s. zhou , _ neutrinos in particle physics , astronomy and cosmology _ , zhejiang university press , 2011 . c. giunti and a. studenikin , `` neutrino electromagnetic properties , '' _ physics of atomic nuclei _ , vol . 12 , pp . 2089 - 2125 , 2009 . c. broggini , c. giunti , and a. studenikin , `` electromagnetic properties of neutrinos , '' _ advances in high energy physics _ , vol . 2012 , article i d 459526 , 47 pages , 2012 . c. giunti and a. studenikin , `` neutrino electromagnetic interactions : a window to new physics , '' arxiv:1403.6344 [ hep - ph ] , 72 pages . s. t. petcov , `` the processes @xmath332 , @xmath333 , @xmath334 in the weinberg - salam model with neutrino mixing , '' _ soviet journal of nuclear physics _ , vol . 25 , p. 340 k. fujikawa and r. shrock , `` magnetic moment of a massive neutrino and neutrino - spin rotation , '' _ physical review letters _ 963 - 966 , 1980 . b. pal and l. wolfenstein , `` radiative decays of massive neutrinos , '' _ physical review d _ , vol . 3 , pp . 766 - 773 , 1982 . r. e. shrock , `` electromagnetic properties and decays of dirac and majorana neutrinos in a general class of gauge theories , '' _ nuclear physics b _ , vol . 359 - 379 , 1982 . j. f. nieves , `` electromagnetic properties of majorana neutrinos , '' _ physical review d _ , vol . 3152 - 3158 , 1982 . b. kayser , `` majorana neutrinos and their electromagnetic properties , '' _ physical review d _ , 26 , no . 7 , pp . 1662 - 1670 , 1982 . b. kayser , `` cpt , cp , and c phases , and their effects , in majorana - particle processes , '' _ physical review d _ , vol . 1023 - 1033 , 1984 . j. f. beacom and p. vogel , `` neutrino magnetic moments , flavor mixing , and the super - kamiokande solar data , '' _ physical review letters _ , vol . 25 , pp . 5222 - 5225 , 1999 . n. f. bell , v. cirigliano , m. j. ramsey - musolf , p. vogel , and m. b. wise , `` how magnetic is the dirac neutrino ? , '' _ physical review letters _ , 15 , article i d 151802 , 4 pages , 2005 . n. f. bell , m. gorchtein , m. j. ramsey - musolf , p. vogel , and p. wang , `` model independent bounds on magnetic moments of majorana neutrinos , '' _ physics letters b _ , vol . 642 , no.4 , pp . 377 - 383 , 2006 . h. t. wong , h. b. li , s. t. lin et al . , `` search of neutrino magnetic moments with a high - purity germanium detector at the kuo - sheng nuclear power station , '' _ physical review d _ , vol . 1 , article i d 012001 , 16 pages , 2007 . a. g. beda , v. b. brudanin , v. g. egorov et al . , `` the results of search for the neutrino magnetic moment in gemma experiment , '' _ advances in high energy physics _ , vol . 2012 , article i d 350150 , 12 pages , 2012 . l. b. auerbach , r. l. burman , d. o. caldwell et al . , `` measurement of electron - neutrino electron elastic scattering , '' _ physcial review d _ , vol . 11 , 11 pages , 2001 . r. schwienhorst , d. ciampa , c. erickson et al . , `` a new upper limit for the tau - neutrino magnetic moment , '' _ physics letters b _ , vol . 513 , no . 1 - 2 , pp . 23 - 29 , 2001 . d. w. liu , y. ashie , s. fukuda et al . , `` limits on the neutrino magnetic moment using 1496 days of super - kamiokande - i solar neutrino data , '' _ physical review letters _ , vol . 2 , article i d 021802 , 5 pages , 2004 . c. arpesella , h. o. back , m. balata et al . , `` direct measurement of the @xmath335be solar neutrino flux with 192 days of borexino data , '' _ physical review letters _ , vol . 101 , no . 9 , article i d 091302 , 6 pages , 2008 . g. raffelt , `` new bound on neutrino dipole moments from globular - cluster stars , '' _ physical review letters _ , 2856 - 2858 , 1990 . r. n. mohapatra and p. b. pal , _ massive neutrinos in physics and astrophysics _ 72 of _ lecture notes in physics _ , world scientific , river edge , nj , usa , 3rd edition , 2004 . e. d. commins and p. h. bucksbaum , _ weak interactions of leptons and quarks _ , thid edition , cambridge university press , cambridge , 1983 . j. f. carlson and j. r. oppenheimer , `` the impacts of fast electrons and magnetic neutrons , '' _ physical review _ , vol . 41 , no . 6 , pp . 763 - 792 , 1932 . h. bethe , `` ionization power of a neutrino with magnetic moment , '' _ mathematical proceedings of the cambridge philosophical society _ , vol . 108 - 115 , 1935 . a. v. kyuldjiev , `` searching for effects of neutrino magnetic moments at reactors and accelerators , '' _ nuclear physics b _ , vol . 387 - 397 , 1984 . g. domogatsky and d. nadezhin , `` modern theory of star evolution and experiments of f. reines on anti @xmath37-scattering detection , '' _ soviet journal of nuclear physics _ , vol . 12 , p. 678 , 1971 . c. l. cowan and f. reines , `` neutrino magnetic moment upper limit , '' _ physical review _ 528 - 530 , 1957 . c. l. cowan , jr . , f. reines , and f. b. harrison , `` upper limit on the neutrino magnetic moment , '' _ physcial review _ , vol . 5 , pp . 1294 - 1294 , 1954 . p. vogel and j. engel , `` neutrino electromagnetic form factors , '' _ physical review d _ 3378 - 3383 , 1989 . h. t. wong , `` low energy neutrino and dark matter physics with sub - kev germanium detectors , '' international journal of modern physics , vol . 20 , no . 8 , pp . 1463 - 1470 , 2011 . h. b. li , h. y. liao , s. t. lin et al . , `` limits on spin - independent couplings of wimp dark matter with a @xmath227-type point - contact germanium detector , '' physical review letters , vol . 110 , no . 26 , article i d 261301 , 5 pages , 2013 . h. b. li , l. singh , m. k. singh et al . , `` differentiation of bulk and surface events in @xmath227-type point - contact germanium detectors for light wimp searches , '' vol . 56 , pp . 1 - 8 , 2014 . v. gaponov , yu . l. dobrynin , and v. n. tikhonov , `` neutrino and antineutrino scattering on a hydrogen - like atom , '' _ soviet journal of nuclear physics _ , vol . 22 , p. 170 v. yu . dobretsov , a. b. dobrotsvetov , and s. a. fayans , `` inelastic neutrino scattering by atomic electrons , '' _ soviet journal of nuclear physics _ , vol . 55 , p. 1180 , 1992 . s. a. fayans , v. yu . dobretsov , and a. b. dobrotsvetov , `` effect of atomic binding on inelastic @xmath336e scattering '' _ physics letters b _ , vol . 291 , nos . 1 - 2 , pp . 1 - 6 , 1992 . v. i. kopeikin , l. a. mikaelyan , v. v. sinev and s. a. fayans , `` scattering of reactor antineutrinos by electrons '' , _ physics of atomic nuclei _ , vol . 1859 - 1864 , 1997 . s. a. fayans , l. a. mikaelyan , and v. v. sinev , `` weak and magnetic inelastic scattering of antineutrinos on atomic electrons , '' _ physics of atomic nuclei _ , vol . 64 , no . 8 , pp . 1475 - 1480 , 2001 . g. j. gounaris , e.a . paschos , p. i. porfyriadis , `` the ionization of h , he or ne atoms using neutrinos or antineutrinos at kev energies , '' _ physics letters b _ , vol . 525 , nos . 1 - 2 , pp . 63 - 70 , 2002 . v. i. kopeikin , l. a. mikaelyan , and v. v. sinev , `` inelastic scattering of tritium - source antineutrinos on electrons of germanium atoms , '' _ physics of atomic nuclei _ , vol . 707 - 711 , 2003 . h. t. wong , h. b. li , and s. t. lin , `` enhanced sensitivities for the searches of neutrino magnetic moments through atomic ionization , '' _ physical review letters _ , vol . 105 , no . 6 , article i d 061801 , 4 pages , 2010 . m. voloshin , `` neutrino scattering on atomic electrons in searches for the neutrino magnetic moment , '' _ physical review letters _ , vol . 20 , article i d 201801 , 4 pages , 2010 . h. t. wong , h. b. li , and s. t. lin,``enhanced sensitivities for the searches of neutrino magnetic moments through atomic ionization , '' http://arxiv.org/abs/hep-ph/1001.2074v3 . k. a. kouzakov and a. i. studenikin,``magnetic neutrino scattering on atomic electrons revisited , '' _ physics letters b _ , vol . 252 - 256 , 2011 . k. a. kouzakov and a. i. studenikin , `` electromagnetic neutrino - atom collisions : the role of electron binding , '' _ nuclear physics b ( proc . suppl . ) 353 - 356 , 2011 . k. a. kouzakov , a. i. studenikin , and m. b. voloshin , `` testing neutrino magnetic moment in ionization of atoms by neutrino impact , '' _ jetp letters _ , vol . 699 - 703 , 2011 , http://arxiv.org/abs/1105.5543 . k. a. kouzakov , a. i. studenikin , and m. b. voloshin , `` neutrino - impact ionization of atoms in searches for neutrino magnetic moment , '' _ physical review d _ 11 , article i d 113001 , 11 pages , 2011 . v. p. martemyanov and v. g. tsinoev , `` ionization of helium atoms under the effect of the antineutrino magnetic moment , '' _ physics of atomic nuclei _ , 12 , pp . 1671 - 1675 , 2011 . k. a. kouzakov and a. i. studenikin , `` on sensitivity of neutrino - helium ionizing collisions to neutrino magnetic moments , '' _ physics of particles and nuclei letters _ , vol . 4 , pp . 458 - 461 , 2014 . l. van hove , `` correlations in space and time and born approximation scattering in systems of interacting particles , '' _ physical review _ , vol . 95 , no . 1 , pp . 249 - 262 , 1954 . u. fano , `` penetration of protons , alpha particles , and mesons , '' _ annual review of nucler science _ 13 , pp . 1 - 66 , 1963 . belki , `` bound - free non - relativistic transition form factors in atomic hydrogen , '' _ journal of physics b : atomic and molecular physics _ , vol . 12 , pp . 1907 - 1914 , 1981 . a. r. holt , `` matrix elements for bound - free transitions in atomic hydrogen , '' _ journal of physics b : atomic and molecular physics _ 2 , no . 11 , pp . 1209 - 1213 , 1969 . a. i. akhiezer and v. b. berestetskii , _ quantum electrodynamics _ , wiley , new york , 2nd edition , 1965 . l. d. landau and e. m. lifshits , _ quantum mechanics ( non - relativistic theory ) _ , pergamon , oxford , 3rd edition , 1977 . k. a. kouzakov , s. a. zaytsev , yu . v. popov , and m. takahashi , `` singly ionizing 100-mev / amu c@xmath337he collisions with small momentum transfer , '' _ physical review a _ , vol . 86 , article i d 032710 , 8 pages , 2012 . b. l. henke , e. m. gullikson , and j. c. davis , `` x - ray interactions : photoabsorption , scattering , transmission , and reflection at @xmath338ev , @xmath339 , '' atomic data and nuclear data tables , vol . 181 - 342 , 1993 . i. p. grant , `` relativistic self - consistent fields , '' _ proceedings of the royal society a _ , vol . 555 - 576 , 1961 . i. p. grant , `` relativistic self - consistent fields , '' _ proceedings of the physical society _ , vol . 523 - 527 , 1965 . v. l. moruzzi , j. f. janak , and a. r. williams , _ calculated electronic properties of metals _ , pergamon , oxford , 1978 . chen , h .- c . n . huang et al . , `` atomic ionization of germanium by neutrinos from an ab initio approach , '' _ physics letters b _ , vol . 159 - 162 , 2014 . huang and w. r. johnson , `` multiconfiguration relativistic random - phase approximation . theory , '' _ physical review a _ , vol . 634 - 649 , 1982 . n . huang , `` relativistic many - body theory of atomic transitions . the relativistic equation - of - motion approach , '' _ physical review a _ , vol . 734 - 739 , 1982 . p. jrgensen , `` molecular and atomic applications of time - dependent hartree - fock theory , '' _ annual review of physical chemistry _ , 359 - 380 , 1975 . amusia and n. a. cherepkov , `` many - body electron correlations in scattering processes , '' _ case studies in atomic physics _ , vol . 5 , pp . 47 - 179 , 1975 . d. l. lin , `` gauge properties of the hartree - fock and random - phase approximations , '' _ physical review a _ , vol . 16 , no . 600 - 605 , 1977 . n . huang , h .- c . chi , and h .- s . chou , `` the mcrrpa theory and its applications to photoexcitation and photoionization , '' _ chinese journal of physics _ , 33 , no . 6 , pp . 565 - 644 , 1995 . j. p. desclaux , `` a multiconfiguration relativistic dirac - fock program , '' _ computer physics communications _ , vol . 9 , no . 1 , pp . 31 - 45 , 1975 . d. z. freedman , `` coherent effects of a weak neutral current , '' _ physical review d _ , vol . 5 , pp . 1389 - 1392 , 1974 . v. gaponov and v. n. tikhonov , `` elastic scattering of low - energy neutrinos by atomic systems , '' _ soviet journal of nuclear physics _ , vol . 26 , p. 314 , 1977 . l. m. sehgal and m. wanninger , `` atomic effects in coherent neutrino scattering , '' _ physics letters b _ , vol . 107 - 112 , 1986 . a. drukier and l. stodolsky , `` principles and applications of a neutral - current detector for neutrino physics and astronomy , '' _ physical review d _ , vol . 2295 - 2309 , 1984 . k. scholberg , `` prospects for measuring coherent neutrino - nucleus elastic scattering at a stopped - pion neutrino source , '' _ physical review d _ , vol . 73 , article i d 033005 , 9 pages , 2006 . j. barranco , o. g. miranda , and t. i. rashba , `` probing new physics with coherent neutrino scattering off nuclei , '' _ journal of high energy physics _ , vol . 02 , article i d 021 , 14 pages , 2005 . j. barranco , o. g. miranda , and t. i. rashba , `` sensitivity of low energy neutrino experiments to physics beyond the standard model , '' _ physical review d _ , vol .76 , article i d 073008 , 9 pages , 2007 . s. davidson , c. pea - garay , n. rius , and a. santamaria , `` present and future b ounds on non - standa rd neutrino interactions , '' _ journal of high energy physics _ , 03 , article i d 011 , 34 pages , 2003 . j. augustin , b. mller , and w. greiner , `` magnetic neutrino scattering by crystals , '' _ physical review d _ , 5 , pp . 1683 - 1686 , 1990 . j. a. formaggio , e. figueroa - feliciano , and a. j. anderson , `` sterile neutrinos , coherent scattering , and oscillometry measurements with low - temperature bolometers , '' _ physical review d _ 85 , article i d 013009 , 14 pages , 2012 .

cosmic strings are linear gravitational stable topological defects which may have been created as a consequence of phase transitions in the early universe and are predicted in the context of the standard gauge field theory of elementary particle physics with extra symmetries @xcite . observations of anisotropies in the cosmic microwave background radiation ( cmb ) by cobe , wmap and more recently by the planck satellite have ruled out cosmic strings as the primary source for primordial density perturbations since they can not explain the acoustic peaks in the angular power spectrum . however , a couple of other effects due to cosmic strings can be investigated through the observations of cmb , such as non - gaussianity signals and temperature anisotropies caused by the gravitational field of moving strings ( see @xcite for a recent analysis of the possible effects of strings on cmb ) . furthermore , the formation of strings can also have astrophysical and cosmological consequences . for instance , emission of gravitational waves and high energy cosmic rays by strings such as neutrinos and gamma - rays , along with observational data , can help to constraint the product of the newton s constant , @xmath2 , and the linear mass density of the string @xmath3 @xcite . all these effects can also be generated by cosmic strings formed in the context of brane inflation , which makes the physics of cosmic strings a wide area of worth investigation . in this paper , we will only focus on the effects generated by the spacetime of a infinitely long straight static cosmic string as explained below . the geometry of the spacetime associated with an idealized cosmic string , i.e. , infinitely long and straight , is locally flat but topologically conical , having a planar angle deficit given by @xmath4 on the two - surface orthogonal to the string . although this object was first introduced in the literature as being created by a dirac - delta type distribution of energy and axial stress along a straight infinity and line , it can also be described by classical field theory where the energy - momentum tensor associated with the maxwell - higgs system , investigated by nielsen and olesen in @xcite , couples to the einstein s equations . this coupled system was further investigated by garfinkle and linet in @xcite and @xcite , respectively . these authors have shown that a planar angle deficit , @xmath5 , arises on the two - surface perpendicular to a string , as well as a magnetic flux running through its core . although the geometry of the spacetime produced by an idealized cosmic string is locally flat , its conical structure alters the vacuum fluctuations associated with quantum fields . as a consequence , the vacuum expectation value ( vev ) of physical observables like the energy - momentum tensor , @xmath6 , gets a nonzero value . the calculation of the vev of physical observables associated with the scalar and fermionic fields in the cosmic string spacetime has been developed in @xcite and @xcite , respectively . furthermore , the presence of a magnetic flux running through the core of the string gives additional contributions to the vevs associated with charged fields @xcite as well as induces vacuum current densities , @xmath7 . this phenomenon has been investigated for massless and massive scalar fields in @xcite and @xcite , respectively . in these papers , the authors have shown that induced vacuum current densities along the azimuthal direction arise if the ratio of the magnetic flux by the quantum one has a nonzero fractional part . the calculation of induced fermionic currents in higher - dimensional cosmic string spacetime in the presence of a magnetic flux has been developed in @xcite . the induced fermionic current by a magnetic flux in @xmath8-dimensional conical spacetime and in the presence of a circular boundary has also been analyzed in @xcite . the presence of compact dimensions also induces topological quantum effects on matter fields . it is well known that the presence of compact dimensions is an important feature in many high - energy theories of fundamental physics , like supergravity and superstring theories . an interesting application of field theoretical models that present compact dimensions can be found in nanophysics . the long - wavelength description of the electronic states in graphene can be formulated in terms of the dirac - like theory in three - dimensional spacetime , with the fermi velocity playing the role of the speed of light ( see , e.g. , @xcite ) . the corresponding effective @xmath9 dimensional field theory , in addition to dirac fermions , involves scalar and gauge fields originated from the elastic properties and describing disorder phenomena , like the distortion of graphene lattice and structural defects @xcite . for instance , a single - walled carbon nanotube is generated by rolling up a graphene sheet to form a cylinder . in this case , the background spacetime for the corresponding dirac - like theory has topology @xmath10 . the combined effects of the non - trivial topology of the cosmic string spacetime , the compactified dimension along the axis of the string and the presence of a magnetic flux running through its core and enclosed by the compactified axis , on the vevs of the energy - momentum tensor , @xmath6 , and current densities , @xmath11 , associated with charged quantum fermionic fields in a four - dimensional cosmic string spacetime have recently been investigated in @xcite and @xcite , respectively . in kaluza - klein type models , the currents along compact dimensions generate magnetic filed in the uncompactified subspace . the magnetic fields induced by the vacuum currents provide an additional channel for interaction of cosmic string with the enviroment . here , in the present paper , we shall continue along the same line of investigation . we shall calculate the induced bosonic current in a higher - dimensional cosmic string spacetime , under the same conditions as considered in these two previous publications . this paper is organized as follows . the section [ wightman ] is devoted to the evaluation of the positive frequency wightman function for a massive charged scalar quantum field in a higher - dimensional cosmic string spacetime . we also consider that the @xmath12-axis along the string is compactified to a circle , by imposing a quasiperiodic boundary condition on the bosonic field with arbitrary phase . moreover we assume the presence of magnetic fluxes running through the string s core and enclosed by its axis . in section [ current ] , by using the wightman function , we evaluate the renormalized vacuum current density induced by the magnetic fluxes and the compactification . as we shall see , the renormalized charge density and the radial current vanish . for the azimuthal current density , the compactification induces it to decompose into two parts : one of them coincides with the corresponding expression in the geometry of a cosmic string without compactification and the other is the contribution due to the compactification itself . moreover , as a consequence of the compactification , a non - vanishing axial current also arises and is a purely topological one . the most relevant conclusions of the paper are summarized in section [ conc ] . we have also dedicated an appendix to provide some important expressions used in the development of our calculation for the induced current densities . throughout the paper we use natural units @xmath13 . in this paper we consider a @xmath14dimensional idealized cosmic string spacetime with @xmath15 . by using the generalized cylindrical coordinates @xmath16 with the string on the @xmath17dimensional hypersurface @xmath18 , the corresponding geometry is described by the line element @xmath19 the coordinates take values in the following intervals : @xmath20 , @xmath21 and @xmath22 for @xmath23 . the parameter @xmath24 codifies the presence of the cosmic string . moreover , we assume that the direction along the @xmath12-axis is compactified to a circle with the length @xmath25 , so @xmath26 . the standard cosmic string space - time is characterized by @xmath27 , with @xmath28 . in this case @xmath29 , being @xmath3 the linear mass density of the string . by a radial disgyration is described by @xmath27 line element ( [ eq01 ] ) with a negative planar angle deficit @xcite . ] in this paper we are interested in calculating the induced vacuum current density , @xmath30 , associated with a charged scalar quantum field , @xmath31 , in the presence of magnetic fluxes running along the core of the string and enclosed by it , considering that the @xmath12-axis is compactified to a circle . in order to do that we shall calculate the corresponding complete set of normalized bosonic wavefunction . the equation which governs the quantum dynamics of a charged bosonic field with mass @xmath32 , in a curved background and in the presence of an electromagnetic potential vector , @xmath33 , reads @xmath34 where the differential operator above is defined by @xmath35 being @xmath36 and @xmath37 . in addition , we have considered the presence of a non - minimal coupling , @xmath38 , between the field and the geometry represented by the ricci scalar , @xmath39 . however , for a thin and infinitely straight cosmic string , @xmath40 for @xmath41 . in the analysis that we want to develop , it will be assumed that the direction along the @xmath12-axis is compactified to a circle with length @xmath25 : @xmath42 . the compactification is achieved by imposing the quasiperiodicity condition on the matter field , @xmath43 with a constant phase @xmath44 , @xmath45 . the special cases @xmath46 and @xmath47 correspond to the untwisted and twisted fields , respectively , along the @xmath12-direction . in addition , we shall consider the existence of the following constant vector potential @xmath48 with @xmath49 and @xmath50 , being @xmath51 and @xmath52 the corresponding magnetic fluxes . in quantum field theory the condition changes the spectrum of the vacuum fluctuations compared with the case of uncompactified dimension and , as a consequence , the induced vacuum current density changes . in the spacetime defined by and in the presence of the vector potential given above , the equation becomes @xmath53\varphi(x)=0 \ . \label{eq06}\ ] ] the positive energy solution of this equation can be obtained by considering the general expression , @xmath54 where @xmath55 represents the coordinates of the extra dimensions , @xmath56 the momentum along these directions and @xmath57 is a normalization constant . substituting into we find that the radial function @xmath58 must obey the differential equation @xmath59 with @xmath60 in the present analysis we shall assume that the wave functions obey the dirichlet boundary condition on the string s core . the regular solution at origin is @xmath61 , where @xmath62 represents the bessel function of order @xmath63 @xmath1 being the quantum flux . then , the general solution takes the form @xmath64 the quasiperiodicity condition provides a discretization of the quantum number @xmath65 as shown below : @xmath66 under this circumstance the energy takes the form @xmath67 where @xmath68 the constant @xmath57 can be obtained by the normalization condition @xmath69=\delta_{\sigma,\sigma ' } \ , \label{eq14}\ ] ] where the delta symbol on the right - hand side is understood as dirac delta function for the continuous quantum number , @xmath70 and @xmath71 , and kronecker delta for the discrete ones , @xmath72 and @xmath73 . from one finds @xmath74 so , the renormalized bosonic wave - function reads , @xmath75^{\frac{1}{2 } } j_{q|n+\alpha|}(\lambda r)e^{-i\omega t+iqn\phi+ik_l z+i{\vec{k}\cdot{\vec{r}}_{\parallel } } } \ . \label{eq16}\ ] ] the properties of the vacuum state are described by the corresponding positive frequency wightman function , @xmath76 , where @xmath77 stands for the vacuum state . having this function we can evaluate the induced bosonic current . for the evaluation of the wightman function , we use the mode sum formula @xmath78 where we are using the compact notation defined as @xmath79 the set @xmath80 represents a complete set of normalized mode functions satisfying the periodicity condition . in our case , the mode functions in eq . are specified by the set of quantum numbers @xmath81 , with the values in the ranges @xmath82 , @xmath83 with @xmath84 , @xmath85 and @xmath86 with @xmath87 . substituting into the sum we obtain @xmath88 where @xmath89 , @xmath90 , @xmath91 and @xmath92 . having the positive frequency wightman function above , we are in position to calculate the induced vacuum bosonic current density , @xmath11 . this calculation will be developed in the next section . the bosonic current density operator is given by , @xmath93 \nonumber\\ & = & ie\left[\varphi^{*}(x)\partial_{\mu } \varphi ( x)-\varphi(x ) ( \partial_{\mu } \varphi(x))^{*}\right]-2e^2a_\mu(x)|\varphi(x)|^2 \ . \label{eq20}\end{aligned}\ ] ] its vacuum expectation value ( vev ) can be evaluated in terms of the positive frequency wightman function as shown below : @xmath94 this vev is a periodic function of the magnetic fluxes @xmath51 and @xmath52 with period equal to the quantum flux . this can be observed if we write the parameter @xmath95 in in the form @xmath96 where @xmath97 is an integer number . in this case the vev of the current density will depend on @xmath98 only . let us start the calculation with the charge density . because @xmath99 , we have , @xmath100 substituting into the above expression , taking the time derivative and finally the coincidence limit @xmath101 , the formal expression for the charge density is : @xmath102 because the above expression is divergent , in order to obtain a finite and well defined result , we have to regularize it by introducing a cutoff function @xmath103 , with the cutoff parameter @xmath104 . at the end of the calculation we shall take the limit @xmath105 . so , using the cutoff function , the integral over the variable @xmath70 can be evaluated with the help of @xcite , and the regularized contribution gives @xmath106 with @xmath107 being the modified bessel function . as to the integral over @xmath71 we have , @xmath108 thus , the regularized charge density reads @xmath109 where @xmath110 in appendix [ summ1 ] , it is shown that the above summation on the quantum number @xmath72 is given by and . substituting this result into we obtain @xmath111 \ , \label{densityreg1}\end{aligned}\ ] ] with @xmath112 . in the above expression @xmath113 $ ] , where @xmath114 $ ] represents the integer part of @xmath115 , and the prime on the sign of the summation means that in the case @xmath116 the term @xmath117 should be taken with the coefficient @xmath118 . the first term in the square bracket of ( [ densityreg1 ] ) corresponds to the charge density for @xmath119 and @xmath120 . the renormalized value for the charge density is given by subtracting from the contribution corresponding to the minkowski spacetime in the absence of magnetic flux . we can do this in manifest form by discarding the first term inside the bracket . the other contributions contain @xmath121 and @xmath122 , inside the integral and summation respectively ; hence in the limit @xmath123 ( @xmath124 ) these terms vanish for @xmath125 . thereby , we conclude that the renormalized value for the charge density is zero , i.e , there is no induced charge density . in the apendix [ extra - dimension ] we explicitly proved that the renormalized induced vacuum current densities along the extra dimensions vanish , i.e. , @xmath126 , for @xmath127 . this result is in agreement with the invariance of the system under a boost along the @xmath128 direction . now let us analyze the radial current density . because @xmath129 , the vev of the @xmath130-component of the current is simply expressed as @xmath131 taking the radial derivatives with respect to @xmath130 and @xmath132 in the wightman function , subtracting both terms and taking the coincidence limit , there appears a cancellation between those terms . thereby , we also conclude that there is no induced radial current density : @xmath133 the vev of the azimuthal current density is given by @xmath134 substituting into the above equation we get the formal expression for the azimuthal bosonic current density below : @xmath135 in order to develop the summation over the quantum number @xmath136 we shall apply the abel - plana summation formula in the form @xcite , which is given by @xmath137 f(u ) \notag \\ & & \qquad + i\int_{0}^{\infty } du\left [ f(iu)-f(-iu)\right ] \sum_{\lambda = \pm 1}\frac{g(i\lambda u)}{e^{2\pi ( u+i\lambda \tilde{\beta } ) } -1 } \ . \label{sumform}\end{aligned}\ ] ] taking @xmath138 and @xmath139 by using this formula , it is possible to decompose the expression to @xmath140 , eq . , as the sum of the two contributions as shown below : @xmath141 where the term , @xmath142 , is the contribution of the first integral on the right - hand side of ( [ sumform ] ) and corresponds to the azimuthal current density in the geometry of the higher dimensional cosmic string spacetime without compactification . the term , @xmath143 , is induced by the compactification of the string along its axis and is provided by the second integral on the right - hand side of ( [ sumform ] ) . as we shall see the latter vanishes in the limit @xmath144 . as we have already mentioned in the beginning of this paper , the calculation of the induced azimuthal bosonic current density by a magnetic flux in the geometry of an idealized cosmic string has been developed in @xcite and @xcite for massless and massive quantum fields , respectively . in @xcite the calculation of the azimuthal vacuum current density was developed for a higher dimensional cosmic string spacetime considering the case where the parameter @xmath145 . however , to our knowledge , a closed expression for the induced azimuthal current considering general values of @xmath146 is missed . in this sense , here we generalize the results in @xcite considering general values of the parameter , @xmath146 , as well as consider the case where there exists a magnetic flux running through the core of a string in which the axis is compactified to a circle . thus , in the present paper , we want to investigate the induced bosonic current as general as possible , combining all the above effects . from the first integral on the right hand side of , eq . gives @xmath147 where we have introduced a new variable @xmath148 . in order to provide a more workable expression , we use the identity . this allows us to integrate over @xmath70 by using @xcite . also the integral over the momentum on the extra dimensions is easily evaluated . finally , writing @xmath95 in the form given in we obtain @xmath149 where we have defined @xmath150 . in appendix [ summ2 ] it is shown that the above summation in the quantum number @xmath72 is given by and . subtituting this result into , we obtain @xmath151\right.\nonumber\\ & + & \left.\frac{q}{\pi } \int_{0}^{\infty}dy \ \frac{g(q,\alpha_{0},2y)\sinh ( 2y)}{\cosh(2qy)-\cos(q\pi)}f_{\frac{(d+1)}{2}}\left[2mr\cosh(y)\right ] \right ] , \label{jphi_cs04}\end{aligned}\ ] ] where we use the notation @xmath152 being @xmath153 the modified bessel function . we can see that @xmath154 vanishes for the case @xmath155 . for @xmath156 the first term on the right hand side of is absent and the result coincides with the one found for the azimuthal induced bosonic current in @xcite . from eq . , we can see that @xmath154 is an odd function of @xmath98 , with period equal to the quantum flux @xmath157 . in fig.[fig1 ] we plot the behavior of the azimuthal current density as a function of @xmath98 for the case where @xmath27 , considering @xmath158 and different values of @xmath146 . one can see that the effect of the cosmic string parameter , @xmath146 , is to amplify the oscillatory nature of the azimuthal current with respect to the parameter @xmath98 . is plotted , in units of `` @xmath159 '' , in terms of @xmath160 for @xmath158 and @xmath161 and @xmath162.,title="fig:",scaledwidth=40.0% ] . for a massless field , from eq . , we obtain the following expression @xmath163 \ . \label{jphi_massless}\end{aligned}\ ] ] taking for the above expression @xmath27 , we have @xmath164 \ . \label{jphi_csmasslessd3}\end{aligned}\ ] ] considering @xmath27 and @xmath165 , the behavior of the @xmath154 at large distance from string , @xmath166 , is dominated by the first term of with @xmath167 . it reads , @xmath168 now let us develop the calculation of the contribution to the azimuthal current induced by the compactification . so , we substitute the second term of into . doing this we obtain : @xmath169 to continue our analysis , we shall use the series expansion @xmath170 in the above expression , and with the help of @xcite we get , @xmath171 at this point , we shall use the integral representation below for the macdonald function @xcite @xmath172 by using this representation , it is possible to integrate over the variable @xmath70 and over the momentum along the extra dimensions @xmath71 . so , we obtain @xmath173-\frac{r^{2}m^{2}}{2w}}\nonumber \\ & & \times\sum_{n=-\infty}^{\infty}q(n+\alpha_{0 } ) i_{q|n+\alpha_{0}|}(w ) \ , \label{jphi_comp03}\end{aligned}\ ] ] where we have used @xmath95 in the form and introduced the new variable , @xmath174 . the sum over the quantum number @xmath72 in can be developed by using the compact result . this allows us to perform the integrals over @xmath175 . our final expression is : @xmath176\right.\nonumber \\ & & \left.+ \ \frac{q}{\pi}\int_{0}^{\infty}dy\frac{g(q,\alpha_{0},2y ) \sinh{(2y ) } } { \cosh(2qy)-\cos(q\pi)}f_{\frac{(d+1)}{2}}\left[ml\sqrt{l^{2}+\eta^{2}(y)}\right ] \right\ } \ , \label{jphi_comp04}\end{aligned}\ ] ] where we have defined @xmath177 from the above expression we can see that the contribution due to the compactification on the bosonic current density is an even function of the parameter @xmath178 and is an odd function of the magnetic flux along the core of the string , with period equal to @xmath157 . in particular , in the case of an untwisted bosonic field , @xmath179 is an even function of the magnetic flux enclosed by the string s axis . also we can see that for @xmath155 the induced current above vanishes . moreover , @xmath180 for @xmath18 . this result is in contrast with the fermionic case where the azimuthal current induced by the compactification does not vanish at @xmath18 , as shown in @xcite . in fig.[fig2 ] we plot the behavior of the compactified azimuthal current density as a function of @xmath98 for @xmath27 , considering @xmath158 , @xmath181 , @xmath182 and different values of the parameter @xmath146 . on the graphs below we can see that the effect of the cosmic string parameter , @xmath146 , is to amplify the oscillatory nature of the azimuthal current with respect to the parameter @xmath160 while the effect of the parameter , @xmath178 , is to change the direction of oscillation as well as diminish the absolute values of @xmath183 . is plotted , in units of `` @xmath159 '' , in terms of @xmath160 for @xmath158 , @xmath181 and @xmath161 and 3.5 . the plot on the left is for @xmath184 while the plot on the right is for @xmath185.,title="fig:",scaledwidth=40.0% ] is plotted , in units of `` @xmath159 '' , in terms of @xmath160 for @xmath158 , @xmath181 and @xmath161 and 3.5 . the plot on the left is for @xmath184 while the plot on the right is for @xmath185.,title="fig:",scaledwidth=40.0% ] for large values of the length of the compact dimension , @xmath186 , assuming that @xmath187 is fixed and considering @xmath27 , the main contribution comes from the @xmath188 term and to leading order we find @xmath189 \ , \end{aligned}\ ] ] where we can see that there appear an exponential decay . so in this limit , the contribution to the total current density is dominated by @xmath190 . for a massless field and also considering @xmath27 , we obtain @xmath191 \ , \label{massless}\end{aligned}\ ] ] where we have defined @xmath192 the summation above can be developed with the help of @xcite . so , after some elementar steps we obtain : @xmath193 + 2\pi \tilde{\beta}x\sinh [ \pi ( 1 - 2\tilde{\beta})x]}{4x^{3}\sinh ( \pi x ) } \ , \label{cbet1}\end{aligned}\ ] ] for @xmath194 . with we can also obtain the dominant behavior of @xmath195 in the region @xmath196 . it reads , @xmath197 \left[\sideset{}{'}\sum_{k=1}^{p}{\sin(2k\pi / q)\sin(2k\pi\alpha_{0})}\right.\nonumber \\ & + & \left.\frac{q}{\pi}\int_{0}^{\infty}dy\frac{g(q,\alpha_{0},2y ) \sinh(2y)}{\cosh(2qy)-\cos(q\pi ) } \right ] \ .\end{aligned}\ ] ] also with it is possible to obtain the dominant behavior of @xmath183 in the region @xmath198 . considering @xmath199 and @xmath200 or @xmath201 , @xmath202 . on the other hand taking @xmath199 and for @xmath203 , @xmath204 . so we conclude that the dominant behavior of @xmath183 depends on the value assumed for the parameter @xmath178 . is plotted , in units of `` @xmath159 '' , in terms of @xmath187 for the values @xmath205 , @xmath206 and @xmath207 . the curves corresponding the finite values of @xmath208 are compared with the solid curve for @xmath209 , in which case the total azimuthal current is completely dominated by the term without compactification in eq . ( [ jphi_cs04 ] ) . the left plot is for @xmath200 while the right plot is for @xmath210.,title="fig:",scaledwidth=40.0% ] is plotted , in units of `` @xmath159 '' , in terms of @xmath187 for the values @xmath205 , @xmath206 and @xmath207 . the curves corresponding the finite values of @xmath208 are compared with the solid curve for @xmath209 , in which case the total azimuthal current is completely dominated by the term without compactification in eq . ( [ jphi_cs04 ] ) . the left plot is for @xmath200 while the right plot is for @xmath210.,title="fig:",scaledwidth=40.0% ] combining eqs . and we can write the total azimuthal current as @xmath211\right.\nonumber \\ & & \left.+ \ \frac{q}{\pi}\int_{0}^{\infty}dy\frac{g(q,\alpha_{0},2y ) \sinh{(2y ) } } { \cosh(2qy)-\cos(q\pi)}f_{\frac{(d+1)}{2}}\left[ml\sqrt{l^{2}+\eta^{2}(y)}\right ] \right\ } \ , \label{jphi_total}\end{aligned}\ ] ] where the prime on the sum over @xmath136 means that the term with @xmath212 should be taken with the weight 1/2 . in fig.[fig3 ] we plot the total azimuthal current density as a function of @xmath187 for @xmath27 , considering @xmath205 , @xmath206 and different values of @xmath208 . in the left plot we consider @xmath200 while in the right one @xmath210 . for @xmath213 , one has @xmath214 , and the cases @xmath215 and @xmath216 are related with the untwisted and twisted bosonic fields , respectively . one can see that the effect of the parameter @xmath178 is to alter the curves for the finite values of @xmath208 compared with the solid curve for @xmath209 . here we shall analyze the bosonic current density along the string . as we shall see , due to the compactification of the direction along the string axis , there appear a non - vanishing current . the vev of the axial current is given by @xmath217 using and the fact that @xmath218 , a formal expression for this current can be provided . it reads , @xmath219 where @xmath220 is given by . to evaluate the summation over the quantum number @xmath136 we shall use the generalized abel - plana summation formula , eq . . for this case we have @xmath221 and @xmath222 is given by . taking these expressions into consideration , we can see that the first term on the right hand side of eq . vanishes due to the fact that @xmath223 is an odd function . thus , it remains only a contribution due to the second term on the right hand side of eq . . this contribution is a consequence of the compactification assumed for the direction along the cosmic string s axis . the axial current density induced by the compactification can be written as @xmath224 using again the series expansion , @xmath225 , for the summation in @xmath226 present in eq . , we find @xmath227 the integral over @xmath228 can be evaluated with help of @xcite . the result is given in terms of the modified bessel function of first order , @xmath229 . on the other hand , using the integral representation for this function , and the fact that @xmath230 we can show that @xmath231 substituting the right hand side of the above identity into , it is possible to develop the integral over the variable @xmath70 and also the integral over the extra momenta . moreover , defining a new variable @xmath232 , we obtain @xmath233-\frac{m^2r^2}{2w}}\nonumber \\ & & \times\sum_{n=-\infty}^{\infty}i_{q|n+\alpha|}(w ) \ . \label{jaxialc1}\end{aligned}\ ] ] substituting the expression involving the summation of the modified bessel function given by into , we obtain : @xmath233-\frac{m^2r^2}{2w}}\left[\frac{e^{w}}{q}\right.\nonumber \\ & & \left.-\frac{1}{\pi}\int_{0}^{\infty}dy\frac{f(q,\alpha_{0},y)e^{-w\cosh y}}{\cosh(qy)-\cos(q\pi)}+ \frac{2}{q}\sum_{k=1}^{p}\cos(2k\pi\alpha_{0})e^{w\cos(2k\pi / q)}\right ] \ , \label{jaxial_comp02}\end{aligned}\ ] ] where we have used @xmath95 in the form and @xmath234 is given by . integrating over the variable @xmath235 , the above expression can be written as @xmath236 the first term inside the bracket in , provides a contribution that does not depend on @xmath160 and @xmath146 . it is a pure topological term , a consequence of the compactification only . for this contribution one has @xmath237 notice that this term is independent on the radial distance , @xmath130 . we can say that the above equation corresponds to the current density in the @xmath238 minkowski spacetime with the spatial topology @xmath239 . the current density in minkowski spacetime wiht the topology @xmath240 is recently investigated in @xcite ( for the corresponding problem in de sitter spacetime see @xcite ) . moreover , from we can see that the axial current density vanishes for interger and half - integer values of @xmath178 . for @xmath27 , eq . reads @xmath241 this value is exactly the half of the corresponding value for the fermionic case found in @xcite . the second contribution to the axial current coming from the magnetic flux and planar angle deficit is : @xmath242\right.\nonumber \\ & & \left.-\frac{q}{\pi}\int_{0}^{\infty}dy \frac{f(q,\alpha_{0},2y)}{\cosh(2qy)-\cos(q\pi)}f_{\frac{(d+1)}{2}}\left[ml\sqrt{l^{2}+\eta^{2}(y)}\right]\right\ } , \label{jaxial2_comp}\end{aligned}\ ] ] where @xmath243 and @xmath244 are given by . we can see that eq . is an odd function of the parameter @xmath178 and is an even function of @xmath160 , with period equal to the quantum flux @xmath157 . in particular , in the case of an untwisted bosonic field , eq . is an odd function of the magnetic flux enclosed by the string s axis . moreover , this contribution vanishes for @xmath120 and @xmath155 . in fig . [ fig4 ] we plot the behavior of @xmath245 as function of @xmath246 for @xmath27 , considering @xmath247 , @xmath181 and @xmath248 and @xmath162 . the left plot is for @xmath155 and the right plot is for @xmath206 . by these plots we can see that the amplitude of the current increases with @xmath146 and the effect of @xmath160 is to change the orientation of the current . in eq . ( [ jaxial2_comp ] ) is plotted , in units of `` @xmath249 '' , in terms of @xmath178 for the values @xmath247 , @xmath181 and @xmath250 . the left plot is for @xmath155 while the right plot is for @xmath206.,title="fig:",scaledwidth=40.0% ] in eq . ( [ jaxial2_comp ] ) is plotted , in units of `` @xmath249 '' , in terms of @xmath178 for the values @xmath247 , @xmath181 and @xmath250 . the left plot is for @xmath155 while the right plot is for @xmath206.,title="fig:",scaledwidth=39.8% ] for @xmath18 and @xmath27 , eq . becomes @xmath251 \ , \label{jaxial2_rzero}\end{aligned}\ ] ] where we can see that this contribution is finite . now , considering @xmath186 and @xmath187 fixed , the main contribution to comes from the @xmath188 term . so , for @xmath27 we have @xmath252 where there appears an exponential decay . for a massless field and @xmath27 , the equation reads , @xmath253 \ , \end{aligned}\ ] ] where we have defined @xmath254 once more with the help of @xcite , we were able to develop the summation . the result is : @xmath255}{\sinh^2(\pi x ) } \ , \label{sum - sin1}\end{aligned}\ ] ] for @xmath194 . clearly we can see that @xmath256 vanishes for @xmath257 . with we can obtain the dominant behavior of @xmath258 for @xmath196 . it reads , @xmath259 \ .\end{aligned}\ ] ] in the oposite limit , we can obtain the behavior of @xmath260 for @xmath198 . taking the limit @xmath199 in , we can verify that @xmath261 for @xmath262 , and @xmath263 for @xmath264 . in both cases the axial current presents an exponential decay , however there appears a different signal . in this paper we have investigated the bosonic current density in a higher dimensional compactified cosmic string spacetime induced by magnetic fluxes , one of them enclosed by the compactified direction and the other running along the string s core . the calculations were performed by imposing the quasiperiodicity condition , with arbitrary phase @xmath0 , on the solution of the klein - gordon equation . the general solution was obtained by considering a constant vector potential in eq . ( [ eq05 ] ) and , after imposing the quasiperiodicity condition and calculating the normalization constant ( [ eq15 ] ) , was presented in its final form in eq . ( [ eq16 ] ) . the positive frequency wightman function ( [ eq18 ] ) , which is necessary to calculate the vev of the bosonic current density in eq . ( [ eq21 ] ) , was constructed by the complete set of normalized wave function ( [ eq16 ] ) . in this context we were able to show that the renormalized charge and radial current densities vanish . moreover , we have seen that the compactification induces the azimuthal current density to decompose into two parts . the first one corresponds to the expression in the geometry of a cosmic string without compactification and is presented in eq . ( [ jphi_cs04 ] ) . the second contribution is due to the compactification and is presented in eq . ( [ jphi_comp04 ] ) . the former is an odd function of @xmath160 , with period equal to the quantum flux @xmath157 , and is plotted for @xmath27 , in units of `` @xmath159 '' , with respect to @xmath160 as it can be seen in fig.[fig1 ] . by this graph we can see that the intensity of the current increases with the parameter @xmath146 . furthermore , we have seen that the azimuthal current density induced by the compactification is an even function of the parameter @xmath178 and is an odd function of the magnetic flux along the core of the string , with period equal to @xmath157 . we have checked that when @xmath215 ( untwisted bosonic field ) , eq . ( [ jphi_comp04 ] ) becomes an even function of the magnetic flux enclosed by the strings axis . we have also checked that this induced current vanishes for the case @xmath155 . in addition , in contrast with the fermionic case investigated in ref . @xcite , the azimuthal current density , @xmath183 , vanishes for @xmath265 . for a massless field and considering @xmath266 , eq . ( [ jphi_comp04 ] ) is further simplified and is given by eq . ( [ massless ] ) . in this case , the dominant behavior of @xmath183 depends on the values assumed for the parameter @xmath178 . in fig.[fig2 ] , we plotted eq . ( [ jphi_comp04 ] ) for @xmath27 , in units of `` @xmath159 '' , with respect to @xmath160 . also by this graph we can see that the intensity of the current increases with @xmath146 , and the effect of the parameter @xmath178 plays an important rule on the sign of direction . for the total azimuthal current density , that is , the sum of eqs . ( [ jphi_cs04 ] ) and ( [ jphi_comp04 ] ) , we have seen that it is dominated by @xmath154 for large values of the length of the compact dimension , @xmath267 , assuming @xmath27 and @xmath187 fixed . a plot of the total azimuthal current density , in units of `` @xmath159 '' , with respect to @xmath187 is presented in fig.[fig3 ] for @xmath266 and for two different values of @xmath178 . from this graph we can see that the relavance of the compactified part of the current depends on the product @xmath208 , decreasing when @xmath208 becomes larger . moreover , the relative intensity of the total current , compared with the @xmath154 , depends on @xmath178 . we have also shown that the vev of the axial current density in eq . ( [ jaxial_comp02 ] ) has a purely topological origin and vanishes when @xmath268 and @xmath269 . this vev can be expressed as the sum of two terms . one of them is given by eq . ( [ jaxial_compmin ] ) and independ of the radial distance @xmath130 , the cosmic string parameter @xmath146 and @xmath160 . this contribution corresponds to the current density in @xmath238 minkowski spacetime with the spatial topology @xmath239 . the other contribution is given by eq . ( [ jaxial2_comp ] ) and is due to the magnetic fluxes and the planar angle deficit . we verified that this contribution is an odd function of the parameter @xmath178 and is an even function of @xmath160 , with period equal to the quantum flux @xmath157 . for the particular case when @xmath215 , eq . ( [ jaxial2_comp ] ) becomes an odd function of the magnetic flux enclosed by the strings axis . a plot of the azimuthal current as function @xmath246 is presented in fig.[fig4 ] for two different values of @xmath160 and considering @xmath27 . by this graph we can see that the amplitude of the current increases with the parameter @xmath146 and the effect of @xmath160 is to change the orientation of the current . we would like to emphasize that the currents analyzed in this paper refer to the vacuum ones , induced by the presence of the magnetic fluxes and the compactification . as we could see , the planar angle deficit associated with the cosmic string spacetime amplifies the oscillatory nature of the azimuthal current density , and the compactification introduces additional contribution to it , creating a new axial current density . additionally to the vacuum induced current , witten has shown that under specific conditions , cosmic strings behave as superconducting wires @xcite , being the particle excitation the superconducting charge carriers , that may be either bosons or fermions ( see also @xcite ) . moreover , in the context of braneworlds , the stability of superconducting cosmic strings was discussed in @xcite , and the equations of states relating the tension and energy density per unit length of such topological defects was investigated in @xcite . the authors e.a.f.b and h.f.s.m thank the brazilian agency capes for financial support . e.r.b.m . thanks conselho nacional de desenvolvimento cientfico e tecnolgico ( cnpq ) for partial financial support . also we want to thank aram a. saharian for a critical reading of the paper . here , we shall develop the summation involving the modified bessel functions in eqs . and . we start first with the expression . let us consider the sum @xmath271 . \label{eq:01}\end{aligned}\ ] ] a very useful integral representation for @xmath270 has previously been considered in @xcite and will also be used here . this representation is given by @xmath272 where in our case @xmath273 . upon substituting eq . ( [ eq:02 ] ) into eq . ( [ eq:01 ] ) , one can work out each term separately . thereby , using eq . ( 06 ) ( subsection 5.4.3 ) from @xcite , the summation in @xmath72 of the first and second terms on the right hand side of ( [ eq:02 ] ) is found to be @xmath274 . \label{eq:05}\end{aligned}\ ] ] which is only valid for @xmath275 . now , lets us consider the summation in @xmath72 of the last term on the right hand side of ( [ eq:02 ] ) . this summation can be rewritten as @xmath276e^{-(n+\alpha_0)qy}\right.\nonumber \\ & & + \left.\sin[(n-\alpha_0)q\pi]e^{-(n-\alpha_0)qy}\right ] . \label{eq:06}\end{aligned}\ ] ] in addition , one can further consider the sum on the right hand side of ( [ eq:06 ] ) in the following form : @xmath277e^{-nqy}&= & e^{\mp\alpha_0qy}\left[\cos(\alpha_0\pi q)\sum_{n=1}^{\infty } \sin(n\pi q)e^{-nqy}\right.\nonumber \\ & & \left.\pm\sin(\alpha_0\pi q)\sum_{n=1}^{\infty}\cos(n\pi q)e^{-nqy}\right ] . \label{eq:07}\end{aligned}\ ] ] thus , we can use eqs . ( 01 ) and ( 02 ) ( subsection 5.4.12 ) from @xcite to perform the sums on the right hand side of eq . ( [ eq:07 ] ) . after doing so , and substituting the result in eq . ( [ eq:06 ] ) , we obtain @xmath278 where @xmath279\cosh(|\alpha_0| qy)+ \sin(|\alpha_0|\pi q)\cosh[(1-|\alpha_0|)qy ] . \label{eq:09}\end{aligned}\ ] ] as it can be seen in eq . ( [ eq:09 ] ) , we are considering only absolute values of the parameter @xmath160 . the absolute value @xmath280 is necessary in order to make eq . ( [ eq:09 ] ) an even function of @xmath160 and , therefore , compatible with eq . ( [ eq:01 ] ) which is also an even function of the same parameter . finally , combining eqs . ( [ eq:01 ] ) , ( [ eq:02 ] ) , ( [ eq:05 ] ) and ( [ eq:08 ] ) , we are able to show that @xmath281}\cos(2k\pi\alpha_0)e^{w\cos(2k\pi / q ) } \ , \label{eq:10}\end{aligned}\ ] ] where @xmath114 $ ] represents the integer part of @xmath115 , and the prime on the sign of the summation means that in the case @xmath116 the term @xmath117 should be taken with the coefficient @xmath118 . note that by taking the summation in @xmath282 from @xmath283 to @xmath284 in ( [ eq:05 ] ) it provides the last term on the right hand side of ( [ eq:10 ] ) after combining eqs . ( [ eq:01 ] ) , ( [ eq:02 ] ) , ( [ eq:05 ] ) and ( [ eq:08 ] ) . we can see now that eq . ( [ eq:10 ] ) is in perfect agreement with eq . ( [ eq:01 ] ) , i.e , both are even functions of @xmath160 . this is only possible by considering the absolute values of @xmath160 in eq . ( [ eq:09 ] ) . for integer values of @xmath146 and @xmath155 , we have : @xmath285 the special case in eq . ( [ eq:12 ] ) has been considered in a number of other contexts ( see , e.g. , @xcite ) . let us now turn to the proof of the summation in @xmath72 presented in eq . , which is considered here in the following form : @xmath287,\nonumber\\ \label{form_grad}\end{aligned}\ ] ] which is an odd function of the parameter @xmath160 . thus , taking firstly only positive values of @xmath160 and using the relations @xcite : @xmath288 in eq . ( [ form_grad ] ) , we get @xmath289\nonumber\\ & & + \frac wq \left[i_{q|\alpha_0|-1}(w ) + \sum_{n\geq 1}i_{q(n+\alpha_0)-1}(w ) + \sum_{n\geq 1}i_{q(n-\alpha_0)+1}(w)\right ] \ . \label{s1}\end{aligned}\ ] ] upon defining , @xmath290 , we are able to rewrite ( [ s1 ] ) as @xmath291 on the other hand , we can also exchange the factors @xmath292 between the relations presented in eq . . this provides @xmath293 where now , @xmath294 . substituting ( [ eq:10 ] ) into ( [ s2 ] ) , one can separate the latter in two terms , i.e @xmath295 the first term is given by @xmath296}\cos(2k\pi / q ) \cos(2k\pi\alpha_0)e^{w\cos(2k\pi / q)}\nonumber \\ & & -\frac{w}{q\pi}\int_{0}^{\infty}dy\cosh y\frac{e^{-w\cosh y } f(q,\alpha_0,y)}{\cosh(qy)-\cos(\pi q ) } , \label{s1}\end{aligned}\ ] ] and the second term , @xmath297 , is given in terms of @xmath298 as shown below : @xmath299}\sin(2k\pi / q)\sin(2k\pi\alpha_0)e^{w\cos(2k\pi / q)}\nonumber\\ & & + \frac{w}{q\pi}\int_{0}^{\infty}dy\sinh y \frac{e^{-w\cosh y}g(q,\alpha_0,y)}{\cosh(qy)-\cos(\pi q ) } , \label{s2}\end{aligned}\ ] ] where the function , @xmath300 , is defined as @xmath301-\sinh(yq\alpha_0)\sin[(1-|\alpha_0|)\pi q ] . \label{s3}\end{aligned}\ ] ] as we took only positive values of @xmath160 , eq . ( [ s3 ] ) does not need to depend on @xmath280 . however , if we took negative values of @xmath160 in eq . ( [ form_grad ] ) we would get a similar expression as in eq . ( [ s3 ] ) but with the opposite sign . this suggests that in order to satisfy both possibilities one has to include absolute values of @xmath160 as showed in eq . ( [ s3 ] ) . therefore , substituting ( [ s2 ] ) into ( [ soma ] ) we finally arrive at @xmath302}\sin(2k\pi / q)\sin(2k\pi\alpha_0 ) e^{w\cos(2k\pi / q)}+\frac{w}{q\pi}\int_{0}^{\infty}dy\sinh y \frac{e^{-w\cosh y}g(q,\alpha_0,y)}{\cosh(qy)-\cos(\pi q)}.\nonumber\\ \label{s4}\end{aligned}\ ] ] note that both expression in eqs . ( [ form_grad ] ) and eq . ( [ s4 ] ) are odd function of @xmath160 , as it should be . this is only possible if we take absolute values of @xmath160 as showed in eq . ( [ s3 ] ) . note also that by using eq . ( [ soma1 ] ) , instead of eq . ( [ s2 ] ) , we would obtain the same expression ( [ s4 ] ) . the vev of the current densities along the extra dimensions , @xmath303 for @xmath304 , are given by @xmath305 which can be written as @xmath306 this expression goes to zero by taking the limit , @xmath310 . moreover , at the coincidence limit , @xmath311 , the integral over @xmath70 provides a result equivalent to . in addition the integrals over the other components of momentum along the extra dimensions , @xmath312 with @xmath313 , provide finite results similar to the one given by . using again the results and for the sum over the modified bessel function and identifying the first term with the contribution of the minkowski space in the absence of magnetic flux , we can renormalize these current densities in a manifest form by discarding this term . by doing this , we can see that the terms inside the summation and integral contain factors @xmath314 and @xmath315 , respectively . consequently , the integrals over @xmath316 of the remaining terms are finite . so , our final conclusion is that , because goes to zero at the coincidence limit and the renormalized values for the other integrals are finite in that limit , there will be no induced vacuum current densities along the extra dimensions . b. allen and e. p. s. shellard , _ on the evolution of cosmic strings _ , in _ the formation and evolution of cosmic strings : proceedings of a workshop supported by the serc and held in cambridge , 3 - 7 july , 1989 _ ( g. w. gibbons , s. w. hawking , and t. vachaspati , eds . ) , ( cambridge ) , pp . 421448 , cambridge university press , 1990 . v. b. bezerra and n. r. khusnutdinov , _ vacuum expectation value of the spinor massive field in the cosmic string space - time _ , _ class.quant.grav . _ * 23 * ( 2006 ) 34493462 , [ http://xxx.lanl.gov/abs/hep-th/0602048[hep-th/0602048 ] ] . j. spinelly and e. bezerra de mello , _ vacuum polarization of a charged massless scalar field on cosmic string space - time in the presence of a magnetic field _ , _ _ * 20 * ( 2003 ) 873888 , [ http://xxx.lanl.gov/abs/hep-th/0301169[hep-th/0301169 ] ] . j. spinelly and e. bezerra de mello , _ vacuum polarization of a charged massless fermionic field by a magnetic flux in the cosmic string space - time _ , _ int.j.mod.phys . _ * d13 * ( 2004 ) 607624 , [ http://xxx.lanl.gov/abs/hep-th/0306103[hep-th/0306103 ] ] . j. spinelly and e. bezerra de mello , _ vacuum polarization by a magnetic flux in a cosmic string background _ , _ _ * 127 * ( 2004 ) 7783 , [ http://xxx.lanl.gov/abs/hep-th/0305142 [ hep - th/0305142 ] ] . j. spinelly and e. bezerra de mello , _ spinor green function in higher - dimensional cosmic string space - time in the presence of magnetic flux _ , _ jhep _ * 0809 * ( 2008 ) 005 , [ http://xxx.lanl.gov/abs/0802.4401[arxiv:0802.4401 ] ] . l. sriramkumar , _ fluctuations in the current and energy densities around a magnetic flux carrying cosmic string _ , _ class.quant.grav . _ * 18 * ( 2001 ) 10151025 , [ http://xxx.lanl.gov/abs/gr-qc/0011074 [ gr - qc/0011074 ] ] . y. sitenko and n. vlasii , _ induced vacuum current and magnetic field in the background of a cosmic string _ , _ _ * 26 * ( 2009 ) 195009 , [ http://xxx.lanl.gov/abs/0909.0405[arxiv:0909.0405 ] ] . e. r. bezerra de mello , _ induced fermionic current densities by magnetic flux in higher dimensional cosmic string spacetime _ , _ class.quant.grav . _ * 27 * ( 2010 ) 095017 , [ http://xxx.lanl.gov/abs/0907.4139[arxiv:0907.4139 ] ] . e. bezerra de mello , v. bezerra , a. saharian , and v. bardeghyan , _ fermionic current densities induced by magnetic flux in a conical space with a circular boundary _ , _ phys.rev . _ * d82 * ( 2010 ) 085033 , [ http://xxx.lanl.gov/abs/1008.1743[arxiv:1008.1743 ] ] . a. h. castro neto , f. guinea , n. m. r. peres , k. s. novoselov , and a. k. geim , _ the electronic properties of graphene _ , _ reviews of modern physics _ * 81 * ( 2009 ) 109162 , [ http://xxx.lanl.gov/abs/0709.1163[arxiv:0709.1163 ] ] . o. oliveira , c. cordeiro , a. delfino , w. de paula , and t. frederico , _ vortex and gap generation in gauge models of graphene _ , _ phys.rev . _ * b83 * ( 2011 ) 155419 , [ http://xxx.lanl.gov/abs/1012.4612 [ arxiv:1012.4612 ] ] . s. bellucci , e. bezerra de mello , a. de padua , and a. saharian , _ fermionic vacuum polarization in compactified cosmic string spacetime _ , _ eur.phys.j . _ * c74 * ( 2014 ) 2688 , [ http://xxx.lanl.gov/abs/1311.5484[arxiv:1311.5484 ] ] . e. bezerra de mello and a. saharian , _ fermionic current induced by magnetic flux in compactified cosmic string spacetime _ , _ eur.phys.j . _ * c73 * ( 2013 ) 2532 , [ http://xxx.lanl.gov/abs/1305.6902 [ arxiv:1305.6902 ] ] . s. bellucci , a. saharian , and v. bardeghyan , _ induced fermionic current in toroidally compactified spacetimes with applications to cylindrical and toroidal nanotubes _ , _ phys.rev . _ * d82 * ( 2010 ) 065011 , [ http://xxx.lanl.gov/abs/1002.1391[arxiv:1002.1391 ] ] . e. bezerra de mello and a. saharian , _ finite temperature current densities and bose - einstein condensation in topologically nontrivial spaces _ , _ phys.rev . _ * d87 * ( 2013 ) , no . 4 045015 , [ http://xxx.lanl.gov/abs/1211.5174[arxiv:1211.5174 ] ] . s. bellucci , a. saharian , and h. nersisyan , _ scalar and fermionic vacuum currents in de sitter spacetime with compact dimensions _ , _ phys.rev . _ * d88 * ( 2013 ) , no . 2 024028 , [ http://xxx.lanl.gov/abs/1302.1688[arxiv:1302.1688 ] ] .

rayleigh - bnard system rotating uniformly about a vertical axis @xcite is an interesting problem for studying pattern - forming instabilities and bifurcations , @xcite chaos , @xcite and turbulence @xcite in addition to its potential applications in rotating cryogenics . @xcite the rotation about a vertical axis introduces coriolis force as well as centrifugal force @xcite in the flow . rubio , lopez & marques @xcite showed interesting effects of the centrifugal force even for small values ( @xmath16 ) of the froude number @xmath17 , which is a ratio of the centrifugal force to the force of buoyancy . the coriolis force is known to break the mirror symmetry of the convective flow patterns @xcite even at small rotation rates . chandrasekhar @xcite analyzed the effects of coriolis force on the onset of convection in a rayleigh - bnard system . the linear theory of chandrasekhar predicted the onset of rayleigh - bnard convection ( rbc ) through a stationary instability for fluids with prandtl number @xmath18 for all rotation rates . the selection of patterns is , however , a purely nonlinear phenomenon , and can not be determined by a linear theory . the non - linear analysis of the problem by kppers and lortz @xcite ( kl ) showed that ( i ) a set of parallel rolls was the steady state solution at the instability onset for the dimensionless rotation rates ( @xmath19 ) below a critical value @xmath20 , and ( ii ) the convection was always unsteady at the instability onset for @xmath21 . this led to interesting pattern dynamics . a set of parallel rolls was found to be replaced by another set of parallel rolls oriented at an angle of about @xmath22 with the old rolls @xcite for @xmath21 . the change in orientation of rolls was because of noise . the linear theory of chandrasekhar also predicted the possibility of oscillatory convection at the onset for @xmath23 at dimensionless rotation rates above a critical value @xmath24 . for a fluid with @xmath25 , the critical value @xmath26 of the taylor number @xmath27 is equal to @xmath28 . however , the oscillatory convection may appear at small or moderate rotation rates for low - prandtl - number fluids . the previous investigations on instabilities and bifurcations above the onset of oscillatory rayleigh - bnard convection in fluids rotating uniformly about a vertical axis @xcite were done in small simulation boxes . low - prandtl - number convection @xcite is relevant for geophysical @xcite and astrophysical @xcite problems . the estimated value of @xmath29 for earth s molten core @xcite is approximately equal to @xmath30 . the molten outer core and the inner solid core are separated by a transition zone , which itself is in liquid state . the other transition zone between the molten core and the lower mantle is also highly viscous and deformable . the transition zones on the two sides of earth s molten core have discontinuities , and they keep drifting non - uniformly . the boundary conditions for the velocity and the temperature fields are complex on these bounding surfaces . the boundaries of the convective zone " in stars ( @xmath31 ) are also complex . the numerical simulations to investigate the phenomenon of thermal convection in earth s molten core , or in the convective zones of a star , are usually done either with _ no - slip _ or _ free - slip _ ( _ stress - free _ ) boundary conditions on the velocity field . the no - slip conditions are appropriate where the bounding surfaces are rigid and stationary . the predictions of the simulations with no - slip boundary conditions have the advantage of being verified in controlled laboratory experiments . however , the stress - free boundary conditions may be a more useful approximation on a boundary between two liquids with a large difference in their viscosities . goldstein and graham @xcite achieved almost stress - free boundaries in experiments . the famous kl instability @xcite was predicted and understood for the first time by considering stress - free boundary conditions . in addition , the idealized boundary conditions are useful in developing simple models to investigate details of bifurcations qualitatively . both types of boundary conditions are useful for better understanding of the natural convection in geophysical and astrophysical problems in the absence of exact boundary conditions . we present , in this article , the results of our investigations of the role of coriolis force on the convective instabilities near the onset of rayleigh - bnard convection ( rbc ) in a low - prandtl - number fluid rotating uniformly about a vertical axis . we have considered @xmath0 , which is relevant for earth s molten core . the onset of convection is then oscillatory @xcite at relatively smaller rotation rates . our aim is to investigate the effects of the coriolis force on the onset of rbc when the conduction state becomes unstable via an oscillatory instability . we have carried out direct numerical simulations ( dns ) of three - dimensional flows in a box with thermally conducting and stress - free top and bottom surfaces for the purpose . the stress - free boundary conditions also allowed us to investigate the effects of coriolis force on the instabilities and bifurcations @xcite observed in low - prandtl - number fluids without rotation . simulations were done in a box of size @xmath1 for a wide range of taylor numbers ( @xmath2 ) . the ratio @xmath3 of the horizontal dimensions @xmath32 and @xmath33 of the simulation box was varied in a range of @xmath34 ( @xmath35 ) . this enabled us to investigate the nonlinear interactions between two sets of rolls of different wavelengths and at different orientations . the simulations done show the appearance of convection in the form of two - dimensional standing waves in smaller boxes ( @xmath6 and @xmath7 ) . we observe fluid patterns , which vary quasiperiodically in time , at the instability onset in simulation boxes with @xmath8 and @xmath9 . we observe three - dimensional ( 3d ) patterns just above the onset of convection for @xmath36 in a simulation box with square cross - section ( @xmath8 ) . these patterns describe a nonlinear superposition of two sets of mutually perpendicular rolls . the two largest fourier modes @xmath37 and @xmath38 show amplitude modulation with time . the amplitude and the frequency of modulation of both the modes are found to be equal . in a small window of taylor numbers ( @xmath39 ) , the primary instability displays temporally chaotic behavior . the chaotic flow becomes once again quasiperiodic at @xmath40 . however , the amplitude and the frequency of modulation of both the modes @xmath37 and @xmath38 are unequal in this case . the resulting patterns are cross - rolls varying quasiperiodically in time . the amplitude of @xmath38 decreases with further increase in @xmath41 . a bifurcation from three - dimensional quasiperiodic flow to two - dimensional ( 2d ) periodic convection occurs at the primary instability for @xmath42 . the 2d oscillatory rolls continue to exist until @xmath43 . we also present a low - dimensional model constructed for a square simulation box . in a box with rectangular cross - section ( @xmath9 ) , we observe the appearance of convection as 3d waves varying quasiperiodically in time for all values of @xmath41 considered here . three - dimensional ( 3d ) temporally chaotic convection is observed at the instability onset in longer boxes ( @xmath44 ) . the convective patterns at the onset show a chaotic competition between two sets of rolls oriented at an angle @xmath15 , as observed in kl instability . however , the sequence of convective patterns during kl type instability consist of rolls , rhombic patterns , and oblique wavy rolls just above the primary instability . we consider a thin layer of a boussinesq fluid of mean density @xmath45 , thermal expansion coefficient @xmath46 , kinematic viscosity @xmath47 and thermal diffusivity @xmath48 confined between two conducting horizontal surfaces separated by a distance @xmath49 . the whole system is supposed to be rotating uniformly about a vertical axis with uniform rotation rate @xmath50 . the bottom surface is uniformly heated , while the top surface is uniformly cooled . an adverse temperature gradient @xmath51 is imposed across the fluid layer by maintaining a temperature difference @xmath52 across it . we have considered convection near the onset in a low - prandtl - number ( @xmath53 ) fluid with the boussinesq parameter @xmath54 for dimensionless rotation rate @xmath55 . the ratio of coriolis force and the force of buoyancy is @xmath56 , while the ratio of centrifugal force and coriolis force is @xmath57 for all values of @xmath41 considered here . all the simulations have been done in boxes of aspect ratio @xmath58 between @xmath59 and @xmath60 . the froude number @xmath61 therefore varies from @xmath62 to @xmath63 . the effects of centrifugal force are therefore ignored here . we have considered the state of steady conduction as the basic state in a rotating frame of reference . we have not considered the situation where the basic state has a mean flow . the dimensionless hydrodynamic equations , which govern the convective flow in a boussinesq fluid , are then given by , @xmath64 & = & { \nabla}^2 \theta + v_3,\label{eq : theta}\\ \nabla{\cdot}{\bf v } & = & 0 , \label{eq : continuity}\end{aligned}\ ] ] where @xmath65 is the velocity field , and the unit vector @xmath66 is directed vertically upward . the deviations in the pressure and temperature fields from their values in the basic state of stationary conduction are denoted as @xmath67 and @xmath68 respectively . lengths , time , velocity and temperature fields are measured in units of the fluid thickness @xmath49 , the viscous diffusion time @xmath69 , @xmath70 and @xmath71 respectively . dimensionless numbers are : prandtl number @xmath72 , rayleigh number @xmath73 and taylor number @xmath41 @xmath74 @xmath75 @xmath74 @xmath76 , where @xmath77 is the gravitational acceleration . the boundary conditions at thermally conducting and stress - free bounding surfaces located at @xmath78 are : @xmath79 all fields are considered to be periodic in horizontal plane . the use of periodic boundary conditions is very popular to simulate flows in an extended layer of fluid . it captures the flow structure nicely away from the lateral walls . the periodic boundary conditions ( pbc ) allow a high resolution simulation in a box . the simulation of a flow in a large container without pbc may require prohibitive grids to resolve the flow structure . the use of pbc is the most appropriate with pseudo spectral method used here , as it facilitates the computation of terms like @xmath80 , @xmath81 exactly in numerical sense . the nature of thermal convection at the primary instability @xcite depends upon the values of the taylor number @xmath41 and prandtl number @xmath29 . the critical rayleigh number @xmath82 and the corresponding wavenumber @xmath83 for stationary convection are : @xmath84,~~\ k_{c}(ta ) = \pi \sqrt{a_{+}+a_{-}-1/2},\nonumber\\ \text{with } a_{\pm } & = & \left ( \frac{1}{4}\left\lbrace \frac{1}{2}+\frac{ta}{\pi^4}\pm \left [ \left ( \frac{1}{2 } + \frac{ta}{\pi^4}\right)^2 -\frac{1}{4}\right]^\frac{1}{2 } \right\rbrace \right)^\frac{1}{3}.\end{aligned}\ ] ] the critical rayleigh number @xmath85 and the corresponding wavenumber @xmath86 for the oscillatory convection are : @xmath87,~~\ k_{\circ } ( ta , pr ) = \pi \sqrt{b_{+}+b_{-}-1/2},\nonumber\\ \text{with } b_{\pm } & = & \left ( \frac{1}{4}\left\lbrace \frac{1}{2}+\frac{pr^2}{(1+pr)^2}\frac{ta}{\pi^4}\pm \left [ \left ( \frac{1}{2}+\frac{pr^2}{(1+pr)^2}\frac{ta}{\pi^4}\right)^2 -\frac{1}{4}\right]^\frac{1}{2 } \right\rbrace \right)^\frac{1}{3}.\end{aligned}\ ] ] the angular frequency @xmath88 of the oscillatory convection is given by the expression : @xmath89^{\frac{1}{2}}.\ ] ] it is always real , and @xmath90 in a fluid with @xmath0 , if @xmath91 . the conduction state therefore loses stability via oscillatory instability , if @xmath92 in a fluid with @xmath0 . the velocity , temperature , and pressure fields are expanded as : @xmath93 where @xmath94 , @xmath95 , @xmath96 , @xmath97 , and @xmath98 are the fourier amplitudes in the expansion of the fields @xmath99 , @xmath100 , @xmath101 , @xmath102 , and @xmath103 respectively . @xmath104 is the horizontal wave vector of the perturbations . the integers @xmath105 can take values consistent with the continuity equation ( eq . [ eq : continuity ] ) . this leads to the condition : @xmath106 we allow all the modes , which are compatible with eq . [ cont_1 ] , in our simulation . perturbations with the critical wavenumber @xmath86 for the oscillatory convection are the most dangerous ones , as soon as @xmath107 is raised above its critical value @xmath108 . in addition , perturbations with long wavelengths are always present in larger containers . we therefore set @xmath109 , and treat @xmath110 as a parameter . simulations are done in a rectangular box of size @xmath1 , where @xmath111 and @xmath112 . the ratio @xmath113 is varied in a wide range ( @xmath35 ) for each value of @xmath41 . varying the parameter @xmath34 allows us to investigate the nonlinear interactions of perturbations of different wavelengths with 2d rolls . we numerically integrate the hydrodynamic system ( eqs . [ eq : momentum]-[eq : continuity ] ) with the boundary conditions ( eq . [ bcs ] ) using pseudo - spectral method . @xcite the code is validated for low - prandtl - number rbc with and without rotation . the limit of zero - prandtl - number rbc @xcite without rotation was investigated in detail by pal et al . @xcite after validating the code with the results of thual @xcite for stress - free boundary conditions . the turbulent flow in low - prandtl - number rbc with uniform rotation was studied using the code by pharasi et al . @xcite after reproducing the results of king et al . @xcite for @xmath114 . we have used three spatial grids ( @xmath115 , @xmath116 and @xmath117 ) for our simulations . a standard fourth order runge - kutta ( rk4 ) scheme with step size @xmath118 was used for time advancement . simulations for different values of @xmath41 were carried out with small random initial conditions and were continued until a steady state was reached . the fourier modes @xmath119 and @xmath120 , which are responsible for the mean horizontal flow , are not generated in the simulations for all the values of @xmath41 considered here . the convective flow was assumed to be steady , if the temporal behavior of the box averaged kinetic energy @xmath121 and the nusselt number @xmath122 showed steady behavior for a period more than three hundred viscous diffusion time @xmath69 . the power spectra of the fourier modes @xmath37 and @xmath38 also confirmed the flow in a steady state . the symbol @xmath123 stands for the average over the simulation box . the final values of all fields in a steady state , in a given run , were also used as initial conditions for the next run when the reduced rayleigh number @xmath124 was raised in small steps ( @xmath125 ) for a fixed value of @xmath41 . the simulations showing temporally quasiperiodic states were first done on @xmath115 grids for more than three hundred viscous diffusion time . then the final values of all the fields were used as initial conditions on @xmath117 grids for more than one hundred dimensionless time unit . similar spatial grids were used for longer simulation boxes . .convective patterns for @xmath53 just above the primary instability ( @xmath126 ) in different simulation boxes . two - dimensional ( 2d ) and three - dimensional ( 3d ) standing waves ( sw ) , quasiperiodic cross - rolls ( qpcr ) , and kppers - lortz ( kl ) type patterns were observed in dns . [ cols="^,^,^,^,^",options="header " , ] the convective patterns , computed from dns , just @xmath127 above the onset of convection are listed in table [ table1 ] . for shorter periodic boxes ( @xmath128 and @xmath7 ) , we observe the onset of convection in the form of two - dimensional ( 2d ) standing waves ( sw ) . the convective patterns consist of a set of straight rolls . the locations of up - flow and down - flow alternate periodically . the patterns computed by dns in small simulation boxes are in agreement with earlier predictions @xcite near the onset of oscillatory rbc with rotation . dawes @xcite called the time - periodic 2d rolls as standing rolls . the boxes with smaller cross - section in horizontal plane do not allow two sets of rolls simultaneously . however , we observe interesting behavior in square ( @xmath129 ) and rectangular ( @xmath130 ) boxes . the convection appears in the form of a quasiperiodic competition between two sets of mutually perpendicular sets of rolls . we label them as quasiperiodic cross - rolls ( qpcr ) . similar patterns are also known as quasiperiodic standing waves @xcite or quasiperiodic standing cross - rolls @xcite in literature . we also observe a chaotic competition between two sets of wavy rolls oriented with each other at an angle near the primary instability in larger simulation boxes ( @xmath131 ) for all values of @xmath41 considered here . rhombic patterns , which are due to the nonlinear superposition of the two sets of wavy rolls , are observed during the change in orientation of the rolls . we label them as kl - type patterns . and @xmath38 for the patterns ( @xmath132 , @xmath133 , @xmath8 ) shown in fig . [ contour_ta_750 ] : ( a ) the variation of the fourier modes @xmath37 [ blue ( black ) curves ] and @xmath38 [ pink ( gray ) curves ] with time , ( b ) the same as in ( a ) but for a shorter time interval , ( c ) the power spectral density ( psd ) of the modes @xmath37 [ blue ( black ) curves ] and @xmath38 [ pink ( gray ) curves ] , and ( d ) the projection of the phase space on the @xmath134 plane.,width=14,height=11 ] we now present the results computed from dns in a box with square ( @xmath8 ) cross - section , and discuss them . the contour plots of the temperature field for @xmath0 at different instants of time are displayed in fig . [ contour_ta_750 ] . it shows different patterns due to time dependent competition between two sets of mutually perpendicular rolls . we observe square patterns whenever the amplitudes of two sets of rolls become equal . the patterns appear as cross rolls for most of the time . the positions of up - flow and down - flow alternate in time . the temporal variations of the two largest fourier modes @xmath37 and @xmath38 for the patterns shown in fig . [ contour_ta_750 ] are displayed in the first row of fig . [ qp_ta_750 ] ( @xmath136 and @xmath132 ) . the amplitudes and frequencies of modulation of both the modes @xmath37 and @xmath38 are equal , but they are out of phase [ fig . [ qp_ta_750](a ) ] . the faster time variation of these two modes are compared in fig . [ qp_ta_750 ] ( b ) . the phase difference between the two modes varies slowly with time . the power spectra [ fig . [ qp_ta_750 ] ( c ) ] of the fourier modes @xmath37 [ blue ( black ) curve ] and @xmath38 [ pink ( gray ) curve ] look similar . the peaks in the power spectra appear at exactly the same frequencies , although their heights are different . the power spectra [ see the inset of fig . [ qp_ta_750 ] ( c ) ] show the two largest peaks at frequencies @xmath137 and @xmath138 ( @xmath139 ) , which differ slightly . the difference between the two frequencies @xmath140 is the frequency of amplitude modulation . there are many smaller peaks in the spectra at frequencies @xmath141 and @xmath142 , where @xmath143 is a positive integer . other frequencies are the higher harmonics of these frequencies . the projection of the phase space on the @xmath134 plane is shown in fig . [ qp_ta_750 ] ( d ) after a long time . the region of the phase space gets continuously filled as time passes . the trajectories in the actual ( @xmath117 dimensional ) phase space never return to any point . the behavior observed in fig . [ qp_ta_750 ] ( b ) -(d ) clearly shows that the convective patterns vary quasiperiodically in time . we label the patterns as quasiperiodically oscillating cross - rolls ( qpcr ) . these patterns are consistent with the results of rihai , @xcite who ruled out any time - periodic three dimensional convection in square box . similar patterns were also predicted by dawes @xcite in a low - prandtl - number fluid ( @xmath144 ) for much higher values of the taylor number ( @xmath145 ) . we observe these patterns at the onset of convection for @xmath146 at much smaller rotation rates . similar patterns are also observed in numerical simulations @xcite with no - slip boundary conditions . ) in a square box ( @xmath147 ) with the variation of taylor number @xmath41 . the plots in the first , second and third columns are for @xmath148 , @xmath149 and @xmath42 respectively . the temporal variation of the fourier modes @xmath37 [ blue ( black ) curves ] is given in the first row for three different values of @xmath41 , while the same is given for the fourier mode @xmath38 [ pink ( gray ) curves ] in the second row . the third row shows the power spectral density for the modes @xmath37 [ blue ( black ) curves ] and @xmath38 [ pink ( gray ) curves ] for three different values of @xmath41.,width=16,height=10 ] as @xmath41 is raised , the temporal variation of these modes becomes more complex , but the convection remains quasiperiodic at the instability onset . the amplitude and the frequency of modulation of the fourier modes @xmath37 and @xmath38 remain similar and out of phase for @xmath150 . the period of amplitude modulation increases with @xmath41 for @xmath150 . the first column of fig . [ psd_sq_box ] shows properties of the two largest fourier modes just above the onset ( @xmath132 ) for @xmath148 . the fourier modes become chaotic ( not shown here ) in a small window of @xmath41 ( @xmath39 ) . these modes become temporally quasiperiodic once again for @xmath151 . the first two rows of the second column of fig . [ psd_sq_box ] display the temporal variations of the two largest modes for @xmath149 and @xmath132 . the amplitudes of modulation of the modes @xmath37 and @xmath38 become different . the largest and the second largest peaks in the power spectra of these two modes appear at slightly different frequencies . however , the modulation frequency for both the modes remains the same . the third column of fig . [ psd_sq_box ] describes the convection near onset for @xmath42 . the fourier mode @xmath38 vanishes and the power spectra of the fourier mode @xmath37 shows the peak at a single frequency . this signifies a bifurcation from a three - dimensional ( 3d ) quasiperiodic convection to a two - dimensional ( 2d ) periodic convection . the state of 2d standing waves at the primary instability continues to exist until @xmath43 . ) for @xmath53 and @xmath8 : the variation of the maximum of ( a ) the nusselt number @xmath152 and ( b ) the box averaged kinetic energy @xmath153 with the taylor number @xmath41 . both @xmath153 and @xmath154 show a bifurcation at @xmath155.,width=16,height=9 ] the variation of the maximum of convective heat flux ( @xmath152 ) and box averaged kinetic energy @xmath153 with @xmath41 near the instability onset ( @xmath132 ) in a square box ( @xmath8 ) are shown for @xmath0 in fig . [ figure4 ] . the heat flux due to convection @xmath156 remains small , as expected at smaller values of the prandtl number @xmath29 [ see , fig . [ figure4 ] ( a ) ] . it increases initially with @xmath41 , and reaches a peak at @xmath157 . it then decreases slightly with increasing @xmath41 , and again starts increasing monotonically with @xmath41 . the variation of the maximum of kinetic energy @xmath153 with @xmath41 [ fig . [ figure4 ] ( b ) ] shows exactly the same behavior . the peaks at @xmath157 in the plots of both @xmath156 and @xmath153 with @xmath41 indicate a bifurcation . the largest and the second largest peaks in the power spectrum of fourier mode @xmath37 ( see , the third row of fig . [ psd_sq_box ] ) are located at frequencies @xmath158 and @xmath159 , respectively . the two frequencies differ only slightly . similarly , the power spectrum of the fourier mode @xmath38 shows the largest and the second largest peaks at frequencies @xmath160 and @xmath161 , respectively . the details of the bifurcation may be understood by studying the effects of rotation on these frequencies . the variations of @xmath158 [ blue ( black ) curve ] and @xmath160 [ pink ( gray ) curve ] with @xmath41 are shown in fig . [ bifurcation_sq ] ( a ) near the primary instability ( @xmath132 ) in a square box ( @xmath8 ) . both the frequencies for two modes remain exactly the same till @xmath162 , and become slightly different at @xmath163 [ see the inset of fig . [ bifurcation_sq ] ( a ) ] . the maxima of the nusselt number and the kinetic energy also showed peaks exactly at @xmath157 ( fig . [ figure4 ] ) . this behavior continues till @xmath11 . only the frequency @xmath158 exists for @xmath14 , which then increases monotonically with @xmath41 . figure [ bifurcation_sq ] ( b ) shows the variation of frequencies @xmath159 and @xmath161 with @xmath41 . they also remain initially equal , and then become unequal at @xmath163 . both the frequencies for two modes exist only for @xmath11 . figure [ bifurcation_sq ] ( c ) displays the variations of the maxima of modes @xmath37 [ blue ( black ) curve ] and @xmath38 [ pink ( gray ) curve ] with @xmath41 . the maxima of modes @xmath37 and @xmath38 remain exactly the same until @xmath162 . they show a peak at @xmath157 and become unequal for @xmath163 . the maximum of @xmath37 first decreases slightly , and then starts increasing once again with @xmath41 , while the maximum of @xmath38 decreases monotonically with increasing @xmath41 . the mode @xmath38 vanishes at @xmath42 . the mode @xmath37 keeps increasing monotonically with @xmath41 , as @xmath41 is raised further . figure [ bifurcation_sq](d ) shows the variation of the modulation frequency @xmath164 @xmath74 @xmath158 @xmath165 @xmath159 of the mode @xmath37 and the modulation frequency @xmath166 @xmath74 @xmath160 @xmath165 @xmath161 of the mode @xmath38 , respectively , with @xmath41 . the modulation frequencies of both mode remain exactly the same , and siege to exist for @xmath14 . the patterns show a series of bifurcations . we observe a quasiperiodic competition between two sets of mutually perpendicular rolls ( qpcr ) of equal maximum intensity at the primary instability for @xmath36 . a bifurcation occurs at @xmath167 , and the flow becomes chaotic . the convection is chaotic in a narrow range of @xmath41 ( @xmath39 ) . another type of convective flow appears for @xmath168 . it now shows a quasiperiodic competition between two set of mutually perpendicular rolls of unequal maximum intensity ( @xmath169 @xmath170 @xmath171 ) . a bifurcation from a three - dimensional quasiperiodic convection ( qpcr ) to 2d periodic convection occurs at @xmath42 . the convection shows 2d standing waves instead of 3d quasiperiodic cross - rolls at the primary instability for @xmath14 . it is interesting to note that the variation of the maximum of @xmath156 and @xmath153 shows only signature of a bifurcation at @xmath157 , when the flow becomes chaotic . figure [ e_ta_1100_1130 ] shows the temporal variations [ blue ( black ) curve ] of the spatially averaged values of @xmath172 , @xmath173 , @xmath174 and the kinetic energy @xmath175 for @xmath148 . for any time dependent flow in a forced dissipative system like rotating rbc , the amount of energy injected in the system is not equal to the energy dissipated at every instant of time . however , the time averaged values of these quantities over a suitable period remain constant . note the time averaged value of the kinetic energy @xmath176 is a finite positive constant , and therefore the same for @xmath177 is equal to zero . as @xmath177 fluctuates in time , the energy @xmath178 also varies from its average value . the energy @xmath153 and the convective heat flux ( @xmath179 ) vary quasiperiodically in time in this case . the period of amplitude modulations of @xmath172 [ fig . [ e_ta_1100_1130](a ) ] and @xmath173 [ fig . [ e_ta_1100_1130](b ) ] is equal to twice the period of amplitude modulation for @xmath174 [ fig . [ e_ta_1100_1130](c ) ] for @xmath180 . although the period of amplitude modulations for @xmath172 and @xmath173 is the same , they are out of phase . the period of amplitude modulation for @xmath153 and @xmath154 ( not shown here ) is similar to that of @xmath174 . the second column of fig . [ e_ta_1100_1130 ] displays similar plots for @xmath149 . the period of amplitude modulation is the same for all the quantities @xmath172 , @xmath173 , @xmath174 , and @xmath153 in this case . the amplitude modulations for @xmath172 and @xmath173 are again out of phase . ) near onset ( @xmath181 ) in a square box ( @xmath147 ) as the taylor number @xmath41 is varied . mid - plane temperature contour plots at different instants [ ( a)(f ) ] of 3d quasiperiodic patterns for @xmath182 display a competition between two sets of mutually perpendicular rolls . different phases [ ( g)(i ) ] of a 2d standing wave for @xmath183 are shown in the last row.,width=12,height=11 ] figure [ e_nu_ta_3000 ] displays the temporal variation of ( a ) @xmath172 , ( b ) @xmath173 , ( c ) @xmath174 , ( d ) @xmath184 , ( e ) @xmath153 and ( f ) @xmath156 for @xmath183 . all the fields become periodic . notice that all the velocity components @xmath185 , @xmath186 and @xmath187 are nonzero . a bifurcation from 3d quasiperiodic convection to 2d periodic convection is displayed in fig . [ contour_3d_to_2d ] . the upper and middle rows [ see fig . [ contour_3d_to_2d](a)(f ) ] show the contour plots of the temperature field for six different instants of time for @xmath182 , while the lower row [ fig . [ contour_3d_to_2d](g)(i ) ] displays the convective patterns at different instants of time for @xmath183 . this shows a bifurcation from 3d convection to 2d convection , which occurs at @xmath188 . a set of parallel rolls in the presence of coriolis force always has all the three velocity components nonzero . this can be easily seen from the expressions for the vertical velocity @xmath187 and vertical vorticity @xmath189 ( @xmath190 ) for a set of straight rolls parallel to the @xmath191 axis . considering only the critical modes , the fields @xmath187 and @xmath189 may be written as : @xmath192 the horizontal velocities @xmath185 and @xmath186 may be obtained using the following relations : @xmath193 the expressions for the horizontal velocities @xmath185 and @xmath186 are given by , @xmath194 this clearly shows that all the velocity components are nonzero , if the vertical vorticity is nonzero . in fact , it is the velocity @xmath186 that breaks the mirror symmetry of the steady @xmath195 rolls parallel to the @xmath191 axis . we now present the results of our simulations in a rectangular box ( @xmath197 ) . we begin with the results of dns for @xmath130 . the temporal variations of the two largest fourier modes @xmath37 and @xmath198 , the convective heat flux ( @xmath156 ) , and the kinetic energy @xmath153 near the instability onset are displayed for @xmath136 and @xmath53 in fig . the convection is temporally quasiperiodic at the onset ( see the first column of fig . [ l2 ] for @xmath199 ) . the amplitude modulations of the fourier modes @xmath37 [ fig . [ l2 ] ( a ) ] and @xmath198 [ fig . [ l2 ] ( b ) ] are out of phase . however , the convective heat flux ( @xmath156 ) [ fig . [ l2 ] ( c ) ] across the fluid layer and the kinetic energy @xmath153 [ fig . [ l2 ] ( d ) ] are in phase . the mid - plane contour plots of the temperature field at the instability onset for the simulation box with @xmath130 are displayed in fig . [ contour_eta_2 ] . the convective patterns ( fig . [ contour_eta_2 ] ) are quasiperiodically oscillating cross - rolls ( qpcr ) , as observed in a square box ( fig . [ contour_ta_750 ] ) . the period of amplitude modulation decreases with increase in @xmath200 [ see fig . [ l2](e ) - ( h ) ] . as @xmath200 is raised to a value just @xmath201 above the threshold for oscillatory convection , the flow becomes chaotic in time ( see the third column of fig . the chaotic behavior was also predicted by dawes @xcite at the onset for much higher values of @xmath41 ( @xmath202 ) in a fluid with @xmath53 . we observe the chaotic flow at a much smaller value of @xmath41 , and it occurs as a secondary instability . we observe an interesting change in the fluid pattern dynamics at the instability onset for @xmath203 ( see fig . [ contour_eta_4 ] ) . two sets of rolls inclined at an angle of @xmath204 @xmath74 @xmath205 @xmath165 @xmath206 compete with each other . the patterns vary in time chaotically . we observe a gradual transition from one set of wavy rolls oriented along the @xmath207 axis to another set of rolls oriented at an angle of approximately @xmath208 with the @xmath207 axis . we observe rhombic patterns during the transition . figure [ signal_details_eta_4 ] gives the details of the two largest fourier modes @xmath209 and @xmath210 for patterns shown in fig . [ contour_eta_4 ] . the temporal variation of both the modes [ see , fig . [ signal_details_eta_4](a ) , ( b ) ] shows chaotic behavior . the time average of both the modes is zero . the phase between the maximum or minimum of these two modes varies with time . the power spectra of @xmath209 [ blue ( black ) curve ] and @xmath210 [ purple ( gray ) curve ] are shown in fig . [ signal_details_eta_4 ] ( c ) . the details of the first largest peak [ see the inset of fig . [ signal_details_eta_4](c ) ] show a narrow band of frequencies instead of a few discrete ones . the largest peak in the power spectra for both the modes @xmath209 and @xmath210 appear at slightly different frequencies . the temporal evolution of any trajectory in the @xmath211 plane shows a chaotic behavior . the convective patterns shown in fig . [ contour_eta_4 ] have similarity with kppers - lortz instability , @xcite where the change in orientation of rolls are determined by noise . the change in orientation of rolls occur here due to the presence of the wavy modes ( e.g. , @xmath210 ) in a longer container . the waves along the roll axis are easily excited in low - prandtl - number fluids . the longer box size allows the distortion of fields easily even at slow rotation rates . the nonlinear interactions of these perturbations with the critical modes lead to the excitation of more than two independent frequencies . this causes temporally chaotic patterns at the onset of convection . the oblique wavy rolls appear as soon as the mode @xmath209 vanishes , but the value of @xmath210 remains relatively large . the wavy rolls are oriented along the @xmath207 axis when the mode @xmath210 is vanishingly small . we observe rhombic patterns when both the modes are nonzero . the angle between the two sets of wavy rolls is fixed , and is equal to @xmath15 , although the dynamics is chaotic . the convective patterns observed at the primary instability are similar in other larger containers ( @xmath131 ) . we now construct a low - dimensional model for the rayleigh - bnard convection with rotation in a square box ( @xmath8 ) . we first recast the hydrodynamic equations in a convenient form for this purpose . we operate by curl twice on eq . [ eq : momentum ] and use the equation of continuity ( eq . [ eq : continuity ] ) . the equations for the vertical velocity @xmath187 and the vertical vorticity @xmath189 are then given by , @xmath212,\label{velocity}\\ \partial_{t}\omega_{3 } & = & \nabla^{2}\omega_{3 } + \sqrt{ta } \partial_{z}v_{3 } + [ ( { \bf{\omega \cdot\nabla}})v_3 - ( { \bf{v\cdot\nabla}})\omega_3 ] . \label{vorticity}\end{aligned}\ ] ] the equation ( eq . [ eq : theta ] ) of temperature field remains unchanged . the spatial dependence of all the fields is expanded in terms of fourier series with appropriate horizontal and vertical boundary conditions . the fourier modes found to carry higher energy in the direct numerical simulations ( dns ) are retained for the construction of the low - dimensional model . the vertical velocity @xmath213 , the vertical vorticity @xmath214 and the temperature @xmath102 fields are expressed as : @xmath215\sin{\pi z } + w_{112}(t)\cos{kx } \cos{ky } \sin{2\pi z},\ ] ] @xmath216\cos{\pi z } + z_{110}(t)\cos{kx}\cos{ky}\nonumber\\ & + & z_{112}(t)\cos{kx}\cos{ky}\cos{2\pi z}+z_{200}(t)\cos{2kx}+z_{020}(t)\cos{2ky}\nonumber\\ & + & z_{130}(t)\sin{kx}\sin{3ky } + z_{310}(t)\sin{3kx}\sin{ky},\end{aligned}\ ] ] @xmath217\sin{\pi z } + \theta_{112}(t)\cos{kx}\cos{ky}\sin{2\pi z}\nonumber\\ & + & \theta_{002}(t)\sin{2\pi z}. \end{aligned}\ ] ] projecting the hydrodynamical equations ( eqs . [ velocity]-[vorticity ] ) and the equation for the temperature field ( eq . [ eq : theta ] ) onto these modes , we arrive at a set of fifteen ordinary nonlinear differential equations . fourier modes @xmath218 , which decay linearly much faster than the other modes , are then adiabatically eliminated . this leads to a nine - mode model given by : @xmath219 , \label{model_eq1a}\end{aligned}\ ] ] @xmath220 , \label{model_eq1b}\end{aligned}\ ] ] @xmath221 @xmath222 @xmath223 @xmath224 @xmath225 @xmath226 @xmath227 where @xmath228 , @xmath229 , @xmath230 , @xmath231 , @xmath232 , @xmath233 , @xmath234 , @xmath235 , @xmath236 @xmath74 @xmath237 , @xmath46 @xmath74 @xmath238 , @xmath239 @xmath74 @xmath240 , @xmath241 @xmath74 @xmath242 , @xmath243 @xmath74 @xmath244 , and @xmath245 @xmath74 @xmath246 . figure [ modes_model ] shows the temporal variation of the leading modes at four different rotation rates @xmath247 , @xmath248 , @xmath249 , and @xmath250 obtained from the low - dimensional model ( eqs . [ model_eq1a]-[model_eq5 ] ) . the results computed from the model are in good agreement with those obtained from dns at the primary instability , and in qualitative agreement with those at secondary and tertiary instabilities . at @xmath136 , both the modes @xmath37 and @xmath38 are found to oscillate quasiperiodically just above the onset of oscillatory convection . this behavior continues to exist until @xmath251 . the primary instability is temporally chaotic in a small range of the taylor numbers @xmath252 . at @xmath253 , a bifurcation occurs and both the leading modes @xmath37 and @xmath38 show quasiperiodic temporal variation with different modulation amplitudes and frequencies . the bifurcation from 3d quasiperiodic behavior into 2d periodic behavior occurs at @xmath183 . one of the roll - modes becomes zero and the other oscillates with zero mean . the threshold values for these bifurcations obtained from the model are slightly higher than those computed from dns . the results in a square box ( @xmath8 ) near the instability onset , computed from dns as well as from the low - dimensional - model , are compared in fig . [ model_dns ] for @xmath0 and @xmath136 . the amplitudes of the temporal modulation of the fourier modes @xmath37 [ blue ( black ) curve ] and @xmath38 [ pink ( gray ) curve ] just above the onset of convection ( @xmath132 ) are shown in the first column of fig . [ model_dns ] . the modulation amplitudes ( frequencies ) of both the modes computed from the model [ fig . [ model_dns ] ( b ) ] are within @xmath254 ( @xmath255 ) of those obtained from dns [ fig . [ model_dns ] ( a ) ] . dns gives larger period of amplitude modulation than that found in the model . as @xmath107 is raised for a fixed value of @xmath41 in dns , the quasiperiodic competition between two sets of rolls bifurcates into stationary square patterns at the secondary instability ( @xmath256 ) . the model also captures the stationary square patterns near the secondary instability . however , the secondary bifurcation occurs at the higher value of @xmath200 in the model . the values of the two largest modes @xmath37 and @xmath38 obtained from the model are compared with those obtained in dns for @xmath257 . they are about @xmath258 smaller than those obtained from dns [ fig . [ model_dns ] ( c ) ] . the stationary square patterns ( @xmath259 ) bifurcate into oscillating square patterns at @xmath260 , as @xmath200 is raised further . we observe qualitatively similar behavior in the model . the secondary and tertiary instabilities occur in the model at much higher values of @xmath200 . the low - dimensional model qualitatively captures the essential features of the pattern dynamics near the onset of oscillatory convection in a slowly rotating rayleigh - bnard system with @xmath8 . the model becomes worse at higher rotation rates . the model and simulations suggest that the amplitude equations need to consider the distortions of fields even at the instability onset in low - prandtl - number fluids in larger boxes and at higher rotation rates . we have investigated the effects of the coriolis force on the convective patterns near onset of oscillatory convection in a rayleigh - bnard system ( @xmath53 ) with stress - free top and bottom surfaces , and rotating uniformly about a vertical axis . we have considered several values of the horizontal aspect ratio @xmath34 . for shorter boxes ( @xmath261 and @xmath262 ) , the primary instability appears in the form of two - dimensional periodic standing waves for a wide range of the taylor number @xmath41 . the locations of up - flow and down - flow keep alternating periodically . the convective patterns at the primary instability in small containers of horizontal aspect ratios @xmath8 and @xmath263 are interesting . the patterns at the primary instability in a square box ( @xmath8 ) depend on the value of @xmath41 . the patterns may be due to a competition between two sets of rolls having identical modulation amplitudes for smaller values of @xmath41 , or due to a competition between two sets of rolls with unequal modulation amplitudes for relatively higher values of @xmath41 , or two - dimensional periodic standing waves for further higher values of @xmath41 . the convective flow also shows stationary as well as oscillatory square patterns for @xmath147 , if the rayleigh number @xmath107 is raised keeping @xmath41 fixed . the patterns at the primary instability , in a rectangular box ( @xmath9 ) , are due to quasiperiodic competition between two sets of rolls in mutually perpendicular directions for a wide range of @xmath41 . the primary instability is always found to be chaotic in rectangular boxes with @xmath131 . the patterns are due to a competition between two sets of rolls oriented at some angle in the horizontal plane . these patterns have similarity with the pattern dynamics observed due to kl instability . however , the patterns near the onset of oscillatory convection appear due to a dynamical chaos rather than the noise as in the kl instability . the chaotic competition between a set of straight rolls and another set of oblique wavy rolls is initiated due to the excitation of the wavy modes in low - prandtl - number fluids . rhombic patterns are observed during the change in orientation of the rolls . the angle @xmath204 between the two sets of chaotically competing rolls is given by @xmath264 . * acknowledgements : * we benefited from fruitful discussions with pinaki pal , priyanka maity , arnab basak , and deepesh kumar . the presentation of this work has been greatly improved by useful comments from two anonymous referees . 99 , the instability of a fluid layer of fluid heated from below and subjected to coriolis forces , " proc . a * 217 * , 306 ( 1953 ) ; y. nakagawa and p. frenzen , a theoretical and experimental study of cellular convection in rotating fluids , " tellus * 7 * , 1 ( 1955 ) . , cellular convection with finite amplitude in a rotating fluids , " j. fluid mech . * 5 * , 401 ( 1959 ) ; motion at subcritical values of rotation rayleigh number in a rotating fluid , " j. fluid mech . * 24 * , 545 ( 1966 ) ; large - amplitude bnard convection in a rotating fluid , " j. fluid mech . * 31 * , 113 ( 1968 ) . , _ hydrodynamic and hydromagnetic stability _ , oxford university press , oxford ( 1961 ) , republished by dover publications ( 1981 ) . , a study of bnard convection with and without rotation , " j. fluid mech . * 36 * , 309 ( 1969 ) . , on the influence of rotation on the convective structure in a deep homogeneous fluid , " dokl . nauk sssr * 273 * , 718 ( 1983 ) . , rotating convection : recent developments , " intl j. engng sci . * 36 * , 1421 ( 1998 ) . , convection in a low prandtl number fluid layer rotating about a vertical axis , " eur . j. mech . b - fluids * 19 * , 213 ( 2000 ) . , scaling laws for rotating rayleigh - bnard convection , " phys . e * 72 * , 056315 ( 2005 ) . , 114101 ( 2012 ) . , transition from laminar convection to thermal turbulence in a rotating fluid layer , " j. fluid . * 35 * , 609 ( 1969 ) . , the stability of steady finite amplitude convection in a rotating fluid layer , " phys . a * 32 * , 7 ( 1970 ) . , convection in a rotating layer : a simple case of turbulence , " science * 208 * , 173 ( 1980 ) . , direct transition to turbulence in rotating bnard convection , " phys . . lett . * 57 * , 2524 ( 1986 ) . , rotating rayleigh - bnard convection : kppers - lortz transition , " physica d * 51 * , 596 ( 1991 ) . , weakly nonlinear oscillatory convection in a rotating fluid , " proc . a * 436 * , 33 ( 1992 ) . , oscillatory convection in a rotating layer , " physica d * 63 * , 213 ( 1993 ) . , stable quasiperiodic solutions in the hopf bifurcation with @xmath265 symmetry , " phys . a * 262 * , 158 ( 1999 ) . , pattern selection in oscillatory rotating convection , " physica d * 147 * , 336 ( 2000 ) . , travelling wave convection in a rotating layer , " geophys . fluid dyn . * 51 * , 195 ( 1990 ) ; j.m . rodrguez , c. prez - graca , m. bestehorn , m fantz , and r. friedrich , pattern formation in convection of rotating fluids with broken vertical symmetry , " phys . rev . a * 46 * , 4729 ( 1992 ) ; t. clune and e. knobloch , pattern selection in rotating convection with experimental boundary conditions , " phys . e * 47 * , 2536 ( 1993 ) ; k. kumar , s. chaudhuri , and a. das , quasiperiodic waves at the onset of zero - prandtl - number convection with rotation , " phys . e * 65 * , 026311 ( 2002 ) . , time and length scales in rotating rayleigh - bnard convection , " phys . * 74 * , 5040 ( 1995 ) ; y. hu , r.e . ecke , and g. ahlers , convection under rotation for prandtl numbers near 1 : linear stability , wave - number selection , and pattern dynamics , " phys . rev . e * 55 * , 6928 ( 1997 ) . square patterns in rayleigh - bnard convection with rotation about a vertical axis , " phys . 81 * , 806 ( 1998 ) . , hexagonal pattern instabilities in rotating rayleigh - bnard convection of a non - boussinesq fluid : experimental results , " phys . e * 69 * , 066311 ( 2004 ) . , square patterns in rotating rayleigh - bnard convection , " phys . e * 72 * , 036307 ( 2005 ) . , bifurcation to oscillations in three - dimensional rayleigh - bnard convection , " phys . e * 56 * , 5511 ( 1997 ) ; j.d . scheel , p.l . mutyaba and t. kimmel , patterns in rotating rayleigh - bnard convection at high rotation rates , " j. fluid mech . * 659 * , 24 ( 2010 ) . , instability of rotating convection , " j. fluid mech . * 403 * , 153 ( 2000 ) . , finite size scaling of domain chaos , " phys . e * 63 * , 045201(r ) ( 2001 ) . , onset of kppers - lortz like dynamics in finite rotating thermal convection , " j. fluid mech . * 644 * , 337 ( 2010 ) . , heat transport scaling in turbulent rayleigh - bnard convection : effects of rotation and prandtl number , " phys . lett . * 79 * , 2257 ( 1997 ) ; k. julien and e. knobloch , fully nonlinear three - dimensional convection in a rapidly rotating layer , " phys . fluids * 11 * , 1469 ( 1999 ) ; r.j.a.m . stevens , j - q zhong , h.j.h . clercx , g. ahlers , and d. lohse , transitions between turbulent states in rotating rayleigh - bnard convection , " phys lett . * 103 * , 024503 ( 2009 ) ; g. ahlers , s. grossmann , and d. lohse , heat transfer and large scale dynamics in turbulent rayleigh - bnard convection , " rev . mod . phys . * 81 * , 503 ( 2009 ) . , boundary layer control of rotating convection systems , " nature * 457 * , 301 ( 2009 ) . , turbulence in rotating rayleigh - bnard convection in low - prandtl - number fluids , " phys . e * 84 * , 047301 ( 2011 ) . , stability and heat transfer of rotating cryogens . influence of rotation on the onset of convection in liquid @xmath266 , " j. fluid mech . * 129 * , 251 ( 1983 ) ; j.m . pfotenhauer , p.g.j . lucas and r.j . donnelly , stability and heat transfer of rotating cryogens . effects of rotation on heat - transfer properties of convection in liquid he , " j. fluid mech . * 145 * , 239 ( 1984 ) ; j.m . pfotenhauer , j.j . niemela and r.j . donnelly , stability and heat transfer of rotating cryogens . effects of finite cylindrical geometry and rotation on the onset of convection , " j. fluid mech . * 175 * , 85 ( 1987 ) ; m.r . ardron , p.g.j . lucas , t. onions , m.d.j . terrett , and m.s . thurlow , rotating cyrogenic platform , " physica b * 165 * & * 166 * , 55 ( 1990 ) . , effect of the centrifugal force on domain chaos in rayleigh - bnard convection , " phys . e * 73 * , 066309 ( 2006 ) . , low - prandtl - number convection in a layer heated from below , " j. fluid mech . * 102 * , 61 ( 1981 ) . , viscous and inertial convection at low prandtl number : experimental study , " europhys . lett . * 4 * , 555 ( 1987 ) . , zero - prandtl - number convection , " j. fluid mech . * 240 * , 229 ( 1992 ) . , experimental and numerical studies of convection in a rapidly rotating spherical shell , " j. fluid mech . * 580 * , 83 ( 2007 ) . , `` bifurcation and chaos in zero - prandtl - number convection , '' europhys . 87 * , 54003 ( 2009 ) . , `` patterns and bifurcations in low - prandtl - number rayleigh - bnard convection , '' europhys . lett . * 89 * , 44003 ( 2010 ) ; p. pal , p. maity , k. kumar , and s.k . dana , `` pattern dynamics near inverse homoclinic bifurcation in fluids , '' phys . e * 87 * , 023001 ( 2013 ) . , evidence against ` ultrahard ' thermal turbulence at very high rayleigh numbers , " nature * 401 * , 885 ( 1999 ) . , on the interaction between convection and magnetic fields , " astrophys . j. * 588 * , 1183 ( 2003 ) . , stability of a horizontal fluid layer with zero shear boundaries , " phys * 12 * , 1133 ( 1969 ) . , benchmarking and scaling studies of pseudospectral code tarang for turbulence simulations , " pramana * 81 * , 617 - 629 ( 2013 )

with decoding of the human genome and other eukaryotic organisms molecular biology has entered into a new era . high - throughput technologies , such as genomic microarrays can be used to measure the expression levels of essentially all the genes within an entire genome scale simultaneously in a single experiment and can provide information on gene functions and transcriptional networks @xcite . the major challenge in microarray data analysis is due to their size , where the number of genes or variables ( @xmath0 ) far exceeds the number of samples or observations ( @xmath1 ) , commonly known as the `` large @xmath0 , small @xmath1 '' problem . this takes it difficult or even impossible to apply class prediction methods ( e.g. , logistic regression , discriminant analysis ) to the microarray data . class prediction is a crucial aspect of microarray studies and plays important role in the biological interpretation and clinical application of microarray data @xcite . for the last few years , microarray - based class prediction has been a major topic in applied statistics @xcite . in a class prediction study , the task is to induce a class predictor ( classifier ) using available learning samples ( i.e. , gene expression profiles ) from different diagnostic classes . given the learning samples representing different classes , first the classifier is learned and then the classifier is used to predict the class membership ( i.e. , diagnostic class ) of unseen samples @xcite . generally , the performance of a classifier depends on three factors : the sample size , number of variables , and classifier complexity @xcite . it was shown that for the fixed sample size , the prediction error of a designed classifier decreases and then increases as the number of variables grows . this paradox is referred to as the peaking phenomenon @xcite . moreover , some well known classifiers are even inapplicable in the setting of high - dimensional data . for example , the pooled within - class sample covariance in linear discriminant analysis ( lda ) is singular if number of variables exceeds the number of samples . similarly , in logistic regression the hessian matrix will not have full rank and statistical packages will fail to produce reliable regression estimates @xcite . therefore , the number of samples must be larger than the number of variables for good prediction performance and appropriate use of classifiers . this naturally calls for the reduction of the ratio of sample size to dimensionality . there are two major ways to handle high - dimensional microarray data in the class prediction framework . the first approach is to eliminate redundant or irrelevant genes ( a.k.a . feature selection ) . the idea is to find genes with maximal discrimination performance and induce a classifier using those genes only @xcite . the most commonly used procedures of feature selection are based on simple statistical tests ( e.g. , fold change , @xmath2-test , anova , etc . ) , which are calculated for all genes individually , and genes with the best scores are selected for classifier construction @xcite . the advantages of this approach are its simplicity , low computational cost , and interpretability . an alternative approach to overcome the problem of high - dimensionality is application of dimension reduction techniques ( a.k.a . feature extraction ) . generally , the aim of dimension reduction procedures is to summarize the original @xmath0-dimensional gene space in a form of a lower @xmath3-dimensional gene components space ( @xmath4 ) that account for most of the variation in the original data @xcite . most commonly used methods for feature extraction with microarray gene expression data are principal component analysis ( pca ) @xcite , partial least squares ( pls ) @xcite , and sliced inverse regression ( sir ) @xcite . although statistical analysis dealing with microarray data has been one of the most investigated areas in the last decade , there are only a few papers addressing the development and experimental validation of new methods and techniques for microarray dimension reduction . as @xcite claimed , the high - dimensional data analysis will be one of the most important research topics in statistics in the nearest future . here , we fill this gap by proposing a latent variable approach for handling high - dimensional microarray data and show its promising potential for class prediction . the conceptual framework on latent variable modeling originates from psychometrics , starting at the beginning of the 20th century @xcite . utility of these models in biomedical research has only quite recently been recognized @xcite . by latent variable model we mean any statistical model that relates a set of observed variables to set of latent variables @xcite . a latent variable is a variable that is not directly observable but does have a measurable impact on observable variables . latent variables describe features that underlie the data . for example , a child s intelligence ( i.e. , latent variable ) is typically assessed by measuring their answers to solving problems or items ( i.e. , observed variables ) on intelligence test . the more items we ask of the child and the wider the breadth of items is included in the assessment , the more our understanding of that child s intellectual ability will be accurate . the rasch model ( rm ) , originally proposed by @xcite , is the simplest latent variable model . the idea behind the rm is that the probability of getting an item correct is a function of a latent trait or ability . for example , a child with higher intellectual ability would be more likely to correctly respond to a given item on an intelligence test . in psychological applications , data are usually given in a matrix , with rows being participants and columns being responses to a set of items . microarray gene expression data can be represented in a similar way : columns are used to represent genes and rows are used to represent expression levels in biological samples . the rm can therefore be used to explain the observed gene expression patterns over different samples . we assume that gene expression levels vary under the influence of @xmath3 latent gene factors . the idea behind our approach is to partition @xmath0 genes into @xmath3 functional subgroups and that covariations between genes with similar expression ( i.e. , genes in the same partition ) could be described with a single gene factor . the factors in the model are latent , unobserved variables that account for the covariation among genes . it is assumed that genes with similar expression patterns might share biological function or might be under common regulatory control @xcite . moreover , it is particularly interesting to model a large set of genes as functions of fewer gene factors , because biologist believe the changes of mrna levels are due to some regulatory factors @xcite . specifically , regulatory factors are proteins that bind certain dna elements to regulate gene transcription to mrna . therefore , the gene factors obtained from latent modeling of gene expression could be interpreted as latent measurements of common regulatory factors of related genes . the number of gene factors is considered to be a meta - parameter and must be estimated or directly supplied based on researcher s prior knowledge . rm can then be used to estimate the magnitude of gene factors . class prediction using standard prediction methods can then be carried out in the reduced space by using constructed gene factors as predictor variables . the main objective of this paper is to evaluate the potential of rm - based dimensionality reduction with microarray gene expression data and investigate its prediction accuracy in the context of class prediction using lda . to validate the proposed approach we apply a parallel pca - based dimension reduction . we propose a framework for dimension reduction and class prediction with application to gene expression data as illustrated in figure [ fig : figure1 ] . our procedure consists of two basic steps : the first step is dimension reduction , in which data are reduced from high @xmath0-dimensional gene space to a lower @xmath3-dimensional gene factor space ; the second step is class prediction , in which response classes are predicted using a class prediction method on the extracted gene factors . we apply our algorithms to two publicly available microarray data sets which have been considered before by several authors . the leukemia data set @xcite contains @xmath5 tissue samples with @xmath6 genes : @xmath7 samples of acute lymphoblastic leukemia ( all ) and @xmath8 samples of acute myeloid leukemia ( aml ) . the prostate data set @xcite contains @xmath9 tissue samples with @xmath10 genes : @xmath11 prostate tumor samples and @xmath12 non - tumor prostate samples . both data sets are from affymetrix high - density oligonucleotide microarrays and are publicly available @xcite . for both data sets , the pre - processing steps are applied as follows @xcite : ( a ) thresholding , floor of @xmath13 and ceiling of @xmath14 ; ( b ) filtering , exclusion of genes with @xmath15 and @xmath16 , where @xmath17 and @xmath18 refer to the maximum and minimum intensities for a particular gene across all samples ; and ( c ) @xmath19 transformation and standardization to zero mean and unit variance . the data were then summarized by a matrix @xmath20 , where @xmath21 denotes the expression level for gene @xmath22 in sample @xmath23 . the data for each sample consist of a gene expression profile @xmath24 and a class label @xmath25 . after data preprocessing the dimension of the matrix @xmath26 was @xmath27 and @xmath28 for leukemia and prostate data set , respectively . although the procedure described here can handle a large number ( thousands ) of genes , the number of genes may still be too large for practical use . the model assessment procedure is very cpu - expensive and therefore time - consuming process , because it requires fitting the data many times due to cross - validation and re - randomization . furthermore , a considerable percentage of the genes do not show differential expression across groups and only a subset of genes is of interest . we used two different methods for feature selection in this study . first we performed an unsupervised random subset selection , consisting of @xmath29 ( @xmath30 ) genes from the set of all genes , as described by @xcite . we selected @xmath29 genes with @xmath31 from both experimental data sets . as the supervised alternative , we applied welch s @xmath2-test @xcite to embed the class information in feature selection process and thus improved prediction accuracy . @xmath2-test provides the measure of the statistical significance of changes in gene expression between classes . @xmath2-test defines the statistic @xmath2 by @xmath32 where @xmath33 , @xmath34 , and @xmath35 are the mean , sample variance , and sample size of the class @xmath36 ( @xmath37 ) for each gene , respectively . feature selection was carried out based on absolute value of the @xmath2-statistic and the top @xmath29 genes with @xmath31 were used for further processing . here we describe the core of feature extraction based on the rms and then briefly outline the pca method , which we used as a benchmark . in this subsection we first give a short overview of the rm theory in its original form , and then present its application to gene expression data . the rm is a simple latent factor model , primarily used for analyzing data from assessments to measure psychological constructs such as personality traits , abilities , and attitudes @xcite . assume that we have @xmath38 persons and @xmath39 items . let @xmath40 be the response of person @xmath23 to the item @xmath22 , where the @xmath40 is ` 1 ' if person @xmath23 answered item @xmath22 correctly and ` 0 ' otherwise . in the rm , the probability of the outcome @xmath41 is given by @xmath42 for @xmath43 and @xmath44 . @xmath45 is the person parameter that denotes the latent factor of the @xmath23th person that is measured by the item @xmath22 and @xmath46 is the item parameter , which denotes the difficulty of the item @xmath22 . difficulty of the item describes the region of the latent trait distribution where the probability of producing a specific response changes from low to high . probability of the response is monotonous in both person and item parameters . figure [ fig : figure2 ] plots the rasch probabilities as a function of the value of the latent factor ( @xmath47 ) for three different items . it can be seen that for a given item , persons with larger @xmath47 value tend to have greater probability of expressing high scores on the latent factor , and for a given person , the response probabilities are different for items with different @xmath48 values . , @xmath49 , and @xmath50 , respectively . ] the final step of the rm - based modeling is to derive latent scores from the item responses . latent score is the total score of person @xmath23 over @xmath39 items . the most common approach to calculate latent scores is to use the expectation of the posterior distribution of @xmath45 given @xmath51 with parameter estimates plugged in @xcite . details on the calculation can be found in @xcite . [ [ application - of - rasch - model - to - gene - expression - data ] ] application of rasch model to gene expression data + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in terminology of rm , we denote each gene as an `` item '' and each sample as a `` person '' . the expression level @xmath21 of gene @xmath22 in sample @xmath23 is the response of a given sample to a given gene . our main assumption is , that one - dimensional latent model may not hold for the complete set of @xmath29 genes selected in the gene filtering step ( see subsection [ subsec : feature_selection ] ) . based on the assumption that expressional similarity implies functional similarity of the genes ( and vice versa ) , we assume that genes with similar expression patterns determine one latent factor @xcite . to identify coexpressed genes we used @xmath52-means clustering @xcite to partition @xmath29 genes into @xmath3 partitions ( @xmath53 ) based on their gene expression profiles over @xmath1 samples . @xmath3-means clustering is a simple and widely used partitioning algorithm . its helpfulness in discovering groups of coexpressed genes has been demonstrated @xcite . the optimal number of @xmath3 is estimated following the procedure described in subsection [ subsec : model_selection ] . to apply the rm to the gene expression data , we need to discretize the gene expression data matrix @xmath26 into binary form . we use the median as a cut - off point for discretization . the intensity of every gene expression value is compared with the median gene expression data of the @xmath26 and assigned a ` @xmath54 ' if it is above and ` @xmath55 ' otherwise . note that gene partitioning is done before discretization step . we fit a rm for genes in each of the @xmath3 partitions respectively and calculate gene factor scores . specifically , we construct @xmath3 latent gene factors on which each gene in the @xmath52 partition is located . the measure for each sample for each gene factor is then estimated . to fit the rm for genes in the @xmath52th partition , let @xmath23 be the sample index , and @xmath22 be the gene index , for @xmath56 , and @xmath57 , and let @xmath58 be the latent gene factor for the @xmath23th sample which is determined by the genes in the @xmath52th partition , and @xmath46 be the gene specific parameter for the @xmath22th gene . class prediction is then carried out in the reduced space by using the gene factors . pca is the most commonly used technique for dimension reduction in microarray data analysis @xcite . the main idea behind the pca is to reduce the dimensionality of a data set , while retaining as much as possible the variation in the original variables @xcite . this is achieved by transforming the @xmath29 original variables @xmath59 $ ] to a new set of @xmath3 predictor variables @xmath60 $ ] , which are linear combinations of the original variables . more formally , pca sequentially maximizes the variance of a linear combination of the original variables , @xmath61 subject to the constraint @xmath62 , for all @xmath63 , where @xmath64 is covariance matrix of the original data . the orthogonal constraint ensures that the linear combinations are uncorrelated . linear combinations @xmath65 are known as the principal components . these linear combinations represent the selection of a new coordinate system obtained by rotating the original system . the new axes represent the directions with maximum variability and are ordered in terms of the amount of variation of the original data they account for . the first principal component accounts for as much of the variability in the original data as possible , and each succeeding component accounts for as much of the remaining variability as possible . computations of the weighting vectors @xmath66 involves the calculation of the eigenvalue decomposition of a data covariance matrix @xmath67 , @xmath68 where @xmath69 is the @xmath23th eigenvalue in the descending order for @xmath70 and @xmath71 is the corresponding eigenvector . the eigenvalue @xmath69 measures the variance of the @xmath23th principal component and the eigenvector @xmath71 provides the loadings for the linear transformation . the number of components @xmath3 is specified on the basis of researcher s prior knowledge or determined using dedicated procedures ( e.g. , kaiser - guttman rule , cattell s scree test , etc . ) . class prediction using standard methods can then be carried out in the reduced space by using the constructed principal components . after dimension reduction , the high dimension of @xmath29 is now reduced to a lower dimension . the original data matrix is approximated by matrix of gene factors ( @xmath72 , where @xmath73 ) , constructed by rms or pca , as described in the previous section . to avoid confusion , we use the term `` factor '' to refer to both latent factor obtained from rm analysis and principal component derived from pca . once the @xmath3 gene factors are constructed we consider prediction of the response classes . to describe the class prediction problem formally , let we have a learning set @xmath74 consisting of samples whose class is known and a test set @xmath75 consisting of samples whose class has to be predicted . denote the data matrix corresponding to @xmath74 as the learning data matrix @xmath76 , and the data matrix corresponding to @xmath75 as the test data matrix @xmath77 . the vector containing the classes of the samples from @xmath74 is denoted as @xmath78 . the goal is to build a rule implementing the information from @xmath76 and @xmath78 in order to predict the class @xmath36 of the @xmath23th sample from the test set given the gene expression profile @xmath79 : @xmath80 because our focus here is on dimension reduction , we fixed the class prediction step with lda , although other methodologies can be used @xcite . a short description of the lda method is given in the following @xcite . suppose we have @xmath3 predictor variables . the random vector @xmath81 is assumed to a multivariate normal distribution within class @xmath82 ( in our procedure @xmath83 ) with mean @xmath84 and covariance matrix @xmath85 . in lda , @xmath85 is assumed to be the same for all classes : for all @xmath36 , @xmath86 . using estimates @xmath87 and @xmath88 in place of @xmath84 and @xmath89 , the discriminant rule assign the @xmath23th new observation @xmath79 to the class @xmath90 lda has been well studied and widely used for class prediction problems . lda relies on a hypothesis of multinormality , and assumes that the classes have the same covariance matrix . although these hypotheses are rarely satisfied with real data sets , lda generally gives good results . studies have demonstrated favorable prediction performances of lda models when compared with more complicated and computationally intensive algorithms such as neural networks and tree method @xcite . for the details of the calculation we refer the reader to @xcite . the number of gene factors @xmath3 is a meta - parameter in the procedure . we estimate @xmath3 on the learning set @xmath74 using leave - one - out cross - validation ( loocv ) . loocv has been shown to give an almost unbiased estimator of the prediction error @xcite , and therefore provide a sensible criterion for our purposes . in a nutshell , a subset of @xmath29 genes is selected ( @xmath30 ) from @xmath74 , and one of the samples is left - out . the feature extraction models are fitted to all but the left - out sample ( see subsection [ subsec : feature_extraction ] ) . the fitted models are then used to predict the class of the left - out sample ( see subsection [ subsec : class_prediction ] ) . this is repeated for all samples in the learning data set @xmath91 with @xmath3 taking successively different values . the mean error rate ( _ mer _ ) over the @xmath91 runs is computed for each value of @xmath3 by @xmath92 where @xmath93 is the predicted response class and @xmath25 is the observed response class . @xmath38 is the indicator function ( @xmath94 if @xmath95 is true , @xmath96 otherwise ) . the value of @xmath3 minimizing the _ mer _ is selected and denoted as @xmath97 . we select @xmath98 in our experiments . to assess the performance of the rm- and pca - based dimension reduction methods in the framework of class prediction we perform a re - randomization study . this evaluation approach was first used by @xcite . the reader may refer to the review of @xcite on this subject . the procedure consists of the six steps as follows . 1 . for each data set , create @xmath99 random partitions into learning data set @xmath74 with @xmath91 samples and a test set @xmath75 with @xmath100 samples ( @xmath101 ) . denote @xmath76 as the learning data matrix of size @xmath102 , and @xmath77 as the test data matrix of size @xmath103 . 2 . select a subset of @xmath29 genes from the set of all genes @xmath0 from matrix @xmath76 using one of the gene selection methods , resulting in @xmath104 matrix of size @xmath105 and @xmath106 matrix of size @xmath107 ( see subsection [ subsec : feature_selection ] ) . 3 . use the learning data matrix @xmath104 to determine the number of latent factors @xmath97 , by loocv ( see subsection [ subsec : model_selection ] ) . perform dimension reduction using rm- or pca - based dimension reduction ( see subsection [ subsec : feature_extraction ] ) . let @xmath108 denote the matrix containing the factor loadings of size @xmath109 . compute the matrix @xmath110 of gene factors for the learning data set ( @xmath111 ) , and the matrix @xmath112 of gene factors for the test data set ( @xmath113 ) . refers to component loadings , and matrix @xmath114 to gene components when pca is performed . ] fit the class prediction model to the learning gene factors @xmath110 . predict the classes of samples in the test set using the fitted classifier and the test gene factors , @xmath112 ( see subsection [ subsec : class_prediction ] ) . repeat all above steps @xmath115 times with re - randomization of the whole data set . the mean error rate ( @xmath116 ) for each method is given by @xmath117 where @xmath93 is the predicted response class and @xmath25 is the observed response class . @xmath38 is the indicator function ( @xmath94 if @xmath95 is true , @xmath96 otherwise ) . although the @xmath116 is the most widely used metric for measuring the performance of the prediction systems , it considers all mispredictions as equally damaging @xcite . it has been demonstrated @xcite that , when the prior class probabilities are different , this measure is not appropriate because it does not consider misprediction cost , is strongly biased to favor the majority class , and is sensitive to class skewness . to overcome these problems the performance evaluation was also carried out via receiver operator characteristic ( roc ) analysis . for comprehensive introduction to roc analysis we refer the reader to @xcite . the roc analysis was performed by plotting true positive rate ( sensitivity ) versus the false positive rate ( 1@xmath118specificity ) at various threshold values , and the resulting curve was integrated to give an area under the curve ( _ auc _ ) value . _ auc _ is a measure of the discriminative power of the classes using the given features and classifier , and varies from @xmath119 for non - distinguishable classes to @xmath120 for perfectly distinguishable classes @xcite . the _ auc _ can be interpreted as the probability that two random samples from the two classes will be predicted correctly , and is invariant to changes in class proportions ( unlike _ mer _ ) . an @xmath121 is generally considered acceptable , @xmath122 as good , and @xmath123 as excellent prediction performance . all computations were carried out in the @xmath124 software environment for statistical computing and graphics @xcite . @xmath3-means clustering was performed using generic ` kmeans ` function . binarization of continuous gene expression values was performed by ` binarize ` function in the ` minet ` package . the function ` generate.split ` of the ` wilcoxcv ` package was used to generate random splitting into learning and test data sets . rm analysis was performed using ` ltm ` package . pca was conducted using generic ` prcomp ` function . lda was carried out using ` lda ` function in the ` mass ` package . roc analysis was performed using the ` caret ` package . the procedures described here can be reproduced using the @xmath124 scripts available from http://www2.arnes.si/~akastr1/rasch/ we illustrate the interest of rm - based dimension reduction by considering applications for the class prediction of microarray data . we compare the results from our procedure with the performance of the pca - based approach . we will consider in turn the leukemia and prostate data sets , as described previously in subsection [ subsec : data_sets ] after data preprocessing , we applied the proposed performance evaluation procedure on the leukemia data set . first , we consider @xmath31 randomly selected genes and used @xmath99 random subsets . we randomly split each subset of genes into two data sets : a learning set with @xmath125 samples and a test set with @xmath126 samples . we used loocv procedure on the learning set to determine the number of gene factors ( components in the case of pca ) , and the test set for evaluating prediction performances . in total , @xmath127 class predictions were calculated using each of the dimension reduction method based on @xmath13 randomization trials . table [ tab : table1 ] gives the estimated mean error rates ( _ mer _ ) , average values of the estimated meta - parameters ( @xmath97 ) , corresponding standard deviations , and areas under the roc curves ( _ auc _ ) with the considering methods , for different number of variables . a roc curve analysis is depicted in figure [ fig : figure3](a ) . it can be seen from table [ tab : table1 ] that _ mer _ decreases with the increase of the size of gene subsets @xmath29 . inversely , the _ auc _ increases when more predictor genes are included in model building . at any of the subsets size , the _ _ mer__s of pca - based class prediction are lower that that of rm - based procedure . _ auc _ scores suggest acceptable prediction performance for rm - based dimension reduction model and excellent performance for the pca - based model . .prediction performances of the rm- and pca - based prediction models using random gene selection on the leukemia data set with @xmath128 split of samples over @xmath13 randomization trials . [ cols=">,^,^,^,^,^,^,^ " , ] legend : rm - lda rm - based class prediction ; pca - lda pca - based class prediction ; @xmath29 number of selected genes ; _ mer _ mean error rate ; @xmath97 estimated number of gene factors ( components ) ; _ auc _ area under the roc curve . in this paper we explored the possibility of rms to solve the course of dimensionality problem arising in the context of microarray gene expression data , and evaluated its performance in class prediction framework using lda . to our knowledge , this is the first extensive validation study addressing rms for microarray data analysis . in terms of using rm - based dimension reduction of microarray data , the evaluated approach appears to be as effective as widely used pca - based dimension reduction . theoretically rms can handle a large number of genes . however , as many other multivariate methods it is challenged by large computational time and danger of over - fitting . therefore , we have used unsupervised random selection of small subset of genes and supervised welch s @xmath2-test gene selection procedure . applying random selection and using the _ mer _ and _ auc _ scores as class prediction performances , overall average values of @xmath129 ( @xmath130 ) and @xmath131 ( @xmath132 ) have been reached for leukemia and prostate data sets , respectively . we demonstrated that simple @xmath2-test improve the prediction performance significantly . considering supervised gene selection procedure , overall average values of @xmath133 ( @xmath134 ) and @xmath135 ( @xmath136 ) have been reached for leukemia and prostate data sets , respectively . the patterns of performance measures between rm- and pca - based procedures were similar , although the results suggested that rms benefit more from preliminary gene selection . compared to other studies aimed at class prediction , such as ones by @xcite , @xcite , or @xcite , our performance values are comparable . slightly better prediction performances in the case of leukemia data set confirm the fact that the biological separation between the two classes is more pronounced in leukemia data set @xcite . we have developed our approach for discretized microarray data because rm scaling assumes a binary response of a gene expression level . although our results indicate that the loss of information due to discretization step in our procedure is minimal , it is still the issue , if it is reasonable to consider gene expression discretely . referring to the work of @xcite , who demonstrated the effectiveness of the gibbs sampling to the biclustering of discretized microarray data , we argue that discretization may improve the robustness and generalizability of the prediction algorithm with regard to the high noise level in the microarray data . following @xcite the discretization of continuous gene expression levels is preferred for three reasons : ( i ) gene transcription occurs in one of a small number of states ( low high , off low high , low medium high , off low medium high , etc . ) ; ( ii ) the mechanisms of cellular regulatory networks can be reasonably approximated by primarily qualitative statements describing the relationships between states of genes ; ( iii ) discretization , as a general rule , introduces a measure of robustness against error . it is a worthwhile future project to study the performance of our method on different levels of dicretization , using polytomous latent variable models ( e.g. , partial credit model ) that could model more than two states of gene expression . the major dilemma coupled with class prediction studies is the measurement of the performance of the classifier . the classifier can not be evaluated accurately when sample size is very low . moreover , feature selection and feature extraction steps should be an integral part of the classifier , and as such they must be a part of the evaluation procedure that is used to estimate the prediction performance @xcite . @xcite reported several studies published in high impact factor journals where this issue is overlooked , and biased prediction performances are reported . to address these issues we applied loocv scheme to estimate the appropriate number of gene factors and re - randomization experimental design to stabilize performance measures . it seems that loocv is appropriate , but a more sophisticated design ( e.g. , subsampling , bootstrap sampling , @xmath137 estimator , etc . ) to determine the number of gene factors could improve the prediction performance of the rm - based approach . the vast amounts of gene expression data generated in the last decade have significantly reshaped statistical thinking and data analysis . the approach presented here can be also applied to many other problems in computational biology ( e.g. , single nucleotide polymorphism ( snp ) array data analysis ) and could be generalized to the other fields of sciences and humanities ( e.g. , health studies , risk management , financial engineering , etc . ) . hence , innovative statistical methods like the one presented in this paper are of great relevance . we have proposed a rm - based dimension reduction approach for the class prediction on microarray gene expression data . our method is designed to address the course of dimensionality and overcome the problem of `` large @xmath0 , small @xmath1 '' so common in microarray data analysis . experimental results showed that our procedure appears to be as effective as widely used pca - based dimension reduction method . we demonstrated that binarization of continuous gene expression levels does not affect prediction performance of the classifier . we showed that appropriate gene selection is crucial before dimension reduction is performed . we restricted our approach to the binary prediction problem , but the methodology can be extended to cover multiclass prediction . the application of our method to other prediction problems ( e.g. , regression , survival analysis ) is straightforward . we are currently working on extending this work on other latent variable models ( e.g. , partial credit model , rasch - andrich model , etc . ) . the first author was supported by junior research fellowship granted by slovenian research agency . 00 alter , o. , brown , p. o. , & botstein , d. ( 2000 ) . singular value decomposition for genome - wide expression data processing and modeling . _ proc natl acad sci unit states am _ , _ 97_(18 ) , 1010110106 . antoniadis , a. , lambert - lacroix , s. , & leblanc , f. ( 2003 ) . effective dimension reduction methods for tumor classification using gene expression data . _ bioinformatics _ , _ 19_(5 ) , 563570 . asyali , m. h. , colak , d. , demirkaya , o. , & inan , m. s. ( 2006 ) . gene expression profile classification : a review . _ curr bioinformatics _ , _ 1_(1 ) , 5573 . bellazzi , r. , & zupan , b. ( 2008 ) . predictive data mining in clinical medicine : current issues and guidelines . _ int j med informat _ , _ 77_(2 ) , 8197 . boulesteix , a .- pls dimension reduction for classification with microarray data . _ stat appl genet mol biol _ , _ 3_(1 ) . retrieved from doi:10.2202/1544 - 6115.1075 boulesteix , l . , & strimmer , k. ( 2007 ) . partial least squares : a versatile tool for the analysis of high - dimensional genomic data . _ brief bioinform _ , _ 8_(1 ) , 3244 . boulesteix , a .- l . , strobl , c. , augustin , t. , & daumer , m. ( 2008 ) . evaluating microarray - based classifiers : an overview . _ canc informat _ , _ 6 _ , 7797 . retrieved from http://www.la-press.com/evaluating-microarray-based-classifiers-an-overview-a577 bura , e. , & pfeiffer , r. m. ( 2003 ) . graphical methods for class prediction using dimension reduction techniques on dna microarray data . _ bioinformatics _ , _ 19_(10 ) , 12521258 . chen , j. j. ( 2007 ) . key aspects of analyzing microarray gene - expression data . _ pharmacogenomics _ , _ 8_(5 ) , 473482 . chiaromonte , f. , & martinelli , j. ( 2002 ) . dimension reduction strategies for analyzing global gene expression data with a response . _ math biosci _ , _ 176_(1 ) , 123144 . cordero , f. , botta , m. , & calogero , r. a. ( 2007 ) . microarray data analysis and mining approaches . _ brief funct genomic proteomic _ , _ 6_(4 ) , 265281 . dai , j. j. , lieu , l. , & rocke , d. ( 2006 ) . dimension reduction for classification with gene expression microarray data . _ stat appl genet mol biol _ , _ 5_(1 ) . retrieved from doi:10.2202/1544 - 6115.1147 de boeck , p. , & wilson , m. ( eds . ) . _ explanatory item response models : a generalized linear and nonlinear approach_. new york , ny : springer . de smet , f. , moreau , y. , engelen , k. , timmerman , d. , vergote , i. , & de moor , b. ( 2004 ) . balancing false positives and false negatives for the detection of differential expression in malignancies . _ br j canc _ , _ 91_(6 ) , 11601165 . dettling , m. ( 2004 ) . bagboosting for tumor classification with gene expression data . _ bioinformatics _ , _ 20_(18 ) , 3583 - 3593 . do , j. h. , & choi , d. k. ( 2008 ) . clustering approaches to identifying gene expression patterns from dna microarray data . _ mol cells _ , _ 25_(2 ) , 279288 . dudoit , s. , fridlyand , j. , & speed , t. p. ( 2002 ) . comparison of discrimination methods for the classification of tumors using gene expression data . _ j am stat assoc _ , _ 97_(457 ) , 7787 . dupuy , a. , & simon , r. m. ( 2007 ) . critical review of published microarray studies for cancer outcome and guidelines on statistical analysis and reporting . _ j natl cancer inst _ , _ 99_(2 ) , 147157 . fan , j. , & li , r. ( 2006 ) . statistical challenges with high dimensionality : feature selection in knowledge discovery . in m. sanz - sol , j. soria , j. l. varona , & j. verdera ( eds . ) , _ proceedings of the international congress of mathematicians _ ( pp . 595622 ) . madrid : european mathematical society publishing house . fawcett , t. ( 2006 ) . an introduction to roc analysis . _ pattern recogn lett _ , _ 27_(8 ) , 861874 . fischer , g. h. , & molenaar , i. w. ( eds . ) . ( 1995 ) . _ rasch models : foundations , recent developments , and applications_. berlin : springer . gan , g. , ma , c. , & wu , j. ( 2007 ) . _ data clustering : theory , algorithms , and applications_. philadelphia , pa : society for industrial and applied mathematics . golub , t. r. , slonim , d. k. , tamayo , p. , huard , c. , gaasenbeek , m. , mesirov , j. p. , et al . molecular classification of cancer : class discovery and class prediction by gene expression monitoring . _ science _ , _ 286_(5439 ) , 531537 . hartemink , a. j. ( 2001 ) . _ principled computational methods for the validation and discovery of genetic regulatory networks_. unpublished doctoral dissertation , massachusetts institute of technology , boston . hastie , t. , tibshirani , r. , & friedman , j. h. ( 2001 ) . _ the elements of statistical learning_. new york , ny : springer . holter , n. s. , mitra , m. , maritan , a. , cieplak , m. , banavar , j. r. , & fedoroff , n. v. ( 2000 ) . fundamental patterns underlying gene expression profiles : simplicity from complexity . _ proc natl acad sci unit states am _ , _ 97_(15 ) , 84098414 . huang , j. , & ling , c. x. ( 2005 ) . using auc and accuracy in evaluating learning algorithms . _ ieee trans knowl data eng _ , _ 17_(3 ) , 299310 . hughes , g. ( 1968 ) . on the mean accuracy of statistical pattern recognizers . _ ieee trans inform theor _ , _ 14_(1 ) , 5563 . jain , a. k. , duin , r. p. w. , & mao , j. ( 2000 ) . statistical pattern recognition : a review . _ ieee trans pattern anal mach intell _ , _ 22_(1 ) , 437 . jeffery , i. b. , higgins , d. g. , & culhane , a. c. ( 2006 ) . comparison and evaluation of methods for generating differentially expressed gene lists from microarray data . _ bmc bioinformatics _ , _ 7 _ , 359 . retrieved from doi:10.1186/1471 - 2105 - 7 - 359 larraaga , p. , calvo , b. , santana , r. , bielza , c. , galdiano , j. , inza , i. , et al . machine learning in bioinformatics . _ brief bioinform _ , _ 7_(1 ) , 86112 . li , h. , & hong , f. ( 2001 ) . cluster - rasch models for microarray gene expression data . _ genome biol _ , _ 2_(8 ) . retrieved from doi:10.1186/gb-2001 - 2 - 8-research0031 nguyen , d. v. , & rocke , d. m. ( 2002 ) . tumor classification by partial least squares using microarray gene expression data . _ bioinformatics _ , _ 18_(1 ) , 3950 . nguyen , d. v. , & rocke , d. m. ( 2004 ) . on partial least squares dimension reduction for microarray - based classification : a simulation study . comput stat data anal _ , _ 46_(3 ) , 407425 . orlando , d. a. , lin , c. y. , bernard , a. , wang , j. y. , socolar , j. e. , iversen , e. s. , et al . global control of cell - cycle transcription by coupled cdk and network oscillators . _ nature _ , _ 453_(7197 ) , 944947 . r development core team ( 2008 ) . _ r : a language and environment for statistical computing_. vienna : r foundation for statistical computing . available from http://www.r-project.org rabe - hesketh , s. , & skrondal , a. ( 2008 ) . classical latent variable models for medical research . _ stat meth med res _ , _ 17_(1 ) , 532 . rasch , g. ( 1966 ) . an item analysis which takes individual differences into account . _ br j math stat psychol _ , _ 19_(1 ) , 4957 . raudys , . measures of data and classifier complexity and the training sample size . in m. basu , & t. k. ho ( eds . ) , _ data complexity in pattern recognition _ london : springer . richards , a. l. , holmans , p. , odonovan , m. c. , owen , m. j. , & jones , l. ( 2008 ) . a comparison of four clustering methods for brain expression microarray data . _ bmc bioinformatics _ , _ 9 _ , 490 . retrieved from doi:10.1186/1471 - 2105 - 9 - 490 ripley , b. d. ( 1996 ) . _ pattern recognition and neural networks_. cambridge : cambridge university press . sheng , q. , moreau , y. , & de moor , b. ( 2003 ) . biclustering microarray data by gibbs sampling . _ bioinformatics _ , _ 19_(suppl . 2 ) , 196205 . simon , r. ( 2003 ) . diagnostic and prognostic prediction using gene expression profiles in high - dimensional microarray data . _ br j canc _ , _ 89_(9 ) , 15991604 . singh , d. , febbo , p. g. , ross , k. , jackson , d. g. , manola , j. , ladd , c. , et al . gene expression correlates of clinical prostate cancer behavior . _ canc cell _ , _ 1_(2 ) , 203 - 209 . slawski , m. , daumer , m. , & boulesteix , a .- cma - a comprehensive bioconductor package for supervised classification with high dimensional data . _ bmc bioinformatics _ , _ 9_(1 ) , 439 . retrieved from doi:10.1186/1471 - 2105 - 9 - 439 tibshirani , r. , hastie , t. , narasimhan , b. , & chu , g. ( 2002 ) . diagnosis of multiple cancer types by shrunken centroids of gene expression . _ proc natl acad sci unit states am _ , _ 99_(10 ) , 65676572 . zhang , c. , fu , h. , jiang , y. , & yu , t. ( 2007 ) . high - dimensional pseudo - logistic regression and classification with applications to gene expression data . _ comput stat data anal _ , _ 52_(1 ) ,

in recent years , there has been increased interest in spectral features that can attributed directly to the recoil atoms undergo on the absorption , emission or scattering of radiation . among topics that have been discussed that fall into this category are recoil - induced resonances ( rir ) @xcite and the collective atomic recoil laser ( carl ) @xcite . these processes appear to have much in common , although they are described quite differently . it is the purpose of this article to compare the rir and carl and to demonstrate that the rir and carl _ formalisms _ are equivalent . the reason that this equivalence is not readily apparent is two - fold . firstly , the rir have been discussed using a density matrix approach in the schrdinger picture , whereas carl has been discussed using an operator approach in the heisenberg picture . secondly , rir and carl have generally been examined for different ranges of the various parameters characterizing the atom - field interactions . in order to facilitate the discussion , it is useful to review briefly the rir and carl . since both the rir and carl relate to probe field absorption or gain in the presence of a pump field , it is perhaps best to recall the features of the probe absorption spectrum , _ neglecting all effects associated with recoil_. consider an ensemble of two - level atoms interacting with both a pump and probe field . the probe absorption or gain is monitored as a function the probe - pump detuning @xmath5 for various pump field strengths , characterized by the pump field rabi frequency @xmath6 . it is assumed that all collisional effects can be neglected and that the two - level atomic system is _ closed _ in the sense that the sum of ground and excited state populations is conserved for _ each _ velocity subclass of atoms . moreover , it is assumed that the pump field detuning @xmath7 from atomic resonance is much larger than any decay rates , rabi frequencies , or doppler shifts in the problem . in this limit , the probe absorption spectrum consists of three features , centered near @xmath8 @xcite . of primary concern here is the structure centered near @xmath9 in the absence of recoil , this line feature has a dispersion - like shape , is proportional to @xmath10 , and has a width of order of the excited state decay rate @xcite . its origin can be traced to an interference effect involving processes in which _ two _ photons are emitted into previously unoccupied vacuum mode states @xcite . as such , it is linked to spontaneous emission , rather than a conservative exchange of energy between the pump and probe fields . the fact that the two - level system is closed plays a critical role here . had the system been open , the amplitude of this component would vary as @xmath11 rather than @xmath12 and its width could be determined by some effective ground state decay rate ( e.g. inverse transit time ) rather than the excited state decay rate @xcite . the rir refer to a class of spectroscopic features in nonlinear spectroscopy that can be attributed to a recoil - induced opening of an otherwise closed , two - level system @xcite . in the presence of recoil , the atomic velocity is changed on the absorption or emission of radiation . as a consequence the sum of ground and excited state populations for specific velocity classes is no longer conserved . in the limit of large detuning @xmath13 , the rir can be interpreted as arising from raman transitions between different center - of - mass states @xcite . if the pump and probe fields have propagation vectors * k*@xmath14 and * k*@xmath15 , respectively , then probe absorption occurs on a raman transition between center - of - mass momentum states @xmath16 and @xmath17 and gain between states @xmath18 and @xmath19 where @xmath20*k*@xmath21*k*@xmath15 . depending on the detuning @xmath5 between the probe and pump fields , one of these processes is favored . for @xmath22 , there is probe gain , for @xmath23 there is probe absorption , and for @xmath24 the nonlinear probe absorption and gain vanish . in general , the rir have been formulated as a _ stationary _ process in which the field achieves a steady - state value at each position @xmath25 - gain occurs as the probe field propagates through the active medium . steady state is achieved by assuming that there is some effective ground state decay rate or , alternatively , by assuming that the doppler width associated with the raman transitions , @xmath26 where @xmath27 is an atomic mass and @xmath3 is the width of the momentum distribution , is larger than the recoil frequency , @xmath28 @xcite . the gain is normally expressed as @xmath29 , where @xmath30 is the probe field s amplitude and @xmath31 is a constant proportional to the atomic density . the point to note here is that the derivative of @xmath30 depends _ locally _ on @xmath32 in contrast to the rir , carl has been formulated as a _ transient _ problem . it is assumed that the pump and probe fields are modes of an optical cavity in which the atoms are located . as a result of the atom - field interaction , the probe field builds up in the cavity as a function of time . in the limit of large @xmath33 one obtains an equation of the form @xmath34 , where @xmath35 where is a constant proportional to the atomic density and @xmath36 is a kernel . if @xmath37 , the time derivative of @xmath30 can depend on the past history of the field , rather than locally on the field this leads to very different dynamics than those encountered in the rir . one finds threshold conditions for probe gain that depend nonlinearly on the atomic density , as does the gain itself . carl is a cooperative effect in the sense that gain occurs only for some minimum atomic density for most values of the detuning @xmath5 . it is possible to have gain for @xmath38 which is not possible for the rir . on the other hand , if @xmath39 , the equation for @xmath40 becomes local and the gain characteristics of carl become identical to those of the rir . thus , although both the rir and carl can be formulated for arbitrary ratios of @xmath41 , one can label the rir limit as one in which @xmath42 and the carl limit as one in which @xmath43 @xcite@xmath44 this is the nomenclature that is adopted in this article . the carl limit can be stated as , @xmath45 , which places it at or near the _ subrecoil limit _ of atom cooling . it should be noted that verkerk and bonifacio @xcite and bonifacio _ et al . _ @xcite have shown that the carl equations reproduce the rir results in the limit that @xmath46 ; however the approach they followed differs considerably from the one to be presented herein . it is important to recognize that it is possible for @xmath40 to depend locally @xmath47 , even in the carl limit , @xmath48 . for example , if one includes ground state decay with rate @xmath49 , a local dependence occurs if @xmath50 as a consequence , in the presence of decay , the rir limit can be stated as one in which @xmath42 _ or _ @xmath51 and the carl limit as one in which @xmath52 _ and _ moreover , the build up of the probe field at very early times always varies linearly with the atomic density and can be viewed as a local process . in this sense , the theory of rir is always valid at early times - the specific conditions are given in sec . however , when @xmath52 and @xmath54 , the gain coefficient in carl depends nonlinearly on the atomic density for times @xmath55 @xcite , and the local dependence no longer holds . in sec . ii , the basic equations are derived and the rir and carl limits of these equations are obtained in sec . iii . an effective potential for ground state atoms is written which helps to categorize the rir and carl limits . the case of equal pump and probe frequencies is analyzed in sec . iv using an effective five - level atom . when expressed in this form the calculation has a range of applicability beyond the recoil problem . the results are discussed in sec . v. special emphasis is placed on the distinction between atom bunching and matter gratings . experimental implications of the results are also explored . the problem under consideration consists of a pump field and a probe field interacting with an ensemble of two level atoms . in some applications in carl , it may be of interest to use a quantized description of the fields to follow the build - up of the probe field from noise , but , in the present discussion , the fields are taken to be classical . the pump field , @xmath56 , \label{1}\ ] ] has polarization * @xmath57*@xmath14 , slowly - varying electric field amplitude @xmath58 , propagation vector @xmath59 and frequency @xmath60 , while the probe field , @xmath61 , \label{2}\ ] ] has polarization * @xmath57*@xmath15 , slowly - varying electric field amplitude @xmath62 , propagation vector @xmath63 and frequency @xmath64 . as a result of the nonlinear interaction with the fields , the probe field can be amplified . if the pump field is initially much more intense than the probe field , as is assumed , pump depletion during the early stages of probe amplification can be neglected . since the calculation in this paper is limited to the early stages of probe amplification , i take the pump field amplitude to be constant , @xmath65 the dynamics of probe field amplification depends on the specific atom - field geometry . one can envision situations in which the probe field amplitude is constant in time but varies in space , or is constant in space but varies in time . the first case is the one considered generally in the rir , in which _ cw _ pump and probe fields interact with atoms in a cell or trap . the probe field amplitude increases in the direction of @xmath63 as it propagates through the medium , but is assumed to have evolved to a stationary state . the second case is the one considered generally in carl , in which the fields correspond to field modes of an optical cavity and the atoms are located in the cavity . for a ring cavity , the probe field intensity is spatially isotropic , but increases in time @xcite . in certain limits ( to be noted below ) , the spatial gain coefficient of the rir multiplied by the speed of light coincides with the temporal gain coefficient of carl . in other limits , the dynamics of carl amplification differs from the spatial build - up of the field in the rir . to compare the rir and carl , it is convenient to adopt the cavity model and assume that the probe field amplitude is a function of time only , @xmath66 all cavity losses are neglected , as is any ground state decay . although carl is referred to as a collective effect since conditions for carl gain depend on the atomic density , each atom in the sample , on average , interacts with the fields in an identical manner . thus , it is sufficient to write the hamiltonian for a single atom interacting with the fields . the dependence of the field gain on atomic density is included automatically in the coupled maxwell - bloch equations . in the dipole and rotating wave approximations , the hamiltonian for our system is @xmath67 , \label{3}\ ] ] where @xmath68 is the atomic center - of - mass momentum , @xmath69 is the transition frequency between the ground state @xmath70 and excited state @xmath71 of the atom , @xmath72 @xmath73 , @xmath74 is the rabi frequency of field @xmath75 , and * @xmath76*@xmath77*@xmath76*@xmath70 is a dipole moment matrix element . terms related to spontaneous emission are not included in the hamiltonian ( [ 3 ] ) , for reasons to be discussed below . the hamiltonian determines the time evolution of the atom , but the atomic evolution must be coupled to the field evolution via maxwell s equations to arrive at a closed set of equations . the time evolution of the ( complex ) probe field amplitude is linked to the polarization of the medium which , in turn , is determined by the atom - field interaction . the medium s polarization can be expressed in terms of atomic density matrix elements as @xmath78 , \label{5}\ ] ] where @xmath79 is the total number of atoms . as a result of the nonlinear atom - field interaction , the density matrix element @xmath80 can be written quite generally ( see below ) as @xmath81 where @xmath82 in the slowly varying amplitude and phase approximation , it follows from maxwell s equations and eqs . ( [ 5 ] ) and ( [ 6 ] ) that the evolution of the probe field is given by [ 8 ] @xmath83 where @xmath84 is the component of the polarization @xmath85 varying as @xmath86 combining eq . ( [ 8b ] ) with eq . ( [ 4 ] ) yields @xmath87 where @xmath88 an equation for @xmath89 is obtained by solving the schrdinger equation for the hamiltonian ( [ 3 ] ) . it is convenient to expand the wave function of the system as @xmath90 where @xmath91 @xmath92 , @xmath93 , @xmath94 , and @xmath95 is the atomic - state eigenfunction associated with state @xmath75 . note that the momentum state amplitudes , @xmath96 , are defined in an interaction representation . the state amplitudes , @xmath97 , in the normal representation are related to those in the interaction representation by @xmath98 and density matrix elements in momentum space are given by @xmath99 where @xmath100 and @xmath101 when the wave function ( [ 11],[12 ] ) is substituted into schrdinger s equation with the hamiltonian ( [ 3 ] ) , one finds that the state amplitudes @xmath96 evolve as [ 15 ] @xmath102\,a_{2}({\bf p+}\hbar { \bf k}_{\mu } , t ) \label{15a } \\ \dot{a}_{2}({\bf p},t ) & = & -i\sum_{\mu = 1,2}\chi _ { \mu } ( t)\,\exp [ -i\delta _ { \mu } t - i\omega _ { k_{\mu } } t+i{\bf k}_{\mu } \cdot { \bf p}t / m]\,a_{1}({\bf p-% } \hbar { \bf k}_{\mu } , t ) , \label{15b}\end{aligned}\ ] ] where @xmath103 is an atom - field detuning and @xmath104 is a frequency associated with atomic recoil . in terms of these state amplitudes , the density matrix element @xmath105 can be written as @xmath106 together , eqs . ( [ 9 ] ) , ( [ 6 ] ) , ( [ 18 ] ) and ( [ 15 ] ) form a closed set of equations which can be solved numerically to obtain the time evolution of the probe field . to simplify the analysis of both the rir and carl , it is assumed that @xmath107 where @xmath108 is the excited state decay rate and @xmath109 is the most probable atomic speed . if inequalities ( [ 19 ] ) hold , both the rir and carl can be interpreted in terms of _ stimulated _ processes involving the pump and probe fields - spontaneous emission plays a negligible role . both the pump and probe fields are turned on and brought adiabatically to their initial values at @xmath110 in a time long compared with @xmath111 , but small compared with all other evolution times in the problem . thus , at @xmath110 , the values for the rabi frequencies are @xmath112 while the density matrix for the atoms is taken as @xmath113 where @xmath114 is the sample volume , @xmath115 is the initial momentum distribution , and the bar indicate an ensemble average @xcite . this initial density matrix corresponds to a wigner function @xmath116 , which is the product of the momentum distribution times a uniform spatial density ( recall that this is a single particle density ) . the calculation can still be carried out using an amplitude picture . the initial condition ( [ 21 ] ) is invoked once the density matrix element @xmath117 has been evaluated . when inequalities ( [ 19 ] ) hold , it is possible to solve eq.([15b ] ) adiabatically for the upper state amplitude in terms of the lower state amplitude . explicitly , one finds @xmath118\,\exp [ -i\delta _ { \mu } t - i\omega _ { k}t+i{\bf k}_{\mu } \cdot { \bf p}t / m]\,a_{1}({\bf p-}\hbar { \bf k}_{\mu } , t ) , \label{22}\ ] ] where @xmath119 and terms of order @xmath120 and @xmath121 have been ignored . substituting this expression back into eq . ( [ 15a ] ) , one obtains @xmath122\,a_{1}({\bf p+}\hbar { \bf k}_{\nu \mu } , t ) , \label{23}\ ] ] where @xmath123 since @xmath124 can take on the values [ 0,@xmath125 only , it is clear that a solution to this equation can be written as @xmath126 combining eqs . ( [ 26 ] ) , ( [ 18 ] ) , ( [ 25 ] ) and ( [ 21 ] ) , one finds that the density matrix element @xmath80 is given by @xmath127 equation ( [ 26 ] ) proves that the general form for @xmath80 is correctly given by eq . ( [ 6 ] ) . extracting the coefficient of @xmath128 provides us with the value of @xmath129 needed in eq . ( [ 9 ] ) . two types of terms in the summation vary as @xmath130 , those with @xmath131 and @xmath132 and those with @xmath133 and @xmath134 denoting the @xmath132 term as @xmath135 , the @xmath136 as @xmath137 , and using the normalization condition @xcite @xmath138 one finds @xmath139 with [ 29 ] @xmath140 . \label{29b}\end{aligned}\ ] ] the first term represents linear dispersion of the medium . the second term , which is at the heart of the rir and carl , represents a scattering of the pump field off the atomic density distribution created by both the pump and probe fields . we now have all the ingredients necessary to derive the rir and carl equations . in the large detuning limit , both the rir and carl equations are most easily derived using an effective hamiltonian for ground state atoms . an effective hamiltonian of the form @xmath141 \label{30}\ ] ] leads to the evolution equation ( [ 23 ] ) for the ground state amplitude @xmath142 the second term in the hamiltonian is the spatially homogeneous light shift potential and is not of interest here . the third term is the potential formed by the pump and probe fields which gives rise to rir and carl . the depth of the potential , @xmath143 , is a critical parameter@xmath44 if the energy spread of the atoms @xmath144 is much larger than @xmath145 , any bunching or spatial modulation of the atomic density will be negligibly small , independent of the detuning @xmath5 . this is generally the case at thermal energies , making it all but impossible to observe rir and carl using atoms in cells at or above the doppler limit of laser cooling ( see discussion in sec . v ) . assuming that @xmath146 , there are still two limiting cases to consider@xmath44 if @xmath144 is much greater than the level spacings in the potential @xmath145 , then the motion of the atoms can be treated classically . by approximating the potential as that of a harmonic oscillator and assuming that the spread @xmath147 is of order @xmath148 , one can estimate that the validity condition for the motion to be treated classically is @xmath149 . this is the limit considered by bonifacio and coworkers @xcite and will be referred to as carlsc , with the sc denoting the semiclassical limit . a theory of carl in which the atomic motion is fully quantized has been given recently by moore and meystre @xcite . it will be referred to as carlq , when it is necessary to distinguish between carlsc and carlq . the theory of rir has been developed using a quantized description of the center - of - mass motion ; consequently , the rir and carlq theories should produce identical results , and the rir and carlsc theories should produce the same results when @xmath150 . it should be noted that , although the atomic motion in the _ optical potential _ is treated classically in carlsc , the gain in carlsc results from recoil - induced processes . this point is discussed in more detail in sec . v. in calculating expectation values of operators that are diagonal in the internal states , one can neglect contributions from the excited state since the excited state population is assumed to be negligibly small . the carl equations are derived using the heisenberg equations of motion while the rir are derived using a density matrix approach . i now show that these methods yield equivalent results , as they must . to make connection with carl , one defines a heisenberg operator , bunching parameter , @xmath151 . using eqs . ( [ 14 ] ) , ( [ 25 ] ) , ( [ 21 ] ) , and ( [ 29b ] ) , one can write the average value of the bunching parameter as @xcite [ 31 ] @xmath152\,e^{-ie_{p,\left| { \bf p}+\hbar { \bf q}\right| } t/\hbar } \\ & = & v^{-1}\sum_{n=-\infty } ^{\infty } e^{i(2n+1)\omega _ { q}t}\int d{\bf p\,}w(% { \bf p}){\bf \,}e^{-i{\bf p\cdot q}t / m}s_{n}({\bf p}+n\hbar { \bf q}% , t)s_{n+1}^{\ast } [ { \bf p}+(n+1)\hbar { \bf q},t ] \label{31e } \\ & = & \frac{\delta ve^{i\delta t}}{\chi _ { 1}}\tilde{\rho}_{21}(t;b ) , \label{31f}\end{aligned}\ ] ] where the brackets denote a quantum - mechanical average and the bar an ensemble average . therefore , by combining eqs . ( [ 9 ] ) , ( [ 28 ] ) , ( [ 29 ] ) , and ( [ 31f ] ) , one finds @xmath153 where @xmath154 is the atomic density . this is one of the basic carl equations in the limit of large @xmath7 @xcite@xmath44 the remaining carl equations , obtained from the heisenberg equations of motion with the hamiltonian ( 30 ) are [ 33 ] @xmath155 . \label{33b}\end{aligned}\ ] ] these equations do not form a closed set since the equation for @xmath156 involves @xmath157 . one must generate a series of equations for these higher moments or use some approximation techniques ( such as assuming classical motion in the effective potential ) to obtain a solution to the equations . the linear dispersion parameter @xmath158 results in a shift of the probe field frequency . by setting @xmath159 in eqs . ( [ 32],[33 ] ) , it is easy to see that the detuning @xmath5 enters only in the combination @xmath160 thus , one can redefine the detuning to include the dispersion shift and replace eq . ( [ 32 ] ) by @xmath161 to examine the small signal gain regime , one can solve eqs . ( [ 33 ] ) in perturbation theory . to zeroth order in the fields , [ 34 ] @xmath162 ^{\dagger } & = & \exp \left\ { -i\left ( \frac{{\bf % q\cdot p}}{m}+\omega _ { q}\right ) t\right\ } b^{\dagger } ( 0)=b^{\dagger } ( 0)\exp \left\ { i\left ( \frac{{\bf q\cdot p}}{m}+\omega _ { q}\right ) t\right\ } , \label{34c}\end{aligned}\ ] ] where the last equalities in eqs . ( [ 34b],[34c ] ) follow from the commutation properties of @xmath68 and @xmath163 note that @xmath164 and @xmath165 are normal schrdinger operators . substituting eqs . ( [ 34b],[34c ] ) into ( [ 33b ] ) , one finds , to first order in @xmath166 , @xmath167 and , when this result is substituted in eq . ( [ 33a ] ) , one obtains [ 36 ] @xmath168 ^{2 } \\ -\chi _ { 2}(t^{\prime \prime } ) \chi _ { 1}^{\ast } \exp ^{i\left ( \frac{{\bf % q\cdot p}}{m}+\delta + \omega _ { q}\right ) ( t^{\prime } -t^{\prime \prime } ) } \end{array } \right ) \label{36b}\end{aligned}\ ] ] where the commutation properties of @xmath169 [ or @xmath170 $ ] and @xmath171 and the relationship @xmath172 have been used to go from ( [ 36a ] ) to ( [ 36b ] ) . the bunching parameter @xmath173 can now be averaged with the initial density matrix ( [ 21 ] ) . the average of eq . ( [ 34b ] ) for @xmath174 vanishes as does the first term in eq . ( [ 36b ] ) for @xmath175 . on averaging the remaining term in eq.([36b ] ) for @xmath175 with the initial density matrix ( [ 21 ] ) , interchanging the order of integration , and carrying out the integration over @xmath176one obtains @xmath177 . \label{37}\ ] ] i will return to this equation after showing that an identical equation is reached using a density matrix approach . the rir are usually calculated in the context of a density matrix approach . from eq . ( [ 31a ] ) , it follows that @xmath178 which gives the alternative form for eq . ( [ 33 m ] ) , @xmath179 to complete the rir equations , one uses the hamiltonian ( [ 30 ] ) , along with eqs . ( [ 11])-([14 ] ) , to obtain density matrix equations @xmath180 \nonumber \\ & & -i\frac{\chi _ { 1}\chi _ { 2}^{\ast } ( t)e^{i\delta t}}{\delta } \left [ \rho _ { 11}({\bf p-\hbar q},{\bf p}^{\prime } ; t)-\rho _ { 11}({\bf p},{\bf p}% ^{\prime } { \bf + \hbar q};t)\right ] , \label{40}\end{aligned}\ ] ] where @xmath181 . it is easily verified that equations for the quantities @xmath182 and @xmath183 , obtained using eq . ( [ 40 ] ) , are identical to eqs . ( [ 33 ] ) . thus , the rir density matrix equations are _ totally equivalent _ to the corresponding operator carl equations . to evaluate @xmath184 in the perturbative limit using eq . ( [ 38 ] ) , one sets @xmath185 in eq . ( [ 40 ] ) and replaces the density matrix elements in the right hand side of that equation by their zeroth order values , @xmath186 to obtain @xcite @xmath187 . \label{42}\end{aligned}\ ] ] this equation is consistent with the rir picture of raman transitions between center - of - mass momentum states differing by @xmath188 . integrating eq . ( [ 42 ] ) over * p * and @xmath189 and using eq . ( [ 38 ] ) , one reproduces eq . ( [ 37 ] ) for @xmath184 . since @xmath190 as @xmath191 @xmath192 , the probe gain in the rir and carl is a recoil - induced effect . the value of @xmath184 given by eq . ( [ 37 ] ) , which determines the small signal gain , depends critically on the ratio @xcite @xmath193 where @xmath194 is the width of the momentum distribution and @xmath109 is the most probable atomic speed . for @xmath195 ( rir limit ) , the integrand in eq . ( [ 37 ] ) is rapidly oscillating except when @xmath196 allowing one to approximate the integral as [ 44 ] @xmath197 where it has been assumed that the momentum distribution can be written as the product of a one - dimensional , symmetric distribution , @xmath198 and a two - dimensional distribution , @xmath199 for momenta @xmath200 transverse to the @xmath201 direction . equation ( [ 44 ] ) is the general rir result , in which the bunching parameter depends locally on the field @xmath202 . by combining eqs . ( [ 44 ] ) and ( [ 33 m ] ) , one sees that , in general , the exponential build up of the field is not linear with time . however , if @xmath203 and @xmath204 , the expression for @xmath184 reduces to @xmath205 where @xmath206 indicates a principal value , enabling one to combine eqs . ( [ 150 ] ) and ( [ 33 m ] ) to obtain the small signal gain @xcite @xmath207 the gain depends linearly on the atomic density and the pump field intensity . the gain is positive for @xmath208 negative ( absorption ) for @xmath209 and vanishes at @xmath210 . this is a single particle gain in that each atom contributes separately to the gain and there is no threshold condition for gain that depends on atomic density . the situation changes in the carl limit @xmath211 . in that limit one can replace @xmath115 in eq . ( [ 37 ] ) by @xmath212 to obtain @xmath213 . \label{46}\ ] ] the bunching parameter now depends nonlocally on the field amplitude @xmath214 ; that is , it depends on the past history of @xmath202 . equation ( [ 46 ] ) is equivalent to the differential equation @xmath215 subject to the initial conditions @xmath216 . the small signal behavior is determined by the coupled equations ( [ 47 ] ) and ( [ 33 m ] ) . these equations have already been analyzed by bonifacio and coworkers for carlsc with neglect of the @xmath217 term in eq . ( [ 47 ] ) @xcite and by moore and meystre @xcite including this term . gain occurs if one of the roots of the cubic indicial equation , obtained from the coupled equations , ( [ 47 ] ) and ( [ 33 m ] ) , has a positive real value . in terms of the quantity @xmath218 the indicial equation is @xmath219 and the condition for gain is @xcite @xmath220 ^{3/2}<27q/2 . \label{49}\ ] ] if @xmath221 , gain occurs in the range @xmath222 and the gain coefficient equals @xmath223 on the other hand , for @xmath224 , gain occurs for @xmath225 in contrast to the rir limit@xmath44 moreover in both cases , the gain depends nonlinearly on the atomic density and there is a threshold condition for all values of @xmath226 as such , carl is a collective effect in the sense that gain does not occur for atomic densities below a certain critical value . for @xmath227 which is the resonance condition for raman transitions between center - of - mass states having momenta @xmath228 and @xmath229 there is gain irrespective of the value of @xmath230 for @xmath231 and @xmath232 , the gain varies as @xmath233 , and the probe field undergoes exponential gain linear in time for @xmath234 for early times , one can evaluate @xmath235 in eq . ( [ 37 ] ) at @xmath236 , and combine eq . ( [ 37 ] ) with eq . ( [ 33 m ] ) to obtain @xmath237 as long as the magnitude of the second term is much less than unity , the build - up of the probe field depends linearly on the density and the dependence of @xmath238 on @xmath239 is approximately local . in this sense , the rir limit is always valid for sufficiently small @xmath240 or atomic density ( since @xmath241 is proportional to the density ) . it follows from eq . ( [ 101 ] ) that the rir limit is always valid if @xmath242 . if @xmath243 and @xmath244 , the rir limit is valid when @xmath245 the condition @xmath246 is less restrictive than @xmath247 for @xmath248 . an interesting limiting case is one in which @xmath250 ( carl limit ) , and the pump and probe field frequencies are degenerate , @xmath251 ; @xmath252 [ recall that the detuning has been redefined to include the shift arising from linear dispersion ] . when @xmath210 , the probe gain vanishes identically in the rir limit @xmath195 , but grows exponentially for carlsc . for carlq , it follows from eq . ( [ 49 ] ) that the probe gain is exponential provided that @xmath253 this qualitative difference between the rir and carl limits is reason enough to consider the @xmath210 case in some detail , but it is not the only reason . the existence of exponential gain when @xmath210 is surprising at first glance . it would seem that processes in which a pump photon is absorbed and a probe photon emitted would be exactly cancelled by processes in which a probe photon is absorbed and a pump photon emitted , owing to the symmetry of the interaction when @xmath243 . it is the purpose of this section to investigate the origin of exponential gain in carl when @xmath210 . calculations are carried out in a perturbative limit , that is , to lowest order in the pump field intensity . as is shown below , the results are also applicable to a wider range of problems . in perturbation theory , there are five states that enter the calculation , starting from atoms in their ground state having @xmath228 . the relevant states are @xmath254 @xmath255 @xmath256 @xmath257 , having energies @xmath258 , @xmath259 , @xmath260 , and @xmath261 , respectively ( recall that @xmath262 . it is convenient to relabel these states as @xmath263 the energy levels associated with these states are shown in fig . the pump field drives the @xmath264 to @xmath70 and @xmath265 to @xmath266 transitions while the probe field drives the @xmath264 to @xmath267 and @xmath70 to @xmath71 transitions . the hamiltonian for the system can be obtained by expanding the hamiltonian ( [ 3 ] ) in a momentum - state basis for the subspace ( [ 51 ] ) . using the relationship @xmath268 , one finds @xmath269 note that this level scheme could equally well describe an atom in a stark field , driven by circularly polarized pump and probe fields . the energy @xmath270 would then correspond to the stark shifts of the @xmath271 ground state , zeeman sublevels . since decay is neglected , the calculation is most conveniently carried out using state amplitudes rather than density matrix elements . no ensemble average is needed here since we start in an eigenstate of momentum , @xmath272 it is convenient to work in a field interaction representation in which the state amplitudes @xmath273 ( @xmath274 are written as [ 53 ] @xmath275^{1/2}\tilde{a}_{0}(t)\,\delta ( % { \bf p } ) \label{53a } \\ a_{1}({\bf p},t ) & = & [ ( 2\pi \hbar ) ^{3}/v]^{1/2}\tilde{a}_{1}(t)\,e^{-i\omega t}\,\delta ( { \bf p}-\hbar { \bf k}_{1 } ) \label{53b } \\ a_{-1}({\bf p},t ) & = & [ ( 2\pi \hbar ) ^{3}/v]^{1/2}\tilde{a}_{-1}(t)\,e^{-i% \omega t}\,\delta ( { \bf p}-\hbar { \bf k}_{2 } ) \label{53c } \\ a_{\pm 2}({\bf p},t ) & = & [ ( 2\pi \hbar ) ^{3}/v]^{1/2}\tilde{a}_{\pm 2}(t)\,\delta ( { \bf p}\mp \hbar { \bf q } ) , \label{53d}\end{aligned}\ ] ] with density matrix elements given by [ 54 ] @xmath276 in eq . ( [ 8a ] ) for the probe field evolution , one need the component of polarization @xmath277 varying as @xmath278 . from eqs . ( [ 53])and ( [ 54 ] ) , it follows that @xmath279\,e^{i({\bf k}_{2}{\bf \cdot r}-\omega t ) } \label{55}\ ] ] where the * @xmath76 * s are dipole matrix elements . for the level scheme corresponding to the states ( [ 51 ] ) , @xmath280 thus , the problem reduces to calculating @xmath281 and @xmath282 to order @xmath283 . = 16 cm = 16 cm before undertaking this calculation , it is useful to obtain an expression for the time evolution of the probe field s energy density @xmath284=@xmath285 . using eqs . ( [ 8a ] ) , ( [ 4 ] ) , ( [ 55 ] ) , and ( [ 56 ] ) , one finds [ 57 ] @xmath286 . \label{57c}\end{aligned}\ ] ] the last line follows from the density matrix equations of motion for the hamiltonian ( [ 52 ] ) and can be given an obvious physical interpretation . population of state @xmath71 implies gain on the probe field , while population in states @xmath267 or @xmath287 implies loss for the probe field . the time rate of change in energy density is simply the difference @xmath288 $ ] multiplied by the product of energy , @xmath289 gained or lost by the probe field in each elementary process and the atomic density . in the adiabatic limit considered in this paper , the excited state population is negligible and one has @xmath290 one might expect that @xmath291 , owing to the symmetry of the level scheme , but we will see that this is not the case . assuming that the detuning of the fields @xmath7 from the ground to excited state atomic resonance is sufficiently large to adiabatically eliminate the excited states , one uses the hamiltonian ( [ 52 ] ) to show that the excited state amplitudes are given by [ 59 ] @xmath292 and that the state amplitudes @xmath293 , @xmath294 evolve as [ 60 ] @xmath295 it is a straightforward exercise to solve eqs . ( [ 60 ] ) in perturbation theory to third order in the fields , starting from @xmath296 . one finds [ 61 ] @xmath297 where the superscripts denote the order of the fields . consider , first , eq . ( [ 58 ] ) for the probe field intensity , which depends on @xmath298 by inspecting eqs . ( [ 61d],[61e ] ) , one can understand the manner in which @xmath299 can grow more rapidly than @xmath300 . suppose @xmath202 acquires a positive , time - dependent phase as a result of the atom - field interaction . in this case , the quantity @xmath301 appearing in the integrand of eq . ( [ 61d ] ) varies more slowly than the quantity @xmath302 appearing in the integrand of eq . ( [ 61e ] ) . as a result , state @xmath71 population builds up more rapidly than that of state @xmath266 , leading to probe gain . in other words , the nonlinear phase modulation of the probe field effectively favors the @xmath264 to @xmath71 transition over the @xmath264 to @xmath266 if the phase is positive . to examine the phase and amplitude build - up of the field , one can combine eqs . ( [ 8a ] ) , ( [ 4 ] ) , ( [ 54])-([56 ] ) , and ( [ 61 ] ) to obtain @xmath303 \nonumber \\ & = & i\frac{\omega d^{2}}{\epsilon _ { 0}\hbar } n\frac{\left| \chi _ { 1}\right| ^{2}}{\delta ^{2}}\int_{0}^{t}dt^{\prime } \chi _ { 2}(t^{\prime } ) \sin \left [ \omega _ { q}(t - t^{\prime } ) \right ] \nonumber \\ & = & iq\omega _ { q}\int_{0}^{\omega _ { q}t}dy^{\prime } \chi _ { 2}(y^{\prime } /\omega _ { q})\sin \left [ ( \omega _ { q}t - y^{\prime } ) \right ] . \label{63}\end{aligned}\ ] ] of course , the integral equation ( [ 63 ] ) is equivalent to the coupled equations ( [ 33m],[37 ] ) with @xmath9 however , this form of the equations is convenient for obtaining the early time development of the field and for determining the conditions under which the field undergoes exponential gain . by solving eq . ( [ 63 ] ) iteratively , one finds @xmath304-iq\sin ( \omega _ { q}t ) \nonumber \\ & & -(q^{2}/2)[4\cos ( \omega _ { q}t)+\omega _ { q}t\sin ( \omega _ { q}t)-4]+o[q^{3}(\omega _ { q}t)^{2 } ) ] . \label{64}\end{aligned}\ ] ] for @xmath305 , @xmath306 . at early times , the phase is positive , favoring probe gain . whether or not the field continues to grow depends on the value of @xmath230 if @xmath307 then @xmath239 is slowly varying compared with @xmath308 and the integral in ( [ 63 ] ) can be evaluated asymptotically to yield @xmath309\chi _ { 2}\text{% , } \label{65}\ ] ] which implies that @xmath310}\chi _ { 2}(0 ) \label{66}\ ] ] for @xmath221 , the intensity of the probe field remains approximately constant . in some sense , this can be viewed as the rir limit , since the time development of the field is approximately local , and there is neither absorption nor gain . on the other hand , for @xmath311 the field build - up occurs sufficiently rapidly to insure that @xmath312 \nonumber \\ & = & 4% % tcimacro{\func{re}}% % beginexpansion \mathop{\rm re}% % endexpansion \left\ { i\frac{\left| \chi _ { 1}\right| ^{2}\chi _ { 2}^{\ast } ( t)}{\delta ^{2}}% \int_{0}^{t}dy^{\prime } \chi _ { 2}(t^{\prime } ) \sin ( t - t^{\prime } ) \right\ } \label{67}\end{aligned}\ ] ] remains positive for all @xmath240 . in this case , there is exponential gain for the probe for times @xmath313 . in summary , the probe gain that occurs for @xmath210 and large @xmath7 is clearly _ not _ a single particle effect . it is more closely related to a propagation effect in which the phase modulation of the probe field produced by the nonlinear atom - field interaction drives the probe gain . it has been shown that the density matrix - rir and heisenberg operator - carl formalisms lead to equivalent equations . the rir and carl limits refer simply to different regions of parameter space of these equations . for a given experimental situation , one must determine whether one is in the rir limit , the carl limit , or neither limit ( as is most often the case ) . the experimental implications of the rir and carl are discussed below , but first i would like to discuss the distinction between the terms matter grating used in discussions of the rir and atomic bunching used in discussions of carl . the term matter grating refers to spatially modulated atom distributions resulting from a nonlinear atom - field interaction . the term atomic bunching refers to a redistribution or focusing of atoms in an optical potential . for the rir and carl , _ these terms are synonymous_. if recoil effects are neglected , that is , if the center - of - mass motion is treated classically from the outset , the _ total atomic density _ is conserved for each velocity subclass of atoms ( neglecting collisions ) . a homogeneous atomic density remains homogeneous to all orders in the atom - field interaction . recoil effects allow for a modification of the total atomic density . whether one calls this atomic bunching or matter grating production is a matter of personal preference @xcite . the key point is that the modification of the total atomic density results entirely from effects related to recoil on the absorption , emission , or scattering of radiation . to lowest order in the atom - field interaction , the matter grating or atomic bunching consists of a spatial modulation of the atomic density having period 2@xmath314 . with increasing field strength , higher order spatial harmonics are produced , corresponding to higher order matter gratings or focusing or atomic bunching . of course it is possible to derive an effective potential of the form @xmath315 \right\ } \left| 1\right\rangle \left\langle 1\right| \label{68}\ ] ] _ without _ quantization of the center - of - mass motion and to consider _ classical _ motion in this potential . however , since the effective potential is proportional to @xmath316 , any changes in the atomic density vanish in the classical limit . this is in contrast to bunching in the free electron laser where the effective potential does not vanish in the classical limit . are there situations where matter gratings are produced by fields without any contribution from recoil ? the answer to this question is yes , provided one considers the matter gratings associated with individual internal atomic states rather than the total atomic density . for example , for an ensemble of _ stationary _ , closed , two - level atoms there is a spatial modulation in the _ population difference_@xmath317between the excited and ground states produced by the pump and probe fields . scattering of the pump field from this spatially modulated population difference leads to a dispersion - shaped probe absorption profile centered at a probe - pump detuning @xmath210 in the limit of large atom - field detuning @xmath318 the amplitude of the dispersion profile varies as the square of the pump field intensity and its width is equal to the excited state decay rate @xcite . it is also possible to have gain profiles with a width corresponding to some effective ground state decay time if one considers open systems in which spontaneous emission also plays a role . for example , if the ground state consists of two hyperfine states and the fields drive transitions between only one of these ground states and an excited state , then , as a result of spontaneous emission , both ground state sublevel populations can be spatially modulated ( although the _ total _ atomic density - the sum of all ground and excited state populations - remains constant in the absence of recoil ) . it is possible to monitor the atomic gratings in specific ground state hyperfine levels by using radiation that couples only the targeted ground state sublevel to an excited state . finally , i would like to discuss some experimental implications of the rir and carl . to observe the spectral features of the rir and carl , it is best to use cold atoms in a collisionless environment . moreover , to isolate the effects of interest , one must use experiments which involve closed , two - level transitions or signals that depend only on total atomic state density . the rir have already been observed in several experiments involving laser - cooled atoms @xcite . the experiments in which evidence for carl was claimed @xcite were all carried out under conditions ( collisions , radiation trapping , large doppler widths ) which do not favor observation of carl . the results of these experiments can be explained by conventional theories in which recoil is neglected @xcite . as was mentioned previously , the carl limit is essentially a subrecoil limit . although the carl limit has been stated as @xmath319 , when @xmath320 a better estimate for the carl regime is @xmath321 @xcite . for densities of order 10@xmath322atoms / m@xmath323 , it is possible to achieve values of @xmath324 of order 100 - 1000 . thus , it may be possible to observe carl for atoms cooled to or somewhat above the sub - doppler limit of laser cooling . the linear gain coefficient in this case is of order @xmath325s@xmath326 , which must exceed any cavity loss . is it possible to observe the rir in a thermal vapor using pump - probe spectroscopy ? the linear absorption rate is @xmath327 which implies that the ratio of the recoil - induced gain @xmath328 [ eq . ( [ 45 ] ) ] to the linear absorption rate is of order @xmath329 , where @xmath330 is the kinetic energy of the atoms . for sub - doppler cooled atoms , this ratio can be greater than unity , but it is small at room temperature . still it might be possible to use modulation techniques to isolate the rir contribution to the probe absorption . for the rir signal to be larger than the background , dispersion - like contribution to the probe absorption that varies as @xmath331 , one requires that the ratio @xmath332 be greater than unity . this can be achieved at room temperature for sufficiently large @xmath333 . perhaps the best way to observe carl would be to use an atomic beam , transversely cooled below the recoil limit . the beam can be passed through a cavity with some transit time @xmath334 it is not difficult to extend the theory to allow for a transit time effects through an effective decay rate @xmath49 for ground state atoms . for subrecoil cooled atoms , eq . ( [ 46 ] ) is replaced by @xmath335 , \label{69}\ ] ] eq . ( [ 47 ] ) by @xmath336 and the indicial equation ( [ 200 ] ) by @xmath337 with the inclusion of decay , the rir limit , in which @xmath338 depends locally on @xmath339 is @xmath340 . thus the carl limit occurs when @xmath341 . decay tends to diminish the gain parameter when @xmath248 , but can actually reduce the gain threshold when @xmath342 . for example , if @xmath343 , the threshold condition is reduced from @xmath344 to @xmath345 if @xmath346 . it is a pleasure to acknowledge helpful discussions with b. dubetsky and j. l. cohen . this work is supported by the u. s. office of army research under grant no . daag55 - 97 - 0113 and the national science foundation under grants nos . phy-9414020 and phy-9800981 . r. bonifacio and l. de salvo , nucl . methods a * 341 * , 360 ( 1994 ) ; r. bonifacio , l. de salvo , l. m. narducci , and e. j. dangelo , phys . a * 50 * , 1716 ( 1994 ) ; r. bonifacio and l. de salvo , opt . commun . * 115 * , 505 ( 1995 ) ; l. de salvo , r. cannerozzi , r. bonifacio , e. j. dangelo and l. m. narducci , phys . rev . a * 52 * , 2342 ( 1995 ) ; r. bonifacio , l. de salvo , l. m. narducci , and e. j. dangelo , unpublished . the line width actually results from the convolution of a lorentzian having width equal to the excited state decay rate , @xmath347 and a gaussian having width @xmath348 , where @xmath349 is the magnitude of the difference of pump and probe field wave vectors and @xmath109 is the most probable atomic speed . a more precise statement of the carl limit is @xmath350 , where @xmath241 is defined by eq . ( [ 48 ] ) . this condition follow from eq . ( [ 37 ] ) , where evaluating @xmath214 in the integrand at @xmath351 requires that @xmath214 be slowly varying with respect to @xmath352 and the fact that @xmath202 varies as @xmath353 for @xmath248 and @xmath354 , where @xmath355 is a constant of order unity . similarly , the region of simple exponential gain for carl occurs for times @xmath356 and , when ground state decay is included , the field dependence is local if @xmath340 .

the dynamics of relaxing quantum systems in sufficiently strong laser fields , which are neither stationary nor monochromatic , is an important aspect of the theory of interaction between atoms and photon fields . this problem is of particular practical significance because of dramatic advances in the precision of measurements presently attainable in spectroscopic experimental studies of hydrogenic and few particle atoms . among the simple atoms that attracted attention in the last few years one should mention : _ hydrogen _ and its isotopes @xcite , _ positronium _ @xcite , denoted @xmath2 , _ muonium _ ( see @xcite and references therein ) , denoted @xmath3 , and _ helium _ atom @xcite . the measurements of the ground state hfs and the lamb shift are of particular interest for these systems . high precision spectroscopy of low lying transitions in hydrogenic atoms offers a unique opportunity to test qed calculations and to refine both the fundamental constants and various properties of respective nuclei . this urges further theoretical developments intended to describe the well established photon - induced processes in simple atoms with much higher precision than techniques used so far are able to provide , in order to allow a proper interpretation of newly available experimental data . in the present paper the stepwise 3-photon ionization of muonium in a _ non monochromatic _ laser field with a _ time dependent _ amplitude is studied . it has been recognized already a long time ago ( see @xcite for a more detailed discussion ) that the @xmath1 transition offers unique opportunities for high precision spectroscopy due to the narrow natural line width , @xmath4 , of the @xmath5-state . in the hydrogen atom , for example , @xmath6 hz which enables the quality factor of @xmath7 to be achieved already today , as well as suggests future resolutions of order of 1 part in @xmath8 . experimentally , the @xmath1 transition can be induced doppler free by absorbing two photons from two counter propagating laser beams . in particular , this scheme is presently used at the rutherford appleton laboratory @xcite for determination of the @xmath1 energy separation in muonium to the @xmath9 mhz accuracy . this study offers an opportunity to improve the present knowledge of the muon / electron mass ratio and , thereby , of the muon mass itself . in the course of this investigation , two counter propagating pulsed laser beams with almost identical intensities , @xmath10 , and the same photon energies , @xmath11 ( @xmath12 nm ) are used . thus , the frequency @xmath13 is such that it is tuned into resonance with the 2photon doppler free @xmath14 transition . unlike the hydrogen case however , the @xmath1 transition in muonium is hard to be observed directly , i.e. by detecting radiation emitted as a result of the @xmath5state deexcitation . this is due to the fact that the appropriate line intensities in muonium happen to be weaker by several orders of magnitude , owing to much lower densities at which muonium atoms can be produced . furthermore , the latter reason necessitates the use of higher laser intensities ( by a factor of @xmath15 , at least ) that are required for a reasonable signal strength . this demands a _ pulsed _ rather than a cw laser source ( as in the case of hydrogen @xcite ) to be employed . the @xmath5state of muonium is detected therefore via its photoionization by the third photon absorbed from one of the laser beams @xcite . schematically , the experiment in muonium is presented in figure [ fig1](a ) where @xmath16 stands for the resonant energy ( frequency ) which drives the 2photon @xmath1 transition . despite the fact that the laser frequency can be calibrated , in principle , to a rather high accuracy , there are a number of systematic error sources each of which proves to be essential for precise determination of the @xmath1 energy separation in muonium . in this paper we report the results of the theoretical study intended to allow for one of the most important among these effects , that is , the time - dependent frequency variation ( chirp ) of the laser field . this phenomenon arises due to the rapid refraction index variation of the _ laser media _ and is hardly avoidable , especially with powerful pulsed lasers , unless some technical developments are made in order to compensate the frequency alteration @xcite . as will be demonstrated below , the chirped laser signal immediately makes its appearance in appreciable spurious shifting and broadening of spectral / photoionization lines , as well as leads to a noticeable distortion of the line shapes . to estimate the scale of the effect considered , one should note , for example , that the chirp induced shift of the center of the ionization line is roughly equal to a characteristic magnitude of the chirp itself . for the @xmath1 experiment in muonium , the latter can typically amount to @xmath17 , thus leading to the relative shift of the line s center by @xmath18 . the effects of such an order are usually completely neglected in atomic physics . nonetheless , these happen to be crucial for studies where an absolute accuracy of @xmath19 is anticipated . it is therefore of primary practical importance to work out a relatively simple theoretical scheme which could allow one to describe certain types of multi - photon ionization phenomena in hydrogenic atoms , where an arbitrary time variation of the frequency of the laser field is accompanied by an arbitrary time modulation of its amplitude . although the problem of the hydrogenic systems interaction with monochromatic resonant laser field with / without the amplitude modulation is investigated quite fully by now ( see @xcite and references therein ) , there have been much fewer studies where both the amplitude and the frequency of the laser signal vary with time . even though , the authors confine themselves in most cases to either 1photon ( rather than 2photon ) resonant transition @xcite , or treat only somewhat special forms ( `` abrupt step '' , `` exponential field pulses '' etc . ) of the field modulation @xcite . under these circumstances , it seems very desirable to reconsider the problem as a whole , while adapting it specifically to the conditions of the @xmath1 experiment in muonium . we consider , without a loss of generality , a set of three levels : @xmath20 and @xmath21 , of a hydrogenic atom with the charge @xmath22 of its nucleus , such that their one particle energies satisfy the inequality : @xmath23 ; this rather general setup is shown in figure [ fig1](b ) . it is supposed that the given 3level system is exposed to two counter propagating ( along the @xmath24axis ) laser waves with equal time dependent ( circular ) frequencies , @xmath25 , polarizations vectors , @xmath26 , and wave vectors , @xmath27 , where @xmath28 is the fine structure constant and @xmath29 denotes the speed of light . the stationary part , @xmath30 , of the field frequency @xmath31 is assumed to be such that it is in the 2 and 1photon resonances with the pairs of states , @xmath32 and @xmath33 , respectively . this implies that @xmath34 with @xmath35 being the differences of the one particle energies . in addition , it is supposed that the @xmath36transition is forbidden between @xmath37 and @xmath38 and is allowed between @xmath38 and @xmath21 . under the resonance conditions assumed above , 3photon ionization of the system , that is , @xmath39 transition , occurs predominantly as a 2step process : ( i ) a 2photon resonant absorption from the state @xmath37 into the state @xmath38 , followed by ( ii ) a 1photon resonant transition between the levels @xmath38 and @xmath21 . this simple physical picture is a consequence of the evident estimate of the probability for an atom to absorb 3 photons of equal energies ( the atomic units , @xmath40 , are used throughout the paper ; in this system , the speed of light is equal to @xmath41 ) : @xmath42 here , @xmath43 are the 2 and 1photon resonant transition probabilities , @xmath44 is the total width of the state @xmath38 , and @xmath45 denotes some characteristic mean energy which a two fold summation over intermediate states in the exact expression to define @xmath46 ( see @xcite for a more detailed discussion ) can be reduced to . equation ( [ 3phot ] ) readily follows then from the estimate , @xmath47 , which is fulfilled for most states of atoms in moderately strong laser fields , except extraordinarily short living ones . a situation when @xmath37 stands for the ground , @xmath38 for the @xmath5 , and @xmath21 for the continuum @xmath48states , respectively , is of our primary concern here . in this case @xmath49 and @xmath50 since @xmath51 is dominated by the photoionization rate of the @xmath5level ( @xmath52 ) . accordingly , we will assume that the natural line widths of the @xmath37- and @xmath38levels are both equal to zero : @xmath53 . a straightforward generalization of the latter condition , that might be required for the treatment of _ excited _ states , can be achieved by adding appropriate imaginary parts to the energies of the levels @xcite : @xmath54 . in the semiclassical approximation , which happens to be accurate enough for our purposes , the electric fields inducing the 2 and 1photon transitions can be taken in the form @xmath55 here , the real functions @xmath56 and @xmath57 describe the ( macroscopic ) spatial inhomogeneity of the laser field and its time - dependent amplitude . these are related direct to the laser intensity @xmath58 which , along with the chirped ( circular ) laser frequency @xmath31 , is obtainable direct from measurements . for the @xmath1 experiment with muonium , @xmath58 and @xmath31 can be well approximated as @xcite : @xmath59 where @xmath60 mj is the energy within one laser pulse ; @xmath61 ns , @xmath62 mm denote the temporal and spatial dispersions of the external laser field ( in the @xmath63 plane perpendicular to the direction of the beams propagation ) . the full temporal width of the laser pulse is expressed then as @xmath64 ns . parameter @xmath65 mhz defines the magnitude of the chirp @xmath66 , being actually detected in the experiment , and @xmath67 stands for the error function @xcite . as usually happens to be true in most practical situations , both the amplitude @xmath68 and the additonal phase @xmath69 vary much slower with time than @xmath70 and @xmath31(see equations ( [ intens])-([omega ] ) ) . in addition , the magnitude of the chirp is normally much smaller than the `` base '' frequency @xmath30 . it is therefore permissible to assume that @xmath31 and @xmath68 are subject to the following general conditions @xcite : @xmath71 within the framework of the conventional time dependent perturbation theory @xcite , the exact time dependent wave function of the quasi stationary state @xmath72 in the field is sought in the form @xmath73 where @xmath74 , being the amplitudes to find an atom in one of the unperturbed states @xmath75 , satisfy the following well known system of coupled differential equations : @xmath76 here , @xmath77 denotes the matrix element of @xmath78 being the operator of the `` particle laser field '' interaction . due to the given below reasons , @xmath78 can be taken in the form @xmath79 where @xmath80 is the dipole operator of an electron ( or a muon ) and @xmath81 denotes the electric field ( [ fields ] ) seen by the atom in its rest frame . for particular laser parameters of interest , the accuracy of the above approximation follows from the evident estimate : @xmath82 . this implies that the contribution of the quadrupole terms @xmath83 , which originate from the spatial _ macroscopic _ inhomogeneity of @xmath81 , is expected to be a factor @xmath84 smaller than the contribution of the ( dipole ) terms in ( [ dip_appr ] ) . almost the same estimate holds also for corrections arising due to the quadrupole component @xmath85 of the photon field itself . this is despite the fact that the states @xmath37 and @xmath38 are coupled , in accord with the initial assumptions of our model and equation ( [ dip_appr ] ) , in only the second order of the perturbation theory in @xmath78 , i.e. by taking into account at least the 2photon absorption / emission . indeed , unlike the second order dipole coupling , the 1photon quadrupole @xmath86 in the amplitude . if @xmath87 , as it is assumed throughout the paper , then the terms of such a type are known to be almost negligible @xcite . in our case , in particular , their relative contribution is of the order of @xmath88 , as can be shown by averaging the transition amplitude over some time interval : @xmath89 . the same `` off - resonant '' argument is fully applicable to magnetic transitions as well , thus elucidating why the @xmath90 and @xmath91contributions are suppressed by a factor @xmath92 as compared to the dipole one . by virtue of ( [ a])-([dip_appr ] ) , one can readily retrieve , after some conventional algebra , the following form for effective quasiclassical one particle operators which couple two given states of reference @xcite : @xmath93 all summations are performed here over complete set of one - particle atomic orbitals . equations ( [ vba])-([vab ] ) describe the real transitions , whereas ( [ vbb ] ) defines the _ ac stark _ shift of the level @xmath38 ( or @xmath37 , if @xmath94 ) which arises due to interaction of the atom with the incident radiation . note that the sum in ( [ vbb ] ) coincides , by definition @xcite , with the dipole dynamic polarizability of the state @xmath38 at the frequency @xmath95 . also , it should be pointed out that only the leading terms , as given by equations ( [ vba])-([vbb ] ) , were retained in the course of derivation . the most important among neglected terms are of the form : @xmath96 } + c.c . \right ) \right|^2 \simeq \frac{1}{\omega_{\rm l}^4 } \left|\dot{e}(t)\right|^2u^2(\bi{r}).\ ] ] apparently , this contribution is smaller than @xmath97 by a factor of @xmath98 , as follows from ( [ intens])-([omega ] ) , in accord with ( [ cond_omega])-([cond_e ] ) . as was assumed above , there are no other discrete resonant atomic levels ( either 1 or 2photon ) except @xmath20 . this enables the system ( [ a ] ) to be reduced to these two states only , and @xmath99 , is ignored , since we are mainly concerned here with the lowest @xmath100- and @xmath5states . ] while describing the 1st stage of the two step 3photon ionization process under consideration . indeed , an inclusion of the off - resonant states would give rise in the right - hand side of equations ( [ ala1])-([cc1 ] ) to the rapidly @xmath101varying terms @xmath102 . after temporal averaging ( see above ) , these happen to be of order of @xmath103 , and can therefore be omitted to the accuracy adopted in this study . on introducing the new functions , @xmath104 \right\ } , \label{alpha_c}\ ] ] and replacing the matrix elements in ( [ a ] ) by equations ( [ vba])-([vbb ] ) , one arrives at the following system of equations : @xmath105 , \\ \fl \label{cc1 } \dot{c}(t ) & = & - \rmi \:d_{b , a } \frac{4\pi}{c } i(\bi{r},t ) - \rmi \ : \left [ d_{b , b } \frac{4\pi}{c } i(\bi{r},t ) - \left ( 4\pi \delta \nu + \dot{\alpha}(t)+2 \dot{\phi}(t)\right ) \right ] c(t),\end{aligned}\ ] ] where @xmath106 is the time independent frequency detuning off the resonance , @xmath107 is the intensity of the single laser beam and @xmath108 denote the effective matrix elements given by the sums in ( [ vba])-([vbb ] ) . it has been taken into account here that the 3level system is actually probed by a superposition of two laser fields ( [ fields ] ) , rather than by a single one ; this results eventually in doubling all matrix elements . in addition , we have neglected a contribution due to the 2photon absorption from _ each _ of two laser beams . this contribution comes from the second order interactions of an atom with only one of two light waves and makes its appearance finally in two doppler broadened terms , i.e. depending on a @xmath109component of atomic velocity in the laboratory frame , in the expression to define ionization probabilities . however , after averaging over @xmath109 , as would be required within the framework of a systematic approach , these terms turn out to be small compared , at the center of the photoionization line @xcite , with the doppler free ones which originate from ( [ ala1])-([cc1 ] ) . these arguments provide the ground for the approximation made , although the effects caused by atomic movement in the media are beyond the scope of our present consideration . it is relevant to point out here that the matrix elements , @xmath108 , in ( [ ala1])-([cc1 ] ) are @xmath110dependent . in particular , @xmath111 has a non zero imaginary part . the latter allows for a non zero photoionization rate of the @xmath38state , due to the action of the photon field , since @xmath30 is supposed to exceed the photoionization threshold @xmath112 . indeed , as was noted above , @xmath111 is proportional to the dynamic ( tensor ) polarizability @xmath113 of the level @xmath38 , which is complex - valued if @xmath114 ( see our recent paper @xcite for a more detailed discussion ) . @xmath115 and @xmath116 define then the _ ac stark _ shift of the level ( see equation ( [ vbb ] ) ) and its decay probability ( @xmath117 ) via the single photoionization . this result is in agreement with the optical theorem @xcite : @xmath118 , where @xmath119 is the fine structure constant and @xmath120 denotes the total photoionization cross section of the state @xmath38 . numerical calculations show that the single photoionization happens to be the main channel of the @xmath38level depopulation . it is therefore crucial for the correct description of the time evolution of the 3level system that this depopulation mechanism is taken in fact into account in ( [ ala1])-([cc1 ] ) through the matrix elements @xmath108 . an allowance for the 2nd step of the process considered can be made by augmenting the system ( [ ala1])-([cc1 ] ) with an additional equation that explicitly describes the @xmath121 coupling through the single photoionization of the state @xmath38 . given by ( [ cont ] ) , this extra equation is nothing else but a differential form of the conservation law which determines the probability balance between the states of reference . moreover , equations ( [ ala1])-([cc1 ] ) can be decoupled by substituting @xmath122 of ( [ ala1 ] ) into ( [ cc1 ] ) to finally get @xmath123 c(t ) \\ \lo+ \rmi d_{b , a } \frac{4\pi}{c } i(\bi{r},t ) c^2(t ) , \nonumber \\ \label{ala2 } \fl \dot{\alpha}(t ) = \frac{4\pi}{c } i(\bi{r},t ) \left [ d_{a , a } + d_{b , a } c(t ) \right],\\ \label{cont } \fl \dot{w}_c(t ) = \frac{2}{\omega_{\rm l } } \sigma^{(\gamma)}_b ( \omega_{\rm l } ) i(\bi{r},t ) \left| c(t)\right|^2 \exp \left\{2\im \alpha(t ) \right\},\end{aligned}\ ] ] where @xmath124 is the probability for an atom to be ionized by the time @xmath101 , via absorption of 3 photons of equal energy . under the actual experimental conditions @xcite , initial population of the level @xmath38 , i.e. before the laser pulse is shot , is usually rather low . this is because the major fraction of the muonium atoms is produced in the _ ground _ state @xmath37 where the population is proportional to @xmath125 . in view of this physical picture and by virtue of ( [ alpha_c ] ) , it is natural to impose the following initial conditions on the functions @xmath126 : @xmath127 these enable @xmath128)$ ] to be determined uniquely . being of our primary concern , this quantity constitutes the required 3photon resonant ionization probability of an atom at infinitely large positive times , i.e. when the interaction between the single laser pulse and the system has already ceased . as a function of the laser frequency detuning @xmath129 , @xmath128)$ ] describes , and will be therefore used as a synonym of , the 3photon ionization line profile / shape . for brevity , the latter will be denoted as @xmath130)$ ] where the argument @xmath131 $ ] is introduced in order to indicate explicitly that the line profile depends on the form of the chirp . in particular , if @xmath132 then @xmath133)$ ] will be denoted as simply @xmath134 . equations ( [ cc2])-([cont ] ) are well suited for numerical calculations with arbitrary functions defining the spatial and temporal distributions of the laser pulse . in addition , the given system is rather convenient for analytic treatment , as will be demonstrated in subsection [ subsec : anal_cons ] . in particular , one can develop further perturbative expansion of equations ( [ cc2])-([ala2 ] ) in terms of the intensity @xmath58 . this yields analytic formulae which , after setting @xmath135 , agree with those formerly derived in @xcite within the framework of a 2level model , for _ unchirped _ laser signals . despite a relatively simple form of the appropriate equations , however , a not very straightforward numerical integration is still required in order to calculate , for example , photoionization line profiles in most practical situations , let alone the fact that the result of the work @xcite is valid for rather weak laser intensities only . note that equations ( [ cc2])-([cont ] ) are free from the latter restriction . furthermore , unlike our approach employing the probability amplitudes , the problems similar to those considered in this paper are usually treated ( e.g. see @xcite ) in the formalism of the density matrix @xcite . under the physical conditions adopted here , these two approaches are equivalent . to facilitate the adequate comparison however , it is relevant to present here an alternative form of ( [ cc2])-([cont ] ) , by rewriting the system in terms of the density matrix elements : @xmath136 and @xmath137 , denoted as @xmath138 . this yields : @xmath139 here , 5 unknown functions are subject to the following initial conditions : @xmath140 it should be pointed out that only real quantities enter equations ( [ r2s])-([r_cont ] ) . this circumstance may be advantageous , especially for numerical calculations , since most standard numerical packages are not applicable direct to systems of ode involving complex valued coefficients . apart from that , ( [ cc2])-([cont ] ) and ( [ r2s])-([r_cont ] ) generate identical results , as has been proved by extensive numerical tests carried out by us for laser signals with various forms of the chirp , pulse envelope , and spatial distribution . however , before we demonstrate appropriate results ( see section [ sec : appl ] ) , it is instructive to present here an approximate but reasonably accurate analytic solution for @xmath141 . this consideration seems to be rather useful for a qualitative description of the process as a whole , since it reveals some key features which make their appearance due to the non - stationarity of both the frequency and the amplitude of the laser signal . let us assume , in agreement with ( [ intens ] ) , that the intensity of the incident radiation is gaussian both in time and space and use the most general form of the chirp , @xmath142 ( cf . equation ( [ omega ] ) ) . to start with , consider riccati type equation ( [ cc2 ] ) to determine the complex valued function @xmath143 , equation ( [ cc2 ] ) takes the form @xmath144 c(x ) \nonumber \\ \label{eq : full } \lo+ \rmi d_{b , a } { \cal i } \e^{-\frac{\tau^2 x^2}{2 \sigma^2_{\rm t}}}=0,\end{aligned}\ ] ] where the following notations were introduced for brevity : @xmath145 note that @xmath146 depends on the time independent parameters of the pulse only , but not on @xmath147 . due to its non - linearity , equation ( [ eq : full ] ) admits only numerically solving , except for the trivial case , @xmath148 , where @xmath141 is of the form @xmath149 - \delta d \biggr\}. \label{c0}\end{aligned}\ ] ] however , the problem can be simplified considerably by linearizing equation ( [ eq : full ] ) . this approximation , which the consideration developed below is stemmed on , is justifiable if @xmath150 holds uniformly for @xmath151 and arbitrary @xmath152 and @xmath153 . to satisfy the latter condition , it is sufficient ( but not necessary ) to assume that @xmath154 as readily follows from equation ( [ c : simp ] ) . if @xmath155 and @xmath156 , for example , then @xmath157 this imposes the following restriction on the pulse energy : @xmath158 , where particular values of the atomic matrix element , @xmath159 ( see table [ tabd ] ) , and the typical laser parameters have been used . the above estimates are illustrated in figure [ fig1a ] where @xmath160 obtained by direct numerically solving equation ( [ eq : full ] ) is plotted versus the dimensionless variable @xmath147 , at several @xmath161values . in addition , for each pulse energy , the appropriate @xmath162value defined by equation ( [ cond_c ] ) is indicated for reference as well . as a function of @xmath147 , for various pulse energies , @xmath161 , and fixed values of the chirp , @xmath163 , and the frequency detuning , @xmath164 . the laser field frequency is assumed to be of the form : @xmath165 ; the rest laser parameters used are : @xmath166.,width=302,height=302 ] the linearized form of equation ( [ eq : full ] ) reads : @xmath167 \widetilde{c}(x ) + \rmi d_{b , a}\ : { \cal i } \e^{-\frac{\tau^2 x^2}{2 \sigma^2_{\rm t}}}=0.\end{aligned}\ ] ] its solution obeying the zero initial condition ( [ cond_inf ] ) at @xmath168 has the form @xmath169 to demonstrate the fair accuracy of this relation for @xmath170 and @xmath171 , for example , the graph of @xmath172 as a function of @xmath147 is presented , for several chirp values , in figure [ fig1b ] , along with appropriate graphs for numerical solution of the non - linear equation ( [ eq : full ] ) . together with plots in figure [ fig1a ] , these graphs provide instructive information about the magnitude of @xmath173 at various time moments . in addition , both figures clearly demonstrate a rather peculiar way in which the values for the above ratio are ( strongly ) distorted due the presence of the chirp . and @xmath174 as functions of @xmath147 , for fixed values of the pulse energy , @xmath175 , frequency detuning , @xmath176 , and for several chirp values : @xmath177 . the frequency of the non - monochromatic laser field is assumed to be of the form : @xmath178 , and the rest laser parameters used are : @xmath179 . + @xmath180 numerical solution of equation ( [ eq : full ] ) ; @xmath181 analytic solution , @xmath182 , of equation ( [ eq : simp ] ) , as given by ( [ c : simp]).,width=302,height=302 ] it is relevant to note that @xmath183 approximates the function @xmath141 qualitatively correctly for even those @xmath161 which violate condition ( [ cond_c ] ) . an elaborate comparison with exact numerical solutions of ( [ eq : full ] ) shows that , for most applications , the region of the validity of equation ( [ c : simp ] ) can be safely extended up to @xmath184 , depending on the rest laser parameters . at higher pulse energies , the contribution of non linear effects becomes so essential that the neglect of the @xmath185term is no longer permissible . this can be seen , for example , in figure [ fig1a ] where , for @xmath186 and @xmath187 , appropriate non linear contribution amounts to 20% . as follows from the method of our derivation , equation ( [ c : simp ] ) can be readily generalized on a wide class of similar multi - photon ionization processes that are induced by not very intense laser fields and which involve the chirps and laser amplitudes , such that @xmath188 and @xmath189 sufficiently fast as @xmath190 . note that the latter condition ensures the rapid convergence of the integral in ( [ c : simp ] ) . in addition , it makes certain that the values of all physical quantities of interest are determined actually by only a somewhat narrow domain of @xmath147 within which the laser intensity is peaked . equation ( [ c : simp ] ) is suitable to be used further for the analytical study of the 3photon ionization with muonium , where @xmath191 and @xmath192 . in particular , @xmath183 readily enables one to obtain analytic formulae for the probabilities : @xmath193 , @xmath194 , and @xmath195)$ ] . however , these expressions are not given explicitly here as they turn out to be too bulky . instead , the formula ( [ c : simp ] ) will be used in the following subsection [ subsec : wood ] , while addressing an interesting complementary problem : a calculation of the integral of the 3photon ionization line profile over the entire range of the frequency detunings , @xmath196 . in what follows , it is shown that , for given @xmath161 and @xmath153 , the integral of the 3photon ionization line profile , @xmath133)$ ] , over the entire range of the frequency detunings @xmath129 , @xmath197)\equiv \int_{-\infty}^{+\infty } w_c^{\infty}(\delta \nu , a_\omega , r;[\phi ] ) \rmd \delta \nu,\ ] ] is almost independent of particular form of the chirp , at least for low pulse energies . to be more precise , equation ( [ eq : rho_gen ] ) defines in fact a non - linear functional of @xmath69 whose `` strength '' depends , though , on the rest laser parameters . within certain range of these parameters ( @xmath161 , in the first instance ) , the functional relation ( [ eq : rho_gen ] ) turns out to be weak , so that @xmath198)$ ] plays in this case the role of a sort of adiabatic invariant for 3photon resonant ionization processes occurring under the action of the fields with slow varying frequencies . hence , an `` almost conservation '' of @xmath198)$ ] provides a simple and easily verifiable approximate criterion which enables the photoionization line profiles corresponding to different values / forms of the chirp to be compared and normalized . the problem discussed here was first addressed , to our knowledge , in @xcite , within the framework of the similar 3level model and for low pulse energies only . in contrast with our present approach , however , the genuine photoionization rate of the level @xmath38 ( that is , @xmath5 ) was set in @xcite equal , respectively , to zero and to some time and energy independent constant , while describing the dynamics of the 1st and the 2nd stages of the resonant 3photon ionization . although this approximation provides acceptable results for @xmath199 , say , it is of interest , both experimental and theoretical , to reconsider the entire problem by releasing the above simplifying assumptions and getting , thereby , a deeper insight into the physics of the process . [ [ section ] ] let us consider first the simplest case of low @xmath161 , such that condition ( [ cond_c ] ) is satisfied . in addition , we will still assume , without a loss of generality , that the intensity of the incident radiation is gaussian both in time and space , without specifying , though , particular form of the chirp for a moment . by making use of equation ( [ cont ] ) , this allows @xmath198)$ ] to be expressed in terms of @xmath183 as @xmath200 ) = \frac{c}{4\pi } \frac{2}{\omega_{\rm l } } \sigma^{(\gamma)}_b { \cal i } \int_{- \infty}^{+ \infty } \rmd \delta \nu \int_{- \infty}^{+ \infty } \rmd x \exp \left\{- \frac { \tau^2 x^2}{2 \sigma^2_{\rm t } } \right\ } \left|\widetilde{c}(x,\delta \nu , a_\omega , r)\right|^2 \nonumber \\ \label{eq : rho - cappr } \lo \times \exp \left\ { 2 d_{b , a } { \cal i } \im \left [ \int_{-\infty}^{x } \exp \left\{- \frac { \tau^2 u^2}{2 \sigma^2_{\rm t } } \right\ } \widetilde{c}(u,\delta \nu , a_\omega , r ) \rmd u \right ] \right\}.\end{aligned}\ ] ] here , our former notations ( [ defin : iab ] ) have been employed . to this end , it must be noted that @xmath201 is satisfied for all @xmath129 and @xmath202 , except those @xmath203 which are in the vicinity of the lineshape s maximum , @xmath204 , and for @xmath161 such that @xmath205 . the latter condition happens to be more restrictive as compared to the one assumed here . this is due the fact that the diagonal matrix elements are usually much bigger than the non - diagonal ones ( see table [ tabd ] for comparison ) . by making use of ( [ c : simp ] ) , ( [ cond_omega ] ) and presuming that the above condition is fulfilled , one can show that @xmath206 \approx \nonumber \\ \label{eq : exp } \fl - \sqrt{2 \pi } d^2_{b , a } { \cal i}^3 \frac{\sigma_{\rm t}}{\tau } \frac{\im(\delta d ) } { \left[4 \pi \tau \delta \nu + 2 \dot{\phi}(0 ) + { \cal i } \re ( \delta d)\right]^2 + { \cal i}^2 [ \im ( \delta d)]^2 } \left [ 1+{\rm erf } \left ( \frac{\tau x } { \sqrt{2}\sigma_{\rm t}}\right ) \right].\end{aligned}\ ] ] note that , for any @xmath147 , arbitrary laser parameters and @xmath207 , this expression is negative , as it should be , since @xmath208 . for small pulse energies considered here , the right hand side of ( [ eq : exp ] ) is proportional to @xmath209 , which permits the appropriate exponent in ( [ eq : rho - cappr ] ) to be expanded in terms of @xmath161 . on retaining two leading terms in this expansion , this yields @xmath200 ) = \frac{c}{4\pi } \frac{2}{\omega_{\rm l } } \sigma^{(\gamma)}_b ( \omega_{\rm l } ) { \cal i } \int_{- \infty}^{+ \infty } \rmd \delta \nu \int_{- \infty}^{+ \infty } \rmd x \exp \left\{- \frac { \tau^2 x^2}{2 \sigma^2_{\rm t } } \right\ } \left|\widetilde{c}(x,\delta \nu , a_\omega , r)\right|^2 \nonumber \\ \label{exp_ser } \fl \times \left\ { 1 - \frac{\sigma_{\rm t}}{\tau } \frac{\sqrt{2 \pi } d^2_{b , a } { \cal i}^3 \im(\delta d ) } { \left[4 \pi \tau \delta \nu + 2 \dot{\phi}(0 ) + { \cal i } \re ( \delta d)\right]^2 + { \cal i}^2 [ \im ( \delta d)]^2 } \left [ 1+{\rm erf } \left ( \frac{\tau x } { \sqrt{2}\sigma_{\rm t}}\right ) \right ] \right\}.\end{aligned}\ ] ] further calculations are straightforward but cumbersome . hence , some technical details will be given below for the first term in the curly brackets in ( [ exp_ser ] ) only ; the corresponding contribution is denoted as @xmath210 ) $ ] . in this simplest case , by substituting ( [ c : simp ] ) for @xmath183 and interchanging the order of integration , one arrives at @xmath211 ) = \nonumber \\ \fl \frac{c}{4\pi } \frac{\sigma^{(\gamma)}_b ( \omega_{\rm l}){\cal i}^3}{\tau \omega_{\rm l } } d^2_{b , a } \int_{-\infty}^{+\infty } \exp\left\ { - \frac { \tau^2 x^2}{2 \sigma^2_{\rm t } } - \sqrt{2 \pi } \frac{\sigma_{\rm t}}{\tau } { \cal i } \im(\delta d ) { \rm erf}\left ( \frac{\tau x } { \sqrt{2}\sigma_{\rm t}}\right ) \right\ } \rmd x \nonumber \\ \lo{\times } \int_{-\infty}^{x } \exp\left\ { - \frac { \tau^2 u^2}{\sigma^2_{\rm t } } + \sqrt{2 \pi } \frac{\sigma_{\rm t}}{\tau } { \cal i } \im(\delta d ) { \rm erf}\left ( \frac{\tau u } { \sqrt{2}\sigma_{\rm t}}\right ) \right\ } \rmd u \nonumber \\ \label{eq : wood1 } \fl = \frac{c}{4\pi } \frac{\sigma_{\rm t } \sigma^{(\gamma)}_b { \cal i}^2}{\sqrt{2 } \tau^2 \omega_{\rm l } } \frac{d^2_{b , a}}{\im(\delta d ) } \left\{\sqrt{\frac{\pi}{2 } } - \int_{-\infty}^{+\infty } \exp\left [ - 2 u^2 - \sqrt{2 \pi } \frac{\sigma_{\rm t}}{\tau } { \cal i } \im(\delta d ) { \rm erfc}(u ) \right ] \rmd u \right\ } .\end{aligned}\ ] ] for two most important cases of interest , @xmath212 and @xmath213 , the integral in ( [ eq : wood1 ] ) can be evaluated , respectively , by expanding the integrand into the power series in @xmath214 and by using the method of the steepest descent . this yields @xmath215 \rmd u \nonumber \\ \label{step : desc } \lo{\asymp } \left\ { \begin{array}[c]{ll } \sqrt{\pi/2 } - \pi ( \sigma_{\rm t } / \tau ) \im(\delta d ) { \cal i } \qquad & { \rm for } \ ; |{\cal i } \im(\delta d ) | \ll 1 \\ \ms \left ( \sqrt{\pi } \tau^2 / 2^{3/2 } \sigma^2_{\rm t } \right ) \left [ \im(\delta d ) { \cal i } \right]^{-2 } \qquad & { \rm for } \ ; |{\cal i } \im(\delta d ) | \gtrsim 1 \end{array } \right . .\end{aligned}\ ] ] accordingly , we are left finally with @xmath211 ) \nonumber \\ \label{rho_0 } \fl = \frac{c}{4\pi } \sqrt{\frac{\pi}{2 } } \frac{\sigma_{\rm t } } { \tau^2 \omega_{\rm l } } \sigma^{(\gamma)}_b ( \omega_{\rm l } ) d^2_{b , a } \times \left\ { \begin{array}[c]{ll } \sqrt{\pi } ( \sigma_{\rm t}/\tau ) { \cal i}^3 \qquad & { \rm for } \ ; |{\cal i } \im(\delta d ) | \ll 1 \\ \ms [ \sqrt{2 } \im(\delta d ) ] ^{-1 } { \cal i}^2 \qquad & { \rm for } \ ; |{\cal i } \im(\delta d ) | \gg 1 \end{array } \right . .\end{aligned}\ ] ] note that in either limiting case , the result is independent , in accord with our initial statement , of particular form of the chirp @xmath216 . to clarify this point , one should emphasize that the only two simplifications made in the course of derivation were : ( i ) the use of @xmath183 given by ( [ c : simp ] ) and ( ii ) the replacement of the appropriate exponent by a unity . both approximations are fully justifiable for @xmath217 assumed here . obviously , the latter condition is consistent with the _ first _ relation in ( [ rho_0 ] ) , which has been formerly obtained by means of a different technique ( and cast in a slightly different analytic form ) in @xcite . apart from being chirp independent , this result demonstrates also that @xmath210 ) \propto a^3_{\omega}$ ] , as opposed to the next limiting case , that is , @xmath218 , where @xmath210 ) \propto a^2_{\omega}$ ] . a gradual decrease of the exponent is clearly evidenced by numerical simulations carried out by us for muonium , while scanning sufficiently broad @xmath161domain . some results of this numerical study are discussed in more detail in section [ sec : appl ] ( see figure [ fig : power ] , for example ) . these show , in particular , that equations ( [ rho_0 ] ) describe @xmath210)$ ] qualitatively correctly for both low and moderately high pulse energies . in a similar manner one can also obtain appropriate contribution , denoted @xmath219)$ ] , coming from the second term in the curly brackets in ( [ exp_ser ] ) . after tedious calculations whose details will be given elsewhere , one ends up with the following expression for a sum of two contributions : @xmath211 ) + \rho^{(1)}(a_\omega , r;[\phi ] ) \nonumber \\ \label{rho_01 } \fl = \left\ { \begin{array}[c]{ll } \frac{c}{4\pi } \frac{\pi}{\sqrt{2 } } \frac{\sigma^2_{\rm t } } { \tau^3 \omega_{\rm l } } \sigma^{(\gamma)}_b ( \omega_{\rm l } ) d^2_{b , a } { \cal i}^3 \left ( 1 - \sqrt{2 } \pi \frac{\sigma^2_{\rm t } } { \tau^2 } d^2_{b , a } { \cal i}^2 \right ) \qquad & { \rm for } \ ; |{\cal i } \im(\delta d ) | \ll 1 \\ \ms \frac{c}{4\pi } \frac{\sqrt{\pi}}{2 } \frac{\sigma_{\rm t } } { \tau^2 \omega_{\rm l } } \sigma^{(\gamma)}_b ( \omega_{\rm l } ) \frac { d^2_{b , a } } { \im(\delta d ) } { \cal i}^2 \left ( 1 - \frac{4 \sqrt{2 } \pi } { 3 } \frac{\sigma^3_{\rm t } } { \tau^3 } d^2_{b , a } { \cal i}^2 \right ) \qquad & { \rm for } \ ; |{\cal i } \im(\delta d ) | \gtrsim 1 \end{array } \right . .\end{aligned}\ ] ] here , according to ( [ cond_omega ] ) , we have discarded the terms in @xmath219 ) $ ] containing the second and higher order derivatives of @xmath220 . this simplification results in the fact that entire equation ( [ rho_01 ] ) turns out to be completely independent of the chirp . the right hand side of equation ( [ cond_omega ] ) depends ( through @xmath202 ) on the radius @xmath153 . to eliminate this dependence and to facilitate , thereby , an adequate comparison with the results of numerical calculations , the left and right hand sides of ( [ cond_omega ] ) can be integrated over the entire @xmath221plane perpendicular to the direction of the laser beam s propagation ( @xmath24axis ) . a clear physical meaning of such an averaging procedure is discussed in the comments following equation ( [ w_avar ] ) . by virtue of the identity , @xmath222 which defines the spatially independent part of @xmath202 , denoted as @xmath223 , this yields eventually the following @xmath153independent result : @xmath224 ) \approx \int_{-\infty}^{+\infty}\!\!\int_{-\infty}^{+\infty } \left(\rho^{(0)}(a_\omega , r;[\phi ] ) + \rho^{(1)}(a_\omega , r;[\phi ] ) \right ) \rmd x \rmd y \nonumber \\ \label{rho_fin } \fl = \left\ { \begin{array}[c]{ll } \frac{c}{4\pi } \frac{\sqrt{2 } \pi^2}{3 } \frac{\sigma^2_{\rm t } \sigma^2_{\rm r } } { \tau^3 \omega_{\rm l } } \sigma^{(\gamma)}_b ( \omega_{\rm l } ) d^2_{b , a } \widetilde{{\cal i}}^3 \left ( 1 - \frac { 3 \sqrt{2}\pi } { 5 } \frac{\sigma^2_{\rm t } } { \tau^2 } d^2_{b , a}\widetilde{{\cal i}}^2 \right ) \qquad & { \rm for } \ ; |\widetilde{{\cal i } } \im(\delta d ) | \ll 1 \\ \ms \frac{c}{4\pi } \frac{\pi^{3/2}}{2 } \frac{\sigma_{\rm t } \sigma^2_{\rm r } } { \tau^2 \omega_{\rm l } } \sigma^{(\gamma)}_b ( \omega_{\rm l } ) \frac { d^2_{b , a } } { \im(\delta d ) } \widetilde{{\cal i}}^2 \left ( 1 - \frac{2 \sqrt{2 } \pi } { 3 } \frac{\sigma^3_{\rm t } } { \tau^3 } d^2_{b , a } \widetilde{{\cal i}}^2 \right ) \qquad & { \rm for } \ ; |\widetilde{{\cal i } } \im(\delta d ) | \gtrsim 1 \end{array } \right . .\end{aligned}\ ] ] , over the entire range of the frequency detunings @xmath225 , as a function of the chirp s magnitude , @xmath226 ; three particular pulse energies are considered : @xmath227 . the chirped frequency of the laser field employed is of the form : @xmath228 . the rest laser parameters used are : @xmath229.,width=283,height=283 ] in order to check this equation , the numerical calculation of @xmath230)$ ] , in a wide range of @xmath161 and for typical values of the rest physical parameters used in the 1s2s experiment in muonium , have been carried out . this was done by an extra numerical integration of the 3photon ionization line profiles @xmath231 ( see equation ( [ w_avar ] ) for definition ) over the entire @xmath129domain . these results are shown in figure [ fig1c ] , for three particular pulse energies . at @xmath175 , being the only value displayed which satisfies condition ( [ cond_c ] ) , the plot demonstrates that @xmath230)$ ] is conserved to the 1% accuracy , within a wide range of the chirp s magnitude . the mean numerical value , @xmath232 ) \approx 0.080\:{\rm mhz}$ ] , is to be compared with that obtained by means of the _ first _ equation in ( [ rho_fin ] ) : @xmath233 ) \approx 0.1242 - 0.0292=0.095\:{\rm mhz}$ ] . to make these results comparable , the latter value includes also an additional weighting factor , @xmath234 . even though parameter @xmath162 is very close to unity : @xmath235 , for @xmath175 and @xmath236 used here , numerical and analytical data are consistent to 20% of relative accuracy . interesting to note that the first / second relation in ( [ rho_fin ] ) overestimates / underestimate the true value of @xmath230 ) $ ] . much better agreement can be anticipated ( and actually takes place ) for lower pulse energies and/or bigger spatial dispersions @xmath237 . for @xmath238 and @xmath239 ( apppropriate line profiles are displayed in figure [ fig2](a ) ) , for example , numerical and analytical results read : @xmath240 ) \approx 0.90\cdot 10^{-4}\:{\rm mhz}$ ] and @xmath241 , respectively . [ [ section-1 ] ] for higher pulse energies and/or smaller spatial dispersions @xmath237 , such that @xmath242 , equation ( [ rho_fin ] ) fails to describe adequately the true function @xmath230)$ ] . this refers both to the absolute values of @xmath230)$ ] and , particularly , to its dependence on the magnitude of the chirp , which makes its appearance at high pulse energies . it has been elucidated above that the main underlying reason for this failure originates from the fact that `` the linear approximation '' ceases to be valid any longer , due to corrections caused by the @xmath243term in equation ( [ eq : full ] ) . this non linear term does not allow any reasonably accurate analytic treatment to be developed , so that the numerical analysis of the problem must be used instead . to simulate the high energy regime , the results of our calculation are presented in figure [ fig1c ] , for two pulse energies , @xmath244 , and for typical values of the rest physical parameters used in the 1s2s experiment in muonium . as in the case of low @xmath161 , this was done by the numerical integration of the 3photon ionization line profiles @xmath245 over the entire @xmath129domain . the plots demonstrate in fact a noticeable deviation of @xmath230)$ ] from a constant value , with an increase of the pulse energy . for example , at @xmath187 and for @xmath226 in the range @xmath246 , this deviation ( that is , the dependence on the chirp s magnitude ) amounts already to 30% ; in absolute units , the data read : @xmath247)=2.265\:{\rm mhz}$ ] for @xmath248 and @xmath247)=2.90\:{\rm mhz}$ ] for @xmath249 . furthermore , the gauge of the given deviation may be expected to be even more pronounced at higher pulse energies , thus preventing us actually from viewing @xmath230)$ ] as a conserving quantity in this case . although such @xmath161 are of no practical interest at the moment for the @xmath100@xmath5 experimental study with muonium , the entire effect might be of relevance for future experimental studies where much higher laser intensities are involved . let us apply the results obtained in preceding sections to particular laser parameters adopted in the 1s-2s experiment in muonium @xcite , by assuming that @xmath250 and setting @xmath251 . it is convenient in this section to measure the energies in the units : @xmath252 , where @xmath253 stands for the reduced mass of the electron and muon . here , the value for the electron to @xmath254 mass ratio , @xmath255 , was used . the quantity @xmath256 stands for the unity of distance , with @xmath257 being the bohr radius . accordingly , one particle energies and dynamic polarizabilities , @xmath258 , are measured in the units of @xmath259 and @xmath260 . as was mentioned above , 3photon ionization probability , @xmath261 , can be obtained by accurate numerically solving either the system ( [ ala2])-([cont ] ) or ( [ r2s])-([r_cont ] ) . in both cases this has been done by the stepwise time integration , starting from @xmath262 , while scanning sufficiently broad @xmath129domain centered at @xmath263 ( i.e. @xmath264 ) , for all @xmath265 on the @xmath153mesh : @xmath266 . numerical values of the matrix elements , @xmath267 , which enter ( [ ala2])-([cont ] ) were calculated in @xcite ; these are compiled in table [ tabd ] , together with the value of the single photoionization cross section of the @xmath5level at @xmath268 . the latter is given explicitly by ( see @xcite and references therein ) @xmath269 where @xmath119 is the fine structure constant and @xmath270 denotes the ionization potential of the @xmath5state . @xmath271 [ tabd ] at the final stage of our numerical procedure , a spatial averaging has been performed in order to obtain @xmath153-independent ionization profiles , @xmath272 . this has been achieved by an extra integration of @xmath273 over entire @xmath221plane perpendicular to the direction of the laser beam s propagation : @xmath274 basically , this newly introduced quantity can be interpreted as an averaged ionization probability related to the entire laser beam spot , since @xmath245 is independent of the distance from the beam s axis . alternatively , in view of the low values of the muon density in the media @xcite ( see below ) , @xmath275 can be called as `` the ionization probability per one ( muon ) atom in the beam '' . an auxiliary factor , @xmath276 , has been introduced in ( [ w_avar ] ) in order to make @xmath277 dimensionless , as the original probability @xmath261 is . this particular choice was adopted according to the so called rule being inherent to various problems involving gaussian law . in the case considered , the `` @xmath278''rule makes its appearance through the fact that the values of @xmath261 happen to be almost negligible for @xmath279 , unless the pulse energies higher than @xmath280 , say , are considered . for such a high energy regime , the probability @xmath273 has a long distance `` tail '' spreading out beyond the _ effective _ beam radius , @xmath281 , within which the values of @xmath261 may be very close to , yet less than , @xmath9 . in this situation the proper normalization factor different from that chosen above must be used so as to prevent @xmath282 from being bigger than @xmath9 . as is clearly evidenced by our calculations , the values of @xmath277 defined by equation ( [ w_avar ] ) exceed 1 starting already from @xmath283 , depending on the rest laser parameters . one should point out , however , that such @xmath161 are hardly attainable currently with proper pulsed lasers sources in the required frequency range . this makes the high energy regime in the resonant 3photon ionization with muonium to be mostly of academic rather than practical interest at the moment , thus justifying the definition of equation ( [ w_avar ] ) . to avoid any misunderstanding , it must be clearly stated once again that the spurious effect mentioned here originates solely from the particular choice of the normalization factor in ( [ w_avar ] ) , and has , therefore , nothing to do with either the model developed in the work or a lack of the numerical accuracy employed by us . the latter was chosen to be equal to @xmath84 which proves to be sufficient to ensure that @xmath284 holds ( as it should ! ) for any particular individual set of all parameters involved in the problem , including @xmath285 , and @xmath237 . the averaged probability @xmath277 readily enables one to estimate the expected number of experimentally observed ionization events , @xmath286 , defined as the number of the muonium atoms which are ionized by a sequence of @xmath287 identical gaussian laser pulses ( [ intens ] ) within the time @xmath288 and detected eventually in apparatus . such a formulation of the problem corresponds direct to the actual experimental situation where the gas media containing muonium atoms interacts with @xmath289 laser pulses per second . under these conditions , the required number of the ionization events to be detected during the observation time @xmath290 can be estimated by means of the following simple relation : @xmath291 here , @xmath292 is the spatial density of the muonium atoms in the chamber , @xmath293 and @xmath294 stand , respectively , for the _ effective _ cross section of the laser beam ( see above ) and the total path which this beam travels in the media , @xmath295 denotes an efficiency of registration of the muonium atoms in apparatus ; finally , @xmath296 is the value of the averaged 3photon ionization probability @xmath297 at its maximum , for particular values of the chirp and the energy of the pulse . provided @xmath226 and @xmath161 are fixed , this maximum is achieved approximately at @xmath298 , almost independently of @xmath161 , as can be seen in figure [ fig2](b ) . in particular , @xmath299 thus leading , for example , to the following expected number of events detectable during the observation time @xmath300 : @xmath301 . it is rather instructive that this value happens to be quite close to that found preliminary in the @xmath1 experiment in muonium being currently underway . , of the ground state of muonium by the laser field with the chirped frequency @xmath302 , versus laser frequency detuning , @xmath303 . two pairs of different values for the pulse energy , @xmath161 , and the spatial dispersion of the laser signal , @xmath237 , are used : ( a ) @xmath304 , and ( b ) @xmath305.,width=529,height=283 ] , of the ground state of muonium by the laser field with the chirped frequency @xmath306 , versus laser frequency detuning , @xmath307 , for two pulse energies : ( a ) @xmath308 and ( b ) @xmath309 . , width=529,height=283 ] . the chirped frequency of the laser field is assumed to be of the form : @xmath310 ; ( b ) - the power dependence of the line profile s maximum value . in both cases the rest laser parameters are the same as in figures [ fig3](a , b ) . , width=529,height=264 ] some results of our simulations are presented in figures [ fig2]-[fig5 ] , for various magnitudes of the chirp @xmath226 and various parameters of the laser pulse which are typically used in the measurements of the @xmath1 energy separation in muonium ; particular values employed are indicated in figures [ fig2],[fig3 ] . both these graphs demonstrate appreciable red shift of the maxima ( i.e. towards lower frequencies , relative to @xmath311 ) of the photoionization profiles with an increase of the chirp . it must be noted , however , that this shift is caused by a combination of two competing mechanisms : ( i ) the shift being due to the combined _ ac stark _ effect for the @xmath100- and @xmath5levels and ( ii ) that arising because of the chirp itself . the curves show that the former mechanism turns out to be small compared with the latter at relatively small pulse energies , @xmath312 . at @xmath313 and @xmath314 , for instance , i.e. in the case of completely unchirped laser signal , the _ ac stark _ shift amounts approximately to @xmath19 . one can anticipate , however , that the relative contribution of the _ ac stark _ shift will increase as the laser intensity increases , since its magnitude is linearly proportional to the laser intensity , whereas the chirp induced shift is almost intensity independent . this behaviour can be seen in figures [ fig3](a , b ) and is summarized in figure [ fig4](a ) showing , in particular , that the _ ac stark _ shift becomes equals to @xmath315 at @xmath316 . in addition to the shift of the maxima , the curves in figures [ fig2],[fig3 ] exhibit a strong dependence on both the spatial dispersion @xmath237 and the power @xmath161 of the laser signal . it should be noted that an appreciable distortion of the photoionization line shapes happens to be particularly enhanced at relatively large chirp values , where the line profiles become asymmetric . generally speaking , this asymmetry is an intrinsic feature of the 3photon resonant photoionization occurring under the action of the _ chirped _ pulsed laser signal . moreover , an asymmetry of the ionization line shapes happens to be one of the most pronounced manifestations of the non - monochromaticity of the laser field . this phenomenon is basically due to the fact that the chirp , even though its relative magnitude amounts to only @xmath317 in our calculations , violates an equivalence of initial ( @xmath318 ) and final ( @xmath319 ) time moments . indeed , the photoionization probability turns out to be either more efficient or suppressed , depending on whether the time dependent chirped laser frequency is in the resonance with the 2photon @xmath320 transition or slightly off it . the process as a whole becomes then somewhat `` time sensitive '' , although the form of the laser pulse envelope used is time invariant . it should be noted , however , that the gauge of the above asymmetry depends strongly on the laser power , as well as particular forms of the laser amplitude and the chirp . this dependence can be seen in figures [ fig3](a , b ) where higher @xmath161values as compared with figures [ fig2](a , b ) are used . the former show that the asymmetry discussed is less pronounced as long as higher laser intensities are involved . two essential characteristics of the line profiles maximum , that are , the shift and the magnitude , are plotted in figure [ fig4](a , b ) as functions of the laser power for various values of the chirp . in particular , the graph [ fig4](b ) provides information about the dependence of the line shapes maximum value , @xmath321 , on the pulse energy @xmath161 . it is quite natural to parameterize this relation by the following simple power law : @xmath322 . the exponent here proves to be a slow varying function of the pulse energy , such that @xmath323 . according to this condition , @xmath324 can be estimated as @xmath325.\ ] ] dependence of the exponent , @xmath324 , defined by the relation : + @xmath326.,width=264,height=264 ] one should note , for example , an appreciable deviation of the @xmath324values from @xmath327 which must be expected for a 2step 3photon ionization of an atom in the state @xmath37 by a _ weak , monochromatic and spatially homogeneous laser field _ , without the relaxation of the intermediate levels . for @xmath314 , the @xmath161dependence of @xmath324 is displayed in figure [ fig : power ] . this graph demonstrates that @xmath328 gradually decreases as the pulse energy increases and tends , for @xmath329 , to an almost constant value , @xmath330 , thus indicating on the presence of saturation in the 3photon transition considered . separation in muonium ; ( c ) - photoionization line profile measured in , versus laser frequency detuning , @xmath307 . , width=566,height=264 ] finally , it should be mentioned that the above consideration has been developed mainly in the attempt to simulate those line profiles of the two step 3photon ionization probabilities that are supposed to be measured soon in the new @xmath1 experimental investigation of muonium . essential physical parameters employed in this study are still not fixed completely and are subject to further changes . it is therefore rather tempting to apply our present technique to appropriate data formerly collected within the framework of the previously employed experimental setup @xcite . these are shown in figures [ fig5](a , b ) , along with the photoionization line profile obtained within the framework of the current theoretical approach , figure [ fig5](c ) . the fact that the corresponding experimental line shape ( not shown ) is virtually indistinguishable from the theoretical one provides an additional and independent check of the overall validity of the model developed in this work . we have developed a simple theoretical scheme intended to describe , to the @xmath19 accuracy , the stepwise 3photon resonant photoionization in hydrogenic systems , induced by the chirped laser field with time dependent amplitude . it has been shown that such an accuracy can be achieved within the framework of a simple 3-level model , by taking into account ( i ) the _ ac stark _ shifts and ( ii ) non - zero ionization rates of the levels involved , together with ( iii ) a spatial inhomogeneity of the laser signal and ( iv ) arbitrary @xmath101-dependencies of its intensity ( pulse envelope ) , @xmath331 , and frequency , @xmath31 . the system of equations ( [ ala2])-([cont ] ) or , equivalently , equations ( [ r2s])-([r_cont ] ) is a key point of the method employed . these are of independent significance as being not specifically restricted to particular states of reference , laser intensities , and chirps , so that the results obtained for @xmath332 and @xmath333states can be generalized on arbitrary @xmath334 levels of hydrogen like atoms . excited @xmath335levels , such that @xmath336 , are of particular current interest for ultra - high precision laser spectroscopy , and it should be expected that an adequate interpretation of experimental data should require an accurate theoretical account of a wide spectrum of light induced effects . some of these effects are beyond the scope of our present consideration and , in the first instance , comprise appropriate corrections due to inevitable motion of atoms in the media and associated _ second _ order doppler shifts , as being of major importance . indeed , even for cw lasers where the signals are usually almost unchirped in the laboratory frame , the amplitude and the frequency of the pulsed laser signals become essentially time dependent in the atomic rest frame , thus leading , as was demonstrated above , to the shifts and distortions of ionization / excitation lines . the relative contribution of the effects discussed is estimated as @xmath337 with @xmath109 being the atomic velocity in the laboratory frame , and these should be incorporated in the theoretical scheme developed here , along with relativistic / radiative corrections @xmath338 to the operator of the particle laser field interaction . this work is currently in progress , and we consider it as a subject of forthcoming publications . the authors are indebted to g. zu putlitz for his constant support and encouragement . this work has been strongly motivated and influenced by discussions within the muonium 1s-2s collaboration , particularly with p.e.g . baird , m.g . boshier , p.g.h . sandars , and w.t . toner . one of us ( v.y . ) , wishes to acknowledge his gratitude to prof . i.p . grant ( frs ) for his kind advice and to the volkswagen stiftung and the royal society for financial support . also , this work has been funded in part by the grants from nato , the royal society and the bundesminister fr bildung und forschung of germany . 99 nez f , plimmer m d and bourzeix s 1993 _ europhys . lett . _ * 24 * 635 bourzeix s 1996 384 berkeland d , boshier m d and hinds e 1995 2470 schmidt - kaler f , leibfried d , seel s 1995 a * 51 * 2789 udem th , huber a 1997 2646 fee m s , chu s , mills a p 1993 a * 48 * 192 mills jr . a p 1993 _ hyperfine interactions _ * 76 * 233 maas f e 1994 247 jungmann k 1995 _ physikalisches bltter _ * 51 * 1167 g. zu putlitz 1996 _ hyperfine interactions _ * 103 * 103 boshier m g , hughes v w , jungmann k and zu putlitz g 1996 _ comments atom . mol . phys . _ * 33 * 17 eikema k s , ubachs u , vassen w and hogervorst w 1997 a * 55 * 1866 ; 1997 1690 giacobino e and cagnac b 1980 _ in progress in optics xvii _ ed e wolf ( amsterdam : north - holland ) pp 86 - 161 zon b a 1974 _ opt . spectrosc . _ * 36 * 838 zon b a and katsnelson b g 1973 _ sov.phys.jetp _ * 65 * 947 zon b a , manakov n l and rapoport l p 1975 _ opt . spectrosc . _ * 38 * 6 beausoleil g r and hnsch t w 1986 a * 33 * 1661 kaplan a e 1973 _ sov.phys.jetp _ * 65 * 1416 kaplan a e 1975 _ sov.phys.jetp _ * 68 * 823 lifshitz e m and pitaevskii l p 1974 _ relativistic quantum theory_. course of theoretical physics , part i , vol 4 ( oxford : pergamon ) landau l d and lifshitz e m 1974 _ nonrelativistic quantum theory_. course of theoretical physics , vol 3 ( oxford : pergamon ) bakule p , cornish p s , lane i , baird p e g , toner w t , sandars p g h , liu w , towrie m , _ private communication _ 1970 _ handbook of mathematical functions _ ed m abramowitz and i stegun ( dover , new york ) . cohen - tannoudji c 1968 _ cargese lectures on physics _ vol 2 ( gordon and breach : n.y . ) levinson m d 1982 _ introduction to nonlinear spectroscopy _ ( academic : new york ) salzman w r 1971 220 vasilenko l s , chebotaev v p and shishaev a v 1970 _ pisma zh . * 12 * 161 bloembergen n and levensen m d 1976 _ high resolution laser spectroscopy _ ed k shimoda ( springer , berlin ) p 315 sargent iii m , scully m o and lamb jr . w e 1974 _ laser physics _ ( reading , mass . : addison - wesley ) bassani f , forney j j and quattropani a 1977 1070 yakhontov v and jungmann k 1996 _ _ d * 38 * 141 woodman j 1992 _ ph.d . thesis _ , university of oxford , clarendon laboratory ( unpublished )

there is much reliable evidence for the existence of dark matter ( dm ) which is mainly associated with the missing mass problem . astronomers and particle physicists seek to probe dm properties by direct laboratory experiments or indirect observations @xcite . indirect techniques include efforts to detect gamma rays from dm annihilation using telescopes such as the fermi large area telescope ( fermi - lat , @xcite ) , the major atmospheric gamma - ray imaging cherenkov ( magic ) telescope @xcite , the high energy stereoscopic system ( h.e.s.s . , @xcite ) and the very energetic radiation imaging telescope array system ( veritas , @xcite ) ( see , e.g. , @xcite for a review ) . it was first suggested by gondolo and silk @xcite that adiabatic growth of a bh at the center of a dm halo whose density had a singular power - law cusp @xmath4 with @xmath5 led to a high density dm region around the central bh , @xmath6 with @xmath7 . this region is called a dm spike . inside the spike , dm annihilations are enhanced and produce the strong gamma - ray photon flux which could be detectable to the telescopes mentioned above . however , subsequent studies pointed out that this spike could be weakened by dynamical processes such as mergers of host - galaxies , sub - halo accretion and passing of molecular clouds @xcite . these processes transfer energy to the dm particles and destroy the structure of the dm spike . then the annihilation rate in the spike is smaller than predicted in @xcite because it depends on the line - of - sight integral of the squared density of the spike . if supermassive black holes ( smbhs ) have experienced mergers , they are unlikely to have surviving spike structure . even this however is controversial because of the uncertainty in whether the final parsec problem for smbh mergers has been resolved phenomenologically @xcite or even theoretically @xcite . on the other hand , formation scenarios of intermediate - mass black holes ( imbh ) which allow dm mini - spikes have been proposed @xcite . if the imbh have never experienced mergers in the past , the dm mini - spike around the central imbh is likely to survive . imbhs may exist in our universe @xcite , and even several hundreds would reside in the halo of the present - day milky way galaxy @xcite . those imbhs in globular clusters are recognized as promising sources for the evolved laser interferometer space antenna ( elisa ) @xcite / the new gravitational wave observatory ( ngo ) @xcite and deci - hertz interferometer gravitational wave observatory ( decigo)@xcite . in our previous work @xcite , we demonstrated that a very tiny effect such as the gravitational pull of a dm mini - spike around an imbh indeed affects detectability of gw by elisa and thereby we could infer presence or absence of a dm mini - spike around an imbh using gw . specifically , when a stellar mass object inspirals into the central imbh , it is affected by the gravitational force of not only the central imbh but also the mini - spike . therefore the inspiral gw is modified by the mini - spike around the central imbh . we found that the very tiny effect from gravitational pull of a dm mini - spike could have a large impact on detectability of the gw , thanks to the huge number of orbital cycles which the binary experienced in the elisa detection frequency band . we also found that gw detectability strongly depends on the density profile of the dm mini - spike . in this paper , we extend our previous work in the following way . we again consider gws emitted from a binary system consisting of a stellar mass object and an imbh harbored in a dm mini - spike , and calculate the gw waveform including the effect of both the gravitational potential and the dynamical friction on the falling stellar mass object in the dm mini - spike . furthermore , we investigate how accurately the dm parameters are determined by the gw observations . we find that the dm information contained in the waveform can be extracted with very good accuracy by gw observations if the central imbh has a steep density mini - spike . we also investigate how the detection accuracy of the dm parameters changes depending on the masses of the binary components and the density profile of the dm mini - spike such as the power index and overall normalization . recently @xcite made clear the importance of the dynamical friction on the gw waveform in a quite different context from ours , namely , a stellar mass object falling in a compact configuration of dm clouds @xcite . also , @xcite has given a wide survey on astrophysical environmental effects on gw signals using order of magnitude estimates , concluding that astrophysical environmental effects such as accretion disks , magnetic fields , and dm halos do not obscure gravitational wave astrophysics , e.g. , precision measurements of binary masses and tests of general relativity @xcite ( see also @xcite ) . to indicate one exception , our paper shall clearly show that , in the recently advocated dm mini - spike scenario , environmental effects do affect gw detectability @xcite and we can measure dm properties quite accurately from elisa gw detection , which will be shown through a detailed study using a matched filtering technique and fisher matrix analysis . we stress that while gamma - ray observations from dm annihilations can only work if the dm is a weakly interacting massive particle ( wimp ) , the gw observations we proposed should be widely applicable to any type of dm particles . furthermore , matter is almost completely transparent to gws unlike electromagnetic waves because of the smallness of their gravitational cross - section . hence , the gws carry pure information on the dm from the mini - spike to the detector . future gw experiments will probe the structure of the dm mini - spike and will even offer a hint on the nature of the dm particle . the rest of the paper is organized as follows . [ sec : mini - halo_model ] presents the dm mini - spike model and candidates for the stellar mass object . in sec . [ sec : gw_waveform ] we derive the gw waveform from the system which we consider and the observational errors of the waveform parameters are calculated in sec . [ sec : parameter_resolution ] . finally our conclusions are given in sec . [ sec : conclusion ] . we assume that the initial dm mini - halo profile which leads to the dm mini - spike after the adiabatic growth of the imbh is approximately described by the navarro - frenk - white ( nfw ) profile @xcite @xmath8 where @xmath9 is the radius , @xmath10 is the mass density and the subscript `` s '' stands for the scaling . navarro , frenk and white obtained this profile via cosmological n - body numerical simulation and numerical fitting of obtained dm halo profiles around clusters of galaxies . surprisingly , their simulations showed that every dm halo around a cluster of galaxies follows the nfw profile when normalized properly by @xmath11 and @xmath12 . later work , however , shows that the inner slope may be slightly steeper than the nfw one ( @xmath13 ) and may not even be universal @xcite . in any case , we refer to the nfw profile and the derived parameters listed in the table [ tbl : dmparameters ] below as our reference model in the following for simplicity . we study how accurately the dark matter parameters can be measured in our reference model in detail , then extend our analysis to different sets of values of the dm parameters to take into account ambiguities in the dm distribution around an imbh . the nfw parameters @xmath11 and @xmath12 are related to the cluster mass and concentration parameters by @xmath14 where @xmath15 and @xmath16 is the virial radius and @xmath17 is the virial mass of the cluster , @xmath18 is the formation redshift of the cluster , @xmath19 is the matter density parameter , @xmath20 is the critical matter density of the universe and the function @xmath21 is the volume integral of the nfw profile @xmath22 ( see , e.g. , @xcite ) . we used the fitting formula given by @xcite for the parameter @xmath23 : @xmath24 where @xmath25 @xcite . the mass - concentration relation is taken from @xcite which fits the profiles of the clusters of galaxies obtained in their n - body simulations . this result for clusters of galaxies may or may not apply for the mini - halo . in any case , concentration parameters between @xmath26 lead to qualitatively similar results and are given by the following relation . @xmath27 where we assume @xmath28 from the result of @xcite . the parameters @xmath29 and so on may be appropriately used when the overdensity @xmath23 equals @xmath30 . however , we here assumed @xmath31 and so on for simplicity . as will be shown later , the gw waveform depends on the dm mini - spike slope @xmath0 and some combination of the radius at which the mini - spike is established , @xmath32 , and the dm density there , @xmath33 . under the assumption of adiabatic growth , while the final power - law index of the dm mini - spike , @xmath0 , depends on the power - law index of the initial inner dm profile , the latter two depend on @xmath11 , @xmath12 and @xmath0 . for concreteness , we adopt @xmath34 , @xmath35 in eqs . ( [ eq : mvir ] ) , ( [ eq : rhos ] ) , and ( [ eq : concentration ] ) @xcite and find @xmath36 , @xmath37 , and @xmath38 . we proceed to discuss the dm profile of the mini - spike . if the dm mini - halo initially has a cuspy profile @xmath4 with @xmath39 , then the adiabatic growth of the central imbh produces the dm mini - spike . hence the dark matter profile becomes @xcite @xmath40 with @xmath41 where @xmath33 is the normalization constant and @xmath32 is empirically defined by @xmath42 . the radius @xmath43 is the distance of the gravitational influence of the central imbh with the mass @xmath44 and is approximately obtained by @xmath45 . instead of @xmath46 to estimate @xmath33 and @xmath32 . ] the slope of the dm mini - spike takes the value @xmath47 for @xmath39 . in the case of an initially nfw profile , @xmath48 , this gives rise to @xmath49 . if the initial profile of the mini - halo is a uniform distribution , then the final profile after the adiabatic growth of the imbh would become a more gentle @xmath50 @xcite . it is important to note that the final profile of the dm mini - halo depends on the formation history of the central imbh . if the imbh has experienced disruptive processes such as mergers in the past , the mini - spike would be weakened or disappear . for this reason , we do not specify the value of the power - law index @xmath0 of the dm mini - spike and treat it as a free parameter within the range @xmath51 . in the following , even if @xmath52 , we will still call the dm distribution close to the central imbh described by eq . ( [ eq : minispikedensity ] ) `` a dm mini - spike '' for the sake of simplicity . indeed we will see that the `` dm mini - spike '' leaves its signature in the gw waveform when @xmath53 , but certainly does not when @xmath54 . we will also assume different values of @xmath33 to study how the ambiguities in @xmath11 and @xmath12 mentioned above affect our results . finally , we take @xmath55 to be the innermost stable circular orbit of the central imbh , @xmath56 . it may be more precise to use @xmath57 @xcite , but such a change of @xmath55 does not alter at all the measurement accuracy of the dm parameters shown below . the parameters of the dm density profile are summarized in the table [ tbl : dmparameters ] below . . our reference model parameters of the imbh , the dm mini - halo and the dm mini - spike . @xmath58 : the total mass of the mini - halo , @xmath59 : the mass of the central intermediate mass black hole , @xmath60 : the formation redshift of the mini - halo , @xmath61 : the concentration of the mini - halo , @xmath16 : the virial radius of the mini - halo , @xmath12 : the nfw @xmath12 parameter of the mini - halo , @xmath11 : the nfw @xmath11 parameter of the mini - halo , @xmath43 : the radius at which @xmath62 , @xmath32 : the radius where the spike forms ( estimated by @xmath63 ) , and @xmath33 : the mini - halo mass density at @xmath32 . [ cols="^,^,^,^,^",options="header " , ] before moving onto the calculation of the gw waveform , we discuss what can be a candidate for `` a stellar mass object '' . let us consider a stellar mass object with mass @xmath64 denoted by @xmath65 , orbiting around an intermediate mass black hole @xmath66 with mass @xmath59 . we consider the inspiral up to the the innermost stable circular orbit @xmath67 @xmath68 hence , the object @xmath65 should have a radius smaller than at most @xmath69 km . at the same time , the tidal radius of @xmath65 orbiting @xmath66 at the orbital radius of @xmath67 is @xmath70 hence , this object must be either a black hole or a neutron star . alternatively , if we assume @xmath65 to be a white dwarf of radius @xmath71 km or a sun - like object of radius @xmath72 , the innermost orbital radius should be replaced by the radius below which the object @xmath65 is tidally destroyed @xmath73 as will be stated , we will consider the orbital radius of order @xmath74 or less , so we can not assume our stellar mass object to be a normal star with radius @xmath75 . a white dwarf may be an interesting candidate since an electromagnetic counterpart may be expected when it is tidally disrupted ( e.g. , @xcite ) . yet , here in this paper we assume a neutron star or a black hole when we refer to a stellar mass object . let us consider a binary system which involves a small compact object with a mass of @xmath76 and an imbh with a mass of @xmath77 . the mass of the stellar mass object @xmath64 is much smaller than the mass of ibmh @xmath78 . so the reduced mass is approximately equal to @xmath64 and the barycenter position is approximately equal to the position of the imbh . by adopting a reference frame attached to the barycenter , the equation of motion of the radial relative separation between the stellar mass object and the imbh describes the motion of the former and is given by @xmath79 where @xmath80 is the angular momentum of the stellar mass object per its mass , and @xmath81 and @xmath82 are defined by @xmath83 the mass @xmath84 denotes the dm mass contained within the isco and is defined as @xmath85 . the first term on the right - hand side of eq . ( [ eq : eom ] ) describes the gravitational potential of the effective mass of the central imbh which is modified by the dm due to the absence of the dm within the isco , the second term accounts for the dm effect , and the third term represents a centrifugal force . here the dynamical friction force and the gw back reaction force are neglected because these effects are much smaller than the gravitational potential of the imbh . we will introduce these effects to include an adiabatic evolution of the orbital radius in the next subsection . we assume that the stellar mass object orbits in a circular manner for simplicity . the orbital radius @xmath86 is obtained by solving @xmath87 in eq . ( [ eq : eom ] ) . the orbital frequency @xmath88 is related to the angular momentum @xmath80 by @xmath89 , so we get @xmath90^{1/2}. \label{eq : orbital_frequency}\end{aligned}\ ] ] when a dm mini - spike is not present around the imbh , @xmath91 and @xmath92 , so eq . ( [ eq : orbital_frequency ] ) leads to the kepler s law @xmath93 . in this subsection , we introduce the gw back - reaction and the dynamical friction into the stellar mass object s orbit by taking the energy balance equation into account . when the stellar mass object orbits around the imbh , a part of its energy @xmath94 is converted into gw emission loss @xmath95 and dynamical friction loss @xmath96 . thus the following energy balance equation is satisfied : @xmath97 as we will see in this subsection , this energy balance equation gives the time evolution of the orbital radius . the resulting orbit can be regarded as a quasi - circular orbit because of the smallness of these dissipative effects . the orbital energy @xmath94 is the sum of the kinetic energy and the gravitational potential of the stellar mass object , so we can calculate @xmath94 using eq . ( [ eq : orbital_frequency ] ) , @xmath98 where @xmath99 is the orbital velocity . when we consider the evolution of the radius @xmath86 , @xmath100 does not vanish . so the time derivative of eq . ( [ eq : orbital_energy ] ) gives the following equation , @xmath101 to the lowest order in the post newtonian expansion , the gravitational radiation energy is given by the quadrupole formula . we apply the formula to the circular newtonian binary and obtain @xmath102 when the stellar mass object moves through the cloud of dm , it gravitationally interacts with dm particles . this effect is called dynamical friction , sometimes referred to as gravitational drag which was first discussed by chandrasekar @xcite . because of dynamical friction , the stellar mass object running through the dm halo is decelerated in the direction of its motion and loses its kinetic energy as well as its angular momentum . the dynamical friction force is given by @xmath103 where @xmath99 is the velocity of the stellar mass object @xcite . the coulomb logarithm @xmath104 is defined by @xmath105 where @xmath106 is the maximum impact parameter and @xmath107 is the typical velocity of the stellar mass object . we take @xmath108 . from the expression of the dynamical friction force , we obtain the friction loss , @xmath109 to find the numerical solution of the energy balance equation ( [ eq : energy_balance_eq ] ) easily , we introduce a dimensionless radius parameter @xmath110 defined by @xmath111 with @xmath112 using the above definition of @xmath110 , the energy balance equation ( [ eq : energy_balance_eq ] ) can be rewritten in the form of the differential equation of @xmath110 with respect to time @xmath113 as @xmath114 } - c_{\text{df}}\dfrac{1}{\left(1+x^{3-\alpha}\right)^{1/2}\left[1+\left(4-\alpha \right)x^{3-\alpha}\right]x^{- 5/2 + \alpha } } , \label{eq : dxdt}\end{aligned}\ ] ] where the coefficients are defined by @xmath115}. \label{eq : c_df}\end{aligned}\ ] ] the coefficient @xmath116 is related to the gravitational radiation energy and the coefficient @xmath117 is related to the dynamical friction . in the case of the initially nfw profile , @xmath118 , the coefficients @xmath116 and @xmath119 are @xmath120 , c_{\text{df } } = 2.1 \times 10^{-8 } \ \left[1/\text{year}\right ] $ ] . note that the dynamical friction coefficient @xmath117 is much larger than the gravitational radiation coefficient @xmath116 . the gw waveform from the binary composed of two compact objects with masses @xmath64 and @xmath59 is given by @xmath121 where @xmath122 is the distance to the source ( luminosity distance for a cosmological source ) , @xmath86 is the orbital radius , @xmath123 is the inclination angle , which is the angle between the line - of - sight and the rotational axis of the orbits , and @xmath124 is the gw frequency which is given by @xmath125@xcite . the waveforms eqs . ( [ eq : gwwaveform_plus_without_backreaction ] ) and ( [ eq : gwwaveform_cross_without_backreaction ] ) are derived on the assumption that the motion of the source is described by a circular newtonian orbit . but in fact , the radius @xmath86 and the frequency @xmath88 are not constant because the orbital energy @xmath94 decreases gradually due to both dynamical friction and the gw back - reaction . including these effects , the orbit shrinks adiabatically and becomes a quasi - circular orbit . so the radius @xmath86 and the frequency @xmath88 should be replaced by @xmath126 , @xmath127 and the phase @xmath128 should also be replaced by @xmath129 as defined by eq . ( [ eq : phase_t_def ] ) below . thus , the gw waveform is expressed by @xmath130 , \label{eq : gwwaveform_plus_including_backreaction } \\ h_{\times}\left(t\right ) & = \dfrac{1}{\sourcedistance } \dfrac{4g\mu \omega_s(t)^2 r(t)^2}{c^4 } \cos \iota \sin\left [ \phi(t)\right ] , \label{eq : gwwaveform_cross_including_backreaction } \\ \phi \left(t\right ) & \equiv \int^t \omega_{\text{gw } } \left(t'\right ) \ dt ' . \label{eq : phase_t_def}\end{aligned}\ ] ] in order to discuss detectability and parameter accuracy in gw observations , it is convenient to work in the frequency domain . the fourier transformation of the gw waveform is given by @xmath131 where @xmath132 is the gw frequency . for simplicity , we consider a gw coming in the detector from the optimal direction for @xmath133 mode . in such a situation , detector pattern function are @xmath134 and @xmath135 . so the response of the detector to the gw is @xmath136 . using eq . ( [ eq : gwwaveform_plus_including_backreaction ] ) , we rewrite the gw waveform as @xmath137 where @xmath138 is the time - dependent amplitude and @xmath139 is the time - dependent gw phase . in the above equations , we have introduced the retarded time @xmath140 . in the range of frequency we are concerned with , the time - dependent amplitude @xmath141 varies slowly , while the time - dependent phase @xmath139 varies rapidly . so , the fourier transformation of the gw waveform can be calculated approximately using the stationary phase method . in this method , the rapidly oscillating term is neglected and only the slowly oscillating term survives . then the gw waveform in the fourier domain becomes @xmath142^{1/2 } , \label{eq : waveformf}\\ & \psi(t ) = 2\pi f \dfrac{d}{c } + \tilde{\phi}(t ) - \dfrac{\pi}{4 } , \label{eq : psit_def}\\ & \tilde{\phi}(t ) \equiv 2\pi ft - \phi(t ) , \label{eq : tilde_phi_def}\end{aligned}\ ] ] where the time @xmath113 is related to frequency by @xmath143 . as we will discuss in appendix [ app : rewriting_waveform ] , the gw waveform eqs . ( [ eq : waveformf ] ) , ( [ eq : psit_def ] ) and ( [ eq : tilde_phi_def ] ) can be rewritten explicitly in the frequency domain as follows : @xmath144^{-1/2 } , \label{eq : gwwaveform01}\\ & \mathcal{a}= \left(\dfrac{5}{24 } \right)^{1/2 } \dfrac{1}{\pi^{2/3 } } \dfrac{c}{d } \left ( \dfrac{gm_c}{c^3}\right)^{5/6 } \dfrac{1+\cos^2\iota}{2 } , \label{eq : gwwaveform02}\\ & \psi\left(f \right ) = 2\pi f\tilde{t}_c - \phi_c -\dfrac{\pi}{4 } - \tilde{\phi}\left(f \right ) , \label{eq : gwwaveform03}\\ & \tilde{\phi } \left(f\right ) = \dfrac{10}{3 } \left ( \dfrac{8\pi gm_c}{c^3}\right)^{- 5/3 } \left [ - f \int_\infty^{f } df ' \ \dfrac{\chi^{11/2}}{f'{}^{11/3 } k \left ( 1 + \tilde{c } j \right ) } + \int_\infty^{f } df ' \ \dfrac{\chi^{11/2 } } { f'{}^{8/3}k \left ( 1 + \tilde{c } j \right ) } \right ] , \label{eq : gwwaveform04}\\ & j\left(x\right ) = \dfrac{4x^{11/2-\alpha}}{\left ( 1 + x^{3-\alpha}\right)^{7/2 } } , \label{eq : gwwaveform05}\\ & k\left(x\right ) = \dfrac{\left ( 1 + x^{3-\alpha } \right)^{5/2 } \left(1+\alpha x^{3-\alpha}/3 \right)}{1 + \left(4-\alpha\right)x^{3-\alpha } } , \label{eq : gwwaveform06}\\ & \chi = \left ( \delta \varepsilon \right)^{1/(\alpha-3 ) } x , \label{eq : gwwaveform07}\\ & \delta = \left(\dfrac{gm_{\text{eff}}}{\pi^2 f^2 } \right)^{(3-\alpha)/3 } , \label{eq : gwwaveform09}\end{aligned}\ ] ] where @xmath145 is the overall amplitude , @xmath146 is the chirp mass defined by @xmath147 , @xmath148 is the sum of the binary coalescence time @xmath149 and @xmath150 , @xmath151 is the phase at coalescence , @xmath0 is the power - law index of the dm mini - spike , @xmath152 is defined by @xmath153 , @xmath154 is a new frequency variable , @xmath110 is defined in eq . ( [ eq : def_x ] ) , and @xmath155 is defined by eq . ( [ eq : def_epsilon ] ) . the dm information is encoded in the waveform eq . ( [ eq : gwwaveform11 ] ) through @xmath156 , @xmath157 , @xmath158 , @xmath152 and @xmath159 . so if we take @xmath160 , @xmath161 , @xmath162 , @xmath163 , then eq . ( [ eq : gwwaveform11 ] ) becomes the waveform without the dm shown in eqs . ( [ eq : gw_without_dm_01])-([eq : gw_without_dm_04 ] ) . as we will discuss in the next section , we consider a five - year observation by elisa which corresponds to @xmath165 . in this setup , @xmath166 is satisfied . for example , we get @xmath167 for @xmath168 , @xmath169 , @xmath170 and the parameters @xmath33 , @xmath32 , and @xmath59 listed in the table [ tbl : dmparameters ] . so @xmath171 can be treated as a small expansion parameter . since the measurement errors of the physical parameters contained in gw are much more sensitive to the gw phase rather than its amplitude , we expand the gw waveform @xmath172 up to the first order in @xmath173 in the phase and up to the zero - th order in the amplitude . using an expansion of @xmath158 in @xmath173 , @xmath174 the gw waveform given by eqs . ( [ eq : gwwaveform01])-([eq : gwwaveform09 ] ) becomes @xmath175 , \label{eq : gwwaveform13}\\ & l\left(f\right ) = 1 + 4c_\varepsilon \tilde{\delta}^ { ( 11 - 2\alpha ) / \left[2\left(3-\alpha\right)\right ] } , \label{eq : gwwaveform14 } \\ & \tilde { \delta } = \left(\dfrac{g}{\pi^2 f^2 } \right)^{(3-\alpha)/3 } , \label{eq : gwwaveform15}\\ & c_\varepsilon = \meff^{\left(11 - 2\alpha\right)/6 } \tilde{c } \varepsilon^{(11 - 2\alpha)/\left[2\left(3-\alpha\right)\right ] } , \label{eq : gwwaveform16}\end{aligned}\ ] ] where the overall amplitude @xmath145 is defined by eq . ( [ eq : gwwaveform02 ] ) and @xmath176 is defined by eq . ( [ eq : gwwaveform03 ] ) . the upper bound of the integration in eq . ( [ eq : gwwaveform13 ] ) , @xmath177 in the elisa frequency band , is the gw frequency when the stellar mass object enters the innermost stable circular orbit . hence , @xmath178 is in essence the gw cycles from the frequency @xmath132 to the coalescence . the post - newtonian ( pn ) effects which are neglected in the above equations must be taken into account in real data analysis . however , the frequency - dependence of the pn effect in the gw phase @xmath179 differs from that of the dm effect which depends on the power - law index @xmath0 . so , the measurement accuracies of the dm parameters as we will discuss later would not be affected seriously by higher order terms in the pn expansion . until the previous sections , we have included both the dynamical friction and the gravitational pull of the dm mini - spike . it is easily shown that the dynamical effect has much more impact on the measurement accuracy of the dm parameters than the dm mini - spike does @xcite , and the above expression indeed includes the dynamical friction but not the gravitational pull of the dm mini - spike . in fact , within the approximation in this subsection and the following , the gravitational potential of the dm mini - spike shows its signature only in the imbh mass redefinition ( @xmath180 in @xmath146 in the above equations ) . we however note that even such a tiny effect as the gravitational pull of the dm mini - spike do affect the detectability of gw thanks to the large number of the gw cycles in the elisa detection band @xcite . it is important to note that the dm parameters appear only in @xmath0 and @xmath181 and that they are contained in the gw phase @xmath182 . we make use of the above equations to calculate measurement errors of the waveform parameters in the next section . we also define the phase difference @xmath183 by @xmath184 where @xmath182 defined by eq . ( [ eq : gwwaveform13 ] ) is the phase including the dm effect and @xmath185 defined by eq . ( [ eq : gw_without_dm_04 ] ) is the phase without the dm effect . @xmath186 is shown in fig . [ fig : phase_diff ] which indicates that the phase difference becomes significant for large @xmath0 and for the large gw frequency @xmath132 . this is because in this case , the dm density near the central bh increases and the effect of the dm on the motion of the stellar mass object is significant . as we discussed in our previous paper @xcite , the phase difference causes the mismatch between the waveform including the dm effect and the waveform without the dm effect . the phase difference @xmath187 typically above 1 indicates the necessity to use the waveform including the dm effect as a template . as can be seen in fig . [ fig : phase_diff ] , if the template without the dm effect is applied to the gw signal including the effect induced by the dm with @xmath188 , the resulting @xmath189 would degrade significantly . against the power - law index @xmath0 , defined by eq . ( [ eq : phase_difference ] ) . in essence , this is the difference between the accumulated phase from gw frequency @xmath132 and the binary coalescence with and without the dm mini - spike . three different curves show @xmath190 for three different values of @xmath0 . for instance , if detecting a binary gw from @xmath191hz to its coalescence , we would observe by a factor of @xmath192 more gw cycles in the case with a @xmath49 dm mini - spike than without any . for this plot , we take @xmath170 and @xmath33 , @xmath32 , and @xmath59 are as listed in the table [ tbl : dmparameters ] . , width=302 ] in this subsection , we give a brief review of parameter estimation ( see @xcite for more details ) . let us consider detecting gws with a single detector . the detector output @xmath193 can be written by the sum of the gw signal @xmath194 and detector noise @xmath195 : @xmath196 assuming that the detector noise is stationary , the correlation between different fourier components of the noise is expressed as @xmath197 where the angled brackets @xmath198 denote an ensemble average , the asterisk is complex conjugation and @xmath199 is a one - sided power spectral density of the detector noise . in this paper , we consider the gw observation using elisa which has the best sensitivity at around @xmath200 $ ] . the noise spectral density of elisa is given by @xmath201 ^ 2 , \label{eq : elisa_sn}\end{aligned}\ ] ] where @xmath202 $ ] is the acceleration noise spectral density , @xmath203 $ ] is the shot noise spectral density , @xmath204 $ ] is the other measurement noise spectral density and @xmath205 $ ] is the separation between the spacecraft which is the length of its arms of the laser interferometer ( see @xcite for details ) . it is convenient to introduce a noise - weighted inner product between two signals @xmath206 and @xmath207 by @xmath208 where @xmath209 denotes the real part and @xmath210 is the initial frequency . assuming that the detector noise is gaussian and stationary , the probability density of the detector noise is described by @xmath211 . we can rewrite this expression in the form of detector signal @xmath193 and gws signal @xmath194 using eq . ( [ eq : detector_output ] ) as @xmath212 . in the above case , @xmath194 is known , while in actual gw experiments , @xmath194 should be replaced with a template @xmath213 , where @xmath214 is a collection of unknown parameters . to determine the waveform parameters @xmath215 , it is necessary to search for the parameters which minimize the logarithm of the maximum likelihood ratio , @xmath216 . as a result of this process , we can infer the values of @xmath215 . however , the expected values have statistical errors because the detector noise is a random process . these measurement errors @xmath217 of the waveform parameters are approximately described by the gaussian probability distribution for large @xmath189 , @xmath218 where @xmath219 is the normalization factor and @xmath220 is called the fisher information matrix defined by @xmath221 the inverse of the fisher matrix gives the root - mean - square ( rms ) errors of the waveform parameters @xmath222 : @xmath223 where @xmath224 denotes the diagonal elements of the inverse fisher matrix . the inspiral gw waveform from the imbh surrounded by the dm mini - spike is described by six parameters which appear in eqs . ( [ eq : gwwaveform11 ] ) - ( [ eq : gwwaveform16 ] ) : the overall amplitude , @xmath145 ; the time constant , @xmath225 , which is the sum of the traveling time @xmath150 and the coalescence time @xmath149 ; the coalescence phase , @xmath151 ; the chirp mass , @xmath146 ; the two dm parameters , @xmath0 and @xmath181 . note that the beam pattern function of elisa is neglected here because we are concerned with how the dm parameters are determined by gw observations but not with the angular resolution of elisa ( see @xcite for discussion of angular resolution ) . the inner product between the derivatives of the waveform with respect to the parameters @xmath226 yields the values of the fisher matrix elements . the derivatives with respect to @xmath227 and @xmath228 are calculated straightforwardly as follows : @xmath229 the derivatives with respect to the dm parameters @xmath230 are obtained by applying the chain rule to the following equations : @xmath231 where @xmath232 is defined in eq . ( [ eq : gwwaveform14 ] ) . however , since the explicit expressions are complicated , we take the derivatives numerically . next , we derive the initial frequency at which the gw observation starts . in the presence of the dm mini - spike , the stellar mass object orbiting the central imbh loses its angular momentum gradually due both to the dynamical friction and gw radiation reaction . so the coalescence arises earlier than the case without the dm . time evolution of the frequency is described by eq . ( [ eq : dfdt ] ) , @xmath233 , \label{eq : dfdtau}\end{aligned}\ ] ] where @xmath234 and @xmath235 is the time to the coalescence . ( note that @xmath236 . ) the lower bound @xmath237 of the integral in eq . ( [ eq : inner_product_def ] ) is required for calculating the inner product in the fisher matrix . given that the gw is observed by elisa for 5 years prior to the coalescence , this bound is obtained by @xmath238\right).\end{aligned}\ ] ] by numerically solving eq . ( [ eq : dfdtau ] ) , we show the dependence of @xmath0 on @xmath210 in fig . [ fig : fini ] . this figure indicates that the dm mini - spike affects more strongly the motion of the stellar mass object for larger @xmath0 . the initial frequency for a 5 year observation is almost constant for small @xmath0 due to the smallness of the effect of the dm . conversely , the initial frequency drops sharply for large @xmath0 due to the dynamical friction from the dm . . we assume that the gw is detected by an elisa 5 year observation . for small @xmath0 , @xmath210 is almost constant . on the other hand , for large @xmath0 , @xmath210 drops sharply due mainly to the dynamical friction . for this plot , we take @xmath170 and @xmath33 , @xmath32 , and @xmath59 are as listed in the table [ tbl : dmparameters].,width=302 ] in this section , we consider the most likely case that the dm mini - spike has an initially nfw profile , @xmath49 . derivatives of the waveform with respect to the parameters given by eqs . ( [ eq : waveform_derivative_01])-([eq : waveform_derivative_06 ] ) are calculated numerically . substitution of these results into eqs . ( [ eq : fisher_matrix_def ] ) and ( [ eq : rms_error ] ) gives rise to the rms errors @xmath239 , @xmath240 , @xmath241 , @xmath242 , @xmath243 and @xmath244 as follows . @xmath245 \left(\dfrac{10}{s / n}\right ) , \\ \delta \phi_c & = 1.3 \left[\text{rad}\right ] \left(\dfrac{10}{s / n}\right ) , \\ \dfrac{\delta m_c}{m_c } & = 3.1 \times 10^{-7 } \left(\dfrac{10}{s / n}\right ) , \label{eq : chirpmass_error}\\ \dfrac{\delta \alpha}{\alpha } & = 1.2 \times 10^{-6 } \left(\dfrac{10}{s / n}\right ) , \label{eq : alpha_error}\\ \dfrac{\delta c_\varepsilon}{c_\varepsilon } & = 5.9 \times 10^{-5 } \left(\dfrac{10}{s / n}\right ) . \end{aligned}\ ] ] here we take @xmath33 , @xmath32 , and @xmath59 from table [ tbl : dmparameters ] and @xmath170 . these measurement errors are inversely proportional to @xmath189 . so the waveform parameters are measurable with better accuracy for larger gw signals . a notable feature of the above results is that the chirp mass @xmath146 and the two dm parameters @xmath0 and @xmath181 are determined much more accurately than the overall amplitude @xmath145 , the coalescence time @xmath148 and the coalescence phase @xmath151 . this fact reflects that @xmath146 , @xmath0 and @xmath181 appear in the phase of the waveform @xmath182 . from eqs . ( [ eq : number_of_circles ] ) , ( [ eq : phase_t_def ] ) and ( [ eq : tilde_phi_def ] ) , the gw phase is proportional to the number of gw cycles which amplify the sensitivity to the parameters which appear in the phase @xmath182 by a factor @xmath246 . thus , the fractional error of the chirp mass which is proportional to the phase is order of @xmath247 and the two dm parameters are also determined very accurately . in fact , fig . [ fig : ncycle ] indicates the value of @xmath247 is about @xmath248 , which is consistent with the value of @xmath249 in eq . ( [ eq : chirpmass_error ] ) . we also investigate the correlation between the parameters which appear in the phase , @xmath146 , @xmath250 and @xmath181 . figure . [ fig : fisher_ellipse ] illustrates the fisher ellipses for @xmath251 and @xmath181 in @xmath252 . from the figures , we observe that @xmath251 and @xmath181 are strongly correlated with each other because all of them are contained in the phase . however , they are not completely degenerate and are determined independently . this fact can be traced to the difference of the frequency - dependence between @xmath251 and @xmath181 . in the case where the initial dm halo has an nfw profile and the final profile has the radial power - law index of @xmath49 through an adiabatic growth . we assume @xmath33 , @xmath32 , and @xmath59 from the table [ tbl : dmparameters ] and @xmath170 . , width=566 ] in the above discussion , the mass of the central imbh @xmath78 and that of the stellar mass object @xmath64 are fixed . next we analyze the measurement errors for various values of @xmath64 and @xmath78 . the results are shown in fig . [ fig : error_virsus_mbh_mu ] . the figure indicates the errors of the parameters in the phase @xmath182 increase linearly with the stellar mass object mass @xmath64 . this behavior comes from the fact that the number of cycles @xmath253 decreases in proportion to the stellar mass object mass @xmath64 . similarly , the larger is the mass of the imbh , the smaller the number of the orbital cycles the stellar mass object experienced in the five years prior to the coalescence within the elisa band . for this reason , the measurement errors in @xmath146 , @xmath0 , and @xmath254 increase for a larger imbh mass as can be seen in fig . [ fig : error_virsus_mbh_mu ] . versus ( a ) the central bh mass @xmath78 and ( b ) the stellar mass object mass @xmath64 for @xmath255 and @xmath49 . for this plot , @xmath33 and @xmath32 are taken from the table [ tbl : dmparameters ] . the other parameter is fixed to be @xmath256 in the left and @xmath257 in the right , respectively . note that the both axes are in the logarithmic scales . the solid line , the dashed line , the dashed - dotted line correspond to @xmath258 respectively . , width=680 ] we now extend the analysis in the previous section where we considered the case of the initially nfw profile . we next consider the general case without specifying the value of @xmath0 with @xmath59 , @xmath33 , and @xmath32 fixed to the values quoted in the table [ tbl : dmparameters ] . the rms errors depend on the dm power - law index @xmath0 . we show @xmath242 , @xmath243 and @xmath259 in fig . [ fig : error_versus_alpha ] . as shown in fig . [ fig : error_versus_alpha ] , the accuracy of the dm parameters @xmath243 and @xmath260 are better for the larger @xmath0 . this is because steeper density distributions contain more dm mass within the orbital radius ( see fig . 2 in @xcite ) . in other words , the steeper density distribution has more impact on the motion of the stellar mass object and the gw waveform is modified more strongly by the dm mini - spike . so the dm information can be extracted from the gw waveform if the dm mini - halo near the bh has a steep profile . on the other hand , the measurement accuracy of the parameters which appear in the phase @xmath261 become worse in @xmath262 . this feature can be explained by the number of gw cycles @xmath253 which will be discussed in the appendix [ app : ncycle ] . there we show that @xmath253 falls sharply at @xmath263 ( see fig . [ fig : ncycle ] ) . the sensitivity to the parameters which appear in the phase @xmath261 is amplified by the number of circles @xmath253 in the frequency bandwidth of elisa . for this reason , the measurement errors of @xmath251 and @xmath264 increase suddenly at @xmath263 , as is shown in fig . [ fig : error_versus_alpha ] . we also note that this figure shows that we can measure the power - law index @xmath0 at 10 % level even for a moderately flat radial distribution with @xmath265 . in fact , when considering the gravitational pull due to the dm potential only , it affects detectability of gw signals only for @xmath266 . it is the dynamical friction that enables us to explore a flatter dm distribution than `` a dm mini - spike '' referred in the literature that has @xmath267 . figure [ fig : error_versus_alpha_varrho ] shows the relative errors of the dm parameters , @xmath0 and @xmath181 for various values of @xmath268 as a function of @xmath0 . as can be seen in fig . [ fig : error_versus_alpha_varrho ] , the relative errors for the fixed @xmath0 become smaller approximately linearly as the dm density increases . this behaviour can be traced to the amount of the dm within the orbital radius of the stellar mass object . it should be noted that the value of @xmath268 we adopt in this paper is derived under the assumption that the initial dm mini - halo profile is the nfw profile as discussed in the section [ sec : mini - halo_model ] . even if the dm density is an order of magnitude more sparse than that indicated by the nfw profile , the power - law index @xmath0 is measurable with an accuracy of @xmath269 for @xmath270 . versus the power - law index of the dm profile for @xmath252 in the case where the dm mini - spike harboring the imbh has a radially power - law profile . the solid line , the dashed line , and the dashed - dotted line corresponds to @xmath271 respectively . for this plot , @xmath170 and the values of the parameters @xmath59 , @xmath33 , and @xmath32 are assumed as in the table [ tbl : dmparameters ] . , width=302 ] and ( b ) @xmath181 versus the power - law index of the dm profile for @xmath272 in the case where the dm mini - spike harboring the imbh has a radially power - law profile . the solid line , the dashed line and the dashed - dotted line corresponds to @xmath268 , @xmath273 , and @xmath274 respectively . the value of @xmath268 is taken from the table [ tbl : dmparameters ] . , width=680 ] in this paper , we have investigated the measurement accuracy of dark matter ( dm ) parameters by gravitational ( gw ) observations . we consider a binary system composed of an intermediate mass black hole ( imbh ) surrounded by a dm mini - spike and a stellar mass compact object . the compact object falling into the central imbh is affected by the gravitational interaction of both the imbh and the dm mini - spike , namely the gravitational potential of both the imbh and the dm mini - spike , gravitational wave back - reaction and dynamical friction . then the resulting inspiral gw is modified by the dm mini - spike in comparison with the case where the imbh has no dm mini - spike around it . such a gw will be detected by future space - crafted detectors such as elisa / ngo . we find that thanks to the dm parameters contained in the gw phase , the measurement errors of the dm parameters are very small for large power - law index of the mini - spike profile . to put it another way , we can extract the dm parameters very accurately from the gw waveform using matched filtering if the dm mini - spike has a steep profile . indeed , in our reference case as originally advocated by @xcite , we could determine the power - law index of the dm mini - spike radial profile with the 1 @xmath1 relative error of @xmath275 for a gw signal with signal - to - noise - ratio 10 and assuming 5 years observation with elisa , as shown in fig . [ fig : fisher_ellipse ] and indicated by eq . ( [ eq : alpha_error ] ) . we also investigated how accurately the dm parameters can be determined for various dm parameters and the masses of the imbh - stellar mass object binary surrounded by a dm mini - spike . we have found that smaller the mass of the stellar mass object , that of the imbh , or the larger the power - law index of the dm mini - spike , we can measure dm parameters to better accuracy as shown in figs . [ fig : error_virsus_mbh_mu ] and [ fig : error_versus_alpha ] . even a moderately flatter mini - spike with the radial distribution proportional to @xmath276 would still allow us to determine the power - law index to 10 % accuracy . indirect dark matter searches in the gamma - ray band and through gw observation proposed in our previous @xcite and current papers are complementary to each other . the gw observations we propose should be applicable to both very weakly annihilating and non - annihilating dm particles . even if the dm particles do not weakly interact with each other , they affect the motion of the stellar mass object gravitationally and the resulting gw is modified by them . gw are insensitive to absorption and scattering in the interstellar medium during the propagation unlike electromagnetic waves . therefore gw observations offer information on the dm mini - spike directly . on the other hand , if the dm particles self - annihilate , an annihilation plateau may develop within a hubble time @xcite and the power - law index @xmath0 of the dm radial profile becomes effectively zero within a radius @xmath277 . for the case of the values of the parameters listed in the table [ tbl : dmparameters ] , the dm mass @xmath30gev , and its cross section @xmath278 @xcite , we find @xmath279pc which is much larger than the initial orbital radius at which the gw frequency from the binary enters the elisa detection frequency band . hence in this case , gamma - ray searches are a better way to explore a dm mini - halo surrounding the imbh , as indicated by fig . [ fig : error_versus_alpha ] . in summary , the combination of gamma - ray observations with future gw observations will enable us to probe the structure of the dm mini - spike and even to offer hints that may clarify the nature of dm particles . moreover , because the dm profile strongly depends on the formation history of the central imbh , both types of observation may shed light on how the imbh evolved with cosmic history . the authors thank junichi yokoyama and enrico barausse for useful comments . this work is supported by the grant - in - aid for jsps fellows 26@xmath2808636 ( ke ) , the grant - in - aid for young scientists no . 25800126 ( yi ) and the mext grant - in - aid for scientific research on innovative areas `` new developments in astrophysics through multi - messenger observations of gravitational wave sources '' ( grant number 24103005 ) ( yi ) . our goal in this appendix is to rewrite the waveform eq . ( [ eq : waveformf ] ) in the form of an explicit function of gw frequency @xmath132 . according to eqs . ( [ eq : waveformf ] ) , ( [ eq : psit_def ] ) and ( [ eq : tilde_phi_def ] ) , the gw waveform from the binary system composed of the stellar mass object and the imbh surrounded by dm mini - halo is expressed by @xmath281^{1/2 } , \label{eq : waveformf_appendix}\\ & \psi \left(t\right ) = 2\pi f \dfrac{\sourcedistance}{c } + \tilde{\phi } \left(t \right ) - \dfrac{\pi}{4 } , \label{eq : psit_def_appendix}\\ & \tilde{\phi } \left(t\right ) \equiv 2\pi ft - \phi \left(t\right ) , \label{eq : tilde_phi_def_appendix}\end{aligned}\ ] ] where @xmath141 is the time - dependent amplitude defined by eq . ( [ eq : amplitude_t_def ] ) , @xmath139 is the time - dependent phase defined by eq . ( [ eq : phase_t_def ] ) , and @xmath122 is the distance to the source . we proceed as follows . we start with the amplitude @xmath282 . the frequency @xmath132 can be expressed in terms of the orbital radius @xmath86 which is related to the time @xmath113 by eq . ( [ eq : dxdt ] ) . so the amplitude can be expressed as a function of the frequency @xmath132 through the relation between @xmath113 and @xmath132 . next , we tackle the phase @xmath283 . finally , combining these results , we find the explicit expression of the gw waveform in the fourier domain . the gw frequency @xmath284 which is defined by eq . ( [ eq : orbital_frequency ] ) is expanded in a taylor series in a power of @xmath86 : @xmath285^{1/2 } \nonumber\\ & = \dfrac{\sqrt{g m_{\text{eff}}}}{\pi } r^{-3/2}\left [ 1 + \dfrac{1}{2 } r^{3-\alpha } \varepsilon -\dfrac{1}{8 } r^{2(3-\alpha ) } \varepsilon^2 + \cdots \right].\end{aligned}\ ] ] inverting this equation , we obtain @xmath86 as a function of gw frequency and expanded in @xmath155 : @xmath286 , \\ & \delta \equiv \left(\dfrac{gm_{\text{eff}}}{\pi^2 f^2 } \right)^{(3-\alpha)/3 } , \label{eq : delta_def } \end{aligned}\ ] ] where we introduce a new frequency variable @xmath154 defined by eq . ( [ eq : delta_def ] ) for convenience . using the definition of @xmath110 given by eq . ( [ eq : def_x ] ) , the dimensionless radius parameter @xmath110 can be expanded in a power of @xmath155 : @xmath287 where we introduce @xmath158 for convenience . note that the function @xmath158 is equal to one when a dm mini - spike is not present around an imbh . for later convenience , we rewrite @xmath288 which is defined by eq . @xmath289 as follows . @xmath290 } - c_{\text{df } } \dfrac{1}{\left(1+x^{3-\alpha } \right)^{1/2}\left [ 1+\left(4-\alpha \right)x^{3-\alpha } \right]x^{-5/2 + \alpha } } \nonumber\\ & = - c_{\text{gw } } f_{\text{gw}}\left ( x\right ) - c_{\text{df } } f_{\text{df}}\left ( x\right ) \nonumber\\ & = - c_{\text{gw } } f_{\text{gw}}\left(x\right ) \left [ 1 + \dfrac { c_{\text{df } } } { c_{\text{gw } } } \dfrac { f_{\text{df}}\left(x\right ) } { f_{\text{gw}}\left(x\right ) } \right ] \nonumber\\ & = - c_{\text{gw } } f_{\text{gw}}\left(x\right ) \left [ 1 + \tilde{c } j\left(x\right ) \right ] , \label{eq : dxdt_x}\end{aligned}\ ] ] where functions @xmath291 and @xmath157 and a coefficient @xmath152 are defined by @xmath292},\\ & f_{\text{df}}\left(x\right ) \equiv \dfrac{1}{\left(1+x^{3-\alpha } \right)^{1/2}\left [ 1+\left(4-\alpha \right)x^{3-\alpha } \right]x^{-5/2 + \alpha } } , \\ & j\left(x\right ) \equiv \dfrac { f_{\text{df}}\left(x\right ) } { f_{\text{gw}}\left(x\right ) } = \dfrac{4x^{11/2-\alpha}}{\left ( 1 + x^{3-\alpha}\right)^{7/2 } } \label{eq : j_def},\\ & \tilde{c } \equiv \dfrac { c_{\text{df } } } { c_{\text{gw}}}.\end{aligned}\ ] ] the coefficient @xmath152 is the ratio of the dynamical friction coefficient to the gravitational radiation coefficient . so @xmath152 includes the dm information . next , we rewrite the second time derivative of @xmath293 , @xmath294 , as a function of @xmath110 . from eq . ( [ eq : phase_t_def ] ) , @xmath294 is expressed by @xmath295 } \dfrac{3+\alpha x^{3-\alpha}}{x^{5/2 } \left(1+x^{3-\alpha}\right)^{1/2}}\dfrac{dx\left(t\right)}{dt}.\label{eq : domegadt}\end{aligned}\ ] ] to move from the first line to the second , we have made use of eqs . ( [ eq : orbital_frequency ] ) and ( [ eq : def_x ] ) . the time derivative of @xmath110 displayed in the right - hand side of eq . ( [ eq : domegadt ] ) can be rewritten as a function of @xmath110 by eq . ( [ eq : dxdt_x ] ) . so we can express @xmath294 as a function of @xmath110 : @xmath296 } c_{\text{gw } } \left [ 1 + \tilde{c } j\left(x\right ) \right ] \times f_{\text{gw}}\left(x\right ) \dfrac{3+\alpha x^{3-\alpha}}{x^{5/2 } \left(1+x^{3-\alpha}\right)^{1/2 } } \nonumber\\ & = \left(gm_{\text{eff } } \right)^{1/2}\varepsilon^{3/\left[2(3-\alpha)\right ] } c_{\text{gw } } \left [ 1 + \tilde{c } j\left(x\right ) \right ] \times \dfrac{3}{4}x^{-11/2 } \dfrac{\left ( 1 + x^{3-\alpha } \right)^{5/2 } \left(1+\alpha x^{3-\alpha}/3 \right)}{1 + \left(4-\alpha\right)x^{3-\alpha } } \nonumber\\ & = \left(gm_{\text{eff } } \right)^{1/2 } \varepsilon^{3/\left[2(3-\alpha)\right ] } c_{\text{gw } } \left [ 1 + \tilde{c } j\left(x\right ) \right ] \times \dfrac{3}{4}x^{-11/2 } k\left(x\right ) , \label{eq : ddphi_rewriting}\end{aligned}\ ] ] where the function @xmath156 is defined by @xmath297 note that @xmath298 is equal to one when a dm mini - spike is not present around an imbh . substitution of eqs . ( [ eq : orbital_frequency ] ) and ( [ eq : def_x ] ) into e. ( [ eq : amplitude_t_def ] ) gives @xmath299 after some algebra and simplification . combining eqs . ( [ eq : a_rewriting ] ) and ( [ eq : ddphi_rewriting ] ) , we finally arrive at the final expression for the amplitude , @xmath300 } c_{\text{gw}}^{-1/2 } \left [ 1 + \tilde{c } j\left(x\right ) \right]^{-1/2 } x^{11/4}k\left(x\right)^{-1/2 } \nonumber\\ & = \sqrt{\dfrac{32\pi^5}{3 } } \dfrac{g\mu}{\sourcedistance c^4 } \left ( gm_{\text{eff}}\right)^{-1/4 } c_{\text{gw}}^{-1/2 } \varepsilon^{5/\left[4\left ( 3-\alpha\right)\right ] } f^{2}x^{19/4 } \left[k\left(x\right ) \left ( 1 + \tilde{c } j\left(x\right ) \right ) \right]^{-1/2 } \nonumber \\ & = \sqrt{\dfrac{5}{24 } } \dfrac{1}{\pi^{2/3 } } \dfrac{c}{\sourcedistance } \left ( \dfrac{gm_c}{c^3}\right)^{5/6 } f^{-7/6 } \chi^{19/4 } \left[k\left(x\right ) \left ( 1 + \tilde{c } j\left(x\right ) \right ) \right]^{-1/2 } , \label{eq : amplitude_final}\end{aligned}\ ] ] where @xmath146 is defined by @xmath147 and is called the chirp mass . from the second line to the third line , we have used eqs . ( [ eq : def_epsilon ] ) , ( [ eq : c_gw ] ) and ( [ eq : x_delta_epsilon ] ) . our next task is to express the phase @xmath283 given by eq . ( [ eq : psit_def_appendix ] ) as a function of @xmath301 . from eq . ( [ eq : ddphi_rewriting ] ) , the time derivative of frequency @xmath302 is expressed by @xmath303 } c_{\text{gw } } x^{-11/2 } \left[k\left ( 1 + \tilde{c } j\right ) \right ] \nonumber \\ & = \dfrac{3}{5}\pi \left ( \dfrac{8\pi gm_c}{c^3}\right)^{5/3 } f^{11/3 } \chi^{-11/2 } \left[k \left ( 1 + \tilde{c } j \right ) \right ] . \label{eq : dfdt}\end{aligned}\ ] ] we used eqs . ( [ eq : c_gw ] ) and ( [ eq : x_delta_epsilon ] ) to go from the second line to the third line . ( [ eq : dfdt ] ) , we get @xmath304 where the constant of integration is determined by the initial condition of the gw phase . collecting the above results , eqs . ( [ eq : amplitude_final ] ) , ( [ eq : phi_f ] ) and ( [ eq:2pift_f ] ) , we finally obtain the gw waveform in the frequency domain : @xmath305^ { -1/2 } , \\ & \mathcal{a}= \left(\dfrac{5}{24 } \right)^{1/2 } \dfrac{1}{\pi^{2/3 } } \dfrac{c}{\sourcedistance } \left ( \dfrac{gm_c}{c^3}\right)^{5/6 } \dfrac{1+\cos^2\iota}{2 } , \\ & \psi\left(f \right ) = 2\pi f\left(t_c + \dfrac{\sourcedistance}{c}\right ) - \phi_c -\dfrac{\pi}{4 } - \tilde{\phi}\left(f \right ) , \\ & \tilde{\phi } \left(f\right ) = \dfrac{10}{3 } \left ( \dfrac{8\pi gm_c}{c^3}\right)^{- 5/3 } \left [ - f \int_\infty^{f } df ' \ \dfrac{\chi^{11/2}}{f'{}^{11/3 } k \left ( 1 + \tilde{c } j \right ) } + \int_\infty^{f } df ' \ \dfrac{\chi^{11/2}}{f'{}^{8/3 } k \left ( 1 + \tilde{c } j \right ) } \right],\end{aligned}\ ] ] where @xmath145 is the overall amplitude , @xmath149 is the coalescence time , @xmath151 is the coalescence phase . note that when a dm mini - spike is not present around an imbh , @xmath161 , @xmath306 , @xmath307 , so the waveform becomes @xmath308 this is consistent with the waveform from the binary composed of two point - like compact object with mass @xmath64 and @xmath59 @xcite . the detector sensitivity to the inspiral gws is closely related to the number of gw cycles . that is because @xmath253 which is defined by eq . ( [ eq : number_of_circles ] ) is proportional to the gw phase which is defined by eq . ( [ eq : phase_t_def ] ) . therefore snr strongly depends on the number of cycles @xmath253 . the number of gw cycles in the frequency range @xmath309 $ ] is defined by @xmath310 where an overdot denotes the time derivative and @xmath302 can be calculated by eq . ( [ eq : dfdtau ] ) ( note that @xmath236 . ) . we define the frequency bandwidth of elisa as the frequency range @xmath311 $ ] within which the square root of the noise spectral density is below half its minimum value : @xmath312 \right),\end{aligned}\ ] ] where @xmath313 is the frequency at which the elisa is most sensitive to the gws . because we assume 5 year observation , depending on the binary configuration , @xmath314 may be smaller or larger than the initial frequency @xmath315 from which the inspiral gw frequency sweeps to the frequency at the innermost stable circular orbit , @xmath316 . taking the initial frequency @xmath210 into account , the frequency bandwidth @xmath317 $ ] in which the inspiral gw sweeps is expressed by and we obtain fig . [ fig : f_min ] from eq . ( [ eq : number_of_circles ] ) . for @xmath319 , the initial frequency @xmath210 at which the gw start to be observed is within the full width at half minimum of @xmath320 . so the minimum frequency @xmath321 which the inspiral gw spend in the detector bandwidth is equal to the initial frequency @xmath210 . on the other hand , for @xmath262 , the initial frequency @xmath210 is out of the full width at half minimum . so @xmath321 is equal to the lower bound of the detector bandwidth of elisa @xmath314 . since @xmath322 is smaller than @xmath323 for all values of @xmath0 and in the cases we studied , @xmath324 using eqs . ( [ eq : dfdtau ] ) , ( [ eq : number_of_circles ] ) , ( [ eq : fmin ] ) and ( [ eq : fmax ] ) , the number of cycles @xmath253 is obtained in fig . [ fig : ncycle ] . the figure shows that @xmath253 is almost constant for small @xmath0 but drops sharply for large @xmath0 . this behavior of @xmath253 is explained by the fact that the dm has more influence on the motion of the stellar mass object for larger @xmath0 . for large @xmath0 , @xmath302 increases sharply as @xmath325 due to the dm effect and the gw frequency of the stellar mass object goes up rapidly through the frequency bandwidth of elisa . it follows from this that larger @xmath0 leads to wider frequency band but to the less number of gw cycles near the best sensitivity of the detector . the sensitivity to the gws is determined by the competition between these two effects .

electronic transport in mesoscopic systems has been at the focus of experimental and theoretical interest during the last decade due to recent development in fabrication of small electronic devices and their interesting equilibrium and non - equilibrium properties . especially interesting are the transport properties of a quantum dot ( qd ) under the influence of external time - dependent fields . the high - frequency signals may be applied to a qd and the time - dependent fields will modify the tunneling current . new effects have been observed and theoretically described , e.g. photon - assisted tunneling through small quantum dots with well - resolved discrete energy states [ 1 , 2 , 3 ] , photon - electron pumps [ 4 , 5 , 6 ] and others . one can investigate the current flowing through a qd under periodic modulation of the qd electronic structure [ 7 ] or periodic ( non - periodic ) modulation of the tunneling barriers [ 6 ] and electron energy levels in both ( left and right ) electron reservoirs [ 8 ] ( see also [ 9 , 10 ] ) . the progress of nanomaterials science has enabled the experimental study of the phase coherence of the charge carriers in many mesoscopic systems . the asymmetric fano line shapes [ 11 ] are observed whenever resonant and nonresonant scattering paths interfere . in some nanostructures , e.g. in single - electron transistors , the fano resonances in the conductance were observed [ 12 ] , which imply that there are two paths for transfer of electrons between a source and a drain . especially the recent experimental and theoretical study with a low - temperature scanning tunneling microscope ( stm ) of the single magnetic atom deposited on a metallic surface showed the asymmetric fano resonances in the tunneling spectra [ 1215 ] . the stm measurements indicate that in tunneling of electrons between stm tip and a surface with a single impurity atom two different paths are present . the electrons can tunnel between the tip and the adsorbate state and directly between the tip and the metal surface . the electronic transport through a qd coupled to the electron reservoirs within a model with two electron tunneling channels was considered in ref . [ 16 ] and it was shown , that transport of electrons through both channels leads to an asymmetric shape of the zero bias voltage conductance curves , which is typical behaviour for a fano resonance resulting from constructive and destructive interference processes for electrons transmitted through both channels . in all papers mentioned above and relating to the electron transport through a qd with the additional ( bridge ) transmission channel the external fields were not applied and the considered systems were driven out of equilibrium only by means of a dc voltage bias . in this paper we address the issue of a qd with a bridge channel between a source and a drain driven out of equilibrium by means of a dc voltage bias and additional time - dependent external fields . in this manner , our paper can be seen as generalization of ref . [ 9 ] to the case of a qd with the additional bridge channel in the presence of external microwave fields which are applied to the dot and two leads , respectively . in literature , different theoretical approaches have been developed to treat the time - dependent , nonequilibrium electron transport processes in the mesoscopic systems . it seems , that the most popular is the nonequilibrium green s function method . however , these green s functions depend on the time arguments and for non trivial quantum models it is a rather difficult task to calculate them . in our treatment of the time - dependent tunneling through a mesoscopic system we use the evolution operator technique ( e.g. [ 17 , 18 ] ) . the final expressions for the qd charge and the current flowing in the system can be described in terms of the corresponding matrix elements of the evolution operator . in our earlier work [ 19 ] we have considered the similar problem solving numerically the corresponding sets of the differential equations satisfied by the matrix elements of the evolution operator . due to the complexity and a large number of these equations we considered only a very limited number of intereting cases although we were able to take into consideration the electronic structure of the lead energy bands and the specific time - dependence of the qd - lead barriers . here we give the analytical expressions for the qd charge and current assuming so called the wide - band limit approximation and the time - independent strength of the qd - lead barriers . as a result , due to these analytical forms , we are able to analyze the required characteristics of the considered system for the very broad class of parameters . additionally , due to the final forms given for some matrix elements of the evolution operator it is possible to build up the expressions for the qd charge or current in the form of the perturbation series . in the next section we present the model and formalism and give the resulting expressions for equations for the corresponding matrix elements of the evolution operator . in section 3 we obtain the approximate solutions for all required matrix elements and give the final forms for the qd charge and the current flowing in the system . the results are presented in section 4 which includes also the summary and a brief discussion . we model the qd coupled to the left and right electron reservoirs with the additional bridge tunneling channel between them by the usually used hamiltonian @xmath0 , where & & h_1 = _ k _ _ k_(t ) a^+_k _ a_k _ + _ d(t ) a^+_d a_d , + & & v = _ k _ v_k_d(t ) a^+_k_a_d + h.c . + _ k_l , k_r v_k_lk_r(t ) a^+_k_l a_k_r + h.c . the operators @xmath1 , @xmath2 are the annihilation ( creation ) operators of the electron in the lead @xmath3 ( @xmath4 ) and in the qd , respectively . the couplings between qd and lead states and between both lead states are denoted by @xmath5 and @xmath6 , respectively . for simplicity , the dot is characterized only by a single level @xmath7 and we have neglected the intradot electron - electron coulomb interaction . we assume the case in which there exist microwave fields applied to the leads and qd . in the adiabatic approximation our time - dependent driven system is described by @xmath8 , @xmath9 , i.e. the energy levels of the leads and qd are driven by the ac field with the frequency @xmath10 and the amplitudes @xmath11 and @xmath12 , respectively . we describe the dynamical evolution of the charge localised on the qd and the current flowing in the system in terms of the time evolution operator @xmath13 ( in the interaction representation ) which satisfies the equation it u(t , t_0 ) = v(t)u(t , t_0 ) , where & & v(t ) = u_0(t , t_0 ) v(t ) u^+_0(t , t_0 ) , + & & u_0(t ) = t(i^t_t_0 dt h_1(t ) ) . here we assume that the interaction between qd and leads and between both leads is switched on in the distant past @xmath14 , i.e. @xmath15 and @xmath16 equal to zero for @xmath17 and takes constant values for @xmath18 . the qd charge and currents flowing in the system can be obtained from the knowledge of the appropriate matrix elements of the evolution operator @xmath13 . the qd charge is given as follows ( cf . [ 18 ] ) : n_d(t ) = n_d(t_0)|u_dd(t , t_0)|^2 + _ k _ n_k _ ( t_0)|u_dk _ ( t , t_0)|^2 , where @xmath19 and @xmath20 denote the matrix elements of @xmath13 calculated within the basis functions containing the electron single - particle functions of the leads and qd , @xmath21 , @xmath22 and @xmath23 , respectively . @xmath24 and @xmath25 represent the initial filling of the corresponding single - particle states . the tunneling current flowing , e.g. from the left lead into the qd and the right lead , @xmath26 , can be obtained from the time derivative of the total number of electrons in the left lead , @xmath27 ( cf . [ 9 ] ) , where n_l(t ) & = & _ k_l n_k_l(t ) = _ k_l(n_d(t_0)|u_k_l d(t , t_0)|^2 + & + & _ q_l n_q_l(t_0)|u_k_l , q_l(t , t_0)|^2 + _ k_r n_k_r(t_0)|u_k_l , k_r(t , t_0)|^2 ) . let us begin with the calculations of the qd charge @xmath28 . then we have to calculate the matrix elements @xmath29 and @xmath30 . using the identity operator @xmath31 the following set of coupled equations can be obtained from eq . ( 3 ) : & & t u_dd(t , t_0 ) = -i_k_d v^*_k_d(t ) u_k_d(t , t_0 ) , + & & t u_k_l d(t , t_0 ) = -iv_k_l d(t ) u_dd(t , t_0 ) - i_k_r v_k_lk_r(t ) u_k_r d(t , t_0 ) , + & & t u_k_r d(t , t_0 ) = -iv_k_r d(t ) u_dd(t , t_0 ) - i_k_l v_k_rk_l(t ) u_k_l d(t , t_0 ) , where v_k_d(t ) & & k_|v(t)|d= v_k_d ( i(_- _ d)(t - t_0 ) + & & - i(_d - _ ) ( t - t_0)/ ) , + v_k_lk_r(t ) & & k_l|v(t)|k_r= v_k_lk_r ( i(_k_l - _ k_r)(t - t_0 ) - + & & - i(_l - _ r)(t - t_0)/ ) . it is easy to show that the equation for @xmath29 can be written as follows : = -^t_t_0dt(k(t , t)u_dd(t,t_0 ) + _ k _ l_k_(t , t ) u_k_d(t,t_0 ) ) , where ( t , t ) = _ k _ v^*_k_d(t ) v_k_d(t ) , + l_k_l(t , t ) = _ k_r v^*_k_r(t ) v_k_rk_l(t ) , and the similar equation can be written for @xmath32 . the formal solution of eq . ( 9 ) written in the form u_k_l d(t , t_0 ) = -i^t_t_0 dt v_k_l d(t ) u_dd(t,t_0 ) - i_k_r ^t_t_0 dt v_k_lk_r(t ) u_k_r d(t ) , ( and the similar equation for @xmath33 ) can be iterated giving + & + & ^_j=2 ( -i)^j _ r_1,l_2,r_3,l_4, ,_j-1 ^t_t_0 dt_1 ^t_j-1_t_0 dt_j v_lr_1(t_1 ) v_r_1 l_2(t_2 ) v__j-1 ( t_j)u_dd(t_j , t_0 ) . + & & here we have introduced the abbreviated form for the vector @xmath34 and replaced it by @xmath3 and @xmath35 . the equation for @xmath36 can be obtained from eq . ( 17 ) by interchanging @xmath37 . inserting these expressions for @xmath38 and @xmath39 into eq . ( 13 ) one obtains the closed integro - differential equation for @xmath29 + & - & ^_j=2 ( -i)^j-1 _ l_1,r_2,l_3, ,_j ^t_t_0 dt_1 ^t_j-1_t_0 dt_j v^*_l_1(t ) v_l_1 r_2(t_1 ) v__j ( t_j ) u_dd(t_j , t_0 ) + & & + ( the second term with the change lr ) . equation ( 18 ) together with the expressions for @xmath40 and @xmath41 , eqs . ( 11 , 12 ) , and for @xmath42 written as follows ( t , t_1 ) = _ = l , r |v_|^2 d_(t - t_1 ) ( i_d(t - t_1 ) + i(_d - _ ) ( t - t_1)/ ) where @xmath43 a fourier transform of the @xmath3-th lead density of states , gives exact , closed equation for @xmath29 . here we have assumed that @xmath44 does not depend on the wave vector @xmath34 and then the similar assumption will be made for @xmath45 . under the wide - band limit ( wbl ) approximation ( e.g. [ 9 ] ) this equation can be analytically solved and such solutions will be considered later . formally , solving eq . ( 18 ) and inserting its solution to eq . ( 17 ) , the solutions for @xmath38 and @xmath39 can be obtained . these functions are needed in the calculations of the first term of @xmath26 ( see eq . 7 ) . in order to calculate @xmath28 we still need @xmath46 . writing down the set of closed equations for @xmath47 , @xmath48 and @xmath49 ( obtained on the basis of eq . ( 3 ) ) and performing similar calculations to those described above , one obtains for @xmath47 ( and similar equation for @xmath50 by interchanging @xmath37 ) : + & - & iv^*_r(t ) + ( -i)^2 ^t_t_0dt_1 _ l v^*_l(t ) v_lr(t_1 ) + & + & ^_j=2,4,6 , ( -i)^j+1 _ r_1,l_2,r_3, ,l_j ^t_t_0dt_1 ^t_j-1_t_0 dt_j v^*_r_1(t ) v_r_1l_2(t_1 ) v_l_jr(t_j ) + & + & ^_j=3,5,7 , ( -i)^j+1 _ l_1,r_2,l_3, ,l_j ^t_t_0dt_1 ^t_j-1_t_0 dt_j v^*_l_1(t ) v_l_1r_2(t_1 ) v_l_jr(t_j ) + & + & ( -i)^2 ^t_t_0 dt_1 _ l v^*_l(t ) v_l(t_1 ) u_dr(t_1,t_0 ) + & + & ^_j=2 ( -i)^j+1 _ l_1,r_2,l_3, ,_j ^t_t_0dt_1 ^t_j-1_t_0 dt_j v^*_l_1(t ) v_l_1r_2(t_1 ) v__j(t_j ) u_dr(t_j , t_0 ) , and for @xmath49 ( needed for calculation of the last term of @xmath51 , eq . ( 7 ) ) : + & + & ^_j=3,5,7 ( -i)^j _ r_1,l_2,r_3, ,l_j-1 ^t_t_0dt_1^t_1_t_0 dt_2 ^t_j-1_t_0 dt_j v_lr(t_1)v_r_1l_2(t_2 ) v_l_j-1r(t_j ) + & - & i^t_t_0 dt_1 v_l(t_1 ) u_dr(t_1,t_0 ) + & + & ^_j=2 ( -i)^j _ r_1,l_2, ,_j-1 ^t_t_0dt_1 ^t_1_t_0dt_2 ^t_j-1_t_0 dt_j v_lr_1(t_1 ) v_r_1l_2(t_2 ) v__j(t_j ) u_dr ( t_j , t_0 ) . the analytical solutions of these equations under the wbl approximation will be discussed in the next section . for calculation of @xmath52 one needs the functions @xmath53 , @xmath49 and @xmath54 . the first two functions are given in eqs . ( 17 ) and ( 21 ) and @xmath54 should be calculated from the set of coupled equations for @xmath50 , @xmath54 and @xmath55 obtained from eq . the result is as follows : + & & - i^t_t_0 dt_1 v^*_l_1(t_1 ) u_dl(t_1,t_0 ) + & + & ^_j=2 ( -i)^j _ r_1,l_2,r_3, ,_j-1 ^t_t_0 dt_1 ^t_1_t_0 dt_2 ^t_j-1_t_0 dt_j v_l_1r_1(t_1 ) v_r_1l_2(t_2 ) v__j-1l(t_j)u_dl(t_j , t_0 ) , where @xmath50 is given by solving eq . ( 20 ) with the replacement @xmath37 . in order to calculate the qd charge @xmath28 or current @xmath26 one has to solve , in the first step , the integro - differential equations satisfied by @xmath29 , eq . ( 18 ) , and @xmath56 , eq . the other needed functions @xmath53 , @xmath54 and @xmath49 can be obtained from eqs . ( 17 , 21 , 22 ) inserting into them @xmath29 , @xmath57 and performing multiple time and @xmath58-vector integrations . unfortunately , it is a very difficult task to solve the integro - differential equations and to perform these integrals in a general case when the leads are characterized by some density of state curves . here , we use the wbl approximation , under which all multiple time and @xmath58-vectors integrals can be performed without difficulties . the wbl approximation has been widely used in calculating of many properties of mesoscopic systems ( e.g. [ 2 , 3 , 5 , 9 , 10 ] ) . it is justified under the conditions that the qd energy level linewidth is much smaller than the bandwidth of the leads and that the density of states and hopping matrix elements vary slowly with energy . furthermore , as we are not going to consider the case of the qd energy level lying close to the edges of the leads energy band or lying close to some singular structure present in the leads density of states , then application of the wbl approximation should be fully justified . the conditions under which we perform our calculations are satisfied in most experimental constructions of mezoscopic systems . as a check , we have performed the direct but time consuming numerical integration of eqs . ( 810 ) ( and similar equations for other functions ) for the rectangular leads density of states and did not find any differences in the results for the time - averaged qd charge or the currents flowing in the considered system . therefore , in this section we consider the electron transport through the qd with the additional bridge tunnel over the dot within the wbl approximation . in this approach the solutions of the integro - differential eqs . ( 19 ) and ( 20 ) in the analytical form can be obtained and the infinite sums of all terms in eqs . ( 17 ) and ( 21 ) can be calculated . let us consider the equation for the matrix element @xmath29 , eq . the function @xmath59 , eq . ( 19 ) , is approximated in wbl as follows + & & + i(_d - _ ) ( t - t_1)/ ) + & & _ 2(t - t_1 ) , where the leads density of states @xmath60 was replaced by the rectangular density of states with the effective bandwidth @xmath61 using similar approximation in calculations of the multiple integrals present in eq . ( 18 ) one obtains u_dd(t , t_0 ) = ( -2 - 2v^2d ^_j=1 ( -iv_rl / d)^j ) u_dd(t , t_0 ) , where @xmath62 , @xmath63 , @xmath64 , @xmath65 . assuming @xmath66 the series can be summed up and finally the equation for @xmath29 reads u_dd(t , t_0 ) = -c_1 u_dd(t , t_0 ) , where @xmath67 . similarly , eq . ( 20 ) becomes u_d(t , t_0 ) = -iv_dd(t)/(1 + iv_rl /d ) - c_1 u_d(t , t_0 ) . the solutions of eqs . ( 25 ) and ( 26 ) are as follows : & & u_dd(t , t_0 ) = ( -c_1(t - t_0 ) ) , + & & u_d(t , t_0 ) = -i/(1 + iv_rl / d ) ^t_t_0 dt_1 v_d(t_1)(-c_1(t - t_1 ) ) and the qd charge @xmath28 can be easily obtained , eq . it can be verified , that the first term of eq . ( 6 ) tends to zero as @xmath68 and finally for the qd charge we have n_d(t ) = 1 ( 1 + ( v_rl / d)^2 ) _ d f_()|a_(,t)|^2 , where a_(,t ) = -^t_t_0dt_1 ( i(_d - ) ( t - t_1 ) - i(_d - _ ) ( t - t_1)/- c_1 ( t - t_1 ) ) in the limit of vanishing bridge over the qd , eq . ( 29 ) reproduces the result of ref . [ 9 ] . the current @xmath26 flowing from the left lead into the qd and the right lead is calculated from the evolution of the total number of electrons in the left lead ( see eq . 7 ) and one can read : j_l(t ) & = & 2 re\{n_d(t_0 ) _ k_l u^*_k_ld(t , t_0 ) ddt u_k_ld(t , t_0 ) . + & + & _ k_l q_l n_q_l(t_0 ) u^*_k_l , q_l ( t , t_0 ) ddt u_k_l , q_l(t , t_0 ) + & + & . _ k_l , k_r n_k_r(t_0 ) u^*_k_l , k_r(t , t_0 ) ddt u_k_l , k_r(t , t_0 ) } . the functions @xmath69 , @xmath70 and @xmath71 are calculated according to eqs . ( 17 ) , ( 22 ) and ( 21 ) , respectively , and after summing up of the corresponding multiple integrals one obtains : u_k_ld ( t , t_0 ) & = & -i1 + ix ^t_t_0dt_1 v_k_ld(t_1 ) u_dd(t_1,t_0 ) , + u_k_l , q_l(t , t_0 ) & = & _ k_l , q_l - v_rl x 1 + x^2 ^t_t_0 dt_1 \{i(_k_l - _ q_l ) ( t_1-t_0 ) } + & - & i1 + ix ^t_t_0 dt_1 v_k_ld(t_1 ) u_dq_l(t_1,t_0 ) , + u_k_l , k_r(t , t_0 ) & = & -i1 + ix^t_t_0 dt_1v_k_ld(t_1 ) u_dk_r(t_1,t_0 ) + & - & i1 + x^2 ^t_t_0 dt_1 v_k_lk_r(t_1 ) , where @xmath29 , @xmath72 and @xmath73 are given in eqs . ( 27 , 28 ) and @xmath74 . one can verify , that the first term in eq . ( 31 ) tends to zero for @xmath68 as we have for this term n_d(t_0 ) |1/(1 + ix)|^2 /2 ( -(t - t_0 ) re c_2 ) , where @xmath75 . the second and third terms of the expression for @xmath26 , eq . ( 30 ) , together with eqs . ( 27 , 28 , 31 - 33 ) give finally the time averaged current : + & & + i m \{1 - x^2(1 + x^2)(1 + ix)^2 _ l df_l ( ) a_l(,t ) } + & & + re \{2x(1+x^2)(1 + ix ) _ l df_r ( ) a_r(,t)}. in the vanishing bridge channel case eq . ( 35 ) coincides with the results of ref . [ 9 ] : j^v_rl=0_l(t)= 2 n_d(t)+ _ l df_l ( ) i m a_l ( , t ) . the current @xmath76 , eq . ( 36 ) , flowing from the left lead to the central region and to the right lead ( through the bridge channel ) is the superposition of four terms . the first term corresponds to the current between two leads and this term is not disturbed by the qd . the form of the second and third terms is the same as for @xmath77 , eq . ( 37 ) , except for the renormalization constants due to the additional tunneling channel . note , that some additional renormalizations also occur due to @xmath78 which enters into the expression for @xmath28 , eq . ( 29 ) , and for @xmath79 , eq . the last term of eq . ( 36 ) is the interference term due to the simultaneous tunneling through two channels . here we show the numerical results of the time - averaged qd charge @xmath80 and the current @xmath76 and its derivatives with respect to the qd energy level position and the chemical potential @xmath81 ( or equivalently , with respect to the gate and source - drain voltages ) for different sets of parameters which characterize our system . we assume the temperature @xmath82 k and the time - average of time - dependent quantities @xmath83 is defined by f(t)= _ 2 12 ^_- dt f(t ) , and because @xmath83 is a periodic function of time , we average it over the period @xmath84 . we take the chemical potential of the right lead @xmath85 as the energy measurement point , @xmath86 . as the potential drop between the left and right leads is given by @xmath87 and @xmath88 is the measured voltage between a source and a drain , then the derivatives of the current @xmath76 with respect to @xmath81 will correspond to the derivatives @xmath89 usually measured in experiments . in experiments the gate voltage controls the position of the qd energy level @xmath7 ( regardless how complicated the relation between the gate voltage and @xmath7 is ) and for that reason to mimic measurements of the qd charge or current vs. the gate voltage we have calculated them vs. the position of the qd energy level @xmath7 . the values of the hybridization matrix elements @xmath5 present in the hamiltonian do not enter the final expressions for the current or qd charge obtained within the wbl approximation . usually the effective linewidth @xmath90 , responsible for the additional tunneling channel for which we should take some values in order to perform numerical calculations . we have taken the values comparable with @xmath5 and estimated @xmath5 ( assuming its @xmath58-independence , @xmath91 ) using the relation @xmath92 , where @xmath61 is the @xmath3-lead s bandwidth and @xmath93 ( @xmath94 , @xmath64 was assumed ) . we assumed the amplitude of the qd energy levels oscillation @xmath12 to be one half of @xmath95 and @xmath96 , if otherwise stated . in our calculations the values @xmath78 were taken from the range ( 010 ) , in @xmath97 units . in the first three figures , figs . 13 , we present the overall shape of the average current @xmath76 and the derivatves of the average current with respect to the qd energy level @xmath7 and the left chemical potential @xmath81 , respectively , against @xmath7 and @xmath81 . the upper panels correspond to the @xmath98 case and the lower ones present the results obtained for the non - vanishing over - dot tunneling channel , @xmath99 . there are quite visible differences between the case of a qd with and without additional bridge tunneling channel ( cf . upper and lower panels in figs 13 ) . for better visualization of the peculiarities of the presented functions and for simpler discussion let us consider the specific cuts of the surfaces given in figs . 1 - 3 . at first , let us consider the dependence of @xmath76 vs. @xmath7 for given values of the left lead chemical potential @xmath81 . in fig . 4 we present such curves for different values of @xmath81 the subsequent curves beginning from the lower one correspond to @xmath100 up to the upper curve ( with the step @xmath101 ) obtained for @xmath102 . the left ( right ) panel corresponds to @xmath98 ( @xmath103 ) . in the case of vanishing over - dot tunneling channel ( the left panel ) the current has a simple structure a single peak localized in the middle between @xmath81 and @xmath85 for smaller values of @xmath81 . the width of this peak increases with increasing @xmath104 ( @xmath86 ) and for greater values of @xmath104 the current is almost independent of @xmath7 localized inside the energy region between @xmath85 and @xmath81 . for non - vanishing over - dot tunneling ( fig . 4 , the right panel ) the curves @xmath76 become asymmetric . with increasing source - drain bias , the current possesses greater values in comparison with the @xmath98 case due to the direct tunneling between both leads . note , however , that due to the interference effects the resulting @xmath76 curves are asymmetric . the interference effects are most visible approximately for @xmath7 lying in the region ( @xmath105 ) . in fig . 5 we show the average current @xmath76 vs. the left lead chemical potential @xmath81 for several values of @xmath7 . for vanishing @xmath78 the corresponding curves are nearly asymmetric with the asymmetry point @xmath106 . with increasing @xmath81 at fixed @xmath7 the current achieves a constant value depending on the position of the qd energy level @xmath107 with respect to the @xmath86 . it means , that electrons which occupy the lead energy levels not too distant from @xmath7 take part in the tunneling process . for greater @xmath81 most of the lead energy levels lying far away from the @xmath7 are inactive in the tunneling between leads through the qd energy level . however , at non - vanishing @xmath78 ( see the right panel of fig . 5 ) these lead energy levels are active and the current @xmath76 vs. @xmath81 is of much richer structure . the current is nearly linearly growing with the increasing @xmath81 ( for larger @xmath81 ) because the tunneling through the qd can be neglected compared with the direct tunneling between both leads . the clearly visible interference effects appear only for @xmath81 not too distant from @xmath7 . in fig . 6 we show the derivatives of the average current vs. the qd energy level @xmath7 obtained for some values of @xmath81 . these are the results of the intersection of the surfaces given in fig . 2 with the planes at constant values of @xmath81 or the results of the differentiation of curves shown in fig . , the most visible differences between the results obtained for @xmath98 and @xmath99 are present for the qd energy level @xmath7 localized approximately between chemical potentials of both leads ( compare , for example , the curves calculated for @xmath102 ) . 7 presents the comparison of the @xmath108 vs. @xmath81 curves calculated for vanishing @xmath78 ( left panels ) and for @xmath103 ( right panels ) for two different values of the amplitudes @xmath95 ( @xmath109 , @xmath96 ) . at the vanishing value of @xmath78 , the shape of the curves is symmetrical in relation to the values @xmath106 although for greater @xmath95 some shoulders appear on both sides of the corresponding peaks in the distance @xmath110 from the curve centres . for non - vanishing @xmath78 , the corresponding curves are approximately asymmetric and for large @xmath81 they tend to constant , non - zero values corresponding to linear increasing of the current at large @xmath81 . it is interesting that with the increasing amplitudes @xmath95 and @xmath12 very clear structures appear on both sides of the corresponding curves . note , that all these curves can be obtained , for example , from the one calculated for @xmath111 and moved along the @xmath81-axis by the corresponding value ( equal to @xmath7 ) . fig . 8 shows the dependence of the average current @xmath76 on the qd energy level @xmath7 and the direct coupling @xmath78 between both leads . the upper and bottom panels correspond to different values of the amplitude @xmath95 ( @xmath109 ) . a very distinct transition from the symmetric to nearly antisymmetric behaviour of the current vs. @xmath7 curves is observed with the increasing value of the over - dot additional coupling between the leads for both values of @xmath95 . the larger amplitudes of the left lead and qd levels oscillations result only in some broadening of the characteristic features of the average current vs. @xmath78 and @xmath7 surface and do not introduce any additional structures on these surfaces ( for the range of parameters where @xmath112 and @xmath113 is not very small ) . fig . 9 presents the average current @xmath76 obtained for the case in which only the qd energy level @xmath7 oscillates with some frequency ( @xmath114 ) and for the small source - drain voltage @xmath115 . the subsequent panels a , b , c and d correspond to the increasing value of the amplitude @xmath12 and the different curves ( broken , thin , thick and very thick ) describe the qd without the over dot channel , @xmath98 , and with this channel at @xmath116 and 10 , respectively . for vanishing @xmath78 ( broken curves ) we observe for small amplitude @xmath12 only a central resonant peak ( fig . 9a ) . with increasing @xmath12 , the subsequent peaks appear and the distance between them and the central peak is an integer multiple of the frequency @xmath10 ( sidebands ) . the location of peaks is independent of the amplitude @xmath12 but their relative intensity values change and with increasing @xmath12 the height of the central peak is reduced . for the relatively large amplitude @xmath12 , the heights of the two neighbouring peaks ( at @xmath117 ) are approximately equal to the height of the central peak . if we take the additional over - dot tunneling channel into consideration , especially for small @xmath12 , the asymmetric shape of the current curve is observed and this asymmetry increases with the increasing strength of the over - dot coupling between both leads . with the increasing amplitude @xmath12 this asymmetry is reduced largely due to the extra , photon - assisted tunneling peaks whose strength increases with the increasing @xmath12 . for sufficiently large values of @xmath12 and @xmath78 the functional dependence of the average current on the qd energy level @xmath7 is nearly the same for vanishing and non - vanishing over - dot tunneling channels . there is only one difference for large @xmath78 the corresponding curve is shifted to the higher values due to the direct channel between both leads . the last two figures 10 , 11 are devoted to the analysis of the average current @xmath76 dependence on the oscillation period @xmath84 of the external fields . in fig . 10 we show the overall dependence of @xmath76 on @xmath84 and the qd energy level @xmath7 . the upper ( lower ) part of the figure presents the results for the vanishing ( non - vanishing ) over dot channel between both leads . the most visible differences between averaged currents calculated for @xmath118 and @xmath99 can be observed for @xmath119 , especially for small values of @xmath84 . more detailed analysis of the @xmath76 dependence on the oscillation period @xmath84 is shown for some chosen values of the parameters @xmath7 and @xmath78 in fig . the thin ( thick ) curves correspond to @xmath120 ( @xmath121 ) and the solid ( broken ) curves correspond to @xmath122 ( @xmath98 ) . additionally , we give the resutls for two values of the amplitude @xmath95 ( @xmath109 , @xmath96 ) , i.e. for @xmath123 and @xmath124 , the left and the right parts of fig . 11 respectively . we observe the characteristic average current oscillations damped with increasing @xmath84 for @xmath7 lying in the central part between the left and right chemical potentials ( see also [ 9 ] ) . these oscillations are present for both @xmath98 and @xmath99 and more visible for greater amplitudes @xmath95 and @xmath12 but the maxima and minima of the oscillating average current are localized at the same value of the oscillation period . note , that the existence of the additional over - dot tunneling channel results approximately in shifting the corresponding curves to higher values without any additional serious modifications . for the qd energy level lying away from the middle point between the left and right chemical potentials , the average current is still an oscillating function of @xmath84 although these oscillations are less transparent and their oscillation period is much longer . to conclude , we have provided a detailed investigation of a qd connected to two leads with an additional over - dot tunneling channel . the harmonic external microwave fields were considered as applied to the qd and two leads which result in time - dependence of the corresponding qd and leads energy levels . the qd charge and the average current flowing in this system were calculated within the evolution operator technique . the corresponding matrix elements of the evolution operator required for the calculation of the qd charge and current were presented in the form of the infinite series of multiple time integrals of the functions containing the information about the coupling between the qd and leads or in the form of the integro - differential equation . applying the wbl aproximation we were able to obtain all required evolution operator matrix elements in closed forms and give the final analytical expressions for the qd charge and current . we have performed the extended numerical calculations for the qd charge , the average current and the derivatives of the current with respect to the gate and source - drain voltages ( fig . the most spectacular influence of the additional bridge tunneling channel is visible in the @xmath76 dependence vs. the position of the qd energy level at the constant source - drain voltage . going from the vanishing values of the over - dot tunneling channel strength to the non - vanishing one due to the interference effects , the corresponding curve transforms from the symmetric to nearly antisymmetric shape . similar influence of the non - vanishing @xmath78 is visible in the dependence of @xmath108 vs. @xmath81 ( fig . the characteristic behaviour of the average current vs. the qd energy level position at the small source - drain voltage is observed for the case when the external oscillating field is applied only to the qd ( fig . 9 ) . for the vanishing over - dot tunneling channel at the small amplitude @xmath12 , the main resonant peak is only observed and with the increasing amplitude @xmath12 the next peaks localized at @xmath7 equal to the multiplicity of @xmath10 appear corresponding to the photon - assisted tunneling . in that case the current is a symmetric curve centered around the main resonant peak . for the nonvanishing over - dot tunneling channel and small amplitudes @xmath12 , the character of the dependence of the average current on the qd energy level position transforms with increasing @xmath78 from the symmetrical to nearly antisymmetrical behaviour . this tendency to the antisymmetrical behaviour with increasing @xmath78 at small amplitude @xmath12 is reduced with increasing @xmath12 . for sufficiently large amplitude @xmath12 the overall behaviour of the average current vs. @xmath7 is very similar for different values of @xmath78 and does not manifest the tendency for the antisymmetry with increasing @xmath78 . 1 . t. h. oosterkamp , l.p . kouwenhoven , a.e.a . koden , n.c . van der vaart , c.j.p.m . harmans , phys . lett . * 78 * , 1536 ( 1997 ) 2 . qing - feng sun , tsung - han lin , phys . * b56 * 3591 ( 1997 ) 3 . qing - feng sun , jian wang , tsung - han lin , phys . rev . * b58 * , 13007 ( 1998 ) 4 . c.a . stafford , n.s . wingreen , phys . lett . * 76 * , 1916 ( 1996 ) 5 . qing - feng sun , tsung - han lin , j. phys . : condens . matter * 9 * , 3043 ( 1997 ) 6 . kouwenhoven , a.t . johnson , n.c . van der vaart , a. van der enden , c.j.p.m . harmans , c.t . foxon , z. phys . b , condens . matter * 85 * , 381 ( 1991 ) 7 . zhao , j. wang , eur . j. * b9 * , 513 ( 1999 ) 8 . y. goldin , y. avishai , phys . rev . * b61 * , 16750 ( 2000 ) 9 . jauho , n.s . wingreen , y. meir , phys . rev . * b50 * , 5528 ( 1994 ) 10 . qing - feng sun , tsung - han lin , j. phys . : condens . matter * 9 * , 4875 ( 1998 ) 11 . u. fano , phys . rev . * 124 * , 1866 ( 1969 ) 12 . j. gres , d. goldhaber - gordon , s. heemeyer , m.a . kastner , h. shtrikman , d. mahalu , u. meirav , phys . rev . * b62 * , 2188 ( 2000 ) 13 . m. plihal , j.w . gadzuk , phys . rev . * b63 * , 085404 ( 2001 ) 14 . v. madhavan , w. chen , t. jamneala , m.f . crommie , n.s . wingreen , science * 280 * , 567 ( 1998 ) 15 . o. ujsaghy , j. kroha , l. szunyogh , a. zawadowski , phys . lett . * 85 * , 2557 ( 2000 ) 16 . buka , p. stefaski , phys . * 86 * , 5128 ( 2001 ) 17 . m. tsukada , n. shima , in : `` dynamical processes and ordering on solid surfaces '' , eds . a. yoshimori and m. tsukada ( springer , berlin 1995 ) , p. 34 grimley , v.c.j . bhasu , k.l . sebastian , surf . * 121*,305 ( 1983 ) 19 . t. kwapiski , r. taranko , physica e , in press against the left lead chemical potential @xmath81 and qd energy level @xmath7 for @xmath98 ( the qd without the over - dot channel , the upper panel ) and for @xmath103 ( lower panel ) . @xmath86 , @xmath125 , @xmath126 , @xmath127 , @xmath96 , @xmath128 and all energies are given in @xmath97 units.,scaledwidth=90.0% ] against @xmath7 for given values of @xmath81 ( beginning from @xmath100 up to @xmath102 ) . the left and right panels correspond to @xmath98 and @xmath103 , respectively , and the other parameters as in fig . 1.,scaledwidth=100.0% ] against @xmath81 for given values @xmath7 ( beginning from @xmath130 up to @xmath131 ) . the broken curves correspond to @xmath111 . the left ( right ) panel corresponds to @xmath98 ( @xmath132 . the other parameters as in fig . 1.,scaledwidth=100.0% ] with respect to the qd energy level @xmath7 for given values of @xmath81 ( beginning from @xmath100 up to @xmath102 ) . the broken curves correspond to @xmath100 . the left ( right ) panel corresponds to @xmath98 ( @xmath103 ) and the other parameters as in fig . 1.,scaledwidth=100.0% ] , @xmath108 , for given values of @xmath7 ( beginning from @xmath130 up to @xmath131 ) . the left ( right ) panels correspond to @xmath98 ( @xmath103 ) and upper ( lower ) panels correspond to @xmath126 ( @xmath133 ) . the other parameters as in fig . 1.,scaledwidth=90.0% ] against @xmath7 for the oscillating qd energy level at @xmath78 = 0 , 4 , 7 , 10 broken , thin , thick and very thick curves , respectively . the panels a , b , c and d correspond to @xmath136 and 7 , respectively . @xmath137 , @xmath138 , @xmath125 , @xmath139 , @xmath140.,scaledwidth=90.0% ] against @xmath7 and the period of the time oscillation of the external field for @xmath98 the upper panel and for @xmath103 the lower panel . @xmath125 , @xmath141 , @xmath124 , @xmath142 , @xmath96.,scaledwidth=90.0% ] against the period of the time oscillation of the external field for @xmath98 the broken curves , and for @xmath122 the solid curves . the thin ( thick ) curves correspond to @xmath121 ( @xmath120 ) . the left ( right ) panel corresponds to @xmath123 , @xmath143 ( @xmath124 , @xmath142 ) . @xmath96 , @xmath125 , @xmath141.,scaledwidth=100.0% ]

despite their overall similarity , the optical light curves of classical novae show a wide variety of timescales and shapes ( e.g. , * ? ? ? * ) . various empirical time - scaling laws have been proposed in the attempt to recognize common patterns and to unify the nova light curves . for example , @xcite proposed a compression of the time scale obtained by normalizing the time to @xmath7 , the time for the brightness to decrease by @xmath8 magnitudes ( usually @xmath9 or 3 ) . the results were however unsatisfactory because slow novae reached their late stages relatively earlier than faster novae . a different approach was adopted by @xcite who , based on a large collection of nova light curves , found that @xmath10 , the visual magnitude at time @xmath11 , is conveniently represented , on average , by @xmath12 at early stages and @xmath13 at later stages with @xmath14 and @xmath15 . the underlying problem of both the above approaches is that , during the decline phase from maximum , the flux in the visual band , initially dominated by free free emission , becomes more and more affected by the increasing contribution from the emission lines , which finally dominates . the presence of these two competing and heterogeneous emission mechanisms is most likely the main cause for the difficulty to find a suitable normalization parameter for the nova light curves . a more appropriate way to monitor the evolution of the visual continuum is to use the strmgren @xmath16 filter , which is designed to avoid strong emission lines and , in particular , the [ ] line ( e.g. , * ? ? ? * ; * ? ? ? * ) . in the case of v1668 cyg , @xcite has indeed shown that the continuum @xmath16 flux declines much faster than the h@xmath17 [ ] fluxes . consequently , the decline rate is much faster in the @xmath16 band than in the visual band . this was demonstrated by @xcite , who showed that the @xmath16 magnitude of v1668 cyg declines with a slope of @xmath18 at early stages and it then changes to @xmath19 at later stages , values that are sensibly larger than those reported by @xcite . the observations of v1500 cyg by @xcite suggest that the visual and infrared continua are well represented by free - free emission during the early decay phase . the same applies to gq mus : @xcite reported that , during the early 40 days ( diffuse enhanced and orion - phases ) , the energy spectrum of @xmath20 was well represented by optically thin free - free emission ( see also * ? ? ? the flux in the optical and infrared appears to decay with time as @xmath21 in various speed classes of novae ( e.g. , * ? ? ? * ; * ? ? ? * for v1500 cyg ) . an interesting attempt to unify the light curves of novae was proposed by @xcite , who suggested that the main parameter determining the shape of the light curve is the radius of the ejected shell , @xmath22 , where @xmath11 is the time from optical maximum , and @xmath23 is the velocity of the ejecta . he plotted various nova light curves in the @xmath24 diagram and found that light curves overlapped each other . the rather large scatter of the data in that diagram is probably due to the fact that the expansion velocity of novae varies with time ( see * ? ? ? @xcite studied the time evolution of the uv continuum flux in twelve co and onemg novae and found a strong correlation between the @xmath25 time and the time of maximum flux in the 1455 continuum ( see fig . 4 in their paper ) . in a subsequent paper , @xcite studied the time evolution of the uv emission lines in seven co novae and found a strong correlation between the line ionization potential and the time of maximum emission normalized to the @xmath25 ( fig . 5 in their paper ) . these results strongly suggest that novae do indeed evolve following a common pattern being time - normalized by the @xmath25 time , which mainly depends on the white dwarf mass ( e.g. , * ? ? ? * ) . recently , ( * ? ? ? * hereafter referred as paper i ) found that the visual and ir light curves of several novae follow a universal decline law , and interpreted that in terms of free - free emission ( paper i ; * ? ? ? * hereafter referred as paper ii ) ; in particular , the time - normalized light curves were found to be independent of the white dwarf mass , the chemical composition of ejecta , and wavelength . they also showed that the uv 1455 light curve , interpreted as blackbody continuum , can also be time - normalized by the same factor as in the optical and ir . the authors , in the above quoted papers , determined the white dwarf mass and other parameters for a number of relatively well - observed classical novae . in the present paper , we apply the above universal decline law to the classical nova gq muscae 1983 . gq mus is the first classical nova in which the supersoft x - ray turnoff was detected with the x - ray satellite _ rosat _ @xcite . this object was also well observed by the uv satellite _ iue _ ( e.g. , * ? ? ? * ) and in near - ir bands of _ jhkl _ ( e.g. , * ? ? ? the next section summarizes the basic observational characteristics of gq mus . in 3 , we revisit the uv light curves of gq mus obtained with the _ iue _ satellite . in 4 , we briefly introduce our method based on our optically thick wind model . light curve fittings of gq mus in the x - ray , uv , optical , and near ir bands are shown in 5 . discussion and conclusions follow in 6 and 7 , respectively . finally , in the appendix , we assess the problem of the contribution of the emission lines to the visual and infrared photometric bands for gq mus and for other well known novae : v1500 cyg , v1668 cyg , and v1974 cyg . gq mus was discovered by liller on 1983 january 18.14 ut ( jd 2,445,352.64 ) at @xmath26 @xcite . @xcite reported two pre - discovery magnitudes , i.e. , @xmath27 obtained by gainsford on 1983 january 15.597 ut ( 2,445,350.097 ) and @xmath28 obtained by pattie on 1983 january 15.628 ut ( 2,445,350.128 ) . @xcite suggested that the outburst took place 34 days before the discovery . in absence of precise estimates , we assume that the outburst took place at @xmath29 jd 2,445,348.0 ( 1983 january 13.5 ut ) , i.e. , 4.6 days before the discovery by liller on january 18.14 , and adopt @xmath30 jd 2,445,348.0 as day zero in the following analysis . early optical _ ubvri _ and near infrared _ jhkl _ observations of gq mus show a fast evolution with @xmath31 days and @xmath32 days @xcite . @xcite observed gq mus over a wide wavelength range from 0.12 10 @xmath33 during the early 40 days ( diffuse enhanced and orion - phases ) . these authors reported an extraordinary large amplitude brightening by @xmath34 mag , with respect to the prenova magnitude of @xmath35 . after about 4 months of slow decline , the brightness stabilized in the range @xmath36 for about 400 days . after that , the nova gradually declined to @xmath37 for another 400 days ( see figures below ) . in this paper we will use the optical and ir photometric data from @xcite and @xcite , which cover about 300 days after discovery . no other systematic photometric observations are available at later stages except for the visual magnitude measurements ( @xmath38 ) obtained with the fine error sensor ( fes ) monitor on board _ iue _ , which cover 500 days after discovery , and the visual photometric data collected by rasnz ( royal astronomical society of new zealand ) and by aavso ( american association of variable star observers ) , which cover 1000 days after discovery . @xcite determined a photometric orbital period of 0.0594 days for gq mus , the shortest known for a classical nova . later on , @xcite revised the ephemeris of gq mus based on the 1989 and 1990 photometric observations , i.e. , @xmath39 we have adopted this ephemeris for our binary model . @xcite also suggested that gq mus is a polar system because of the presence of x - ray emission and high ( @xmath40)/h@xmath41 ratio . based on the optical observation from 1984 march to 1988 march ( 400 1800 days after the outburst ) , @xcite suggested that the observed progressive increase in the ionization level of the nova shell was due to a very hot radiation source with a temperature of @xmath42 k , increasing with time while the bolometric luminosity remained constant . @xcite studied the soft x - ray _ exosat _ light curve from optical maximum to about 900 days after it , and suggested a model of a very hot white dwarf remnant with a maximum temperature of @xmath43 k evolving at constant bolometric luminosity . _ rosat _ observations clearly show that the supersoft x - ray flux decayed about 10 yr after the outburst @xcite . this information is very important because , compared with our model predictions , it allows one to determine the white dwarf mass quite accurately ( see , e.g. , paper i ) . one of the most puzzling features of gq mus is its long lasting brightness in the uv range , which made possible its regular monitoring by _ iue _ over as much as 11.5 yr . because of satellite constraints , the _ iue _ observations could not start before jd 2,445,385.07 , i.e. , 32.4 days after discovery . the early _ iue _ observations of gq mus are discussed in detail by @xcite . in the following we revisit the problem of the color excess @xmath44 of gq mus and describe the long term evolution of the ultraviolet continuum and of the emission lines , as well as the secondary outburst which took place about 151 days after the main one . the ultraviolet spectra were retrieved from the _ iue _ archive through the ines ( _ iue _ newly extracted spectra ) system , which also provides details of the observations . the use of _ iue _ ines data is particularly important for the determination of the reddening correction because of the implementation of upgraded spectral extraction and flux calibration procedures compared to previously published uv spectra . the color excess of gq mus has been determined from the hydrogen balmer lines by @xcite and @xcite , who found @xmath45 and @xmath46 , respectively . about the same range of values has been reported by @xcite and @xcite , who found @xmath47 and 0.50 , respectively , based on the strength of the dust absorption band around 2175 in the early _ iue _ spectra . given the criticality of the reddening correction , we have attempted to improve its accuracy by using two independent methods based on the shape of the uv continuum and on the emission line intensities . the galactic extinction curve @xcite shows a pronounced broad maximum around 2175 due to dust absorption , but it takes the same value @xmath48 at @xmath49 , 1878 , and 2386 , so that the slope of the straight line passing through the continuum points at these wavelengths is insensitive to @xmath44 in a @xmath50 plot . this circumstance can be used to get a reliable estimate of @xmath44 as that in which the stellar continuum becomes closely linear in the 15122386 region , and passes through the continuum points at the above wavelengths . from 8 pairs of short and long wavelength _ iue _ spectra taken during the nebular phase we have in this way found @xmath51 . a different way to estimate the @xmath44 color excess consists in using the observed relative intensities of the 1640 balmer line , and the 2734 and 3203 paschen recombination lines , and compare these with theoretical ratios @xcite . the intensity ratios @xmath52 and @xmath53 were measured in 8 pairs of short and long wavelength _ iue _ low resolution spectra obtained during the nebular phase . from a comparison with the theoretical values for an electron temperature and density of ( @xmath54 , @xmath55)=(20,000 k , 10@xmath56 @xmath57 ) we have obtained 22 independent determinations of the color excess leading to a mean value of @xmath58 . in the following we will adopt the error weighted mean of the two above determinations , i.e. @xmath5 . examples of _ iue _ spectra of gq mus corrected with @xmath59 are reported in figure [ ebvplot ] for the following days ( after the outburst ) : 108 , 151 ( secondary outburst ; see later ) , 202 and 538 . we have measured the mean flux in two narrow bands 20 wide centered at 1455 and 2855 , selected to best represent the uv continuum because little affected by emission lines ( cassatella et al . figure [ plotcont ] shows the time evolution of the @xmath60 ) and @xmath61 ) fluxes and of the uv color index @xmath62 )@xmath63 ) ] . the measurements were made on well exposed low resolution large aperture spectra . unfortunately , the _ iue _ spectrum of day 37 ( swp19299 ) , very important because it was the first obtained , was heavily saturated around 1455 . to make a crude estimate of the corresponding flux we have determined the scaling factor between this and the next spectrum obtained on day 49 ( swp 19383 ) , in a region where both were well exposed ( 1520 - 1620 ) , and assumed that the spectral slope remained the same in the two spectra . the value so obtained , reported as an open circle in figure [ plotcont ] , likely represents an upper limit to the true value because , as shown in the same figure , the uv color was comparatively cooler in the first observations . figure [ plotcont ] reports also , for comparison , the visual light curve obtained from the fine error sensor ( fes ) counts on board _ iue _ , once corrected for fes time dependent sensitivity degradation ( see cassatella et al . 2004 for details on the fes calibration ) . it appears from figure [ plotcont ] that the gradual fading of the uv and visual fluxes was interrupted , around jd 2,445,499 ( day 151 ) , by a secondary outburst , as also noticed by @xcite . the secondary outburst had actually the appearance of a `` uv flash '' because of its especially large amplitude at short wavelengths . indeed , compared with the _ iue _ low resolution observations obtained just before and after this event ( days 108 and 202 ) , the uv flux increased by a factor of 9 at 1455 and by a factor 2.2 at 2885 , while the visual flux increased only by a factor of 1.5 . the consequent hardening of the uv continuum is also reflected by the upturn of the @xmath64 color index , as shown in the same figure . llllll solar composition & 0.7068 & 0.0140 & 0.0018 & 0.0034 & @xcite + gq mus 1983 & 0.37 & 0.24 & 0.0023 & 0.0039 & @xcite + gq mus 1983 & 0.27 & 0.40 & 0.0034 & 0.023 & @xcite + gq mus 1983 & 0.43 & 0.19 & & & @xcite also the uv emission lines suffered from important changes in coincidence with the secondary outburst of day 151 , as shown in figure [ plotlines ] , which reports , as a function of time , the flux in the most prominent permitted uv emission lines , as measured by us from the available _ iue _ low resolution spectra . the figure shows that , on day 151 , the flux of the high ionization emission lines ( 1640 , 1240 and 1550 ) had decreased substantially while , on the contrary , it increased in the low ionization resonance lines of and , so implying a drastic change of the ionization structure of the emitting regions . the other important feature of the secondary outburst was the appearance of a strong p cygni profile in the doublet @xmath65 1393.74 , 1402.77 , shown in figure [ pcygni ] . the figure reports also , for comparison , the profile of the neon nova v1974 cyg obtained 45 days after the outburst . a comparison of observed with theoretical p cygni profiles computed with the sei method @xcite indicate that the terminal velocity of the wind was about the same for the two novae ( @xmath66 3200 km s@xmath67 ) . the special interest of the observations in figure [ pcygni ] is that gq mus had entered the nebular phase well before day 151 , so that the presence of p cygni profiles was not expected to be detectable at this stage . indeed , p cygni profiles were not detectable in an earlier _ iue _ high resolution spectrum of day 49 , despite the factor of two longer exposure time , nor they were at later stages . this strongly suggests that a short duration mass ejection episode took place around day 151 . this episode caused efficient recombination to take place in the emitting region , observed as a substantial , although transitory , fainting of the high ionization emission lines ( cf . fig.[plotlines ] ) . the fast recovery of the emission line spectrum by day 202 suggests that the ejected mass must have been small . also the presence of a flare at uv wavelengths rather than in the optical ( cf . fig . [ plotcont ] ) requires a small opacity in the uv and then a small ejected mass . we present a unified model for the ir , optical , uv , and supersoft x - ray light curves of the 1983 outburst of gq mus . as in paper i , our models are based on the optically thick wind theory of nova outbursts described in @xcite . after a thermonuclear runaway sets in on a mass - accreting wd , its envelope expands greatly to @xmath68 , where @xmath69 is the photospheric radius , and it then settles onto steady - state regime . the decay phase of the nova can be followed by a sequence of steady state solutions ( e.g. , * ? ? ? using the same method and numerical techniques as in @xcite , we have followed the nova evolution by connecting steady state solutions along the envelope mass - decreasing sequence . the equations of motion , radiative diffusion , and conservation of energy are solved from the bottom of the hydrogen - rich envelope through the photosphere under the condition that the solution goes through a critical point of a steady - state wind . the winds are accelerated deep inside the photosphere so that they are called `` optically thick winds . '' we have used updated opal opacities @xcite . one of the boundary conditions for our numerical code consists in assuming that photons are emitted at the photosphere as a blackbody with a photospheric temperature @xmath70 . x - ray and uv fluxes are estimated directly from the blackbody emission , but infrared and optical fluxes are calculated from free - free emission by using physical values of our wind solutions as mentioned below in [ free - free_light_curve ] . physical properties of these wind solutions have already been published ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? optically thick winds stop after a large part of the envelope is blown in the winds . the envelope settles into hydrostatic equilibrium , while its mass decreases in time by nuclear burning . then we solve the equation of static balance instead of the equation of motion . when the nuclear burning decays , the wd enters a cooling phase , in which the luminosity is supplied with heat flow from the ash of hydrogen burning . in the optically thick wind model , a large part of the envelope is ejected continuously during a relatively long period ( e.g. , * ? ? ? after maximum expansion , the photospheric radius gradually decreases keeping the total luminosity ( @xmath71 ) almost constant . the photospheric temperature ( @xmath70 ) increases with time because of @xmath72 . the maximum emission shifts from the optical to supersoft x - ray through ultraviolet ( uv ) and extreme ultraviolet ( euv ) . this causes the luminosity to decrease in the optical and to increase in the uv , until it reaches a maximum . the following decay in the uv is accompanied by an increase of the supersoft x - ray range . these timescales depend crucially on the wd parameters such as the wd mass and the chemical composition of the envelope ( e.g. * ? ? ? * paper i ) . thus , we can follow the development of optical , uv , and supersoft x - ray light curves by a single modeled sequence of steady wind solutions . spectra of novae show blackbody features at very early stages , but free - free emission from optically thin plasma will eventually dominate ( e.g. , * ? ? ? * ; * ? ? ? * ) . during the optically thick wind phase ( see paper i for more details ) , the extended regions outside the photosphere are optically thin . the optical and ir free - free emission fluxes arising from these regions can be estimated from @xmath73 where @xmath74 is the flux at the frequency @xmath75 , @xmath55 and @xmath76 are the number densities of electrons and ions , respectively , @xmath77 is the volume of the optically thin region , @xmath69 is the photospheric radius , @xmath78 is the wind mass loss rate , and @xmath79 is the photospheric velocity . in equation ( [ free - free - wind ] ) we assume that the electron temperature and the degree of ionization are constant in the free - free emitting region . we also assume that @xmath80 , @xmath81 , and use the continuity equation , i.e. , @xmath82 , where @xmath83 and @xmath84 are the density and velocity of the wind , respectively . finally , we assume that @xmath85 in the optically thin region . after the optically thick wind stops , the total mass of the ejecta remains constant in time . the flux from such homologously expanding ejecta is approximately given by @xmath86 where @xmath87 , @xmath88 , and @xmath89 are the density , volume , and total mass , respectively , of the ejecta . we assume that the ejecta are expanding at a constant velocity , @xmath23 . so we have the radius of the ejecta of @xmath90 , where @xmath11 is the time after the outburst . the proportionality constants in equations ( [ free - free - wind ] ) and ( [ free - free - stop ] ) can not be determined a priori because radiative transfer is not calculated outside the photosphere : these were determined using the procedure described below in [ light_curve_fitting ] . the light curves of our optically thick wind model are parameterized by the wd mass ( @xmath91 ) , the chemical composition of the envelope ( @xmath92 , @xmath93 , @xmath94 , and @xmath95 ) , and the envelope mass ( @xmath96 ) at the time of the outburst ( jd 2,445,348.0 ) . for the metal abundance we adopt @xmath97 , which also includes carbon , nitrogen , oxygen , and neon with solar composition ratios . three different sets of abundance determinations for the ejecta of gq mus are available from the literature , but their values are rather scattered , as it appears from table [ gqmus_chemical_abundance ] . we here consider three cases , in order of decreasing hydrogen content in the envelope , not directly corresponding to the three sets of abundance determination in table [ gqmus_chemical_abundance ] : a ) @xmath98 , @xmath99 , @xmath100 , and @xmath97 ; b ) @xmath101 , @xmath102 , @xmath100 , and @xmath97 ; c ) @xmath103 , @xmath104 , @xmath100 , and @xmath97 . these composition sets correspond to assuming 25% , 55% , and 100% mixing of c+o wd matter with a hydrogen - rich envelope with solar composition . if we change @xmath93 while keeping the hydrogen content @xmath92 constant , the light curves hardly change provided @xmath105 . we have searched for the best fit model by changing the wd mass in steps of @xmath106 for the above three sets of chemical compositions . we apply to gq muscae 1983 the model light curves of classical novae described in the previous section , and evaluate its fundamental parameters . the optical @xmath77 magnitude and the near infrared @xmath107 , @xmath108 , @xmath109 , and @xmath110 magnitudes are calculated from free - free emission , while the uv 1455 and supersoft x - ray fluxes are obtained from blackbody emission , as illustrated in figure [ sed_uv_opt_ir_filter ] . models and observations are compared in figures [ all_mass_gq_mus_x55z02c10o10_compare ] [ all_mass_gq_mus_x35z02c10o20_compare_uv_x ] which refer to , respectively , the three following chemical compositions sets : @xmath111 @xmath112 , @xmath113 , and @xmath114 . the white dwarf mass range explored is @xmath115 to @xmath116 , in @xmath106 steps . in each figure , the thick solid line represents the white dwarf mass value that leads to the best representation of the data for the above chemical composition sets . the decay time of the supersoft x - ray flux is a good indicator of the wd mass ( paper i ) . through a careful visual inspection to the data , we have selected the models that best fitted the _ observations @xcite . the model and the observed supersoft x - ray light curves are reported in figures [ all_mass_gq_mus_x55z02c10o10_compare ] [ all_mass_gq_mus_x35z02c10o20_compare_uv_x ] . the light curves in the uv 1455 continuum are a good indicator of photospheric temperature during the early decay phase of novae . such curves are in general quite smooth , as shown by @xcite for several objects observed with _ iue_. a comparison of models with observed uv light curves ( see figs . [ all_mass_gq_mus_x55z02c10o10_compare ] [ all_mass_gq_mus_x35z02c10o20_compare_uv_x ] ) shows that this is also the case of gq mus , except for the two earliest observations on day 37 and 49 , and that of day 151 . leaving these data apart , we find that the model that best fits the data is that of a white dwarf with mass @xmath117 , 0.7 , and @xmath118 , for the three above chemical composition sets , respectively . as for the observation of day 151 , the large deviation from the model s smooth trend is clearly due to the occurrence of the uv flash discussed in 3.2 . the otherwise good agreement between models and observations shown in the above figures may suggest that also the observations of day 37 and 49 are due to a uv flash ( see discussion in [ uv_flushes ] ) . the visual light curves computed from free - free emission are compared with the observed ones in figures [ all_mass_gq_mus_x55z02c10o10_compare ] [ all_mass_gq_mus_x35z02c10o20_compare_uv_x ] for the above chemical composition sets . as discussed in 1 and in papers i and ii , visual magnitudes are contaminated by strong emission lines which will eventually dominate over the continuum , causing an increasing deviation from our free - free models . in gq mus , the forbidden [ ] @xmath65 4959 , 5007 emission lines already appeared 39 days after the outburst @xcite . at about this date the observed visual light curve did actually start to show an increasing deviation from the free free emission model until , by about day 500 , the contribution from the emission lines stabilized so that the observed light curve recovered the the shape of the model curve . this effect is approximatively taken into account by lifting up the template light - curve of gq mus by about 4.5 mag , as shown in figures [ all_mass_gq_mus_x55z02c10o10_compare ] [ all_mass_gq_mus_x35z02c10o20_compare_uv_x ] . on the other hand , the near infrared @xmath119 bands are not so heavily contaminated by emission lines , as shown in figure [ gq_mus_m0700_x45z02c15o20_i_j_h_k ] , where the @xmath119 photometric data from @xcite are compared with our theoretical light curve . the figure shows that the model fits the observations reasonably well until about day 100 ( thick solid line in the figure ) , being anyhow the deviations after that date sensibly smaller than in the visual . the contribution from emission lines is particularly small in the @xmath107 , @xmath109 , and @xmath110 bands , whereas the emission line of @xmath120 contributes somewhat to the @xmath108 band from the very early phase @xcite . to mimic the different contribution by the emission lines at different phases of the nova development , it is sufficient to lift up the template free - free line two times ( starting from the dash - dotted line representing the very early phases , to the thick solid line representing the intermediate phases , and then finally to the thin solid line , representing the late phases ) . to summarize the results in this section , the white dwarf mass that best reproduces the x - ray , uv , optical , and near ir is @xmath121 , where the error bar reflects mainly the uncertainty on chemical composition . in this subsection , we analyze the overall development of the nova outburst based on our model of @xmath122 , @xmath123 , @xmath102 , and @xmath97 ( fig . [ all_mass_gq_mus_x45z02c15o20_compare_uv_x ] ) . figures [ gq_mus_softxray_r_t ] [ gq_mus_t3_uv1455_m0700_x45z02c15o20 ] report some relevant model predictions and observations as a function of time . first we point out two important epochs of the nova development : one is the end of optically thick wind phase on day 1000 , and the other is the end of hydrogen shell burning on day 3300 as summarized in table [ properties_gq_mus ] . because of the rapid shrinking of the the photospheric radius ( @xmath69 ) , also the photospheric temperature ( @xmath70 ) rapidly increases until the wind stops , as indicated in figure [ gq_mus_softxray_r_t ] . also the wind mass loss rate ( @xmath78 ) decreases rapidly from @xmath124 yr@xmath67 to @xmath125 yr@xmath67 during the first 1000 days . thus the total mass ejected by the winds amounts to @xmath126 . after the wind stops , the photospheric radius becomes smaller than @xmath127 , and the photospheric temperature eventually becomes larger than 20 ev , so that the nova enters a supersoft x - ray phase . after the hydrogen shell burning ends , the nova cools down rapidly , and a fast decay phase of the supersoft x - ray flux follows , as already shown in the previous section . in what follows , we summarize some relevant observational aspects and make a comparison with our model predictions . from the coronal lines in the optical spectrum @xcite deduced a luminosity of @xmath128 ergs s@xmath67 and a temperature of @xmath129 k. these values are consistent with our model of a @xmath130 wd , which provides @xmath131 k and @xmath132 ergs s@xmath67 about 2700 days after the outburst , as it may be appreciated from figure [ gq_mus_softxray_r_t ] . @xcite estimated a central source temperature of 164,000 k , a luminosity of @xmath133 , and an ejecta mass of @xmath134 , about 4000 days after the outburst . for our best - fit model of a @xmath135 white dwarf with an initial envelope mass of @xmath136 , the total mass lost by the optically thick wind is @xmath137 . we also obtain @xmath138 k and @xmath139ergs s@xmath140 , 4000 days after the outburst , as shown in figure [ gq_mus_softxray_r_t ] . our values are then very consistent with those obtained by @xcite , except for their larger ejected mass ( see below for another estimate ) . the envelope mass itself depends mainly on the white dwarf mass and slightly on the chemical composition . for a @xmath141 white dwarf , @xmath142 , and @xmath104 , we obtain a slightly larger envelope mass , @xmath143 , and a larger wind mass loss , @xmath144 . on the other hand , for a @xmath145 white dwarf with @xmath146 , and @xmath99 , we obtain a slightly smaller envelope mass , @xmath147 , and a smaller wind mass loss , @xmath148 . using their photoionization model , @xcite estimated the temperatures during the static phase and at the hydrogen burning turnoff to be @xmath149 k and @xmath150 k , respectively , and an ejected shell mass of @xmath151 . these values are consistent with our model , i.e. , @xmath152 k , @xmath153 k , and @xmath154 . the above authors determined the duration of the wind phase and of the static hydrogen burning phase to be @xmath155 yr and @xmath156 yr , to be compared with our values of 2.7 yr and 6.3 yr , respectively . the much longer duration of the wind phase obtained by us is likely due to the rapid decrease of the wind mass loss rate with an e - folding time of 0.24 yr , corresponding to a decrease by a factor of 10 in 0.7 yr . figure [ uv_outburst ] shows the reddening corrected uv color index @xmath64 as a function of the photospheric temperature obtained from our best - fit model at the time of the individual observations ( filled circles ) . the figure indicates that the color index decreases smoothly with increasing temperature except for its abrupt decrease at the time of the uv flash on day 151 ( note that the temperature increases with time in these early phases ) . it is interesting to compare the color@xmath70 values reported in figure [ uv_outburst ] with those derived from @xcite and @xcite , who calculated early - time nova spectra using non - lte atmosphere codes with winds having density and velocity laws within the nova envelope . for this comparison , we have taken the values of @xmath64 from figure 10 of @xcite for @xmath157 , 20,000 , 25,000 , and 30,000 k , and from figure 8 of @xcite for @xmath158 k. the model values of @xmath64 so obtained are plotted as a function of the corresponding effective temperature in figure [ uv_outburst ] ( open triangles ) . it clearly appears from figure [ uv_outburst ] that the two sets of values are very consistent with each other , except for the uv flash on day 151 . this result supports our assumption that the uv 1455 flux is reasonably well accounted for by blackbody emission . indeed , line blanketing is rather small at this wavelength , so that the blackbody model does not deviate much from hauschildt et al.s non - lte models , as already discussed in paper i. the mass of the donor star ( the secondary component ) can be estimated from the orbital period . @xcite obtained @xmath159 days ( @xmath160 hr ) from the orbital modulations with an amplitude of 0.6 mag . using warner s ( 1995 ) empirical formula @xmath161 we get @xmath162 . the orbital separation is then @xmath163 for @xmath164 , the effective radius of the roche lobe for the primary component ( white dwarf ) is @xmath165 , and the effective radius of the secondary is @xmath166 . in our model , the companion emerges from the white dwarf envelope when the photospheric radius of the white dwarf shrinks to @xmath167 ( the separation ) or @xmath168 ( the roche lobe ) . this happens on about day 330 or 500 , respectively , in our best - fit model with @xmath164 , as shown in figures [ gq_mus_softxray_r_t ] and [ gq_mus_uv1455_r_t ] . according to @xcite , strong infrared coronal lines appeared 57 days after optical maximum in v1500 cyg , but were not present in gq mus as late as 97 days after maximum . their appearance roughly coincides with the emergence of the companion from the white dwarf photosphere : this suggests that these strong coronal lines arise from the shock between the white dwarf wind and companion star . indeed , the emergence of the companion in v1500 cyg took place about 50 days after optical maximum ( paper i ) , which is consistent with the appearance of the line 57 days after maximum . in the case of gq mus , the absence of strong coronal lines on day 97 days is consistent with our model in which the companion emerged about 330 or 500 days after maximum . in their study of gq mus , @xcite reported that the [ ] @xmath169 coronal line first appeared 2.2 to 3.4 yrs after the outburst , and became the strongest about 4 years after it . they argued that the coronal line was due to photoionization from a hot radiation source rather than to collisional excitation . in our @xmath135 white dwarf model , the optically thick wind stopped about 960 days ( 2.7 yrs ) after the outburst and then the photospheric temperature gradually increased to above @xmath170 ev 4 yrs after the outburst , as shown in figures [ gq_mus_softxray_r_t ] and [ gq_mus_uv1455_r_t ] . these authors suggested that @xmath171 k in 1984 , which is consistent with our photospheric temperature of @xmath172 k 2.2 yr after the outburst ( figs . [ gq_mus_softxray_r_t ] and [ gq_mus_uv1455_r_t ] ) . @xcite reported that the count rate of the _ exosat _ low - energy telescope was about @xmath173 c s@xmath67 400 days to 700 days after optical maximum , a value that gradually decreased to @xmath174 c s@xmath67 about 900 days after it . if the x - rays originates from the shock between the wind and the companion star , the soft x - ray behavior is consistent with our @xmath135 white dwarf model ( as well as @xmath145 and @xmath141 white dwarf models ) , because the emergence of the companion star is predicted to happen on day 330 , and the wind stopped at 960 days after optical maximum . lllll discovery & ... & 2,445,352.6 & jd & @xcite + nova speed class & ... & moderately fast & & @xcite + @xmath175 & ... & 18 & days & @xcite + @xmath25 & ... & 48 & days & @xcite + @xmath176 from @xmath25 & ... & @xmath177 & mag & @xcite + distance from @xmath25 & ... & 5 & kpc & @xcite + dust & ... & no & & @xcite + orbital period & ... & 1.425 & hr & @xcite + @xmath44 & ... & 4.5 & & @xcite + wind phase & ... & @xmath178 & yr & @xcite + h - burning phase & ... & 3300 & days & @xcite llll outburst day & ... & 2,445,348.0 & jd + opt . maximum & ... & 2,445,352.6 & jd + @xmath175 & ... & 67 & days + @xmath25 & ... & 122 & days + @xmath176 from @xmath25 & ... & @xmath179 & mag + distance from @xmath25 & ... & @xmath180 & kpc + secondary mass & ... & 0.1 & @xmath181 + @xmath44 & ... & @xmath182 & + distance by uv fit & ... & @xmath183 & kpc + @xmath184 & ... & 275 & day + wd mass & ... & @xmath185 & @xmath181 + wd envelope mass & ... & @xmath186 & @xmath181 + mass lost by wind & ... & @xmath187 & @xmath181 + wind phase & ... & @xmath188 & days + h - burning phase & ... & @xmath189 & days + separation & ... & 0.6 & @xmath190 + companion s emergence & ... & @xmath191 & days + uv flash luminosity & ... & @xmath192 & @xmath193 @xcite and @xcite have estimated the distance to gq mus from the apparent and the absolute @xmath77 magnitudes at maximum , @xmath194 and @xmath195 . this latter was computed from the @xmath25 time through the schmidt kaler maximum magnitude rate of decline ( mmrd ) relation @xcite : @xmath196 despite the sensibly different input values adopted of 48 and 40 days for @xmath25 , 0.4 and 0.45 for @xmath44 , and 7.2 and 6 for @xmath197 , these authors obtained quite similar values for the distance , 5 kpc and 4.8 kpc , respectively . a re - analysis of the merged visual magnitude data present in the literature ( see plot in fig . [ gq_mus_t3_uv1455_m0700_x45z02c15o20 ] ) confirms @xmath198 days , in close agreement with @xcite , so that @xmath199 [ cf . equation ( [ schmidt - kaler - law ] ) ] . if , together with @xcite , we take @xmath200 as the best estimate of the apparent magnitude at optical maximum @xcite , we finally obtain the following relation between distance and @xmath44 : @xmath201 which is labeled `` mmrd1 '' in figure [ gq_mus_distance_reddening_m0700_x45z02c15o20 ] . in particular , the value @xmath202 obtained in 3.1 , corresponds to a distance of @xmath203 kpc . following the same procedure outlined in paper i and @xcite , an independent estimate of the distance to gq mus can be obtained by comparing the observed light curve in the 1455 continuum with the corresponding model fluxes in figures [ all_mass_gq_mus_x55z02c10o10_compare ] [ all_mass_gq_mus_x35z02c10o20_compare_uv_x ] , [ gq_mus_uv1455_r_t ] , and [ gq_mus_t3_uv1455_m0700_x45z02c15o20 ] . the calculated flux at @xmath204 = 1455 at a distance of 10 kpc for our adopted model ( @xmath135 , @xmath123 , @xmath102 ) is @xmath205 = 2.85 @xmath206 ergs @xmath207 s@xmath67 @xmath67 on jd 2,445,455.6 . the observed flux at the same date is @xmath208= 1.78 @xmath209 ergs @xmath207 s@xmath67 @xmath67 . from these values one obtains the following relation between distance and reddening : @xmath210 where @xmath211 , and @xmath212 @xcite . this curve , labeled `` uv 1455 , '' is reported in figure [ gq_mus_distance_reddening_m0700_x45z02c15o20 ] . in particular , equation ( [ eq : ebv ] ) provides a distance of @xmath183 kpc for @xmath213 , which is sensibly larger than that obtained from mmrd1 . an alternative way to determine the distance of gq mus is to compare the shape of the theoretical visual light curve ( based on the free free model ) with the observed visual fluxes . such a comparison , done in figure [ all_mass_gq_mus_x45z02c15o20_compare_uv_x ] is very instructive because it shows that the object was up to 1.5 magnitudes brighter than predicted by our universal light curve model during the earliest phases ( until day 10 ) . it is then legitimate to consider gq mus as a super bright nova . an extreme case of this type was v1500 cyg , which , near peak luminosity was about 4 magnitudes brighter than the eddington limit for a 1.0 @xmath181 white dwarf @xcite . it is interesting to note that if such a super bright phase is ignored in gq mus , the @xmath25 and @xmath175 times of the theoretical light curve that best fits the observations are considerably longer than the observation indicates at face value : we in fact estimate @xmath25 = 122 days and @xmath175= 67 days . from equation ( [ schmidt - kaler - law ] ) we then obtain @xmath214 at the time of maximum light ( @xmath215 days ) . since the apparent magnitude at visual maximum of our theoretical light curve is @xmath216 mag ( see fig . [ all_mass_gq_mus_x45z02c15o20_compare_uv_x ] ) , the distance to gq mus can be estimated to be @xmath180 kpc for @xmath5 . if the reddening is not fixed , the following relation applies : @xmath217 which is plotted in figure [ gq_mus_distance_reddening_m0700_x45z02c15o20 ] ( labeled `` mmrd2 '' ) . the two curves uv 1455 and mmrd1 cross at @xmath218 kpc and @xmath219 , whereas the curves uv 1455 and mmrd2 cross at @xmath220 kpc and @xmath221 , as it appears from figure [ gq_mus_distance_reddening_m0700_x45z02c15o20 ] . since @xmath222 is the weighted mean of the reddening determinations done in 3.1 , we take this value together with @xmath223= 5 kpc as the best estimates compatible with the observations . the errors on the reddening and the distance are @xmath66 0.05 dex and @xmath66 0.5 kpc , respectively . here we estimate the total luminosity of the uv flash on day 151 and discuss its outburst nature . from figure [ plotcont ] , we have obtained the excess of energy above the smooth decline for each of the three bands , i.e. , 1455 , 2885 , and @xmath38 , which are plotted in figure [ sed_uv_outburst ] . these fluxes are consistent with the energy distribution from a blackbody with a temperature of @xmath224 k. @xcite also estimated the temperature from a zanstra - like method , based on the @xmath225h@xmath41 ratio , to be 85,000 to 100,000 k. the photospheric temperature of our @xmath135 white dwarf model are about 30,000 k ( on day 108 ) before the uv flash and about 60,000 k ( on day 202 ) after the uv flash . assuming blackbody emission , we can estimate the size of the emitting region from @xmath226 for @xmath227 k. taking @xmath228 kpc , the radius of the emission region is @xmath229 . the total flux is estimated to be @xmath230 erg s@xmath67 ( @xmath231 ) from @xmath232 . on day 151 , the photospheric radius of our @xmath135 white dwarf model is as small as @xmath233 and the photospheric temperature is as low as 42,000 k. therefore , the estimated radius and blackbody temperature , @xmath229 and @xmath234 k , suggest an episodic expansion and strong heating of the photosphere . this was probably due to a dynamical mass ejection episode and to strong shock heating . the uv flash is anyhow a short - lived event that hardly contributes to the total ejecta mass , as suggested by the fast recovery of the emission line spectrum by day 202 and by the negligible effect on the optical light curves . the first observation of _ iue _ on day 37 also shows a high uv flux compared with our white dwarf model . a close look at the optical brightness in figure [ gq_mus_t3_uv1455_m0700_x45z02c15o20 ] indicates a small peak around day 37 . this event may also be due to another mass ejection episode . note that these mass ejection episodes add to the major underlying continuous mass ejection process from the optically thick wind , which endures until day @xmath235 , as shown in figure [ gq_mus_softxray_r_t ] . we have applied the `` universal decline law '' of classical novae described in [ model_nova_outburst ] to gq muscae 1983 and derived various parameters of the nova . our main results may be summarized as follows : \1 . we show that the `` universal decline law '' reproduces well the observed light curves of gq mus in the optical and in the near infrared @xmath107 , @xmath108 , @xmath109 , and @xmath110 bands . \2 . our blackbody light curve model for the uv 1455 band can reproduce the observed uv 1455 fluxes except two uv flashes on day 37 and 151 . the uv flash on day 151 described in [ uv_evolution_cont ] was accompanied by a mass ejection episode seen as a fast wind with a terminal velocity of about 3200 km s@xmath67 . an analysis of the _ iue _ reprocessed data of gq mus indicates @xmath202 , a value that is larger than previously reported . we find that the mass of the wd component of gq mus is @xmath236 for the adopted envelope chemical composition of @xmath1 , @xmath237 . this value has been derived by comparing predicted fluxes from our models with the observations in the supersoft x - ray , in the uv , and in the optical and near infrared , obtained at different times . we have estimated a mass of @xmath238 lost by the wind . we have estimated a distance of @xmath239 kpc . we thank albert jones ( rasnz ) for providing us with their machine readable visual estimates of gq mus 1983 and also aavso for the visual data of gq mus 1983 . we are also grateful to the anonymous referee for useful comments that helped to improve the paper . this research has been supported in part by the grant - in - aid for scientific research ( 16540211 , 16540219 , 20540227 ) of the japan society for the promotion of science and by inaf prin 2007 . in this and previous papers we have modeled the time evolution of novae in the continuum and compared the results with observed light curves . such a study is hampered by the presence of emission lines , which give rise to sensible distortions in most photometric bands ( see 1 ) , especially at some nova phases . to evaluate the contamination by emission lines , we take advantage of the fact that the intrinsic colors of free - free emission are constant with time because of @xmath240 for free - free emission . here , we consider four classical novae that are well observed in various photometric bands : v1500 cyg , v1668 cyg , v1974 cyg , and gq mus . intrinsic @xmath241 , @xmath242 , @xmath243 , @xmath244 , and @xmath245 colors from free - free emission , as calculated from @xmath246 , are reported in the second column of table [ color_of_free - free ] . if reddening is known , observed colors are obtained from @xmath247 where @xmath248 is the intrinsic color and @xmath249 is the reddening coefficient listed in the second and third columns of table [ color_of_free - free ] , respectively . in the following subsection , we will compare the observed colors with those calculated from equation ( [ color_reddening_relation ] ) for classical novae v1500 cyg , v1668 cyg , v1974 cyg , and gq mus . figures [ v1500_cyg_yvijhk_color][gq_mus_vijhk_color ] demonstrate that the shape of light curves is almost independent of the wavelength , which is a characteristic feature of free - free emission ( see equation [ [ free - free - wind ] ] ) . in these figures , we shift each observed light curve down by @xmath250 and overlap it on the observed @xmath16 ( or @xmath77 ) light curve . then , we obtain the color of @xmath251 ( or @xmath252 ) from the definition of @xmath253 ( or @xmath254 ) . the colors thus obtained are listed in each figure and plotted in figure [ magnitude_calibration3 ] . for v1500 cyg , we use the @xmath16 photometry in @xcite and the @xmath77 data in @xcite , @xmath107 data in @xcite . the @xmath108 , @xmath109 , and @xmath110 observations were taken from @xcite , @xcite , and @xcite . these light curves are plotted in figure [ v1500_cyg_yvijhk_color ] with each light curve being overlapped on the @xmath16 magnitude . the spectrum became that for free - free emission about 45 days after the outburst while it was that for blackbody during the first 3 days @xcite . so , we shift each observed light curve down by @xmath250 and overlap it on the observed @xmath16 light curve between day 6 and 30 . then , we obtain the color of @xmath251 as listed in figure [ v1500_cyg_yvijhk_color ] . the intrinsic color indices , @xmath248 , for @xmath241 , @xmath242 , @xmath243 , @xmath244 , and @xmath245 of free - free emission ( table [ color_of_free - free ] ) are plotted in figure [ magnitude_calibration3]a ( open circles ) together with @xmath256 ( filled circles ) calculated from equation ( [ color_reddening_relation ] ) . the @xmath242 , @xmath243 , @xmath244 , and @xmath245 color indices obtained from figure [ v1500_cyg_yvijhk_color ] are plotted in figure [ magnitude_calibration3]a ( open triangles ) . if strong emission lines are present in the spectrum , observed colors deviate from those calculated from equation ( [ color_reddening_relation ] ) . in fact , the @xmath77 band is not emission - line free , being contaminated by strong emission lines even between day 6 and 30 . the amount of contamination can be estimated from @xmath257 because the @xmath16 band is almost emission - line free . this value tells us that strong emission lines contribute about 30% of the energy flux to the @xmath77 band . the @xmath258 , @xmath259 , @xmath260 , and @xmath261 color indices are also plotted in figure [ magnitude_calibration3]a ( filled triangles ) . a somewhat larger contamination is present in the @xmath107 and @xmath108 bands while the @xmath109 and @xmath110 bands are not so heavily contaminated by emission lines . for v1500 cyg , the @xmath260 and @xmath261 colors are in a reasonable agreement with equation ( [ color_reddening_relation ] ) with @xmath256 . figure [ v1668_cyg_yvijhk_color ] shows @xmath16 , @xmath77 , @xmath107 , @xmath108 , @xmath109 , and @xmath110 light curves of v1668 cyg . here the @xmath16 magnitude observations are taken from @xcite , the @xmath77 data are from @xcite and @xcite , the @xmath107 magnitudes from @xcite , and the @xmath108 , @xmath109 , and @xmath110 magnitudes from @xcite . each light curve is shifted down to overlap with the @xmath16 light curve between day 6 and 30 because the spectrum was no longer that for free - free emission after day 35 due to dust formation . the colors are summarized in this figure and are also plotted in figure [ magnitude_calibration3]b . the @xmath107 band is strongly , the @xmath108 band is slightly , while the @xmath109 and @xmath110 bands are not heavily contaminated by emission lines . for v1668 cyg , the @xmath260 and @xmath261 color are in reasonable agreement with equation ( [ color_reddening_relation ] ) together with @xmath262 @xcite . note also that the @xmath77 band is contaminated by emission lines as suggested from @xmath263 . for v1974 cyg , no @xmath16 and @xmath107 photometry is available . figure [ magnitude_calibration3]c shows the @xmath243 , @xmath244 , and @xmath245 colors obtained from figure [ v1974_cyg_vjhk_color ] , where @xmath77 magnitudes are taken from @xcite and the @xmath108 , @xmath109 , and @xmath110 magnitudes are from @xcite . open triangles denote the observed colors , which follow quite well the calculated ones with @xmath264 @xcite . however , the @xmath77 band is contaminated by emission lines even at early phases as already shown in v1500 cyg and v1668 cyg . if we subtract this contribution of @xmath265 ( estimated from the spectrum in fig . 2 of * ? ? ? * ) , i.e. , @xmath257 , we obtain the corrected colors ( filled triangles ) as shown in figure [ magnitude_calibration3]c . the @xmath108 band is heavily contaminated by , paschen @xmath41 , and @xmath266 , while the @xmath109 and @xmath110 bands are not so heavily contaminated @xcite . this explains why the @xmath260 and @xmath261 colors , corrected by @xmath267 , are in good agreement with the expected values from free - free emission . in figure [ magnitude_calibration3]d we plot ( as open triangles ) the observed @xmath242 , @xmath243 , @xmath244 , and @xmath245 colors from figures [ gq_mus_vijhk_color ] , and compare them with the intrinsic colors from free - free emission calculated from equation ( [ color_reddening_relation ] ) for @xmath222 ( filled circles ) . the observed colors are in reasonably good agreement with the calculated ones . however , the @xmath77 band is not emission - line free , being contaminated by strong emission lines even in the early phase ( e.g. , @xmath268% from fig . 2 of * ? ? ? this contributes @xmath269 to the colors , so we subtract this from the original value of @xmath77 , and obtain the corrected colors of @xmath258 , @xmath259 , @xmath260 , and @xmath261 ( filled triangles ) as shown in figure [ magnitude_calibration3]d . the @xmath107 band is strongly contaminated by line and the @xmath108 band is also very strongly contributed by paschen @xmath41 line ( e.g. , fig . 3 of * ? * ) . consequently , both the @xmath258 and @xmath259 colors ( corrected by @xmath270 ) are 0.8 and 1.3 mag below the calculated ones , respectively . on the other hand , the @xmath109 and @xmath110 bands are not heavily contaminated by strong emission lines ( e.g. , fig . 3 of * ? ? ? this explains why both the @xmath260 and @xmath261 colors corrected by @xmath270 are in good agreement with equation ( [ color_reddening_relation ] ) of @xmath222 . andreae , j. , & drechsel , h. 1990 , _ physics of classical novae _ , a. cassatella and r. viotti ( berin : springer - verlag ) , 204 balman , . , & krautter , j. 2001 , , 326 , 1441 balman , . , krautter , j. , & gelman , h. 1998 , , 499 , 395 barger , a. j. , gallagher , j. s. , bjorkman , k. s. , johansen , k. a. , & nordsieck , k. h. 1993 , , 419 , l85 bateson , f. m. 1998 , inf . variable stars , 2316 , 1 cassatella , a. , altamore , a. , & gonzlez - riestra , r. 2002 , , 384 , 1023 cassatella , a. , lamers , h. j. g. l. m. , rossi , c. , altamore , a. , gonzlez - riestra , r. 2004 , , 420 , 571 cassatella , a. , altamore , a. , gonzlez - riestra , r. 2005 , , 439 , 205 chochol , d. , hric , l. , urban , z. , komzik , r. , grygar , j. , & papousek , j. 1993 , , 277 , 103 de freitas pacheco , j. a. , & codina , s. j. 1985 , , 214 , 481 deroux , j. k. 1978 , inf . bul . stars , 1519 , 1 . diaz , m. p. , & steiner , j. e. 1989 , , 339 , l41 diaz , m. p. , & steiner , j. e. 1994 , , 425,252 diaz , m. p. , williams , r . e. , phillips , m. m. , & hamuy , m. 1995 , , 277 , 959 diaz , m. p. , williams , r . e. , phillips , m. m. , & steiner , j. e. 1992 , cataclysmic variable stars , ed . n. vogt , ( asp conf 29 ) , 362 . dinerstein , h. l. 1986 , , 92 , 1381 duerbeck , h. w. , rindermann , r. , & seitter , w. c. 1980 , , 81 , 157 ennis , d. , beckwith , s. , gatley , i. , matthews , k. , becklin , e. e. , elias , j. , neugebauer , g. , & willner , s. p. 1977 , , 214 , 478 ferland , g. j. , lambert , d. l. , & woodman , j. h. 1986 , , 60 , 375 gallagher , j. s. , & ney , e. p. 1976 , , 204 , l35 gallagher , j. s. , kaler , j. b. , olson , e. c. , hartkopf , w. i. , & hunter , d. a. 1980 , , 92 , 46 gehrz , r. d. , hackwell , j. a. , grasdalen , g. i. , ney , e. p. , neugebauer , g. , & sellgren , k. 1980 , , 239 , 570 grevesse , n. , & anders , e. 1989 , cosmic abundances of matter , ed . c. j. waddington ( new york : aip ) , 1 groenewegen , m. a. t. , & lamers , h. j. g. l. m. 1989 , , 79 , 359 hachisu , i. , & kato , m. 2001a , , 553 , l161 hachisu , i. , & kato , m. 2001b , , 558 , 323 hachisu , i. , & kato , m. 2004 , , 612 , l57 hachisu , i. , & kato , m. 2006 , , 167 , 59 ( paper i ) hachisu , i. , & kato , m. 2007 , , 662 , 552 ( paper ii ) hachisu , i. , kato , m. , kato , t. , & matsumoto , k. 2000 , , 528 , l97 hachisu , i. , kato , m. , & nomoto , k. 1996 , , 470 , l97 hachisu , i. , kato , m. , & nomoto , k. 1999a , , 522 , 487 hachisu , i. , kato , m. , nomoto , k. , & umeda , h. 1999b , , 519 , 314 hachisu , i. , kato , m. , & schaefer , b. e. 2003 , , 584 , 1008 hassall , b. j. m. , et al . 1990 , _ physics of classical novae _ , eds . a. cassatella and r. viotti ( berin : springer - verlag ) , 202 hauschildt , p. h. , shore , s. n. , schwarz , g. j. , baron , e. , starrfield , s. , & allard , f. 1997 , , 490 , 803 hummer , d. g. , & storey , p. j. 1987 , , 224 , 801 iglesias , c. a. , & rogers , f. j. 1996 , , 464 , 943 kaler , j. b. 1986 , , 98 , 243 kato , m. 1983 , , 35 , 507 kato , m. 1997 , , 113 , 121 kato , m. 1999 , , 51 , 525 kato , m. , & hachisu , i. , 1994 , , 437 , 802 kato , m. , & hachisu , i. , 2005 , , 633 , l117 kato , m. , & hachisu , i. , 2007 , , 657 , 1004 kawara , k. , maihara , t. , noguchi , k. , oda , n. , sato , s. , oishi , m. , & iijima , t. 1976 , , 28 , no . 1 , 1976 , p. 163 krautter , j. , beuermann , k. , leitherer , c. , oliva , e. , moorwood , a. f. m. , deul , e. , wargau , w. , klare , g. , kohoutek , l. , van paradijs , j. , & wolf , b. 1984 , , 37 , 307 krautter , j. , & williams , r. 1989 , , 341 , 968 lamers , h. j. g. l. m. ; cerruti - sola , m. ; perinotto , m. 1987 , , 314 , 726 liller , w. , & overbeek , m. d. 1983 , , livio , m. 1992 , , 393 , 516 lockwood , g. w. , & millis , r. l. 1976 , , 88 , 235 mallama , a. d. , & skillman , d. r. 1979 , , 91 , 99 mclaughlin , d. b. 1942 , , 95 , 428 morisset , c. , & pquignot , d. 1996a , , 312 , 135 morisset , c. , & pquignot , d. 1996b , , 313 , 611 gelman , h. , krautter , j. , & beuermann , k. 1987 , , 177 , 110 gelman , h. , orio , m. , krautter , j. , & starrfield , s. 1993 , , 361 , 331 orio , m. , covington , j. , gelman , h. 2001 , , 373 , 542 payne - gaposchkin , c. 1957 , the galactic novae ( amsterdam : north - holland ) pquignot , d. , petitjean , p. , boisson , c. , & krautter , j. 1993 , 271 , 219 rosenbush , a. e. 1999a , astrophysics , 42 , 43 rosenbush , a. e. 1999b , astrophysics , 42 , 140 schmidt , th . 1957 , z. astrophys . , 41 , 181 seaton , m. j. 1979 , , 187 , 73 shanley , l. , gelman , h. , gallagher , j. s. , orio , m. , & krautter , j. 1995 , , 438 , l95 short , c. i. , hauschildt , p. h. , & baron , e. 1999 , , 525 , 375 stickland , d. j. , penn , c. j. , seaton , m. a. , snijders , m. a. , j. , storey , p. j. 1981 , , 197 , 107 whitelock , p. a. , carter , b. s. , feast , m. w. , glass , i. s. , laney , d. , menzies , j. w. , walsh , j. , & williams , p. m. 1984 , , 211 , 421 woodward , c. e. , gehrz , r. d. , jones , t. j. , lawrence , g. f. , & skrutskie , m. f. 1997 , , 477 , 817

helium is the second most abundant element in normal stars and has a pronounced effect on their structure and evolution . unfortunately , helium abundance can not be determined spectroscopically for low - mass stars because of their low temperatures . as a result , the initial helium abundance of stellar models depends on an assumption about galactic chemical evolution . for instance , the yale - yonsei isochrones @xcite were constructed assuming @xmath3 , @xmath4 and @xmath5 being respectively the initial helium and heavy - element abundances . the dartmouth tracks @xcite assume @xmath6 , while the padova tracks of @xcite assume @xmath7 . such ad hoc prescriptions could easily lead to large errors in results for single stars that are determined using such sets of isochrones and tracks . when additional data , such as frequencies of stellar oscillations and other spectroscopically determined stellar parameters are available , the helium abundance is often treated as one of the free parameters that are adjusted to obtain the best fit to the data ( see e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? while such methods give better constraints on the initial helium abundance , there are possible systematic errors that are introduced because of our inability to model the near surface layers of a star properly . normal one - dimensional models treat convection only in gross approximations and do not treat the dynamical effects of convection at all . this leads to large differences in the structure of the near surface layers that introduces a frequency - dependent error in the frequencies ( see , e.g. , * ? ? ? * ) that is usually referred to as the `` surface effect '' . removal of the surface effect can introduce errors in the determined stellar properties . there are however , more direct ways of determining the helium abundance of a star using its oscillation frequencies . the process of ionization of helium locally depresses the adiabatic index , @xmath8 , which in turn affects the sound speed in that region , and consequently , the frequencies of acoustic modes . in the case of the sun , the signature that helium ionization leaves on the sound - speed profile has been used successfully to determine the helium abundance in the solar convection zone ( e.g. , * ? ? ? * and references therein ) . for the sun , the helioseismic estimation of helium abundance was facilitated by the fact that the frequencies of about a few thousand modes of degrees ranging from @xmath9 ( the radial mode ) to @xmath10 have been determined precisely . this is not the case for other stars , where only modes of @xmath92 and sometimes @xmath11 can be determined . thus the estimation of helium abundance has to be done with a different technique . to estimate the helium abundance in other stars , one can exploit the fact that the steep variation in the sound speed in the second helium ionization zone introduces an oscillatory component , @xmath12 , in the oscillation frequencies as a function of the radial order , @xmath13 , of the modes @xcite . the signal varies as @xmath14 , where @xmath15 is the acoustic depth of the ionization layer as measured from the stellar surface , i.e. , @xmath16 @xmath17 is the sound speed , @xmath18 the radial distance to the ionization layer , @xmath19 is the stellar radius , @xmath20 is the frequency of a mode with radial order @xmath13 and degree @xmath21 , and @xmath22 is the phase of the oscillatory term . the amplitude of the signal is predominantly a function of the helium abundance higher the abundance , higher the amplitude ; but there is also an effect of mass , effective temperature , and luminosity , which is why it is important to choose close enough calibrating models . @xcite , @xcite , and @xcite explored theoretically how this signature may be used to determine the helium abundance of other stars . while thus far no one has used this signature to determine the helium abundance of any star other than the sun , it has been used to determine the acoustic depth of the helium ionization zone . using data from _ corot _ , @xcite determined the location of the second helium ionization zone for the red giant hr 7349 , they lacked sufficient signal to noise to determine the helium abundance . again using _ corot _ data , @xcite and @xcite determined the position of the heii ionization zone of the solar - type star hd 49933 . @xcite have extended this study to 19 stars observed by _ note that a similar oscillatory signature , but at a larger acoustic depth @xmath23 arises from the discontinuity in the second derivative of the sound speed at the base of the convection zone and can be used to determine the depth of the convection zone @xcite . in this work we determine the helium abundance in @xmath0 ( hd 186408 and 186427 ; kic 12069424 and 12069449 ) using oscillation frequencies determined from 2.5-years of observations by the _ kepler _ satellite . these stars form a binary system and are among the brightest stars in the _ kepler _ field of view . analysis of a three month time series to constrain the properties of the stars have shown that the stars are slightly more massive than the sun and somewhat older too . using the frequencies of oscillations , along with spectroscopic constraints on the effective temperature and metallicity of the stars , @xcite found the stars to be @xmath24 gyr old with an initial composition to be @xmath25 and @xmath26 . they estimated the masses of the a and b components to be @xmath27 @xmath28 and @xmath29 @xmath28 , respectively . using similar technique @xcite found a somewhat larger helium abundances for these stars . however , these values of helium abundance depend on input physics of stellar models and on how the surface effects are corrected for . a more direct determination of the current helium abundance in the envelope , @xmath30 , is thus in order . the rest of the paper is organized as follows . we describe the methods we have used to determine the helium abundance from the signature of the ionization zone in section [ sec : tech ] , the models used for the calibration of the helium signature are described in section [ sec : mod ] , the results are presented in section [ sec : res ] and we discuss the results in section [ sec : con ] . the amplitude of the oscillatory term , introduced by the localized depression of @xmath8 in the helium ionization zone , is quite small , and different techniques have been used to extract this signal . in this work we use three techniques , the first ( method a ) relies on taking the second differences of the frequencies of a given degree with respect to the radial order @xmath13 to suppress the smooth variation in frequencies . the other two methods ( methods b and c ) rely on directly fitting the oscillatory signature while simultaneously modelling the predominantly smooth behavior of the frequencies as a function of @xmath13 . the difference between the methods b and c is in the form of oscillatory function used in fitting and how the smooth component is adjusted . we describe these methods in more detail below . to extract the oscillatory signal from the frequencies we take the second difference @xcite of the frequencies with respect to the radial order @xmath13 , i.e. , @xmath31 the advantage of taking the second difference is that the contribution from the smooth variation of the frequencies with @xmath13 is greatly reduced since the dominant variation of @xmath32 with @xmath13 is linear . we fit the second differences to a suitable functional form that represents the oscillatory signal from the heii ionization zone as well as that from the base of the convection zone @xcite . for this purpose we use the following form which has been adapted from @xcite @xmath33 where @xmath34 are free parameters which are fit to the observed second differences . in eq . ( [ eq : fit_eq_a ] ) the first two terms represent the smooth part of the function , which remains after the second differences are calculated and includes a contribution from the hydrogen and first helium ionization zone ; the third term represents the contribution from the base of the convection zone , while the last term represents the contribution from the heii ionization zone . it is possible to add more terms to the smooth part using a higher degree polynomial , but for both stars considered in this work , the reduction in @xmath35 due to additional terms was found to be statistically insignificant ; the linear term was found to have a significant effect . hence , we have restricted the smooth part to linear form in this work . the parameters of eq . ( [ eq : fit_eq_a ] ) are determined by a nonlinear least - squares fit . since the second differences of a given value of @xmath21 and neighboring @xmath13 have correlated errors , we use the full covariance matrix to calculate the value of @xmath35 . we assume that the errors in the frequencies themselves are not correlated with each other to obtain the covariance matrix . since the nonlinear minimization does not always converge to the global minimum of the parameter space , we repeat the minimization 100 times with different initial guesses which are randomly selected in some range of possible values . the minimum of the @xmath35 values obtained for the 100 trials is accepted as the best fit value . to estimate the uncertainties in the fitted parameters we repeat the whole process for 1000 realizations of the data obtained by adding random perturbations to the frequencies with standard deviation equal to @xmath36 uncertainty in the corresponding frequencies . the median value of each parameter for 1000 realizations was accepted as the fitted value , while the @xmath37 uncertainty in the fitted value was estimated from the range covering 34% area on either side of the median in the distribution function of the fitted parameter . for the two stars being analyzed we find that while the signal from heii ionization zone is robust and all realizations of the data result in values of @xmath15 in a reasonable range , the distribution of the fitted values of @xmath23 is extremely wide with multiple peaks . we find that the oscillatory signal from the base of the convection zone is weak for both @xmath0 . for some of the realizations we find that fitting @xmath23 is difficult . an analysis of the peaks reveals that one of the secondary peaks corresponds to @xmath38 where @xmath39 is the acoustic radius of the star . this can happen due to aliasing @xcite . we discard all realizations for which the fitted value of @xmath23 does not lie in the dominant peak of the distribution of @xmath23 . the median of the dominant peak is used as the estimate for @xmath23 . for determining other fitted parameters too we use only those realizations that have @xmath23 in the dominant peak . note that this paper is about determining the helium abundance of these stars , and not the position of the convection - zone base , though that is by itself an interesting question . to this effect we have tested whether our inability to fit @xmath23 reliably affects the helium results by fixing the value of @xmath23 at different values around the main peak . the fit to the heii signal is unaffected , and therefore we are certain that our results concerning helium are not biased . the amplitude of the helium signal in the second differences can be converted to that in the frequencies by dividing the second - difference amplitude by a factor @xmath40 , where @xmath15 and @xmath41 are the acoustic depth and the average large frequency separation respectively . this factor is derived under the assumption that the amplitude of oscillatory signal varies slowly over the ` wavelength ' of oscillation . the amplitude of the oscillatory signal depends on the helium abundance in the ionization zone which in stars that have an outer convection zone is the helium abundance of the convective envelope . as shown by @xcite , the amplitude of the signal can be calibrated using models with known helium abundance . @xcite used the average amplitude over the frequency interval that is used in fitting the signal to measure the helium abundance ; we do the same . to extract directly the signal from the oscillation frequencies , we adapt a method first proposed by @xcite . the approach is , first , to fit a smooth function of radial order @xmath13 to all the modes @xmath42 of same degree @xmath21 , which removes the slowly varying trend . for each degree @xmath21 , we fit a polynomial @xmath43 using least - squares method with third derivative smoothing . the resulting smooth function removes mainly variation of @xmath42 at much longer scales than the characteristic scales of the oscillatory signals , filtering out the signatures of the acoustic glitches . the filtered signal is then fitted simultaneously for all degrees to the following expression : @xmath44 where @xmath45 , @xmath23 , @xmath46 , @xmath47 , @xmath48 , @xmath15 , @xmath49 are 7 free parameters , while @xmath50 is a reference frequency . we have used a modified version of the method used in previous works @xcite . it differs mainly in three aspects : ( 1 ) a global minimization method , based on the pikaia implementation of a genetic algorithm @xcite , is used for the least - squares fit providing independence of initial conditions ; ( 2 ) the robustness of the fit against outliers is improved by performing an iteratively reweighted least - squares regression ( e.g. , * ? ? ? * ) ; and ( 3 ) both signals from the base of the convection zone and the heii ionization zone are fitted simultaneously . the third derivative smoothing procedure depends on a parameter @xmath51 which determines how closely the polynomial @xmath43 interpolates the points ( * ? ? ? * their eq . c5 ) . here , we no longer iterate on this parameter but set it to a value adequate for isolating the longest period signal from the heii ionization zone . the same value of @xmath51 is used for both stars , either observations or model frequencies . with this method , all available data points are used ( frequencies of consecutive orders are not required to construct combinations ) and we do not _ a priori _ assume a specific functional form for the smooth component . to estimate the uncertainties in the fitted parameters we perform 5000 monte carlo realizations of the frequencies by adding random perturbations to them with standard deviation equal to @xmath36 uncertainty in the corresponding frequencies . the median value of the distribution of each parameter was accepted as the fitted value and the @xmath52 uncertainty in that parameter was estimated from the standard deviation of the distribution . to calibrate the helium abundance , we used the amplitude of the signal at a reference frequency @xmath50 ( @xmath53 mhz ) for both components . this choice provides us a simple measure of the amplitude , @xmath54 ( see eq . [ eq : fit_eq_b ] ) , and hence a measure of the amount of helium present in the ionization zone . this technique is a modification of the previous one in the sense that it fits smooth and oscillatory signal together . for each @xmath21 , we take a fourth degree polynomial in @xmath13 , @xmath55 , for the smooth component . although , a third degree polynomial takes care of smooth component for the observed data , it is not enough for the models , as the surface term introduces a smooth component which varies approximately as fourth power of the frequency @xcite . we fit directly the frequencies to the function @xmath56 where @xmath57 . the @xmath58 elements of @xmath59_{l , i}$ ] along with @xmath45 , @xmath23 , @xmath46 , @xmath47 , @xmath60 , @xmath15 , @xmath49 are 27 free parameters which need to be determined . the parameters of eq . ( [ eq : fit_eq_c ] ) are determined by minimizing a @xmath35 defined as : @xmath61 ^ 2 + \lambda_\mathrm{c}^2 \sum_{n , l}\left[\frac{d^2p_{l,\mathrm{c}}(n)}{dn^2}\right]^2\,,\ ] ] where @xmath62 is the uncertainty in the mode frequency and @xmath63 is a smoothing parameter . a very small value of @xmath63 corresponds to the standard weighted least - squares fitting , while a very large value tends to fit the frequencies to a straight line . to make an appropriate choice of @xmath63 , we note that as @xmath63 increases , the fit to the observation stabilizes ( uncertainty in the fitting parameters decreases ) and saturates at some point . we accept a value of @xmath63 when the stabilization occurs . the value of @xmath63 is found to be the same for both stars . to avoid systematic errors in the calibration , we use the same @xmath63 for the models as well . while the number of parameters to be fitted in this method is large , it should be noted that the use of regularization introduces additional constraints that enable more parameters to be determined . similar to what has been done in method a , we repeat the minimization 100 times with random initial guesses covering a reasonable parameter space to get the global minimum . we simulate 1000 realizations of the data in the same way as was done in method a to get the fitted value along with @xmath64 uncertainty . we calculate an average amplitude over the frequency range used in the fitting for the calibration . to determine the helium abundance of the stars we need to compare the observed amplitudes of the helium signal to that calculated for stellar models with different helium abundances . we used three sets of models : two constructed with the mesa code @xcite for two different metallicity mixture , and one with the yale stellar evolution code ( yrec ; * ? ? ? * ) . the models in each set were constructed following different principles and the models for calibration were selected using different criteria . the models and the selection process are described below . the mesa code allows users to explore different input physics and data . our models were constructed using opal equation of state @xcite . we used op opacities @xcite for models with metallicity mixture of @xcite ( hereinafter gs98 ) and opal opacities @xcite for models with metallicity mixture of @xcite ( hereinafter agss09 ) . low temperature opacities were used from @xcite assuming gs98 mixture . reaction rates from nacre @xcite were used for all reactions except @xmath65n(@xmath66,@xmath67)@xmath68o and @xmath69c(@xmath70,@xmath67)@xmath71o , for which updated rates of reaction from @xcite and @xcite were used . convection was modelled using the standard mixing length theory @xcite , and diffusion of helium and heavy elements was incorporated using the prescription of @xcite . we constructed models assuming two different values of the solar metallicity to convert the observed @xmath72 $ ] to @xmath73 needed for the models . recall that there is currently an uncertainty about what the solar metallicity really is . the solar abundances of gs98 , with @xmath74 , had been used extensively for many years and solar models constructed with these abundances satisfied helioseismic constraints quite well ( see e.g. , * ? ? ? * ) . however , @xcite redetermined the solar abundances using 3d model atmospheres and incorporating nlte corrections for many lines and claimed that the solar abundance of heavy elements was much lower , @xmath75 . the relative abundances of different elements were also changed . @xcite revised the abundances further to @xmath76 . the lower abundances are not consistent with helioseismic constraints ( see e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and this discrepancy has not been resolved . to estimate the effect of this uncertainty on stellar models we used both gs98 and agss09 mixtures of heavy elements in this work . opacities were calculated assuming the different heavy - element mixtures , and @xmath73 for the models were derived for a given @xmath72 $ ] assuming that @xmath77}}=0 $ ] corresponds to @xmath74 for the gs98 mixture and @xmath76 for the agss09 mixture . we constructed models on a uniform grid of stellar parameters ( mass @xmath78 , initial helium abundance @xmath4 , initial metallicity @xmath77}}_0 $ ] , and mixing length parameter @xmath70 ) for both stars with two different metallicity mixtures , gs98 and agss09 . the range of values in the grid for these parameters were : @xmath79@xmath80 , @xmath81@xmath82 , @xmath77}}_0 : 0.12$]@xmath83 , @xmath84@xmath85 . the step sizes were @xmath86@xmath28 , @xmath86 , @xmath86dex and @xmath87 , respectively . a reduced @xmath35 between the frequencies of the model ( @xmath88 , corrected for the surface term as per the formulation of @xcite ) and the observed frequencies ( @xmath89 ) is defined as : @xmath90 where @xmath91 is the @xmath36 uncertainty in the observed frequency and the sum runs over @xmath92 radial modes of the concerned star . a model with a particular @xmath78 , @xmath4 , @xmath5 , and @xmath70 is selected as an acceptable model of the star only if the evolutionary track enters in a 4-dimensional box formed by the 2@xmath93 uncertainties in the observed effective temperature , luminosity , surface @xmath72 $ ] , and large separation as given in table [ tab : obs ] . further degeneracy in the age is lifted by minimizing the @xmath94 ( eq . [ eq : chisq_nu ] ) . in other words , the `` best '' model on each evolutionary track is chosen through a matching of the frequencies , provided the parameters mentioned above agree within very conservative limits . the surface metallicity given by @xcite , @xmath95 dex for @xmath1 and @xmath96 dex for @xmath2 , is not quite the same for both components . however , since the metallicities are quite similar , we assume the metallicity of the two stars to be equal , @xmath97 dex , with increased uncertainty . this may be justified as the stars form a binary system and hence their initial metallicities should be the same . since we use multiple values of initial conditions @xmath98 , we get two ensembles each , corresponding to the two abundance choices of gs98 and agss09 , for @xmath1 and @xmath2 . these are not , however , the final sets of calibration models . the dispersion in the age of the models in each of these ensembles is nearly 5 gyr . we use a further selection procedure described below which takes into account the fact that @xmath0 should have the same age . models constructed with the agss09 mixture are treated separately from those constructed with the gs98 mixture so that the effect of the uncertainty in the solar metallicity on our results can be quantified . we constrain the ages using the small frequency separation ( @xmath99 ) , which is known to be a good indicator of the evolutionary stage of main sequence stars ( see e.g. , * ? ? ? we use the weighted average of the small separation over 12 pairs , @xmath100 , for both components . however , we do not select the models with @xmath100 alone , but consider other observables namely , the effective temperature ( @xmath101 ) , luminosity ( @xmath102 ) , surface gravity ( as @xmath103 ) , surface metallicity ( as @xmath72 $ ] ) , and the large separation ( @xmath41 ) as well . although , the radii of the two components have been measured with interferometry @xcite , we do not use the radii to constrain the ages , but it should be noted that the interferometric radii are found to be consistent with the final model radii for both stars . we estimate the average large frequency separation ( @xmath104 ) of the radial modes by using a linear fit to the frequencies @xmath105 as a function of @xmath13 . the same 12 radial modes are used that were used for calculating the average small separation . the same set of modes are used to calculate the average large separation for the observation and the models . as mentioned earlier , the inadequacy of modelling the near - surface layers of the star introduces a frequency - dependent shift in the calculated eigenfrequencies , @xmath42 of the models , making the frequencies systematically larger than observed frequencies for all but the lowest - frequency modes . this error propagates into the large separation too . we find that for our acceptable models @xmath41 is about 1@xmath106hz larger than the observed value . we estimated the exact shift using the differential form of the duvall law , as formulated by @xcite , in which the frequency differences between a model and observation are decomposed into two parts , one that depends on the interior structural differences and another that is essentially the surface term . the @xmath41 values of the model are corrected for the surface term before comparing them to the observed values of @xmath0 . we estimated independently the age of @xmath1 and @xmath2 using models of two different metallicity mixture . to estimate the age of a star using models of a particular metallicity mixture , we define a @xmath35 for each model as : @xmath107 the term @xmath108 in the summation represents 6 observable quantities ( @xmath41 , @xmath100 , @xmath101 , @xmath102 , @xmath103 , and @xmath72 $ ] ) and @xmath109 denotes the uncertainty in the estimate of the respective quantity . the quantity @xmath110 is the age of the model under consideration and @xmath111 is a reference age . for a given @xmath111 , we can find a model that represents the star most closely by minimizing the @xmath112 ( eq . [ eq : age1 ] ) over all the models in the ensemble . we denote the minimum by @xmath113 , and minimize it with respect to @xmath111 in a range spanned by the model ages in the ensemble , to get the individual age of the star . for clarity we show the @xmath113 in fig . [ fig : cage](a ) in a limited range of @xmath114 . clearly the curves have a well defined minimum that gives the best model as well as the best estimate of the age of each individual star for a specific assumed metallicity mixture . the ages are respectively @xmath115 gyr and @xmath116 gyr for @xmath0 when gs98 mixture is used . for the agss09 mixture the ages are respectively @xmath117 gyr and @xmath118 gyr . thus the ages of the two stars are found to be close to each other and the values are consistent with those of @xcite and @xcite . note that the calculation of @xmath112 requires @xmath119 ; we start with a suitable guess and then whole process is iterated until we get the same uncertainty in the estimated age as @xmath119 . the individual ages of the two stars , independently determined as explained above , need not turn out to be the same . however , since the stars form a binary system it is more natural to assume that their ages are the same . we can determine this common age by modifying the definition of @xmath35 to include both stars . thus for a given pair of models , with one model drawn from the model ensemble of @xmath1 and the other from that of the @xmath2 , we define a combined @xmath35 as : @xmath120 the first two terms correspond to the model of the a component while the last two correspond to the model of the b component . the summations in the first and third terms again have 6 terms each corresponding to the same observables as described above . the quantities @xmath121 and @xmath122 are the ages of the models of @xmath1 and @xmath2 under consideration . the quantity @xmath123 is a reference age as before , but common to both stars . for a given @xmath123 , we can find a pair of models that represents the stars most closely by minimizing the @xmath124 ( eq . [ eq : age ] ) over all possible pairs of the two ensembles . we denote the minimum by @xmath113 , and minimize it with respect to @xmath123 in a range spanned by the model ages in the two ensemble , to get the common age of the star . for clarity we show the @xmath113 in fig . [ fig : cage](b ) in a limited range of @xmath125 . again , the resulting curves have a well defined minimum that gives the best pair of models as well as the best estimate of the common age of the stars . the individual models of such a pair are essentially the best models of @xmath0 . the age of the binary system is estimated to be @xmath126 gyr for gs98 mixture , and @xmath127 gyr for agss09 mixture . thus we find that the mixture of heavy elements does not affect the age significantly . the difference in the age between models of two mixtures may become significant when the stellar parameters and the oscillation frequencies are known more precisely . @xcite have pointed out that along with @xmath128 we can use coefficients of higher order asymptotic formula for oscillation frequency to determine the age more precisely as these are more sensitive to the sound speed variation in the stellar core . however , it is not clear if these terms can be determined more accurately from the currently observed frequencies . at the end of this selection procedure we are left with approximately @xmath129 models , nearly @xmath130 models for a star with a particular metallicity mixture , with a range of mass , chemical composition , and mixing length parameters , and age within @xmath36 of the value determined above . for the two metallicity mixtures , gs98 and agss09 , we have two such set of calibrating models . in addition to the models described above , we also used the yale stellar evolution code to model the two stars . the input physics includes the opal equation of state ( eos ) tables of @xcite , and opal high temperature opacities @xcite supplemented with low temperature opacities from @xcite . all nuclear reaction rates are from @xcite , except for that of the @xmath131 reaction , for which we use the rate of @xcite . all models included gravitational settling of helium and heavy elements using the formulation of @xcite . models were constructed assuming the heavy element mixture of gs98 . the yrec models were constructed with an approach that was quite different from the mesa models . the starting point of the modelling was the mass and radius of the two stars as determined by @xcite . we did a monte carlo around this mass and radius , and the spectroscopically determined @xmath101 and metallicity to obtain a set of values for mass , final radius , final @xmath101 and final metallicity . yrec was used in an iterative mode . in this mode the final @xmath101 and radius for a star of a given mass and metallicity are specified . the code iterates over the initial helium abundance @xmath4 in case the mixing length parameter @xmath70 is specified , or alternatively over @xmath70 if @xmath4 is specified until a model with the specified @xmath101 and radius is found . note that this is similar to the construction of standard solar models , though in the case of the sun the iteration is performed simultaneously over both mixing length parameter and @xmath4 ( since the solar age is a known constraint ) . we made the models both ways specifying @xmath70 and determining @xmath4 and specifying @xmath4 and determining @xmath70 . the initial metallicity of the models was kept higher than the observed metallicity in order to account for the depletion of metals by diffusion . the monte carlo exercise assumed that the width of the distribution in mass and radius was three times the quoted uncertainty in @xcite . these models , by construction , satisfy the temperature constraint . models of @xmath0 were constructed independently . the final selection of models was done as follows . all models with initial helium abundance lower than the big bang nucleosynthesis value of @xmath132 @xcite were rejected . also rejected were models with final metallicities that were different from the observed metallicity @xcite by more than 3@xmath93 . we calculated the frequencies for the remaining models . the frequencies were used to calculate the frequency ratios @xmath133 , @xmath134 , and @xmath135 @xcite . these ratios are insensitive to the surface term and hence do not require any _ ad hoc _ corrections to account for near - surface uncertainties in the models . all models that matched the observed ratios to within 3@xmath93 were selected . the common age of the @xmath136 system was estimated to be @xmath137 gyr in the same way as described in section [ subsec : mesa ] . this resulted in 231 models for @xmath1 and 95 models for @xmath2 for calibration after applying the age constraint . we applied all three techniques described in section [ sec : tech ] to estimate the helium abundance of both components of the @xmath136 system . we also estimated the acoustic depths of the heii ionization zone and made preliminary estimates of the acoustic depths of the base of the convection zone . in this work , we used observed frequencies computed following the procedure described by @xcite from 2.5 years of _ kepler _ simple - aperture - photometry light curves corrected following @xcite and high - pass filtered with a 4-days triangular smooth . the observed frequencies as used in this work are given in table [ tab : freq ] . all frequencies listed in the table were used in the fits except for the last two modes for @xmath1 which were not used in methods a , c as these modes are beyond the cut - off frequency in many stellar models and hence can not be determined reliably in stellar models which are used for calibration . the method b is not expected to be sensitive to presence of these modes in the observed frequency set as the amplitude of helium signal is compared at a reference frequency instead of the average over entire range covered in the fit , which is the case in methods a , c . further , the two isolated @xmath11 modes can not be used in method a , which needs to compute second order differences . the upper panels of fig . [ fig : fit_fig_a ] , [ fig : fit_fig_b ] , and [ fig : fit_fig_c ] show respectively the fit to eq . ( [ eq : fit_eq_a ] ) ( method a ) , eq . ( [ eq : fit_eq_b ] ) ( method b ) and eq . ( [ eq : fit_eq_c ] ) ( method c ) of @xmath0 ; whereas the lower panels show the histograms of distribution of @xmath15 ( blue ) and @xmath23 ( red ) obtained by fitting 1000 realizations ( 5000 for method b ) of data perturbed by adding random errors . the fitted values of the various physical parameters are listed in table [ tab : fitpar ] . it should be noted that for methods a and c , the amplitude is averaged over the fitting interval , while for method b , it is calculated at a reference frequency . as explained in section [ subsec : techa ] , the amplitude of oscillatory term in second differences is converted to that in frequencies . from the table it can be seen that the amplitudes of oscillatory signals obtained with methods a and c are similar . this demonstrates the validity of conversion factor . the fitted parameters using all three methods are close to each other and generally within @xmath36 , except for @xmath15 using method b. even in this case the difference is only about 6% of the acoustic depth . this demonstrates that the results are not particularly sensitive to different form of oscillatory term or to different techniques for removing the smooth component in the frequencies as a function of @xmath13 . the frequencies for the models were fitted in the same manner , and using the same modes and weights as the observations . amplitude of the heii signal obtained using method a is shown in fig . [ fig : ampa ] for each mesa model . also shown in the figure is the observed amplitude . note from the figure that the amplitude is predominantly a function of the current helium abundance in the models . there is however , some scatter due to other model parameters , primarily due to variations in mass , effective temperature , and luminosity . the helium abundance , @xmath30 , obtained by calibrating the model amplitude with observation are listed in table [ tab : yresults ] for both stars . the helium abundance obtained with mesa and yrec are consistent with each other within errors . there is a difference , barely within errors , between the helium abundance of @xmath1 and @xmath2 . for @xmath1 all 3 methods give results that are within error bars of each other , but for @xmath2 method a tends to give results that are on higher side , while method b tends to give lower values . the difference between the two methods is up to @xmath138 . the difference in results between various methods can be considered to be a measure of systematic errors . the estimate of @xmath30 does nt directly depend on @xmath15 and hence is not affected by uncertainties in it . the precision of determination of helium abundance depends critically on the modes at the low end of the spectrum . addition of a few low order modes or any significant improvement in the precision of these modes , improve the precision of estimate of helium abundance significantly . to understand this improvement we repeated the fit using method c after removing successively the lowest order modes , or after increasing the uncertainties in the lowest three modes of degree 0 , 1 , and 2 . we found that the uncertainty in the amplitude of the helium signal increases rapidly in each case , by a factor of two or more . this may be due to the fact that the amplitude of the signal decreases very rapidly with increasing frequency , and the low order modes play a role in stabilizing the fit as in that region the amplitude of helium signal is larger . therefore a precise determination of the helium abundance of a star would particularly require a precise set of low order modes . in method b , the amplitude of the oscillatory signal at a reference frequency ( instead of the mean amplitude used in other methods ) is used to calibrate the helium abundance . the stability of this approach has been tested by @xcite for the case of the sun . they found that the use of a reference frequency well inside the frequency range of the observations was a better , more stable option , to calibrate @xmath30 . this is so because the approach used by this method to account for the smooth component is fairly robust in the inner region of the frequency range ( does not change regardless of the details for smooth component that is adjusted ) . in this way , this method can avoid any contamination from the boundaries due to the smoothing . the oscillatory signal in the observed frequencies contains information about the location of the acoustic glitches . [ fig : tau ] shows the acoustic depths of the base of the convection zone and the heii ionization zone obtained using all 3 methods . the @xmath139 for the models were estimated in two different ways ; one by calculating the acoustic depth from the known sound - speed profile of the model ( eq . [ eq : tau ] ) , and the other by fitting the second differences of the model frequencies ( eq . [ eq : fit_eq_a ] ) or by fitting the model frequencies ( eq . [ eq : fit_eq_b ] or [ eq : fit_eq_c ] ) . the first estimate using eq . [ eq : tau ] , is affected by the uncertainties in the definition of stellar surface . this issue has been extensively discussed by @xcite and will affect the acoustic depth of all layers by the same amount . however , the uncertainties in @xmath139 do not affect the determination of @xmath30 which depends on the amplitude of the oscillatory signal . as can be seen in the figure , for @xmath23 the two estimates for a model match well , but for @xmath15 the value obtained from eq . ( [ eq : tau ] ) is systematically higher than that obtained by fitting the second differences or the frequencies . this systematic shift is present in the results of all the three fitting methods , and could be due to the fact that for @xmath23 the location of the discontinuity in the sound - speed gradient is well - defined and there is little ambiguity in its position ; for @xmath15 on the other hand there is , strictly speaking , no real discontinuity at all , but merely a sharp variation in @xmath8 that gives rise to the oscillatory signal in the frequencies . while using eq . ( [ eq : tau ] ) we used the minimum in @xmath8 in the heii ionization zone as the location of the ` discontinuous ' layer , and it is possible that the effective location of the ` discontinuity ' is at a layer above this . the value of @xmath23 obtained by fitting the observed frequencies is close to what we get for our models ; @xmath15 obtained from the observed frequencies lies within about @xmath140 of that obtained by fitting the frequencies of the models . the amplitude of signal due to the base of the convection zone in the observed frequencies is consistent with those in the models frequencies . we examined the variation of the average amplitude of the heii signal with the input parameters of the models and their seismic characteristics . [ fig : par ] shows the variation of the average amplitude calculated using the fitting method a for @xmath1 . we can see that the average amplitude of the heii signal is not particularly sensitive to the metallicity and mixing length parameter ( see fig . [ fig : par]a , [ fig : par]b ) ; however , it shows systematic variation as a function of effective temperature , luminosity , large frequency separation , and age ( see fig . [ fig : par]c - f ) . much of the trend is essentially due to variation in @xmath30 with these parameters rather than variation of the amplitude for a given @xmath30 . the vertical dispersion in each of these figures is due to the dispersion in mass , which also changes @xmath30 at a given set of parameters . the variation of the amplitude was very similar when calculated using method b and c. we have used the oscillatory signature in the oscillation frequencies of @xmath0 caused by the depression of @xmath8 in the heii ionization zone to determine the current helium abundance and the depth of the heii ionization zone . for this task we have used frequencies obtained with 2.5 years of _ kepler _ observations . we have used three different techniques combined with three sets of stellar models for calibrating the signal . we have demonstrated that the helium abundance of these stars can be reliably determined using the observed frequencies . for @xmath1 the three methods give results that are consistent with the error bars , while for @xmath2 the results can differ , though the difference is only @xmath138 in the worst case . the helium abundance for @xmath1 is found to lie in the range 0.231 to 0.251 and that of @xmath2 lies in the range 0.218 to 0.266 . the error bars quoted in table [ tab : yresults ] are the random errors arising from those in the observed frequencies . in addition , there would be systematic uncertainties due to approximate form of oscillatory term as well as uncertainties in the used stellar models . the first contribution can be estimated from the differences in values obtained using the three techniques and is already included in the values quoted above . the stellar - model uncertainty should be mainly from the eos which translates the helium abundance to @xmath8 . from extensive tests for the sun , it is known that the opal eos is close to that of the sun ( see e.g. * ? ? ? * ) and hence we do not expect much uncertainty on that count in these stars where the uncertainties in the frequencies are much larger than those for the sun . this technique for determining helium abundance is not particularly sensitive to the presence of surface effect the surface corrections are smooth function of @xmath13 and will mainly contribute to the smooth part of the frequency and can be separated from the oscillatory part . this is particularly true when the observed frequencies are available over a wide range of @xmath13 values as is the case with both stars considered here . if the observed frequency range is limited , there could be some difficulty in separating the helium signal from smooth part since the oscillatory signal can also be approximated with moderately high degree polynomial . this is the reason why we restrict the degree of polynomial in method a to the minimum value that is statistically significant . the regularization used in methods b and c allows us to use a more complex function to account for the smooth component . @xcite have used similar technique to estimate the helium abundance in the solar envelope using only low degree modes and find results which are consistent with the known value obtained from more detailed seismic inversions . the estimates of the helium mass fraction are those of the current abundance in the outer envelope of the stars . this value is , of course , lower than the initial helium abundance of both stars because of the gravitational settling of helium . we can determine the initial helium abundance from the current one by assuming that the models of the two stars correctly represent the amount of helium depleted due to settling . mesa models show a depletion of @xmath141 ( for gs98 mixture ) and @xmath142 ( for agss09 mixture ) for @xmath1 . depletion of helium is lesser for @xmath2 models , @xmath143 for gs98 and @xmath144 for agss09 mixture . since the two components of the binary system are believed to have formed from the same gas cloud , the initial composition should be the same for both stars . in that case the current helium abundance in the envelope of @xmath2 should be higher than that in @xmath1 by about 0.005 . the values that we find are consistent with this difference . this difference arises because of small difference in masses of the two stars . similarly , the difference between gs98 and agss09 mixtures arise due to the difference in the thickness of convection zones in the two cases . since gs98 models have deeper convection zone , the depletion of helium is lower . thus the initial helium abundance of the @xmath136 system is between 0.28 and 0.31 . this is somewhat larger than the value found by @xcite and close to the values found by @xcite . the solar initial helium abundance was determined to be @xmath145 @xcite . thus our estimates of @xmath4 for the @xmath136 system is consistent with solar values particularly considering the fact that these stars have higher metallicity . the fitted value of @xmath15 obtained from observed frequencies for @xmath2 matched that obtained from fitting the model frequencies , but for @xmath1 the observed value is larger than that in all selected models . this could be due to some systematic errors in modelling or in the other observed quantities that were used to constrain the models . the two sets of mesa models constructed with different heavy - element mixtures were used to check if the observed frequencies could distinguish between the different mixtures . for both sets of models , the value of @xmath30 , as well as @xmath15 and @xmath23 are very similar , although , as expected , there is a difference in the depth of the convection zone between the two sets with models constructed with the agss09 having a shallower convection zone depth and hence lower @xmath23 than models with the gs98 mixture . but there is considerable overlap in the range for each set of models . for @xmath2 the uncertainty in the value of @xmath23 obtained from observed frequencies is too large to distinguish between the two sets of models . for @xmath1 the error in @xmath23 from observed frequencies is reasonably small and if the systematic errors in modelling can be sorted out , it may be possible to use the models to distinguish between the two sets of models and get an independent handle on the controversy over the solar heavy - element abundance . sb acknowledges support from nsf grant ast-1105930 and nasa grant nnx13ae70 g . am acknowledges support from the nius programme of hbcse ( tifr ) . mjpfgm and jpf acknowledge support from the ec project spaceinn ( fp7-space-2012 - 312844 ) and jpf was also supported by the european research council ( ec - fp7 ) through a fellowship of the starting grant erc-2009-stg-239953 . wjc acknowledges support from the uk science and technology facilities council ( stfc ) . funding for the stellar astrophysics centre ( sac ) is provided by the danish national research foundation . , g. 2004 , in american institute of physics conference series , vol . 731 , equation - of - state and phase - transition in models of ordinary astrophysical matter , ed . v. celebonovic , d. gough , & w. dppen , 193207 , a. , michel , e. , antia , h. m. , & deheuvels , s. 2011 , in transiting planets , vibrating stars and their connection , ed . a. baglin , m. deleuil , e. michel , c. moutou , & t. seman , proceedings of second _ corot _ symposium , 197 cccc @xmath101 & 5839@xmath14642 k & 5809@xmath14639 k & 1 + @xmath103 & 4.29@xmath1460.02 dex & 4.36@xmath1460.02 dex & 1 + @xmath102 & 1.56@xmath1460.05 @xmath147 & 1.27@xmath1460.04 @xmath147 & 2 + @xmath72 $ ] & 0.096@xmath1460.040 dex & 0.096@xmath1460.040 dex & 3 + @xmath41 & 102.9@xmath1460.2 @xmath106hz & 116.5@xmath1460.2 @xmath106hz & + @xmath100 & 5.82@xmath1460.03 @xmath106hz & 6.70@xmath1460.03 @xmath106hz & [ tab : obs ] cccccccccc 11 & & 1334.392@xmath1460.034 & 1384.368@xmath1460.040 & & & & & & + 12 & 1391.639@xmath1460.037 & 1437.520@xmath1460.043 & 1488.353@xmath1460.100 & & & & & 1686.327@xmath1460.057 & + 13 & 1494.996@xmath1460.058 & 1541.914@xmath1460.050 & 1591.242@xmath1460.115 & & & 1695.066@xmath1460.056 & 1749.184@xmath1460.050 & 1804.197@xmath1460.235 & + 14 & 1598.695@xmath1460.066 & 1645.058@xmath1460.095 & 1694.193@xmath1460.175 & & & 1812.420@xmath1460.094 & 1866.525@xmath1460.102 & 1921.176@xmath1460.168 & + 15 & 1700.909@xmath1460.083 & 1747.154@xmath1460.082 & 1795.757@xmath1460.105 & 1838.283@xmath1460.355 & & 1928.900@xmath1460.068 & 1982.587@xmath1460.073 & 2036.677@xmath1460.123 & 2085.517@xmath1460.260 + 16 & 1802.316@xmath1460.068 & 1848.980@xmath1460.054 & 1898.284@xmath1460.100 & 1940.756@xmath1460.413 & & 2044.274@xmath1460.058 & 2098.076@xmath1460.056 & 2152.405@xmath1460.100 & + 17 & 1904.611@xmath1460.054 & 1951.997@xmath1460.050 & 2001.647@xmath1460.070 & 2045.922@xmath1460.229 & & 2159.580@xmath1460.060 & 2214.159@xmath1460.058 & 2268.966@xmath1460.082 & 2319.208@xmath1460.225 + 18 & 2007.572@xmath1460.045 & 2055.527@xmath1460.046 & 2105.314@xmath1460.051 & 2149.914@xmath1460.143 & & 2275.946@xmath1460.045 & 2331.138@xmath1460.044 & 2386.267@xmath1460.060 & 2436.792@xmath1460.176 + 19 & 2110.915@xmath1460.040 & 2159.153@xmath1460.045 & 2208.907@xmath1460.067 & 2253.539@xmath1460.172 & & 2392.718@xmath1460.040 & 2448.250@xmath1460.042 & 2503.575@xmath1460.061 & 2554.149@xmath1460.150 + 20 & 2214.226@xmath1460.049 & 2262.536@xmath1460.047 & 2312.545@xmath1460.091 & 2357.368@xmath1460.181 & & 2509.668@xmath1460.038 & 2565.398@xmath1460.042 & 2620.566@xmath1460.064 & 2671.754@xmath1460.155 + 21 & 2317.326@xmath1460.056 & 2366.260@xmath1460.059 & 2416.301@xmath1460.137 & 2462.077@xmath1460.373 & & 2626.399@xmath1460.042 & 2682.409@xmath1460.048 & 2737.741@xmath1460.081 & 2789.187@xmath1460.270 + 22 & 2420.931@xmath1460.089 & 2470.322@xmath1460.081 & 2520.695@xmath1460.266 & & & 2743.327@xmath1460.061 & 2799.735@xmath1460.062 & 2855.682@xmath1460.137 & 2906.915@xmath1460.371 + 23 & 2525.141@xmath1460.197 & 2574.639@xmath1460.138 & 2624.147@xmath1460.366 & 2669.376@xmath1460.869 & & 2860.773@xmath1460.110 & 2917.799@xmath1460.100 & 2973.470@xmath1460.248 & 3024.940@xmath1460.941 + 24 & 2629.212@xmath1460.212 & 2679.951@xmath1460.188 & 2727.410@xmath1461.643 & & & 2978.481@xmath1460.162 & 3036.084@xmath1460.156 & 3093.204@xmath1460.517 & 3144.365@xmath1460.679 + 25 & 2733.199@xmath1460.470 & 2784.178@xmath1460.369 & & & & 3097.575@xmath1460.492 & 3154.418@xmath1460.271 & & + 26 & 2837.442@xmath1460.682 & 2891.303@xmath1460.841 & & & & 3214.348@xmath1460.562 & 3272.941@xmath1460.741 & & + 27 & 2944.587@xmath1461.226 & 2996.533@xmath1461.022 & & & & & & & [ tab : freq ] ccccl + a & @xmath148 & @xmath149 & @xmath150 & @xmath151 + b & @xmath152 & @xmath153 & @xmath154 & @xmath155 + c & @xmath156 & @xmath157 & @xmath158 & @xmath159 + + a & @xmath160 & @xmath161 & @xmath162 & @xmath163 + b & @xmath164 & @xmath165 & @xmath166 & @xmath167 + c & @xmath168 & @xmath169 & @xmath170 & @xmath171 [ tab : fitpar ] cccc + a & @xmath172 & @xmath173 & @xmath174 + b & @xmath175 & @xmath176 & @xmath177 + c & @xmath178 & @xmath179 & @xmath180 + + a & @xmath181 & @xmath182 & @xmath183 + b & @xmath184 & @xmath185 & @xmath186 + c & @xmath187 & @xmath188 & @xmath189 [ tab : yresults ]

the path integral method within the lagrange formalism is a fundamental tool to formulate quantum field theory . certainly , the treatment of the same theory within the canonical operator formalism is somewhat more cumbersome . in fact , additionally to the operator - ordering problems of gauge theories @xcite , the formal invariance properties of a generic theory admit additional transformation laws which are not present in the lagrange formulation . the `` momentum fields '' in such a case are not related to the fundamental fields by means of the canonical equations , see , _ e.g. _ , ref.@xcite , but constitute separate degrees of freedom with independent transformation properties and thus symmetries . as a consequence the number of variables in the path integral increases in comparison to the lagrange formalism . this introduces for most theories considerable additional complications in perturbative as well as non - perturbative calculations @xcite . on the other hand , some recent studies indicate that the first order formalism proves to be a successful tool to study the required complete cancellation of the energy divergences @xcite that emerge in a perturbative treatment of coulomb gauge yang - mills theory within the standard lagrange formalism @xcite . coulomb gauge yang - mills theory has attracted attention since one possible solution to the confinement problem in qcd is provided by the `` gribov - zwanziger '' scenario @xcite ( _ cf . _ also ref.@xcite ) . however , the problem of renormalizing coulomb gauge yang - mills theory is still unsolved since these energy divergences can not be regularized using any of the standard procedures . recently it was illustrated how these energy divergences cancel at each order in perturbation theory @xcite . in order to perform explicit calculations a number of methods have been applied as _ e.g. _ introducing a novel method to regularize feynman integrals in non - covariant gauges @xcite , employing algebraic renormalizability @xcite , and trying to recover coulomb gauge yang - mills theory as a limit of an interpolating gauge @xcite . on the other hand , there are several indications that the canonical or first order formalism is better suited for studying coulomb gauge yang - mills theory @xcite . yet , even if such an approach would be successful , due to the dramatically complicated form of the functional equations in the first order formalism , an explicit non - perturbative study is far too involved to be computationally feasible . therefore , an analysis in the second order formalism would be highly desirable . to this end we provide general connections between the greens functions in the two formulations that should help to perform the renormalization in the lagrange framework according to the insight in the renormalization procedure obtained in the hamilton framework . in order to establish general relations between dressed correlation functions in the different formulations we exploit the following properties . at vanishing sources associated to the momentum fields the effective action ( _ i.e. _ the generating functional of one - particle - irreducible ( 1pi ) green s functions ) of the hamilton formalism reduces to the one of the lagrangean approach . this allows to reduce the set of dyson - schwinger equations ( dses ) @xcite in the first order formalism to the corresponding set derived from the standard path integral representation . an important special case is given if the hamiltonian is quadratic in the momentum fields . for a corresponding generating functional the integration over the momentum fields can be performed and any @xmath1point function involving this field as the average of a polylocal function of the quantum canonical momentum fields can be determined . this means that the full correlation functions involving these canonical variables can be found as a functional of those that usually appear in the standard path integral representation . the procedure to find these connections is closely related to the functional method used in the derivation of the dses ( see _ e.g. _ ref . @xcite ) . in this paper we give the explicit form of these relations for a general four - dimensional renormalizable theory . this includes the case that the interaction terms involve the time derivative of the fields . in a final step we resolve the relations between the proper two - point functions in both frames by considering the inverse of the matrix - valued propagators in the hamiltonian approach . moreover , we show that similar connections arise for theories where not the kinetic but the interaction part is linearized . such a bosonization procedure is an important technique used in many parts of physics ranging from hadronic physics to condensed matter systems . our results explicitly verify , in agreement with other approaches , that there is no double counting in bosonized theories , but instead a given correlation function in the underlying theory is exactly given by the sum of all possible contributions involving both the fundamental and the composite degrees of freedom in the bosonized theory . this paper is organized as follows : in sect . ii the functional equations for dses and a symmetry - related identity ( like ward - green - takahashi , resp . slavnov - taylor , identities ( stis ) of a gauge theory @xcite ) in the phase space formulation are given . in sect . iii we present the first order dses of theories which are quadratic in the momentum fields . in addition , we derive a method to determine the correlation functions and the stis including momentum fields from the respective quantities that usually appear in the standard lagrange representation . diagrammatic rules and the general decomposition of the full proper functions in the lagrange formalism are given in sect . iv , while in sect . v and sect . vi we detail the explicit decomposition of the connected and proper two - point functions in the hamilton framework , respectively . in sect . vii we show that the formalism can also be utilized in the case composite of auxiliary fields , and last but not least in sect . viii we apply the formalism to the case of coulomb gauge yang - mills theory . in the last section we conclude while essential steps of many calculations have been deferred to several appendices . we start our analysis by considering a generic quantum field theory formulated within the first order formalism . the hamiltonian density @xmath2 depends on the fundamental fields @xmath3 and the conjugate momentum fields @xmath4 , respectively . as will be discussed in sect . iv.d , for fermionic fields the relation between the two formalisms is trivial . therefore we will treat only the case of bosonic fields in this and the next section . in the case of gauge invariant theories we will suppose that the hamiltonian @xmath5 includes the additional terms that arise when the constraints associated to such theories , like _ e.g. _ gauss law in quantum electrodynamics , are localized . in this context each lagrange multiplier necessary to impose the constraints will be treated as a fundamental field . certainly the presence of these terms leads to the main differences between gauge theories and the more conventional hamiltonian systems . in addition , in a fixed gauge , ghost fields appear . the usual path integral representation of these grassmannian variables is formulated within the first order formalism ( see also sect . iv.c ) . thus , in the case of a gauge theory they will be analysed in an independent way ( for details see sect . [ sec : coulomb ] ) . in this section , to set up the problem , we will disregard for the moment these potential complications . let us suppose , in addition , that the fields @xmath4 and @xmath3 are coupled to a set of classical sources given by @xmath6 and @xmath7 , respectively . under such conditions the source dependent vacuum - to - vacuum transition amplitude between the asymptotic states @xmath8 and @xmath9 looks like @xmath10&=&\langle\mathrm{vac},\mathrm{out}|\mathrm{vac } , \mathrm{in}\rangle_{j}\nonumber\\&= & \int \mathcal{d}[q]\mathcal{d}[p ] \exp\left\{\frac{i}{\hbar}\left(i\left[q , p , j\right]+i\epsilon\textrm{-terms}\right)\right\}\;\;\;\label{gen}\end{aligned}\ ] ] where @xmath11 $ ] denotes the collection of all fundamental fields , whereas @xmath12 $ ] the corresponding momentum fields . for more details see section @xmath13 of the ref . @xcite . here the argument in the exponential has the structure @xmath14=i_0\left[q , p\right]+\int d^4x j(x)\cdot\phi(x)\ ] ] where @xmath15 include both momentum and fundamental fields and sources . the components of the sources in this notation are distinguished by upper labels , whereas @xmath16&= & \int_{-\infty}^\infty d\tau \bigg ( \int d^3\textbf{x}\big ( p_m(x)\dot{q}_m(x ) \nonumber \\ & & \qquad \qquad \qquad -\mathcal{h}\left(q(x),p(x)\right)\big ) \bigg ) \label{acv } \end{aligned}\ ] ] looks like the classical expression for the hamiltonian action . note that the part concerning to the @xmath17 terms in eq . ( [ gen ] ) have the function to produce the necessary @xmath18 in the denominators of all propagators such that the correct boundary conditions of the fields at asymptotic times , @xmath19 , are implemented ( see again ref . @xcite ) . the fact that @xmath20 $ ] looks like the action expressed in terms of canonical variables is somewhat misleading since the momentum fields @xmath4 are independent variables and therefore * not yet * related to the fundamental fields @xmath3 or their derivatives . in particular , since the path integral is not saturated by its saddle point(s ) they are not constrained to obey the equations of motion of classical hamiltonian dynamics @xmath21 where @xmath22 the path integrals , however , contain the information about these equations of motion . indeed , let us consider the operator version of eq . ( [ me1 ] ) . its vacuum expectation values in the presence of the external classical sources can be expressed as @xmath23\mathcal{d}[p]\frac{\hbar}{i}\frac{\delta}{\delta\phi } \exp\left\{\frac{i}{\hbar}i\left[q , p , j\right]\right\ } } { \langle0_\mathrm{out}\vert0_\mathrm{in}\rangle_{j } } -j.\ ] ] assuming the absence of any boundary terms , the integral of such a functional derivative vanishes . then by substituting each of the elementary objects present in eq . ( [ me1 ] ) by the derivative with regard to the respective classical sources , the following functional differential equation arises @xmath24=-j(x)\mathcal{z}[j ] . \label{partida1}\end{aligned}\ ] ] it contains all equations of motions fulfilled by all green s functions . introducing the generating functional of connected green s functions , @xmath25=-i\ln\left(\mathcal{z}[j]\right)$ ] , allows us to rewrite eq . ( [ partida1 ] ) as @xmath26 where in the last step we made use of the following identity @xmath27 for arbitrary functionals @xmath28 and @xmath29 . this set of equations constitutes the dses for the connected green s functions . in the next step we introduce the vacuum expectation values of the momentum and the fundamental fields in the presence of sources @xmath30=\frac{1}{i}\frac{\delta\mathcal{w}^h}{\delta j(x)}\equiv \left(\begin{array}{c } \bar{p}[j]\\ \\\bar{q}[j ] \end{array}\right)=\left(\begin{array}{c } \frac{1}{i } \frac{\delta\mathcal{w}^h}{\delta j^p_{m}}\\ \\ \frac{1}{i } \frac{\delta\mathcal{w}^h}{\delta j^q_{m } } \end{array}\right).\ ] ] we assume that it is possible to invert these relations such that the sources are expressed as functionals of the vacuum expectation values of @xmath31 and @xmath32 . hereby the replacements of variables in eq . ( [ wsde1 ] ) becomes @xmath33(x)+ \frac{\hbar}{i}\int d^4x^\prime\mathbb{d}[j](x , x^\prime ) \frac{\delta}{\delta \bar{\phi}(x^\prime)}\label{r11}\ ] ] where @xmath34(x , x^{\prime})\equiv\left(\begin{array}{cccc } \delta_{ml}^{pp}[j](x , x^{\prime } ) & \delta_{ml}^{pq}[j](x , x^{\prime } ) \\ & \\ \delta_{ml}^{qp}[j](x , x^{\prime } ) & \delta_{ml}^{qq}[j](x , x^{\prime } ) \end{array}\right)\label{cpro}\ ] ] are the source - dependent two - point green s function where @xmath35 } { \delta j^{p}(x)\delta j^{q}(x^\prime)}=\frac{\delta \bar{q}_l[j](x^\prime ) } { \delta j^{p}(x)}\ ] ] and @xmath36,$ ] @xmath37 $ ] and @xmath38 $ ] are analogously defined . the effective action in the first order formalism @xmath39 $ ] can be defined by the legendre transform of @xmath25 $ ] with respect to the associated `` averaged '' fields of the theory @xmath40\stackrel{def}{\equiv}\mathcal{w}^h[j]- \int d^4x j(x)\cdot\bar{\phi}(x).\label{ea}\ ] ] remember that the compact notation implies a twofold legendre transformation in the two independent components @xmath41 and @xmath42 . by considering the functional derivative of @xmath43 $ ] with respect to @xmath44 and @xmath45 we obtain as expected @xmath46 substituting eq . ( [ eq : ecmov ] ) into eq . ( [ wsde1 ] ) , however , we obtain with the replacement eq . ( [ r11 ] ) the desired functional differential equations for @xmath43,$ ] @xmath47(x)+ \frac{\hbar}{i}\int d^4x^{\prime}\mathbb{d}[j](x , x^\prime ) \frac{\delta}{\delta \bar{\phi}(x^\prime ) } } \ , .\label{sdev}\ ] ] this relation generates all dses for the 1pi green s functions of the first order formalism . the method to obtain such a functional relation will be employed several times in the following . the derivation of the proper propagators is straightforward when keeping in mind that they are embedded in a matrix . in fact , taking a derivative with regard @xmath48 in eq . ( [ sdev ] ) we obtain @xmath49}(x^{\prime\prime},x^\prime)\equiv\left(\begin{array}{cccc } \frac{\delta^2\gamma^h}{\delta\bar{p_m}(x^{\prime\prime } ) \delta\bar{p}_n(x^\prime ) } & \frac{\delta^2\gamma^h } { \delta\bar{p_m}(x^{\prime\prime})\delta\bar{q}_n(x^\prime ) } \\ & \\ \frac{\delta^2\gamma^h}{\delta\bar{q_m}(x^{\prime\prime } ) \delta\bar{p}_n(x^\prime ) } & \frac{\delta^2\gamma^h } { \delta\bar{q_m}(x^{\prime\prime})\delta\bar{q}_n(x^\prime ) } \end{array}\right).\label{pgf}\ ] ] this expression and eq . ( [ cpro ] ) are related via the identity @xmath50(x , x^{\prime\prime } ) \mathbb{g[\phi]}(x^{\prime\prime},x^\prime)=-\mathbb{i},\label{unidad}\ ] ] which can be obtained by taking a derivative with respect to the source @xmath51 in eq . ( [ eq : ecmov ] ) . hereby @xmath52 denotes the identity in the considered space . any other connected or proper green s function can be derived by considering higher order derivatives with respect to @xmath51 and @xmath53 in eq . ( [ cpro ] ) and eq . ( [ pgf ] ) , respectively . in general , these functions are constrained by the symmetry properties of the initial `` action '' . in the case of a gauge theory a corresponding derivation leads to the stis . for coulomb gauge yang - mills theory it is presented in sect . [ sec : coulomb ] . here we will illustrate the potential complications arising in the first order formalism . for this it sufficient to assume that @xmath54 is invariant under the simultaneous infinitesimal transformations @xmath55\equiv\epsilon \cdot \left(\begin{array}{c } \mathfrak{g}_m^q\left[p , q\right ] \\ \\ \mathfrak{f}_m^p\left[p , q\right]\end{array}\right)\label{transfi}\ ] ] and also assume that they leave the path integration measure invariant . performing this substitution in eq . ( [ gen ] ) we can obtain the following `` symmetry - related functional identity '' @xmath56\mathcal{d}[p]\int d^4 x j(x)\cdot\delta\phi(x ) \exp\left[\frac{i}{\hbar}i\left[\phi , j\right]\right].\label{slavnov}\ ] ] in the next step we substitute the fundamental and momentum fields by the respective derivatives with respect to the classical source . following a procedure similar to the derivation of the dses we rewrite the symmetry - related identity eq . ( [ slavnov ] ) and express it in terms of the full generating functional @xmath57 $ ] : @xmath58\mathcal{z}[j].\label{comparar}\ ] ] analogously , the above relation can be rewritten for connected green s functions @xmath59\vert_{\phi\to\frac{\delta\mathcal{w}^h}{\delta j}+\frac{\hbar}{i}\frac{\delta}{\delta j } } .\label{comparar1}\ ] ] as before , eq . ( [ slavnov ] ) can be reformulated in terms of the effective action @xmath60\right\vert_{\phi\to\bar{\phi}[j]+ \frac{\hbar}{i}\int d^4y\mathbb{d}[j](x , y)\frac{\delta}{\delta \bar{\phi}(y ) } } .\label{comparar2}\ ] ] this identity verifies that the effective action in first order formalism preserves the continuous symmetries of the initial canonical quantum action . at this point the dses and the symmetry - related identity eq . ( [ slavnov ] ) of a qft seem to be more cumbersome than the usual one which appear in the standard path integral formulation . however , in the next section we shall show that under certain conditions the above system can be reduced to the latter one . from here on we will denote all the variables a field depends on by a single latin index . for instance in case of a gauge field such indices will indicate the space - time point @xmath61 , the vectorial index @xmath62 , and the adjoint gauge group index @xmath63 . repeated indices are summed and integrated over for discrete and continuous variables , respectively . also all fundamental fields will be denoted by the same letter @xmath41 . similarly , @xmath42 will represent all momentum fields . let us now consider the path integral representation of a @xmath0-point function involving only momentum fields @xmath64\mathcal{d}[p]p_{i_1}\ldots p_{i_n}\exp\left[\frac{i}{\hbar}i[q , p , j]\right ] } { i^{1-n}\int\mathcal{d}[q]\mathcal{d}[p ] \exp\left[\frac{i}{\hbar}i[q , p , j]\right]}\;\;\ ; .\nonumber\end{aligned}\ ] ] here @xmath65 is the time ordering operator , and @xmath66 are the hermitian momentum field operators in the heisenberg picture . the expression ( [ delta ] ) can be written via functional derivatives with respect to @xmath67 as @xmath68\mathcal{d}[p]\frac{\hbar}{i}\frac{\delta}{\delta j^p_{i_1}}\ldots\frac{\hbar}{i}\frac{\delta}{\delta j^p_{i_n}}\exp\left[\frac{i}{\hbar}\left\ { i[q , p , j]\right\}\right]}{i^{1-n}\int\mathcal{d}[q]\mathcal{d}[p ] \exp\left[\frac{i}{\hbar}\left\{i[q , p , j]\right\}\right]}.\label{ass}\ ] ] in the following we will consider only the important class of hamiltonians which are at most quadratic in the momentum fields . note that this case is realized for all renormalizable and most non - renormalizable theories . for later use we write the hamiltonian in the form @xmath69p_m+\mathcal{b}_i[q]p_i+ \mathcal{c}[q],\label{hamiltoniand1}\ ] ] which defines a real , symmetric , positive and non - singular matrix @xmath70 as well as the real functionals of @xmath41 , @xmath71 $ ] and @xmath72 $ ] . for the considered class of hamiltonians the gaussian integration over @xmath42 can be performed analytically . as shown in appendix [ app : p - int ] it yields an expression of the form @xmath68\frac{\hbar}{i}\frac{\delta}{\delta j^p_{i_1}}\ldots\frac{\hbar}{i}\frac{\delta}{\delta j^p_{i_n}}\exp{\left[\frac{i}{\hbar}\tilde{\mathcal{s}}[q , j]\right ] } } { i^{1-n}\int \mathcal{d}[q]\exp{\left[\frac{i}{\hbar}\tilde{\mathcal{s}}[q , j]\right]}}. \label{ass111}\ ] ] the general result for the new action arising in the exponential is given in eq . ( [ gac ] ) . for simplicity we will discuss first the standard hamiltonian with @xmath73=\delta_{ij}$ ] . the general case where @xmath70 is non - trivial is realized _ e.g. _ in the case of coulomb gauge qcd which will be discussed in sect . [ sec : coulomb ] . ( for the moment we only note that then the inverse matrix , @xmath74 $ ] , appears in the following expressions as a prefactor of @xmath75 making in general the action @xmath76 non - local . ) in the considered simpler standard case the new action possesses the structure @xmath77=\mathcal{s}_0 + \frac{1}{2}j^p_{i}j^p_{i}+j^p_{i}\frac{\delta\mathcal{s}_0}{\delta\dot{q}_i}+ j^q_{i}q_i \label{actionlf}\ ] ] where @xmath78 is the standard action @xmath79=\frac{1}{2}\dot{q}^i \dot{q}^i -\dot{q}^i \mathcal{b}^i[q]\ + \frac{1}{2}\mathcal{b}^i[q]\mathcal{b}^i[q]- \mathcal{c}[q ] \ , .\label{act}\ ] ] the application of eq . ( [ especial ] ) on the integrand present in eq.([ass111 ] ) makes it possible to write @xmath68\left\{\hat{\mathcal{o}}_{i_1\ldots i_{n-1}}^{p\ldots p } \left(j^p_{i_n}+\frac{\delta\mathcal{s}_0}{\delta \dot{q}_{i_n}}\right)\right\}\exp{\left[\frac{i}{\hbar}\tilde{\mathcal{s } } \right]}}{i^{1-n}\int \mathcal{d}[q]\exp{\left[\frac{i}{\hbar}\tilde{\mathcal{s}}\right]}}. \label{ass11}\ ] ] where the operator @xmath80 only acts on the function inside the curly brackets . note that the integrand in the numerator is a functional depending on the fields @xmath41 as well as the sources @xmath81 and @xmath75 . analogous to eq . ( [ wsde1 ] ) we can write the partially integrated expression eq . ( [ ass11 ] ) in terms of the generating functional of connected green s functions . let us specialize to the case of one - point green s functions , _ i.e. , _ the vacuum expectation values of the momentum fields @xmath82=\left.\frac{\delta\mathcal{s}_0}{\delta \dot{q}_i } \right\vert_{q\to\frac{\delta\mathcal{w}^h}{\delta j^q}+ \frac{\hbar}{i}\frac{\delta}{\delta j^q}}+j^p_{i}= \frac{\delta \mathcal{w}^h}{\delta j^p_{i}(\textbf{x},\tau)}.\label{mge1}\ ] ] eq . ( [ eq : ecmov ] ) allows to express eq . ( [ mge1 ] ) as @xmath83+\frac{\hbar}{i } \delta^{qq}[j]\frac{\delta}{\delta \bar{q}}}.\label{mge11}\ ] ] here the dependence on the average momentum fields is made explicit and the action in the lagrange formalism appears which depends only on the @xmath41-fields . from a functional point of view it is more convenient to express this via eq . ( [ mge1 ] ) as @xmath84_{\phi\to\bar{\phi}[j]+ \frac{\hbar}{i}\delta^{\phi\phi}[j]\frac{\delta}{\delta \phi}}.\label{mge11v}\ ] ] now , employing the identity @xmath56\frac{\delta}{\delta q_m}\int\mathcal{d}[p ] \exp\left[\frac{i}{\hbar}i[q , p , j]\right]\label{ef1sd}\ ] ] we perform the @xmath85integration , so that the above expression reads @xmath86\left(\frac{\delta\mathcal{s}_0}{\delta q_i}+ j^p_{m}\frac{\delta^2\mathcal{s}_0}{\delta q_i\delta \dot{q}_m}+j^q_{i}\right ) \exp\left[\frac{i}{\hbar}\tilde{\mathcal{s}}[q , j]\right],\ ] ] which makes it possible to obtain an analogous form of the field equations @xmath87+\frac{\hbar}{i}\delta^{\phi\phi}[j ] \frac{\delta}{\delta \bar{\phi}}}.\label{sdeqqq}\ ] ] eqs . ( [ mge11v ] ) and ( [ sdeqqq ] ) represent alternative forms of the first order dses ( [ sdev ] ) . this alternative form of the dses allows us to find corresponding equations for the fundamental proper green s functions by taking derivatives with respect to the momentum and/or fundamental fields . this way the proper momentum propagator can be written as @xmath88+\frac{\hbar}{i}\delta^{qq}[j ] \frac{\delta}{\delta \bar{q}}}\right),\label{mge12}\ ] ] whereas the corresponding mixed green s function is @xmath89+\frac{\hbar}{i}\delta^{qq}[j ] \frac{\delta}{\delta \bar{q}}}\right).\label{mge13}\ ] ] in a similar way , the other mixed propagator can be written as @xmath90+\frac{\hbar}{i}\delta^{qq}[j]\frac{\delta } { \delta \bar{q } } } \right)\nonumber\\&-&\left.\frac{\delta^2\gamma^h } { \delta \bar{p}_i\delta\bar{p}_m}\frac{\delta^2\mathcal{s}_0 } { \delta q_j\delta \dot{q}_m}\right\vert_{\phi\to \bar{\phi}[j ] + \frac{\hbar}{i}\delta^{\phi\phi}[j]\frac{\delta}{\delta \bar{\phi } } } \label{sppdeqqqpr},\end{aligned}\ ] ] whereas the fundamental field propagator is given by @xmath91 + \frac{\hbar}{i}\delta^{qq}[j]\frac{\delta}{\delta \bar{q } } } \right)\nonumber\\&- & \left.\frac{\delta^2\gamma^h}{\delta \bar{q}_i\delta\bar{p}_m } \frac{\delta^2\mathcal{s}_0}{\delta q_j\delta \dot{q}_m } \right\vert_{\phi_j\to \bar{\phi}[j]+\frac{\hbar}{i}\delta^{\phi\phi}[j ] \frac{\delta}{\delta \bar{\phi } } } .\label{sppdeqqq}\end{aligned}\ ] ] hereby , in the derivation of eqs . ( [ sppdeqqqpr ] ) and ( [ sppdeqqq ] ) we have discarded those terms that vanish when the sources are set to zero . in addition , note that the proliferated occurrence of green s functions involving the @xmath85field arises from functional derivatives on @xmath92 $ ] . as will become clear below , despite the different appearance , the structure of eqs . ( [ sppdeqqqpr ] ) and ( [ sppdeqqq ] ) is related to the one of eqs . ( [ mge12 ] ) and ( [ mge13 ] ) . in this subsection we derive expressions for first order correlation functions entirely in terms of second order correlation functions . to this end , we start from eq . ( [ ass11 ] ) . subsequently , we again follow a procedure similar to that used in the derivation of eq . ( [ wsde1 ] ) thus arriving at a general expression for arbitrary momentum correlation functions , @xmath93_{q\to \frac{\delta\mathcal{w}}{\delta j^{q}}+\frac{\hbar}{i}\frac{\delta}{\delta j^{q}}}\label{mge}\ ] ] in contrast to the corresponding equation for the one - point function , eq.([mge1 ] ) , ( which did lead to the alternative set of dses ) here the connected generating functional in the standard lagrange formalism appears . note that the latter does not depend on @xmath75 anymore . from the above functional equation , any other @xmath94-point green s function with @xmath0-external legs associated to the @xmath95 s is obtained by taking @xmath96 derivatives with regard to @xmath97 and setting them to zero . in particular , the quantum average of the momentum fields becomes a functional of the averaged field @xmath98 , @xmath99=\left.\frac{\delta\mathcal{s}_0}{\delta \dot{q}_i } \right\vert_{q_m\to\bar{q}_i[j^q]+\frac{\hbar}{i}\delta_{ij}^{qq}[j^q ] \frac{\delta}{\delta q_j}}\label{momenta}.\ ] ] already at this point we want to point out that the averaged momentum field is generally not given by the usual definition of a canonical momentum field @xmath100 on the level of the effective action @xmath101=\left < \frac{\delta s}{\delta \dot q } \right > \neq \frac{\delta \gamma } { \delta \dot{\bar q } } \equiv \mathfrak{p}_i^{\mathrm{can } } \ , .\ ] ] this will be shown explicitly in subsection [ sub : proper - hamilton ] . in addition , the relation between the @xmath102correlation functions and those that appear in the standard formalism can be expressed as @xmath103+\frac{\hbar}{i}\delta^{qq}[j^q ] \frac{\delta}{\delta q}},\label{fomp}\ ] ] whereas the corresponding mixed @xmath104correlation function can be written as @xmath105=\frac{\delta}{\delta j^q_{i}}\left.\left(\frac{\delta\mathcal{s}_0}{\delta\dot{q}_{j } } \right)\right\vert_{q_m\to\frac{\delta\mathcal{w}}{\delta j^{q}}+\frac{\hbar}{i}\frac{\delta}{\delta j^{q}}}.\label{mix}\ ] ] by using the chain rule @xmath106 , eq . ( [ mix ] ) can also be given in the form @xmath107=\delta_{il}^{qq}\frac{\delta}{\delta \bar{q}_l } \left(\left.\frac{\delta\mathcal{s}_0}{\delta\dot{q}_j } \right\vert_{q\to\bar{q}[j^q]+\frac{\hbar}{i}\delta^{qq}[j^q ] \frac{\delta}{\delta q}}\right ) = \delta_{il}^{qq } \frac{\delta { \bar p}_j}{\delta { \bar q}_l}. \label{fopp}\ ] ] the other mixed correlator @xmath108 can be immediately inferred from the bosonic nature of the fields , @xmath109 . similarly , it is possible to determine the connected three point correlation functions @xmath110&=&\delta_{il}^{qq}\frac{\delta}{\delta \bar{q}_l}\delta_{jk}^{pp},\\ \delta_{ijk}^{qqp}[j^{q}]&=&\delta_{im}^{qq}\delta_{jl}^{qq}\left \{\frac{\delta^3\gamma}{\delta\bar{q}_l\delta\bar{q}_m\delta\bar{q}_n } \delta_{nk}^{qp}+\frac{\delta^2\bar{p}_k}{\delta\bar{q}_m\delta\bar{q}_l } \right\},\nonumber \\ \label{fopp1}\\ \delta_{ijk}^{ppp}[j^{q}]&=&i\delta_{ij}\bar{p}_k+i\delta_{ik}\bar{p}_j + i\delta_{jk}\bar{p}_i+\nonumber\\&+&i^2\left.\left(\frac{\delta\mathcal{s}_0 } { \delta \dot{q}_{i}}\frac{\delta\mathcal{s}_0}{\delta\dot{q}_{j } } \frac{\delta\mathcal{s}_0}{\delta\dot{q}_{k}}\right ) \right\vert_{q\to\bar{q}[j^q]+\frac{\hbar}{i}\delta^{qq}[j^q ] \frac{\delta}{\delta q}}\end{aligned}\ ] ] where in the last expression we have used the previous results . employing eqs . ( [ fopp ] ) and eq . ( [ fopp1 ] ) the first and second derivatives of @xmath111 with respect to @xmath98 can be expressed as functionals of the remaining elements . in this way the vacuum expectation value of @xmath42 can be expanded . in fact , up to second order in the classical field it reads @xmath112\bar{q}_i\bar{q}_j \nonumber\\ & + & \ldots \ , .\label{mepn}\end{aligned}\ ] ] since @xmath113 can be computed using the procedure detailed above , the expression ( [ mepn ] ) determines @xmath111 as a function of the second - order dressed correlation functions . the determination of the proper functions within the canonical formalism as a functional of those appearing in the standard lagrange framework is rather cumbersome . this task involves connected tensors of rank larger than two and depends on the possibility to invert the propagator @xmath114 . once the individual elements of this propagator are computed , the proper two - point function is completely determined in terms of the elements of the lagrange formalism . in case the external sources associated to the @xmath85fields vanish , the inverse of @xmath115 is the proper green s function that arises in the standard path integral representation . by direct inversion of the @xmath116 block matrix one obtains @xmath117 where @xmath118 and @xmath119 according to eq . ( [ generalg ] ) there are several equivalent representations of the latter expression , _ e.g. _ @xmath120 however , in what follows we will consider the most simple form given by @xmath121 where we have introduced a unity in the form @xmath122 in eq . ( [ shrepr ] ) , and furthermore the expressions for the off - diagonal elements of eq.([generalg ] ) have been used . the method to obtain @xmath123 can be generalized to any proper @xmath1point function . for instance , let us suppose that we want to compute the proper three - point function . we denote it as @xmath124 and the correspondingly connected version as @xmath125 by considering the action of the symbolic functional derivative @xmath126 on eq . ( [ unidad ] ) we get the equation @xmath127 by inversion of @xmath114 in the second term we obtain the desired form for the proper 3-point green s function @xmath128 proceeding in an analogous way it is possible to express the proper four - point green s function @xmath129 , and so on . thereby , the number of variables within the first order formalism can be reduced . from that point of view , the initially cumbersome problem becomes simpler . let us now return to the symmetry identity . to this end we write eq . ( [ slavnov ] ) in the following form @xmath130\left\{j^q_{m}\delta q_m\left[\frac{\hbar}{i } \frac{\delta}{\delta j^p},q\right]+j^p_{m}\delta p_m\left[\frac{\hbar}{i } \frac{\delta}{\delta j^p},q\right]\right\}\nonumber\\ & \times & \int \mathcal{d}[p]\exp\left[\frac{i}{\hbar}i[\phi , j]\right ] . \label{importanteqq}\end{aligned}\ ] ] in order to simplify the following analysis let us consider theories where the transformations are linear in the momentum fields . under this condition and restricting to the class of hamiltonians analyzed so far we obtain @xmath130\left\{j^q_{m}\delta q_m\left[j^p+ \frac{\delta \mathcal{s}_0}{\delta \dot{q}},q\right ] \right . \nonumber\\ & & + \left.j^p_{m}\delta p_m\left[j^p+\frac{\delta \mathcal{s}_0}{\delta \dot{q}},q\right]\right\}\exp\left[\frac{i}{\hbar}\tilde{\mathcal{s}}[q , j ] \right].\label{importanteqqq}\end{aligned}\ ] ] in particular , if @xmath131 we find that the action @xmath78 is invariant under a symmetry transformation @xmath132 which is a functional of @xmath41 only , _ i.e. _ a symmetry transformation @xmath133 \to \delta q_m \left [ q \right]$ ] . the structure of eq . ( [ importanteqqq ] ) allows to write the symmetry identities in a similar form to eq . ( [ comparar1 ] ) and eq . ( [ comparar2 ] ) , @xmath134\right.\\ & & + \,\frac{\delta\gamma^h}{\delta \bar{q}_m } \left.\!\delta q_m\!\left[\frac{\delta \mathcal{s}_0}{\delta \dot{q}}\!-\!\frac{\delta \gamma^h}{\delta \bar{p}},q\right]\right\}_{\phi\to \bar{\phi}[j]+\frac{\hbar}{i}\delta^{\phi\phi}[j]\frac{\delta}{\delta \bar{\phi}}}. \nonumber\end{aligned}\ ] ] this concludes the formal discussion of the functional symmetry identities . a complete derivation , especially for constrained systems as coulomb gauge yang - mills theories , is presented below in sect . [ sec : coulomb ] . in this section we will show how a general proper correlation function in the lagrange formalism can be decomposed into correlation functions in the hamilton approach . so far we have presented the formalism for a general field theory . in the present and forthcoming sections we will restrict ourselves to the most important class of renormalizable field theories . in four dimensions the most general renormalizable `` canonical action '' for a pure bosonic theory can be expressed as a functional taylor expansion , @xmath135&=&i_{0ji}^{qp}p_iq_j+\frac{1}{2}i_{0ij}^{pp}p_ip_j+ \frac{1}{2}i_{0ijk}^{pqq}p_iq_jq_k \nonumber\\ & + & \frac{1}{2}i_{0ij}^{qq}q_iq_j + \frac{1}{3!}i_{0ijk}^{qqq}q_iq_jq_k+ \frac{1}{4!}i_{0ijkl}^{qqqq}q_iq_jq_kq_l \ , .\nonumber\\ \label{can - action}.\end{aligned}\ ] ] here @xmath136 is such that @xmath137 clearly , the coefficients @xmath138 are field independent . they are given by the functional derivatives of @xmath54 evaluated at @xmath139 namely @xmath140 we remark that @xmath141 as well @xmath142 are dimensionless tensor couplings whereas @xmath143 and @xmath144 have mass dimension @xmath145 and @xmath146 , respectively , and do not involve time derivatives of the fields . nevertheless , depending on the assumed theory the latter two might depend on @xmath147 and @xmath148 . we are considering bosonic theories , so that all coefficients are symmetric . in particular , the first term in eq.([acv ] ) can be written as @xmath149 where @xmath150 which leads to the functional relation @xmath151 we can identify the coefficients present in the hamiltonian density eq.([hamiltoniand1 ] ) as @xmath152 and @xmath153 substituting eqs . ( [ ceff1 ] ) and ( [ ceff2 ] ) in eq . ( [ act ] ) and collecting the terms of the same order in @xmath41 we get that the action @xmath78 can be written as a polynomial functional , @xmath154 in this context the following relations between the bare coefficients arise @xmath155 @xmath156 @xmath157 substituting the above equations in eq . ( [ a ] ) and considering the relation given by eq . ( [ dothg ] ) allows to find additional relations . in fact , taking the second , third and fourth functional derivatives of @xmath78 and evaluating at @xmath158 , respectively , we obtain @xmath159 @xmath160 @xmath161 @xmath162 in the last two relations we have introduced @xmath163 denoting a partial functional differentiation acting just on those terms in the action that do not involve the time derivative of the field . to complete our analysis we point out that from eq . ( [ cvbr ] ) the general form of the quantum canonical momentum fields in four - dimensional renormalizable theories is given by @xmath164 as these polynomial representations allow to identify a priori the bare elements , the expansions given above prove to be very convenient to derive the dses for a general bosonic theory in both formulations . the relations between the correlation functions in both formulations will be given in the following via explicit diagrammatic expressions . to enable this , we will first introduce a graphical representation in terms of the fundamental objects that characterize the theory in both formulations . * as before we consider fields that involve all the irreducible representations in the theory . the fundamental fields are represented by solid lines whereas the corresponding momentum fields are denoted by zigzag lines . + * dressed propagators are denoted by thick , whereas bare propagators and external lines by thin lines , respectively . off - diagonal propagator components are represented by a thick , half solid and half zigzag line . + + to keep the representation of the dses in the first order formalism concise we also include the matrix propagator eq . ( [ cpro ] ) represented by a double line . + * all proper correlation functions ( including proper 2-point functions ) in the first and second order formalism are denoted by small and large filled blobs , + + whereas bare vertex functions are represented by open blobs . + * in our analysis it will become useful to introduce the inverse proper momentum 2-point function @xmath165 which will be represented by a thick dotted line . since it has the form of an alternative momentum propagator it can connect as an internal momentum line to proper vertices . similarly all connected correlators which are one - particle - irreducible ( 1pi ) in the fields but merely connected via the `` propagator '' @xmath166 are called _ p - connected _ and are denoted by a blob labeled by a p. + to express proper functions in the lagrange formalism in terms of those of the hamilton formalism , we exploit the underlying equivalence between the first and second order formalism . due to the equivalence of the generating functionals of connected green s functions at vanishing sources @xmath95 , the effective actions in the two formalisms are identical when @xmath167 is a functional of @xmath168 which in turn is implicitly given by the stationarity in @xmath167 , @xmath169\equiv\gamma^h\left[\bar{p}(\bar{q}),\bar{q}\right ] \label{equivalenciaquantica}\ \ \mathrm{whenever}\ \ \frac{\delta \gamma^h}{\delta\bar{p}}=-j^p=0.\ ] ] here @xmath170 still depends explicitly on @xmath171 and @xmath111 , where the quantum average of the momentum fields is in general a complicated functional of the fundamental field . any proper @xmath0-point function in the standard formalism can be determined by taking @xmath0 derivatives with respect to the fields in eq . ( [ equivalenciaquantica ] ) and evaluated at the vacuum expectation value . in particular applying the chain rule and eq.([equivalenciaquantica ] ) the first derivative reads @xmath172 in the above defined graphical representation this equation reads . as usual in the standard formalism , a differentiation with respect to a field is equivalent to attaching an external leg in the graphical representation ( _ cf . , e.g. , _ ref . next we consider the second derivative of the action given by @xmath173 the field derivative of the momentum field can alternatively to the previous result ( [ mepn ] ) also be obtained from a field derivative of the constraint equation in eq . ( [ equivalenciaquantica ] ) , @xmath174 inserting this in eq . ( [ fundamenta2 ] ) it is expressed entirely in terms of proper first order green s functions and takes the symmetric form @xmath175 in accordance with eq . ( [ shreprqq ] ) . the above equation yields the graphical representation of the decomposition of the proper two - point green s function . interestingly , in terms of the first order correlation functions there is in addition to the proper part also a connected contribution due to the fact that the fundamental field mixes with the corresponding momentum field . this is evident from the off - diagonal elements eq . ( [ mix ] ) in the propagator matrix in the first order formalism . due to the mixing the arising propagator in eq . ( [ complementary1 ] ) is not the elementary @xmath42-propagator but the propagator for a collective mode described by the inverse of the proper two - point momentum correlation function @xmath166 which is represented by the dotted line and related to the actual momentum propagator via eq . ( [ ppifg ] ) . the result for the fundamental field derivative of a proper correlation function in the hamilton formalism yields the replacement rule . in the next step a field derivative can also act on the propagator @xmath166 . its derivative is obtained from the derivative of the inverse of an operator as @xmath176 which yields the graphical replacement rule . applying these two replacement rules in all possible ways on the right hand side of the above equation for the two - point vertex provides immediately the corresponding decomposition of the proper 3-point vertex . this yields directly a symmetric result , whereas the computation without the replacement eq . ( [ complementary1 ] ) in each step would produce an asymmetric result involving higher derivatives of @xmath111 . to derive the decomposition of higher order lagrange correlation functions it is useful to note that both of the above replacement rules involve the external legs in the form of the composite expression denoted by a dashed line . by introducing the composite external leg into the graphical representation , the decomposition of the 3-point function is given by a single graph with three of these new external legs . via the previous rules it is easy to obtain a corresponding replacement rule for the composite external leg . thereby the extension of a general @xmath0-point function by an additional leg can be obtained from the simplified set of rules in fig . [ fig : rules ] . , scaledwidth=20.0% ] starting from the 3-point function these rules allow to derive the decomposition of arbitrary proper @xmath0-point functions in the lagrange formalism . in the appendix we show that the graphs generated by these replacement rules have a very simple structure that can be summarized by the following general statement : _ a proper @xmath0-point function in the lagrange formalism can be decomposed into the sum of all @xmath42-connected @xmath0-point functions with composite external legs in the hamilton framework . _ this is shown in graphical form in fig . [ fig : lag - dec ] . the graphical representation manifestly shows that the propagation , decay and dispersion of particles are more complicated when analyzed in the hamilton framework . at the tree level the arising propagators @xmath166 are constant and correspondingly the dotted internal lines between the proper vertices represent merely contact terms . similarly the external leg corrections involve no additional poles at tree level and the explicit energy dependence arising from the mixed correlator in eq . ( [ complementary1 ] ) just cancels the one from the momentum field derivative of the initial vertex . in the fully dressed action , however , there could arise additional poles due to the actual propagation of the momentum fields via higher order kinetic terms , as well as an explicit energy dependence of the terms that cancels only after summing over all contributions . via graphical rules , one can therefore generate in a systematical way the connection between any proper green s function in the standard second - order formalism to the appropriate correlation functions in the first - order formalism . we point out already here that the inverse relations for green s functions in the latter context are more complicated and can involve loops related to connected correlation functions with both mixed and pure momentum fields . the method used in the last section allows us to represent them in terms of those appearing in the lagrange formalism . as already mentioned we will consider for completeness also quantum field theories involving grassmannian fields @xmath177 and @xmath178 fulfilling the anti - commutation rules @xmath179 the grassmannian action for a renormalizable theory in four dimensions has the general form @xmath180 the quantum canonical momentum fields associated to the @xmath177 are given by @xmath181 , where the suffix _ r _ denotes differentiation from the right . it is clear that the usual path integral representation for grassmannian field is already of first - order form since the momentum fields are treated as independent variables . _ purely fermionic correlation functions are therefore trivially identical in the two formulations . _ let us now study mixed correlation functions involving bosonic fields represented by the @xmath182 in eq . ( [ fp2 ] ) when the bosonic path integral is written in the canonical form as well . generally the tensors @xmath183 @xmath184 and @xmath185 are real and field independent . the kinetic tensor reads in particular @xmath186 where @xmath187 since it is a bosonic variable , @xmath111 is a functional of fermionic bilinears @xmath188 and the field @xmath98 . as a consequence the second term in eq . ( [ fundamenta2 ] ) vanishes identically , which confirms the equality of the propagators in both formulations @xmath189 analogous to the bosonic case the three - point vertex can be decomposed as @xmath190 although the propagators are the same in both formulations the vertices can be different . the above chain rule again results in the appearance of composite legs for all bosons in the graphical representation . the decomposition of a general @xmath0-point correlation function is then again given by all p - connected graphs where only the bosonic propagators @xmath166 are involved and only bosonic external legs are composite . so far we gave in subsection [ gen - ham - con ] general expressions for correlation functions in the hamilton formalism . the goal in this section is to evaluate the general decomposition of the two - point functions in the hamilton framework in case of a generic four - dimensional renormalizable quantum field theory in terms of lagrange correlation functions . to do this , we start by computing the elements of @xmath191 by considering eq . ( [ fopp ] ) and eq . ( [ gqcm ] ) the mixed connected two - point function in the first order formalism can be written as ( from now on we skip the explicit factor @xmath192 ) @xmath193 using partial differentiation for the identity @xmath194 we obtain @xmath195 and therefore in configuration space @xmath196 where eq . ( [ lamarmotadormida ] ) and eq . ( [ lamarmota ] ) have been used . according to the diagrammatic representation the above function has the decomposition given in fig . [ fig : mixed - dec ] . , scaledwidth=47.0% ] via the bosonic symmetry of the propagator it is easy to see that @xmath197 clearly , eq . ( [ qppp ] ) and eq . ( [ pppq ] ) fulfill the condition @xmath198 . our convention for the fourier transform of a general two - point function ( connected or proper ) obeying translational invariance is @xmath199 with @xmath200 and @xmath201 . as a consequence of this convention and the equivalence @xmath198 we obtain the relation @xmath202 where @xmath203 and @xmath204 represent the remaining internal indices . this yields the corresponding equation in momentum space @xmath205 in the derivation of the latter equation we have taken into account the fourier transformation of the proper 3-point function @xmath206 where the @xmath207function expresses the momentum conservation . the pure @xmath102correlator can be computed using eqs . ( [ fomp ] ) and ( [ gqcm ] ) . neglecting those terms that will eventually vanish when the sources are set to zero , it reads @xmath208.\end{aligned}\ ] ] by iterated application of eq . ( [ const ] ) and considering eqs.([lamarmotadormida ] ) and ( [ lamarmota ] ) this expression yields @xmath209 here we have already suppressed disconnected expressions involving tadpoles that cancel since eq . ( [ mge11 ] ) gives @xmath210 the graphical representation of the decomposition of the momentum propagator in the hamilton framework is shown in fig . [ fig : mom - dec ] . note that the loop graphs in the first line represent precisely the vacuum graphs of the @xmath0-particle irreducible ( @xmath0pi ) actions in the lagrange formalism of order one and two with attached external @xmath42-legs . in addition to these proper contributions this result involves again connected graphs in the second line . these graphs correspond to the second and third line in eq . ( [ ppexpf ] ) which can alternatively be expressed by the known mixed correlation functions derived in the last subsection as @xmath211 and correspondingly represented as a single graph . after fourier transformation of eq . ( [ ppexpf ] ) we find the representation of @xmath212 in momentum space @xmath213 the summation over all field components indicated by the subindices in eq . ( [ sadad ] ) ( see also eq . ( [ ppexpf ] ) ) leads to multiple possibilities . yet many of these might not be allowed due to the symmetries of the theory . the decomposition of higher order functions involving one or two external momentum fields is simply obtained by further derivatives of these generating equations . via the chain rule a derivative w.r.t . to the source can be transformed into a derivative with respect to the averaged fields yielding in addition external propagators . correspondingly the graphical decomposition of such correlation functions can be obtained recursively via the graphical replacement rules in the lagrange framework given in @xcite . on the other hand , from eq . ( [ mge ] ) it is clear that higher order correlation functions involving @xmath0 momentum fields involve graphs of loop order @xmath0 . therefore it is not possible to give a close result for all @xmath0-point functions . to conclude this section we remark that the multiplication of @xmath214 on the left hand side amputates the green s function @xmath215 according to eq . ( [ mepn ] ) , this operation allows to derive @xmath216 considering this expression and the diagrammatic rules in the standard path integral representation we get @xmath217 which can be represented graphically as we note in passing , that similar to the composite collective propagator and composite leg in the last section this `` vertex '' allows to write the graphs in the first line by a single loop graph . yet , here this is then surely no explicit decomposition in terms of second order correlation functions anymore . while in the context of the canonical formulation it is entirely possible to deduce the complete set of dses directly from eqs . ( [ mge12]-[sppdeqqqpr ] ) we will follow a slightly less obvious path here . we proceed to give the diagrammatic and analytic expressions for the proper propagators , typically the ones of the first order formalism . subsequently , we will show that such representations can be encoded into the usual ones , _ i.e. _ eqs . ( [ mge12]-[sppdeqqqpr ] ) , which then completes the proof of equivalence between both types of derivations . the result given by eq . ( [ generalg ] ) allows to analyze the structure of the inverse propagator . by considering the fourier transformation of this equation we find that @xmath218 the presence of the second term in eq . ( [ ftpp ] ) leads to the cancelation of the corresponding term in eq . ( [ ppexpf ] ) . the complete analytic result when inserting the corresponding expressions is rather lengthy and therefore deferred to appendix [ app : exp - form ] . it involves the inverse of sums of different terms . graphically these expressions are represented in fig.[fig.proper ] . for a clearer presentation the subsequent correlation functions are represented in terms of the previous ones . whereas the actual momentum propagator can only be given as an inverse and thereby by an infinite number of graphs , the `` alternative propagator '' @xmath166 , that featured prominently in the decomposition of lagrange correlation functions , is given by a finite number of graphs in terms of second order functions and differs from the ordinary one by the absence of the connected pieces in fig.[fig.proper ] . for a field theory without a three - point interaction vertex involving the time derivative of the fields , one has @xmath219 , and eq . ( [ generalg ] ) reduces to @xmath220 this simplified expression holds for theories like qed and/or self - interacting @xmath221theory , where the vacuum expectation value of the momentum field is completely determined in terms of @xmath41 , @xmath222 . actually , the expressions for @xmath223 , @xmath224 and @xmath225 in eq . ( [ generalgk ] ) encode a more general result because they could be obtained using eq . ( [ mge12 ] ) and eq . ( [ mge13 ] ) without setting variables to zero . as a consequence they show that dressed proper functions involving more than two external momentum legs are not present in such theories . this means that the mass and the coupling constant receive no contribution coming from the higher order corrections besides the usual ones given in the standard framework . on the other hand the absence of quantum corrections to @xmath226 @xmath224 , and @xmath225 means that only wave - function renormalization contributes to these kinetic terms . as in the present framework the quantum corrections to the propagators and vertices depend in general on temporal derivatives this shows that within the standard formalism there are several pieces in the effective action that depend on the time derivative of the averaged field . as a consequence , the canonical momentum fields defined on the level of the effective action differ from those given by the quantum average of @xmath42 since the involved limiting processes do not commute . although the decomposition of proper hamilton functions given above involves inversions that can not be omitted in terms of second order correlation functions it is possible to transform these equations into a form where such matrix inversions are eliminated . this is done by explicitly introducing first order correlation functions on the right hand side again . as we will show in this subsection , this leads precisely to the first order dses . we start from eq . ( [ 3dpqq ] ) and use it to express @xmath223 as @xmath227^{-1}.\label{newrpp}\ ] ] the multiplication of @xmath228 from the left hand side allows to write the above equation as @xmath229\gamma_{sj}^{pp}=-\delta_{ij},\nonumber\end{aligned}\ ] ] where eq . ( [ fopp1 ] ) has been taken into account . the introduction of a @xmath230 as @xmath231 in the last term in the bracket and the use of eq . ( [ fopp1 ] ) makes it possible to express the above relation in the following form @xmath232 which can be translated to @xmath233 this presents the dse for the proper momentum 2-point correlation function . it is possible to obtain a similar equation for the @xmath234propagator : @xmath235^{-1 } \nonumber\\ & & \quad \times \left(i_{0lj}^{pq } - \frac{i}{2}i_{0luk}^{pqq}\delta_{um}^{qq } \gamma_{mjn}^{qqq}\delta_{nk}^{qq}\right ) . \label{sofr}\end{aligned}\ ] ] the multiplication of @xmath228 from the right hand side allows to rewrite eq . ( [ sofr ] ) as @xmath236 this can be rewritten in the following form @xmath237 the introduction of a @xmath230 as @xmath231 in the first and second term inside the brackets allows to write this as @xmath238 by considering eq . ( [ shrepr ] ) we get @xmath239 which yields the result @xmath240 we remark that the last terms of eqs . ( [ sombra111 ] ) and ( [ sombra222 ] ) include several combinations of fundamental and momentum fields . the relations given by eq.([sombra111 ] ) and eq . ( [ sombra222 ] ) are the dses of the proper @xmath241- and @xmath242-propagators expressed in terms of the usual elements of the canonical formulation . it is straightforward to prove that eq . ( [ sombra111 ] ) coincides with the corresponding equation derived using eq . ( [ mge12 ] ) and eq . ( [ gqcm ] ) . this is different for eq . ( [ sombra222 ] ) . in fact , as discussed previously , we could in principle calculate these propagators from eq . ( [ sppdeqqqpr ] ) , nevertheless by considering the action eq . ( [ a ] ) and the relations between the bare elements we find that the structure is more cumbersome than that given in eq . ( [ sombra222 ] ) . instead , the latter one is in correspondence with eq . ( [ mge13 ] ) via the symmetry relation @xmath243 in case both @xmath98 and @xmath111 are bosonic fields . the fact that we recover the standard first order propagator dses from the decomposition of proper hamilton correlation functions has its origin in the equivalence between the canonical and lagrange equations of motion at the quantum level . in appendix [ motion - equivalence ] it is shown that whenever they describe the same dynamical processes , the dses derived from them will be equivalent too . however , as we argued in the last subsection , the classical canonical momentum fields defined from the effective action and those given by the quantum average of @xmath42 are not the same . this means that the equations expressed in term of @xmath111 and @xmath98 do not correspond to the classical canonical ones . based on this statement , the derivation of @xmath244 is considerably simpler using eq . ( [ sppdeqqq ] ) than via the procedure performed in the last two cases . indeed , by considering eqs . ( [ gqcm ] ) and ( [ a ] ) as well as the relations between the bare elements , we arrive at @xmath245_{q\to \bar{q}+\frac{\hbar}{i}\delta^{q\phi } \frac{\delta}{\delta \bar{\phi } } } .\label{tdasd}\end{aligned}\ ] ] now , in order to give the explicit form of @xmath244 , we rewrite eq . ( [ tdasd ] ) as @xmath246-\frac{i}{2}i_{0ijk}^{qqq}\delta_{jk}^{qq}[j]\nonumber\\ & -&\frac{1}{6}i_{0ijkl}^{qqqq}\delta_{jm}^{q\phi}[j]\delta_{ku}^{q\phi}[j ] \gamma_{mun}^{\phi\phi\phi}[\bar{q},\bar{p}]\delta_{nl}^{\phi q}[j_q ] \nonumber\\ & -&\frac{i}{2}i_{0ijkl}^{qqqq}\bar{q}_j\delta_{kl}^{qq}[j].\end{aligned}\ ] ] taking the functional derivative with respect to @xmath98 and setting the vacuum expectation values of the fundamental fields to zero we get @xmath247 the fourier transformation of eqs . ( [ sombra111 ] ) , ( [ sombra222 ] ) and ( [ ph4vt ] ) yields finally the corresponding dses in momentum space @xmath248 @xmath249 @xmath250 we show the graphical representation of the complete set of dses in fig.[f1 ] where for conciseness we use the matrix propagator which yields all possible loop graphs involving physical vertices in accordance with the symmetries of the action as a consequence of the implied summation over repeated indices . before we come to the main application of the developed formalism in the context of coulomb gauge qcd , we will show in this section that the above derived results also apply to theories involving auxiliary fields , _ i.e. _ the typical treatment of such theories represent a special case of the discussion presented here . however , it is not the kinetic but the interaction terms which are linearized . in the case of fermionic theories this is also referred to as bosonization @xcite . the action is by construction only quadratic in the auxiliary fields and lacks kinetic terms for them . therefore , as we detail below the above described analysis directly applies and gives general relations between correlation functions in the fundamental theory and the linearized form involving auxiliary fields . we will illustrate this in the case of fermionic theories with non - renormalizable , quartic interactions . examples for this class of theories are the nambu@xmath251jona - lasinio model @xcite or the bcs theory of superconductivity @xcite . the functional integral is given by @xmath252 with the lagrangian containing local quartic interactions @xmath253 here the @xmath254 are dirac matrices . the linearization of the fermionic interaction can be performed by formally introducing a one of the form @xmath255 into the fermionic path integral , where this path integral over @xmath256 enforces a functional @xmath257-function that allows to rewrite the non - linear fermionic interaction in terms of @xmath258 . integrating then over the @xmath258 yields the path integral of the corresponding linear sigma model @xmath259 where we have introduced additional sources for the auxiliary fields @xmath256 . after this bosonization procedure the lagrangian of the corresponding `` linear @xmath260-model '' reads @xmath261 where @xmath262 . here , there is , in contrast to the first order formalism , by construction no mixing between the fundamental and the auxiliary fields . analogous to eq . ( [ equivalenciaquantica ] ) the effective actions of the two theories are again identical at vanishing sources @xmath263 @xmath264=\gamma_\sigma[{\bar \psi}^\dagger , \bar \psi,\bar \sigma_i[{\bar \psi}^\dagger,\bar \psi]]\ ] ] where the auxiliary fields are implicitly given by @xmath265 via the chain rule of functional differentiation we obtain analogous expressions for eqs . ( [ opstart])-([propder ] ) with @xmath266 replaced by @xmath267 , respectively . correspondingly , the same graphical rules apply . with the simplification that there can be no mixing due to the different statistics of the fields , the external legs are not composite and the arising inverse 2-point function is the ordinary @xmath260-propagator . denoting as before connected correlation functions that are 1pi with respect to the fundamental fields and connected in the auxiliary fields as _ @xmath260-connected _ , this leads to the analogous general result : _ a proper @xmath0-point function in the fundamental theory can be decomposed into the sum of all @xmath260-connected @xmath0-point functions in the linearized theory . _ in particular the decomposition of the proper 4-fermion vertex in the fundamental theory eq . ( [ fermact ] ) reads where we represent in an analogous way the proper vertices in the fundamental and linearized theory by large and small blobs and the ordinary @xmath260-propagator by the dotted line . this result confirms a well - known fact , namely , that no double - counting occurs in the linearized theory although the original fermion field and the auxiliary bosonic field are employed both . this redundancy in the description can also be prevented from the outset by partial re - bosonization @xcite in the context of the functional renormalization group . in this approach the contribution of the fundamental degrees of freedeom is entirely absorbed into the bosonized interactions even at the level of the effective action . since the lagrangian of the linearized theory is at most quadratic and lacks kinetic terms for the @xmath260-fields by construction , these fields can be trivially integrated out retaining their sources at this point . this leads to an analogous expression to eq . ( [ mge ] ) for the correlation function of @xmath0 auxiliary fields @xmath268 after the sources are set to their vacuum expectation value . due to the absence of 3-point interactions and mixing , the decomposition of the auxiliary @xmath260-propagator is again simplified compared to the result given in fig . [ fig : mom - dec ] , where the different prefactors of the loop correction arise due to the fermionic nature of the fields . in contrast to the decomposition of the hamilton propagators there is no difference between proper and connected 2-point functions in the case of the auxiliary field due to the absence of mixing , and the proper correlation functions is simply the inverse of the above equation . this concludes the demonstration of the developed formalism to the case of theories with auxiliary fields . as detailed in the introduction the main motivation for developing the presented formalism is given by the fact that the first order formalism might be better suited for non - perturbative studies of coulomb gauge yang - mills theory . as a first step into this direction we will apply the formalism developed so far to give the explicit relations between the two - point correlation functions of the transversal and longitudinal components of the conjugate momentum to the ones of the gauge field . the starting point is the hamiltonian density of coulomb gauge yang - mills theory : @xmath269 where @xmath270 is the part of the hamiltonian density of pure yang - mills theory @xmath271 here , @xmath272 represents the chromomagnetic field , @xmath273 the time - component of the gluon field , whereas @xmath274 is the covariant derivative in the adjoint representation , with the structure constants @xmath275 of the color - group @xmath276 , and @xmath277 denotes the gauge coupling . on the other hand , @xmath278 is the piece of the hamiltonian density related to gauge fixing terms . above @xmath279 is the conjugate momentum of the gauge field @xmath280 @xmath281 and @xmath282 are the grassmann - valued faddeev - popov ghost fields introduced by fixing the gauge , and @xmath283 is a colored " lagrange multiplier field . note that the `` canonical action '' in the present case is given by @xmath284 it is remarkable that the ghost sector of the hamiltonian is fully disconnected from @xmath285 by identification of @xmath286 the above `` action '' can be written as a functional taylor expansion involving a pure bosonic piece like eq . ( [ can - action ] ) and one containing the ghost field as eq . ( [ fp2 ] ) . this fact allows therefore to use the general expressions derived so far . except for the inverse , bare ghost two - point function @xmath287 the remaining inverse tree level propagators of the theory are presented in table [ tab : w1 ] . note that @xmath54 involves four - vertices that have the following form in momentum space @xmath288,\\ i_{0ijlm}^{abcd(aaaa ) } & = & -g^2\left\{\delta_{ij}\delta_{lm}\left[f^{ace}f^{bde}-f^{ade}f^{cbe}\right ] \right.\nonumber\\ & & + \delta_{il}\delta_{jm}\left[f^{abe}f^{cde}-f^{ade}f^{bce}\right]\nonumber\\ & & + \left.\delta_{im}\delta_{jl}\left[f^{ace}f^{dbe}-f^{abe}f^{cde}\right ] \right\},\qquad\end{aligned}\ ] ] with all momenta defined as incoming . the decomposition of the conjugate momentum into transverse and longitudinal parts , @xmath289 , makes it convenient to study the required complete cancellation of the energy divergences @xcite that emerge in a perturbative treatment of coulomb gauge yang - mills theory . certainly , such a decomposition increases the number of fields of the theory and leads to @xmath290 where @xmath291 is a colored langrange multiplier field which appears via the transversality condition of @xmath292 in this context , the general structure of the connected two - point functions is given in table [ tab : w2 ] . the latter may be found independent of any approximation but based solely on the principles of brst invariance , spatial and time - reversal symmetries and transversality properties of the vector propagators . each dressing function @xmath293 is a dimensionless scalar function of @xmath294 and @xmath295 except the ghost propagator @xmath296 depending only on @xmath295 . the tree - level propagators are obtained when @xmath297 for a complete description , the reader is referred to ref . @xcite . the general results of the preceeding sections can in principle directly applied to coulomb gauge qcd as well by decomposing the momentum field into its individual components . however , since the longitudinal momentum field features more complicated bare correlation functions that involve additional derivative operators we derive the corresponding expressions of subsection [ gen - ham - con ] once more taking now the general action ( [ gac ] ) into account . to this end we first have to relate the bare momentum correlation functions with the corresponding ones for the individual momentum components . we formally express the decomposition via a the operator @xmath298 @xmath299 expanded in a local series in terms of the individual momentum components the general canonical action takes the form @xmath300=i_{0ji}^{qp}\pi_iq_j+\frac{1}{2}i_{0ij}^{pp}\pi_i\pi_j+ \frac{1}{2}i_{0ijk}^{pqq}\pi_iq_jq_k\\ & & \quad+i_{0ji}^{qp}\mathcal{x}_{il}\omega_l q_j\nonumber+ \frac{1}{2}\mathcal{x}_{il}i_{0ij}^{pp}\mathcal{x}_{jk}\omega_l\omega_k+ \frac{1}{2}i_{0ijk}^{pqq}\mathcal{x}_{il}\omega_l q_jq_k\\ & & \quad+\frac{1}{2}i_{0ij}^{qq}q_iq_j + \frac{1}{3!}i_{0ijk}^{qqq}q_iq_jq_k+ \frac{1}{4!}i_{0ijkl}^{qqqq}q_iq_jq_kq_l \nonumber \label{can - action2}\end{aligned}\ ] ] with @xmath301 the above expression yields for the transverse tree level correlators @xmath302 and @xmath303 as well as analogously for the longitudinal expressions @xmath304 @xmath305 and @xmath306 the gaussian integration over @xmath307 yields @xmath308=\mathcal{s}_0+\frac{1}{2}j^\pi_ij^\pi_i + j^\pi_i\frac{\delta\mathcal{s}_0}{\delta \dot{q}_i } \nonumber \\ & & + i_{0ji}^{q\omega}\omega_iq_j+\frac{1}{2}\omega_ji_{oij}^{\omega\omega } \omega_i+\frac{1}{2}i_{0ijk}^{\omega qq}\omega_i q_jq_k \nonumber \\ & & + j_i^{\omega}\omega_i + j_i^{q}q_i \label{can - actionvar}.\end{aligned}\ ] ] where @xmath78 is given by eq . @xmath309 and @xmath310 are the corresponding sources associated to the momentum fields . similarly , the subsequent gaussian integration over @xmath311 gives @xmath312&=&\mathcal{s}_0+\frac{1}{2}j^\pi_ij^\pi_i+ j^\pi_i\frac{\delta\mathcal{s}_0}{\delta \dot{q}_i}+j_i^qq_i -\frac{1}{2}j^\omega_i \left(i_{0}^{\omega\omega}\right)_{ij}^{-1}j^\omega_i \!-\ ! j^\omega_i\left(i_{0}^{\omega\omega}\right)_{ij}^{-1}\ ! \left(\!i_{0uj}^{q\omega}q_u+\frac{1}{2}i_{0juk}^{\omega qq}q_uq_k\!\right ) \nonumber \\ & -&\frac{1}{2}i_{0ji}^{q\omega}\left(i_{0}^{\omega\omega}\right)_{il}^{-1 } i_{0ml}^{q\omega}q_jq_m-\frac{1}{2}i_{0ji}^{q\omega}\left(i_{0}^{\omega\omega } \right)_{il}^{-1}i_{0lmu}^{\omega qq } q_jq_mq_u -\frac{1}{8}i_{0imu}^{\omega qq } \left(i_{0}^{\omega\omega}\right)_{ij}^{-1 } i_{0jkl}^{\omega qq } q_mq_uq_lq_k . \nonumber \\ \label{can - actionvar2}\end{aligned}\ ] ] by collecting the terms of the same order in @xmath41 we obtain the bare vertices in the second order formalism in terms of those that arise in the first order formalism . thereby , the master equation for the momentum propagator eq . ( [ fomp ] ) becomes in the case of the @xmath311-field @xmath313\left(i_{0}^{\omega\omega } \right)_{mj}^{-1 } \nonumber\end{aligned}\ ] ] whereas the corresponding mixed version eq . ( [ fopp ] ) is given by @xmath314 here the @xmath41 fields in eqs . ( [ eq : omegaomega ] ) and ( [ eq : omegaq ] ) again have to be replaced by @xmath315+\frac{\hbar}{i}\delta^{qq}[j^q ] \frac{\delta}{\delta q } .\ ] ] the corresponding equations involving the @xmath307 field are identical to eqs . ( [ fomp ] ) and ( [ fopp ] ) . performing the same algebraic steps as in sect . [ sec.con-ham ] yields the corresponding decomposition of the hamilton propagators in coulomb gauge qcd . the result is displayed in diagrammatic form in fig . [ fig : coulomb - decomposition ] , it constitutes the main result of this section . we have also checked the structure of these expressions by an explicit projection of the general equations on the corresponding momentum components in appendix [ app : projection ] . next we will consider the other direction of the connection discussed in sect . [ sec : lagrange - decomposition ] , the one which expresses proper lagrange correlators in terms of hamiltonian ones . this is interesting in the context of coulomb gauge qcd , since in the hamilton framework a complete proof of the renormalizability of the theory seems possible due to explicit cancellations ensured by powerful ward identities @xcite . here again it is the particular definition of the longitudinal momentum that complicates the issue and does not allow to give these equations explicitly . nevertheless , one can easily convince oneself that these equations are given entirely by tree graphs that introduce no additional divergences . therefore , once the renormalizability of the theory in the first order formalism is established , the corresponding connection immediately implies renormalizability of the theory also in the lagrange framework . yet the cancellation mechanism of arising divergences might be far from obvious in the latter framework and could nevertheless prevent simple truncation schemes . c|c c|c c @xmath54 & @xmath316 & @xmath317 & @xmath260 & @xmath318 + ' '' '' @xmath319&@xmath320 & @xmath321&@xmath322 & @xmath323 + @xmath324&@xmath325 & @xmath326&@xmath323&0 + @xmath327&0&@xmath328&@xmath329 & @xmath329 + @xmath330&@xmath328&0&0 & 0 + c|c c|c c|c c & & & & & + @xmath331 & @xmath316 & @xmath332 & @xmath260 & @xmath311 & @xmath318 & @xmath291 + ' '' '' @xmath333&@xmath334 & @xmath335&0&0 & @xmath336&0 + @xmath337&@xmath338 & @xmath339&0&0&0 & @xmath336 + @xmath260&0&0&@xmath340 & @xmath341 & @xmath342&0 + @xmath311&0&0&@xmath341 & @xmath343 & @xmath344&@xmath345 + @xmath318&@xmath346&0&@xmath347 & @xmath348&0&0 + @xmath291&0&@xmath346&0&@xmath345&0&0 + next we consider the functional symmetry identity . the canonical action @xmath54 is brst - invariant , _ i.e. _ invariant under @xmath349,\ \ \delta\lambda^a=0 . \nonumber\end{aligned}\ ] ] here @xmath350 is a grassmannian infinitesimal parameter , whereas @xmath351 on the other hand @xmath352 is the chromoelectric field . note in addition that @xmath353 is some color - singlet constant which in general could be some function of position . by considering the above transformation the slavnov - taylor identity reads @xmath354\ ! \!\int \ ! \ ! d^4x \left\{-\frac{1}{g}\rho^ad^{0ab}c^b+ \frac{1}{g}\vec{j}^a\cdot\vec{d}^{ab}c^b-\frac{1}{g}\lambda^a\eta^a\right . \nonumber\\ & -&\left.\frac{1}{2}f^{abc}\bar{\eta}^ac^bc^c+f^{abc}c^b\left[(1-\alpha ) \vec{p}^c+\alpha\vec{e}^c\right]\cdot\vec{j}_p\right\}\nonumber\\ & \times&\exp\left\{i i_0+ii_s\right\},\end{aligned}\ ] ] where @xmath355 employing a procedure analogous to the one used in sect . [ sec : srifs ] we obtain @xmath356\ ! \!\int \ ! \ d^4x \left\{-\frac{1}{g}\rho^ad^{0ab}c^b+\frac{1}{g}\vec{j}^a\cdot\vec{d}^{ab}c^b -\frac{1}{g}\lambda^a\eta^a\right.\nonumber\\&-&\left.\frac{1}{2}f^{abc } \bar{\eta}^ac^bc^c+f^{abc}c^b\left[(1-\alpha)\vec{j}_p^c-\vec{e}^c\right ] \cdot\vec{j}_p\right\}\nonumber\\&\times&\exp\left\{i s_0+is_s\right\},\end{aligned}\ ] ] where @xmath357 $ ] denotes the remaining integration measure of the fields @xmath358 and @xmath260 . expressed in the field strength tensor @xmath359 one has @xmath360 in addition we decompose the source @xmath361 into transversal and longitudinal components . employing this decomposoition the slavnov - taylor identity can be written as @xmath362\int d^4x\left\{-\frac{1}{g}\rho^ad^{0ab}c^b+ \frac{1}{g}\vec{j}^a\cdot\vec{d}^{ab}c^b- \frac{1}{g}\lambda^a\eta^a\right.\\&-&\left . \frac{1}{2}f^{abc}\bar{\eta}^ac^bc^c + f^{abc}c^b(1-\alpha)\left[\vec{j}^\pi_c\cdot\vec{j}^\pi_a+\frac{\vec{\nabla } } { -\nabla^2}j^\omega_c \frac{\vec{\nabla}}{-\nabla^2}j^\omega_a\right]\right . -\left.f^{abc}c^b\left[\vec{e}_t^c\cdot\vec{j}^\pi_a-\vec{e}_l^c \frac{\vec{\nabla}}{-\nabla^2}j^\omega_a\right]\right\}\exp\left\{i s_0+is_s\right\},\nonumber\end{aligned}\ ] ] where @xmath363 this completes the application of the developed formalism to coulomb gauge yang - mills theory . we conclude with our main result that given a quantum field theory in the context of the first order formalism it is possible to decompose all green s functions in terms of those obtained from the second order formalism and vice versa . whereas proper lagrange correlation functions are given explicitly to all orders and involve only tree graphs involving dressed hamilton correlation functions , the decomposition of hamilton @xmath0-point functions involves loop graphs of loop order @xmath0 . although the structure of the latter equations seems to be somewhat cumbersome , they are still more compact and simple than the usual dses within the hamilton formalism . we have discussed the connection between the hamilton and the lagrange formalism for a general quantum field theory and illustrated the detailed structure of the arising relations in the important case of a generic four - dimensional renormalizable field theory . in accordance with the obtained equations we have argued that in theories where the quantum average of the momentum fields is completely determined as @xmath364 the proper @xmath85propagators receives quantum corrections only via wave - function renormalization . additionally , it has been shown that the canonical momentum fields which can be defined from the effective action and those given by the quantum average of @xmath42 are in general different . as a demonstration of the general nature of the presented formalism we also showed that the results obtained in this paper can also be applied to the case of theories involving auxiliary fields from a linearization of the interaction part of the action . this yields general relations between correlation functions in the fundamental theory and the linearized form involving auxiliary fields . a major difference is that in this case the fundamental and the auxiliary fields do not mix which simplifies the connection considerably . clearly , the determination of green s functions of yang - mills theories in the first order formalism is considerably more complicated as this case , or even when compared to scalar or abelian gauge theories . the presence of a coupling between the gauge fields and their time derivative as well as the subtleties of gauge fixing present major complications . in particular the renormalizability still poses a major challenge . the presented general connection of all green functions in the two formulations implies that no additional divergences arise when going from one to the other . however , what still needs to be shown is that the theory is renormalizable in both formulations once a proof is given for either of them . on the other hand , renormalizing the theory is probably simpler in the first order formalism where energy divergences explicitly cancel . the obtained connections should then help to renormalize the second order theory where numerical calculations are far simpler . a detailed analysis of the renormalizability in coulomb gauge yang - mills theory will therefore be presented in a forthcoming publication . it is our pleasure to thank peter watson for helpful discussions . s. v .- c . is supported by the doktoratskolleg `` hadrons in vacuum , nuclei and stars '' of the austrian science fund ( fwf ) under contract w1203-n08 . k. s. acknowledges support from the fwf under contract m979-n16 , and r. a. from the german research foundation ( dfg ) under contract al 279/5 - 2 . in this appendix we perform the integration over the momentum fields in the vacuum - vacuum transition amplitude in presence of the classical sources . in order to do this , we assume a general quadratic hamiltonian density given by eq.([hamiltoniand1 ] ) . in order to evaluate the gaussian functional integral over the momentum fields we first consider a finite dimensional generating functional @xmath365=\mathcal{n}_j\int_{-\infty}^{\infty } \prod_{j=1}^{n-1 } \ ! \ ! d^4q^j \prod_{j=0}^{n}\ ! \ ! d^4p^j \exp\left[\frac{i}{\hbar}(\tilde{i}_0-\mathcal{c}+j_q^iq^i)\right].\ ] ] with the momentum field dependent part of the action @xmath366 by completing the squares in the above expression we arrive at @xmath367\mathcal{a}_{ij}^{-1 } \left [ \dot{q}^j-\mathcal{b}^j+j_p^j\right ] , \end{aligned}\ ] ] where @xmath368 . by considering the change of the integration variables from @xmath369 to @xmath370 we obtain @xmath365=\tilde{\mathcal{n}}_j\int\prod_{i=1}^{n-1}d^4q^j \exp\left[\frac{i}{\hbar}\tilde{\mathcal{s}}[q , j]\right ] \label{wqe}\ ] ] with @xmath371\ ] ] and @xmath372\ ! & = & \!\mathcal{s}_0[q ] \\ & + & \frac{1}{2}\!j_p^i \mathcal{a}^{-1}_{ij}[q]j_p^j+j_q^i q^i+j_p^i\mathcal{a}^{-1}_{ij}[q ] \!\left(\dot{q}^j\!-\!\mathcal{b}^j[q]\right).\nonumber \label{gac}\end{aligned}\ ] ] here , the action @xmath78 can be obtained by considering the limit of @xmath373 $ ] when @xmath374 . explicitly it reads @xmath79=\frac{1}{2}\dot{q}^i\mathcal{a}^{-1}_{ij}[q]\dot{q}^j- \dot{q}^i\mathcal{a}^{-1}_{ij}[q]\mathcal{b}^j[q]-\int d^4x \mathcal{v}[q ] , % \label{act}\ ] ] where @xmath375=-\frac{1}{2}\mathcal{b}^i[q]\mathcal{a}^{-1}_{ij}[q]\mathcal{b}^j[q]+ \mathcal{c}[q].\ ] ] the canonical momentum fields defined as @xmath376\left(\dot{q}^j-\mathcal{b}^j[q]\right)\label{cvbr}\ ] ] allow us to identify and substitute the last term in @xmath377 $ ] by its respective definition , and to eventually obtain the desired form of eq . ( [ actionlf ] ) . in this appendix we prove the statement made in subsection [ sub : proper - decomposition ] that the replacement rules precisely generate the @xmath42-connected correlators . as usual for a statement over the integers this is done by induction . the statement is trivially fulfilled in the case @xmath378 given explicitly before . now let us assume it is fulfilled for all integers @xmath379 and show that this implies its validity for @xmath380 . in the @xmath42-connected diagrams with @xmath0 legs we single out the proper vertex we started the iteration with . this vertex can be connected to other connected clusters so that all arising graphs have the general form here we suppress all external legs and only give their number next to the corresponding vertex . the sum over @xmath381 counts the connected clusters attached to the considered proper vertex which in addition has @xmath382 external legs and the implicitly given sum labeled @xmath383 runs over all ways to distribute the remaining @xmath94 external legs to the @xmath381 indistinguishable connected vertices ( taking into account that a vertex has at least 3 legs ) . next we consider the attachment of an additional external leg via the rules given in fig . [ fig : rules ] , applied in all possible ways . the above representation then goes over to a @xmath380-point function of the following form here the connected cluster with @xmath380 legs in the first class of graphs is labeled by the additional index @xmath384 ( for reduced ) since one of the legs is connected to the @xmath42-propagator instead to a composite external leg . the derivative of this propagator is explicitly present via the second class of graphs . when the new proper vertex is absorbed in the connected cluster , according to the induction assumption ( @xmath385 ) these two terms together precisely yield the full @xmath42-connected vertex with @xmath386 external legs and without the index @xmath384 . since each connected cluster had before already at least two external legs , together with the new one there are now at least three . terms where besides the new external leg there is only one other external leg at a connected cluster are explicitly given by the fourth class of graphs involving a new `` connected cluster '' that consists only of the proper 3-point vertex . finally , the general expression at order @xmath380 contains also graphs where the new external leg is attached to the proper vertex itself , given by the third class of graphs . altogether , the sum of the different classes of graphs precisely yields all necessary graphs at order @xmath380 , which completes the proof . in this appendix we give the complete results for the decomposition of the proper 2-point function in the hamilton formalism in terms of those in the lagrange formalism . we represent the appearing operator inverses in terms of schwinger proper time integrals . the momentum correlation function is given as @xmath387\right\}.\end{aligned}\ ] ] whereas the mixed correlation function reads @xmath388\right\}\nonumber\\ & \times & \left\{-ik_0\delta_{lj}- \frac{i}{2}\int\dbar\sigma i_{0fhl}^{qqp}(-\sigma - k,\sigma , k ) \delta_{fg}^{qq}(\sigma+k)\gamma_{gjw}^{qqq}(\sigma+k ,- k,-\sigma ) \delta_{wh}^{qq}(\sigma)\right\}.\label{sofr1}\end{aligned}\ ] ] the fundamental field correlation function is finally given by the lengthy expression @xmath389\right\ } \nonumber\\ & \times&\left\{-ik_0\delta_{lj}- \frac{i}{2}\int\dbar\sigma i_{0fhl}^{qqp}(-\sigma - k,\sigma , k ) \delta_{fg}^{qq}(\sigma+k)\gamma_{gjw}^{qqq}(\sigma+k ,- k,-\sigma ) \delta_{wh}^{qq}(\sigma)\right\}.\label{ph42}\end{aligned}\ ] ] in this appendix we derive the dses in the first order formalism directly . in order to do this we compute eq . ( [ sdev ] ) with the canonical action given in eq . ( [ can - action ] ) @xmath390_{\phi_\to \bar{\phi}+\frac{\hbar}{i}\delta^{\phi\phi } \frac{\delta}{\delta \bar{\phi}}}.\label{mty}\end{aligned}\ ] ] we can identify the sum of the last two terms inside the bracket as the quantum canonical momentum fields , which allows to write the above relation as eq . ( [ mge11 ] ) . + the substitution of eq . ( [ mty ] ) in eq . ( [ sdeqqq ] ) , allows to write the latter as @xmath391_{\phi\to \bar{\phi}+\frac{\hbar}{i}\delta^{\phi\phi } \frac{\delta}{\delta \bar{\phi}}}\label{gfhs}\ ] ] where @xmath392 and @xmath393 in the last two equations we have used the elementary decomposition of the bare elements and eq . ( [ gqcm ] ) . plugging eqs . ( [ rtyy ] ) and ( [ rty ] ) into eq . ( [ gfhs ] ) we arrive at the dse ( [ tdasd ] ) . defining the transverse projector @xmath394 one has @xmath395 and @xmath396 . as no time derivative appears one can obtain @xmath397 @xmath398 @xmath399 and @xmath400 by projecting @xmath42 in the two - point functions @xmath401 , @xmath402 and @xmath403 . for instance to obtain @xmath404 we decode the result given in eq . ( [ qppp ] ) . let us first identify the fundamental field @xmath41 with index @xmath203 with @xmath405 we then expand the remaining sums over the fields involved in the @xmath41 s considering all possible combinations which generate the bare vertex @xmath406 . the bosonic symmetry of the latter allows to write @xmath407 here no internal propagator like @xmath408 appears since no proper vertex functions with @xmath409derivative exist . the relation between this function and @xmath404 is given by @xmath410 obviously the case corresponding to @xmath411 is obtained by replacing in eq . ( [ qpppvrie ] ) @xmath412 where appropriate and considering the relation @xmath413 note that the sum over @xmath414 in eq . ( [ lachacha ] ) and over @xmath204 in eq . ( [ lachachacha ] ) must be understood over the discrete spatial indices only . the bare greens functions involving the full momentum field can now be expressed by the corresponding expression for the individual components via the explicit expressions given in section [ sec : coulomb ] which results in the equations given in diagrammatic form in fig . [ fig : coulomb - decomposition ] . 99 j. s. schwinger , phys . * 127 * ( 1962 ) 324 . n. h. christ and t. d. lee , phys . d * 22 * ( 1980 ) 939 [ phys . scripta * 23 * ( 1981 ) 970 ] . s. weinberg , `` the quantum theory of fields . vol . 1 : foundations , '' _ cambridge , uk : univ . ( 1995 ) 609 p _ p. watson and h. reinhardt , phys . d * 75 * ( 2007 ) 045021 [ arxiv : hep - th/0612114 ] . p. watson and h. reinhardt , phys . d * 76 * ( 2007 ) 125016 [ arxiv:0709.0140 [ hep - th ] ] . a. andrasi , eur . j. c * 37 * ( 2004 ) 307 [ arxiv : hep - th/0311118 ] . a. andrasi and j. c. taylor , eur . j. c * 41 * ( 2005 ) 377 [ arxiv : hep - th/0503099 ] . d. zwanziger , nucl . b * 518 * ( 1998 ) 237 . r. n. mohapatra , phys . d * 4 * ( 1971 ) 378 ; 1007 . v. n. gribov , nucl . b * 139 * ( 1978 ) 1 . r. alkofer and j. greensite , j. phys . g * 34 * ( 2007 ) s3 [ arxiv : hep - ph/0610365 ] . a. niegawa , m. inui and h. kohyama , phys . d * 74 * ( 2006 ) 105016 [ arxiv : hep - th/0607207 ] . g. leibbrandt and j. williams , nucl . b * 475 * ( 1996 ) 469 [ arxiv : hep - th/9601046 ] . g. heinrich and g. leibbrandt , nucl . b * 575 * ( 2000 ) 359 [ arxiv : hep - th/9911211 ] . l. baulieu and d. zwanziger , nucl . b * 548 * ( 1999 ) 527 [ arxiv : hep - th/9807024 ] . a. p. szczepaniak and e. s. swanson , phys . d * 65 * ( 2002 ) 025012 [ arxiv : hep - ph/0107078 ] . a. andrasi and j. c. taylor , arxiv:0704.1420 [ hep - th ] . p. watson and h. reinhardt , phys . d * 77 * ( 2008 ) 025030 [ arxiv:0709.3963 [ hep - th ] ] . f. j. dyson , phys . * 75 * ( 1949 ) 1736 . j. s. schwinger , proc . nat . acad . * 37 * ( 1951 ) 452 ; 455 . r. alkofer and l. von smekal , phys . * 353 * ( 2001 ) 281 [ arxiv : hep - ph/0007355 ] . j. c. ward , phys . * 78 * ( 1950 ) 182 . h. s. green , proc . a * 66 * ( 1953 ) 873 . y. takahashi , nuovo cim . * 6 * ( 1957 ) 371 . j. c. taylor , nucl . b * 33 * ( 1971 ) 436 . a. a. slavnov , theor . * 10 * ( 1972 ) 99 [ teor . mat . * 10 * ( 1972 ) 153 ] . r. alkofer , m. q. huber and k. schwenzer , arxiv:0801 . 2762 [ hep - th ] ; comput . commun . * 180 * ( 2009 ) 965 [ arxiv:0808.2939 [ hep - th ] ] . j. hubbard , phys . rev . lett . * 3 * ( 1959 ) 77 . y. nambu and g. jona - lasinio , phys . rev . * 122 * , 345 ( 1961 ) . j. bardeen , l. n. cooper and j. r. schrieffer , phys . * 108 * ( 1957 ) 1175 . h. gies and c. wetterich , phys . d * 65 * ( 2002 ) 065001 , hep - th/0107221 .

the semiclassical quantisation of non - integrable systems using properties of their periodic orbits was triggered by m. gutzwiller @xcite and extended by several groups @xcite . it allows one to express the oscillating part @xmath7 of the quantum - mechanical density of states , given exactly in terms of a quantum spectrum @xmath8 and separated into two terms by g(e ) = _ n ( e - e_n ) = g(e)+g(e ) , [ gqm ] through a semiclassical trace formula of the form g_sc(e ) _ po a_po(e ) [ s_po(e)/-_po/2 ] . [ dgsc ] the sum is over all periodic orbits @xmath9 of the classical system , @xmath10 are their action integrals , the amplitudes @xmath11 depend on their stabilities and degeneracies , and @xmath12 are some phases called maslov indices . the average part @xmath13 of the density of states , which by definition varies smoothly with energy , is obtained by the extended thomas - fermi ( etf ) model ( see , e.g. , @xcite , chapter 4 ) . then , the sum @xmath14 usually turns out to be a good approximation to the exact quantity , although the sum over the @xmath15 in is only an asymptotic one , correct to leading order in @xmath16 , and in chaotic systems is hampered by convergence problems @xcite . for integrable systems , berry and tabor @xcite and later creagh @xcite showed how a trace formula of the form can quite generally be obtained from the einstein - brillouin - keller ( ebk ) quantisation @xcite . this method will be used in the present paper for a spherical system . the semiclassical theory , often referred to as periodic orbit theory ( pot ) , has been very successful not only in ( partial ) quantisation of a given hamiltonian , but also in interpreting many experimentally observable quantum oscillations in finite fermion systems so - called `` shell effects '' in terms of classical mechanics . examples are atomic nuclei @xcite , metallic clusters @xcite , semiconductor quantum dots @xcite , and metallic nanowires @xcite . an overview over many aspects of the pot and illustrative applications are given in @xcite . one problem that remains with all the trace formulae developed in the work quoted above is that the amplitudes @xmath17 diverge in situations where a continuous symmetry is broken or restored under the variation of a system parameter ( which may also be the energy @xmath18 ) , or where bifurcations of periodic orbits occur . in such situations one has to go beyond the stationary phase approximation which is underlying the semiclassical approach . this has , besides @xcite , been developed most systematically by ozorio de almeida and hannay @xcite for both situations , leading to local uniform approximations with finite amplitudes @xmath17 . so - called global uniform approximations , which yield finite amplitudes at symmetry - breaking and bifurcation points and far from them go over into the standard ( extended ) gutzwiller trace formula , were developed for the breaking of u(1 ) symmetry in @xcite , for some cases of u(2 ) and so(3 ) symmetry breaking in @xcite , and for various types of bifurcations in @xcite . ( details and further references may be found in @xcite , chapter 6.3 . ) in this paper we investigate an anharmonically perturbed three - dimensional spherical harmonic oscillator ( ho ) hamiltonian ( for a particle with mass @xmath19 ) : h ( , ) = 12p^2 + 12 ^ 2 r^2 + 14r^4 = h_0(r , p ) + h(r ) , [ hpert ] where @xmath20 and @xmath21 are the absolute values of the three - dimensional momentum and radius vectors . @xmath22 is the strength of the perturbation which first is assumed to be small , but later may assume arbitrary positive values . the unperturbed ho system has u(3 ) symmetry @xcite which is broken by the anharmonic term , leading to the so(3 ) symmetry of the perturbed spherical system . the hamiltonian suggested itself in a recent study @xcite of harmonically trapped fermionic atoms with a short - range repulsive two - body interaction , treated self - consistently in the hartree - fock approximation . as a result , very pronounced shell effects in the total energy @xmath23 of the interacting system as a function of the number @xmath24 of atoms were found , which remind about the so - called super - shells predicted @xcite and observed @xcite in metallic clusters . in a first attempt to interpret these shell effects semiclassically @xcite , we parameterised the self - consistent mean field of the interacting system by the hamiltonian and applied the semiclassical perturbation theory of creagh @xcite to explain qualitatively the shell structure of the hf results . in the present paper we describe some of the mathematical details of the semiclassical perturbation theory for the symmetry breaking u(3 ) @xmath25 so(3 ) and develop a uniform trace formula valid for arbitrary strengths @xmath1 in the hamiltonian . in section [ secpert ] we present the perturbative trace formula for the density of states which is valid in the limit of small @xmath1 and has been presented shortly in @xcite . it already puts the dominance of the shortest periodic orbits , namely the circles and diameters , into evidence . although their individual contributions to the perturbative trace formula diverge in the limit @xmath26 , their sum restores to the unperturbed ho trace formula with u(3 ) symmetry in this limit , which at the same time is the limit of zero energy . in section [ secuni ] we develop the uniform trace formula that includes the contributions of the diameter and circle orbits for arbitrary values of @xmath1 . for this purpose , we start from the ebk quantisation of an arbitrary system with spherical symmetry and apply the poisson summation formula . we obtain a one - dimensional trace integral for the semiclassical density of states , whose end - point contributions correspond to the diameter and circle orbits , valid for an arbitrary spherical one - body potential . for the hamiltonian , the gross - shell structure of the density of states is at low energies always dominated by the families of diameter and circle orbits , although these orbit families only have a two - fold degeneracy . at sufficiently high energies and repetition numbers , `` rational tori '' with frequency ratios @xmath27 with @xmath28 bifurcate from the circle orbits with repetition numbers @xmath29 , as discussed in section [ sectori ] . these rational tori have the highest ( three - fold ) degeneracy possible in a three - dimensional spherical system , but due to their length they only affect the finer quantum structures of the density of states . this situation is completely different from that of a spherical billiard @xcite where the contribution of the diameter orbits to the density of states is practically negligible . in the limit @xmath30 , becomes a purely quartic oscillator which in many respects is easier to handle . in particular , in this limit the system acquires the `` scaling property '' that its classical dynamics becomes independent of the energy which can be absorbed by a rescaling of coordinates and time . this limit will be discussed separately in section [ secquart ] . one interesting classical aspect is that no bifurcations of the circle orbits occur . the same rational tori , which bifurcate from the circle orbit in the perturbed ho system , exist here at all energies . but , again , they affect only the finer quantum structures of the density of states , while its gross - shell structure is largely dominated by the circles and diameters . in the appendices [ app1 ] - [ app3 ] , we collect some mathematical details about the integration over the manifolds @xmath2p@xmath3 and s@xmath31 and some explicit analytical expressions for action integrals and periods in terms of elliptic integrals . finally , in the appendix [ appothers ] , we re - derive from our general trace integral the known trace formulae for two popular spherical systems : the spherical billiard and the coulomb potential . we first write down the exact density of states of the unperturbed three - dimensional harmonic oscillator ( ho ) , given by the hamiltonian h_0(r , p ) = 12 p^2 + 12 ^ 2 r^2 . [ hho ] its quantum - mechanical density of states can be written in the form g_0(e ) = _ n=0^d_n ( e - e_n ) = g_0^etf(e ) + g_0(e ) , [ gsum ] using the spectrum @xmath32 with the degeneracy factor @xmath33 . here @xmath34 is the smooth part given by the etf model g_0^etf(e ) = , [ getfho ] and @xmath35 is the oscillating part given by the following exact trace formula @xcite g_0(e ) = 2g_0^etf(e)_k=1^(-1)^k ( 2k ) = 2g_0^etf(e)_k=1^(-1)^k . [ dg3ho ] it can be understood as a sum over all periodic orbits of the system . hereby @xmath36 represents the repetition number of the primitive classical periodic orbit which has the action @xmath37 . keeping only the leading tf term of @xmath34 in , the density of states can also be written in the form g_0(e ) = e _ -^(-1)^ke^i2ke/ + o(^-1 ) , [ g3hoc ] since the imaginary parts cancel upon summation . note that the smooth tf part in comes from the contribution with @xmath38 in . next we follow creagh @xcite by writing the perturbed trace formula in the form g_pert(e ) = e _ -^(-1)^k m(k/)e^i2ke/ , [ gpert ] where @xmath39 is a modulation factor that takes into account the lowest - order perturbation of the action of the ho orbits . here @xmath40 is a small action that , quite generally , depends on @xmath1 and the energy @xmath18 . the factor in front of the sum in takes into account only the leading term of the smooth part @xmath34 ; this is consistent with the fact that the perturbation theory only deals with the terms of leading order in @xmath41 of the semiclassical trace formula . as we will see in section [ secqm ] , the contributions of terms of relative order @xmath42 in @xmath34 are , indeed , negligible . as described in detail in @xcite , one obtains @xmath39 from an integration of a perturbative phase function over the manifold that describes the classical degeneracy of the ho orbits . the unperturbed ho has u(3 ) symmetry , so that @xmath39 is formally obtained by the average ( k/ ) = e^iks(o)/_ou(3 ) , [ mav ] where @xmath43 is an element of the group u(3 ) characterising a member of the unperturbed ho orbit family , and @xmath44 is the lowest - order primitive action shift brought about by the perturbation @xmath45 in . in the present case @xmath44 is nonzero already in first - order perturbation theory and therefore @xmath46 is proportional to @xmath1 . we need , however , not integrate over the full nine - dimensional space of group elements of u(3 ) . it is sufficient to consider only the subset of symmetry operations which transform the orbits within a given degenerate family into each other ( without changing their actions ) . the dimension of this subset is the degree of degeneracy @xmath47 of that family . for the three - dimensional spherical ho we have @xmath48 . this can most easily be seen by the following argument which also will allow us to find a suitable parametrisation for the four - fold integration . the full phase space is six - dimensional ; due to energy conservation it is reduced to a five - dimensional energy shell which has the topology of a five - sphere s@xmath31 . of the five parameters that specify a point on s@xmath31 , one can be chosen as a trivial time shift along the periodic orbits corresponding to a simple phase factor @xmath49 . this parameter forms a subgroup u(1 ) , so that its elimination restricts us to the four - dimensional manifold ( see appendix [ app1 ] for its mathematical definition ) s^5/u(1 ) = ^2 , [ coset ] which is neither a four - torus nor a four - sphere @xcite . ( note that for the two - dimensional ho one is led to the manifold @xmath2p@xmath50 which happens to be homomorphous to the two - sphere s@xmath3 , see @xcite . ) a suitable parametrisation of @xmath2p@xmath3 is given in appendix [ app1 ] in terms of four angles @xmath51 whose meaning will become clear in a moment . the modulation factor therefore becomes ( k/ ) = d_^2 e^iks/ = _ 0^ d _ 0^ ^3d _ 0 ^ 2 d_2 _ 0 ^ 2 d_3 e^iks(,,_2,_3)/. [ mfac ] in the zero - perturbation limit @xmath26 , where the action shift @xmath52 and hence also @xmath46 is zero , the modulation factor becomes unity , as it should for to approach . next we have to determine the action shift @xmath52 of a primitive periodic orbit of the ho , caused by the perturbation @xmath53 . the harmonic solutions of the classical equations of motion of the unperturbed system are well known : x_i = ( t+_i ) , i=1,2,3 ( x_1,x_2,x_3)=(x , y , z ) . [ xi ] hereby @xmath54 are the three conserved energies in the three dimensions . depending on the values of @xmath54 and the phases @xmath55 , describes circles , ellipses or librations through the origin ; we will in the following call the latter the `` diameter orbits '' . we now re - parameterise the @xmath56 in the following way : x(t ) = rn_1(t),y(t ) = rn_2(t+_2),z(t ) = rn_3(t+_3 ) , [ xi3d ] where @xmath57 , and @xmath58 is given by r=/,e = e_1+e_2+e_3 . since the @xmath59 must lie on the sphere @xmath60 , due to the conservation of energy , we can also write them as : n_1 = , n_2 = , n_3 = . [ s2par ] actually , since the @xmath61 must be restricted to positive definite values in order to cover once all classically allowed values of the @xmath62 , we only have to integrate over the first octant of s@xmath3 , so that @xmath63 $ ] . the four angles @xmath64 , with their ranges of definition , are precisely the parameters describing the manifold @xmath2p@xmath3 as explained in appendix [ app1 ] , and the correct integration measure is that used in . we might have kept the phase angle @xmath65 of @xmath66 in , too ; integration over it is tantamount to a trivial integration over time along the orbits . the resulting five - dimensional integral becomes equivalent to that over the s@xmath31 sphere discussed in appendix [ app2 ] . according to creagh @xcite , the first - order action shift brought about by a perturbation @xmath67 is given by _ 1 s = - _ po h(t ) dt , where @xmath15 stands for a particular member of the unperturbed periodic orbit family , and the perturbating hamiltonian @xmath68 has to be evaluated along the periodic orbit . inserting - into the perturbation @xmath69 leads to elementary integrals over powers of the trigonometric functions . the result is _ 1 s(,,_2,_3 ) & = & -(3r^4 - 4l^2/^2 ) = -(3e^2-^2l^2 ) + & = & -[3-^2(2 ) ( ^2 ^ 2_2+^2 ^ 2_3 ) -^4 ^ 2(2)^2(_2-_3 ) ] , [ ds3d ] where @xmath70 is the conserved squared angular momentum , and the first - order action unit @xmath46 is given by = e^2/4 ^ 5 . [ sigma ] the integral , with the function given in , has been integrated numerically and found to be identical with the corresponding five - dimensional integral over the 5-sphere s@xmath31 , expressing @xmath71 in terms of the five hyperspherical angles of the six - dimensional polar coordinates ( cf . appendix [ app2 ] ) . however , even the four - dimensional integration in is more than needed . in fact , the same result is obtained if we replace the phase function under the @xmath2p@xmath3 integration as follows : _ 1 s(,,_2,_3 ) _ 1 s ( ) = -(3-^2 ) . [ ds3da ] the integrals over @xmath72 , @xmath73 and @xmath74 then become trivial and the result is found ( with @xmath75 ) to be ( k/ ) & = & 2 _ 0^ d e^-ik(3-^2)/ = _ 0 ^ 1 dze^-ik(3-z)/ = . [ modfex ] the replacement is not obviously justifiable in a direct way , but a suitable reduction of the five - dimensional s@xmath31 integral yields exactly the integral , as demonstrated in appendix [ app2 ] . identically the same result is also obtained independently from the ebk quantisation of the perturbed hamiltonian using poisson summation and keeping only first - order terms in @xmath1 ; it is given in eq . at the end of sect . [ secpois ] . the form of the modulation factor suggests a simple physical interpretation : the resulting two terms correspond to two separate families of orbits that survive the breaking of the u(3 ) symmetry . as we will see below , these are the circle and the diameter orbits with maximal and minimal ( i.e. , zero ) angular momentum , respectively , at fixed energy . for @xmath76 these are actually the shortest periodic orbits found in the perturbed system . in contrast to the four - fold degenerate unperturbed ho orbits , they only have a two - fold degeneracy , since only rotations about two of the euler angles change their orientation . in general , the most degenerate periodic orbits in a three - dimensional system with spherical symmetry so(3 ) can be rotated about three euler angles . however , for each of the circle and diameter orbits one of the three rotations is redundant in the sense that it maps these orbits onto themselves : for the circles , it is the in - plane rotation around their centre , and for the diameters , it is the rotation about their own axis of motion . the loss of two degrees of degeneracy of these two orbit types with respect to the unperturbed ho orbits appears in the form of the factor @xmath77 in their amplitudes in , in agreement with the general counting of the powers of @xmath77 in semiclassical amplitudes @xcite . we will see in the next section that for values @xmath78 and large enough values of @xmath1 , there exist fully three - fold degenerate orbits , namely orbits with rational ratios @xmath79 of radial and angular frequencies with @xmath80 . they are created at bifurcations of the @xmath36-the repetitions of the circle orbits with @xmath81 . this happens , however , only at finite values of @xmath1 , so that these orbits do not contribute in the limit @xmath26 . inserting into , we find the following perturbed trace formula for the oscillating part of the level density : g_pert(e ) & = & _ k=1^ + & = & _ k=1^ , [ dgp ] where @xmath82 and @xmath83 are the perturbed primitive actions of the diameters and the circle , respectively : s_d = s_0 - 3 , s_c = s_0 - 2,s_0=2e/. [ sorbits ] these values will be confirmed from the general expressions derived in the next section . the perturbative trace formula uniformly restores the oscillating part of the exact ho trace formula in the limit @xmath26 . in section [ secqm ] we will test its range of validity comparing against quantum - mechanical results . note that as is usual in perturbation theory the semiclassical amplitudes and actions of the orbits contributing to are correct only in the small-@xmath1 limit . in general , one has to generalise the perturbative trace formula to a uniform version that , in the limit of large perturbation , goes over into the corresponding ( extended ) gutzwiller trace formula . this uniformisation has been done for the breaking of u(1 ) symmetry in @xcite , and for some cases of u(2 ) symmetry breaking in @xcite ; one of the latter result applies , with suitable changes , also to some cases of so(3 ) @xmath25 u(1 ) symmetry breaking . the symmetry breaking u(3 ) @xmath25 so(3 ) under study here has not been treated in the literature so far . however , the simplicity of our above results makes the uniformisation particularly easy in the present case : since the modulation factor already has its own asymptotic form or , inversely speaking : since the asymptotic expansion of the integral in in the limit of large @xmath46 happens to be exact also for @xmath84 it will be sufficient to replace in the perturbed actions @xmath82 , @xmath83 and their semiclassical amplitudes by those valid for all values of @xmath1 , which will be derived in the next section . in this section we will calculate the full actions and semiclassical amplitudes of the diameter and circle orbits of the hamiltonian , valid for arbitrary values of @xmath1 . the general trace formula for an arbitrary spherical potential in three dimensions has been given by creagh and littlejohn @xcite . we choose a different approach here , which is in spirit that of berry and tabor @xcite , but goes beyond their leading - order approximation in taking into account the end - point corrections to a trace integral which yield precisely the circle and diameter orbit contributions . the classical hamiltonian of the system ( with mass @xmath19 ) h ( , ) = e = 12p^2 + v(r ) , v(r ) = 12 ^ 2 r^2 + 14 r^4 , [ hrad ] is integrable due to the spherical symmetry of the potential , so that we can apply the standard ebk ( or radial wkb ) approximation to it . writing @xmath85 in terms of polar coordinates @xmath86 and the associated canonical momenta @xmath87 , we have the usual form involving an effective potential @xmath88 that includes a centrifugal term : e = h(r , p_r , l ) = 12 p_r^2 + v_eff(r ) , v_eff(r ) = v(r ) + , [ heff ] where @xmath89 is the radial momentum . the three independent ( and poisson - commuting ) constants of the motion are the energy @xmath18 , the total angular momentum @xmath90 , and its @xmath91 component @xmath92 . the momentum @xmath93 can be expressed in terms of the latter two as p_= . [ pthet ] before we specialise to our particular potential @xmath94 in , we briefly recall the radial ebk quantisation and derive from it a general trace formula for an arbitrary spherical potential , starting from the hamiltonian . in the standard radial ebk method @xcite , one quantises the three following action integrals s_r & = & p_rdr = 2(n_r+1/2 ) , n_r = 0,1,2, [irad ] + i _ & = & p_d= ( n_+1/2 ) , n_= 0,1,2, [ithet ] + i _ & = & l_z = m , m=0,1,2, [iphi ] where the maslov index 1/2 in is correct only for smooth potentials . since @xmath93 in does not depend on the potential , the integral for @xmath95 can be done once for all and yields i_= l-|l_z| 0 , [ ithet1 ] so that the quantisation condition for @xmath96 is given by l = ( n_+|m|+1/2 ) = ( + 1/2 ) , = n_+|m|=0,1,2 , [ lquant ] in terms of a single angular momentum quantum number @xmath97 . the relation includes the so - called langer correction ; the quantised squared angular momentum @xmath98 agrees with the exact quantum - mechanical value @xmath99 in the limit of large @xmath97 . solving for @xmath89 , we can write the radial action as s_r(e , l ) = dr = 2(n_r+1/2 ) , [ irad1 ] showing that the quantised energies will only depend on the radial and angular momentum quantum numbers @xmath100 and @xmath97 ; they have , of course , the usual @xmath101-degeneracy @xmath102 , since @xmath103 . inverting the relation , we may rewrite the hamiltonian in the form e = ( s_r , l ) . [ ham ] inserting the right - hand side of and into , we obtain the ebk - quantised eigenenergies : e_n_r^ebk = ( 2(n_r+1/2),(+1/2 ) ) , , n_r = 0,1,2, [eebk ] they can in general only be obtained by numerical iteration after doing the radial action integral over @xmath104 within the classical turning points . before continuing , we simplify the situation by a scaling of the energy . in principle we have to vary the three parameters @xmath18 , @xmath105 , and @xmath1 to study the dynamics of our present system . however , we can introduce a scaling of the energy in such a way that the classical dynamics only depends on one single parameter , a dimensionless scaled energy @xmath106 . if we multiply by the factor @xmath107 , we can write the r.h.s . in terms of scaled coordinates @xmath108 and momenta @xmath109 and a scaled time @xmath110 : q_i = x_i,_i = p_i , = t , [ scal ] so that the scaled energy @xmath106 becomes e = e = 12**^2+v(q ) = 12q^2+v(q)+ , v(q)=12q^2 + 14q^4 . [ escal ] where the dimensionless scaled angular momentum @xmath111 is given by l = l / s , s = . [ sscal ] in @xmath112 is the scaled radial variable and @xmath113 the scaled radial momentum , and the dot means the derivative with respect to the scaled time @xmath110 . @xmath114 is the action unit . this brings about a considerable simplification of the classical dynamics : we only need to vary one parameter @xmath106 ; at the end of our calculations we just have to remember that energies are measured in units of @xmath115 , angular momentum and actions in units of @xmath116 , times in units of @xmath117 , etc . in the following we shall give all quantities as functions of the dimensionless scaled variables @xmath106 and @xmath111 . ( other scaled dimensionless quantities such as actions , periods etc . will be denoted by lower - case letters . ) the scaling in is specific for our present hamiltonian . it can , however , easily be modified for ho potentials perturbed by arbitrary central potentials which are pure power laws in @xmath112 . the results of this subsection can therefore easily be generalized to the corresponding suitably scaled potentials @xmath118 . we now take the spectrum as a starting point to write down the density of states in the ebk approximation : g_ebk(e ) = _ n_r=0^_=0^ ( 2 + 1)(e - e_n_r^ebk ) . [ gebk ] next we apply poisson summation @xcite to convert the sums over @xmath100 and @xmath97 into integrals : g_ebk(e ) = _ n=-^_m=-^ _ 0^d ( 2 + 1 ) _ 0^dn_r ( e - e_n_r^ebk ) e^i2(nn_r+m ) + [ gebk1 ] due to the finite lower limits of the summations in , there are boundary corrections to , which we have indicated by the dots . we shall comment on them in a moment . using and , we substitute the variables @xmath97 and @xmath100 by the classical actions @xmath96 and @xmath119 , using @xmath120 and @xmath121 . then @xmath122 becomes , expressed in terms of the dimensionless scaled variables , g_ebk(e ) = _ n=-^ _ m=-^_0^l_m(e ) dl2l _ 0^s_r(e , l ) ds_r(e-(s_r , l ) ) e^is[ns_r+2ml]/-i(n+m ) + [ gebk2 ] hereby @xmath123 is the scaled radial action integral and @xmath124 the scaled hamiltonian . note that the integration limits have been imposed by the energy conservation ; @xmath125 is the maximum scaled angular momentum at fixed energy . the lower limits are , strictly speaking , @xmath126 for the integral over @xmath111 and @xmath127 for the integral over @xmath128 . replacing them by zero corresponds to neglecting corrections of higher order in @xmath77 . in the following , we keep terms up to order @xmath77 with respect to the leading factor @xmath129 in . there are exactly two corrections of relative order @xmath77 in what is indicated by the dots above : one is a boundary correction from @xmath130 to , and the second is coming from the lower limit @xmath131 of the @xmath128 integral in . a short calculation shows that these two corrections cancel identically . all other terms neglected in correspond to corrections of relative order @xmath42 or higher . we now use the relations ( e-(s_r , l ) ) = |s_r(e , l)|(s_r - s_r(e , l ) ) and s_r(e , l ) = = t_r(e , l ) . here @xmath132 is the scaled period of the classical motion at fixed values of @xmath106 and @xmath111 , which is the energy derivative of the corresponding scaled action integral @xmath133 : s_r(e , l ) & = & _ q(e , l)dq = 2_q_1^q_2 dq , [ srad ] + t_r(e , l ) & = & = 2_q_1^q_2 dq . [ trad ] here @xmath134 and @xmath135 are the scaled lower and upper turning points of the radial motion , respectively , which both depend on @xmath106 and @xmath111 . the above integrals can in many cases be expressed in terms of complete elliptic integrals . ( exceptions are the ho and coulomb potentials where they become simple algebraic functions of @xmath106 and @xmath111 . ) for our present potential @xmath118 in , the analytical expressions for , and other quantities of interest are given in the appendix [ app3 ] . one important result derived there is the taylor expansion of @xmath133 around @xmath136 , whose first terms are s_r(e , l ) = s_r(e,0 ) - l + a(e)l^2 + o(l^3 ) , [ srltay ] where @xmath137 is given in . the structure of but not the explicit form of @xmath137 appears to be a general result valid for all regular central potentials @xmath118 with a minimum at @xmath138 ( and hence not for the coulomb potential , see appendix [ appcoul ] ) , but we were not able to prove it in the general case . the relevance of the linear term @xmath139 in will become clear in the next subsection . using the above relations we can now do the integral over @xmath128 in exactly , due to the delta function , and obtain the following `` ebk trace integral '' : g_ebk(e ) = _ n=-^ _ m=-^(-1)^n+m _ 0^l_m(e ) dllt_r(e , l ) e^i[ns_r(e , l)+2msl]/ + o(^-1 ) . [ gebkint ] the phase of the integrand of can be interpreted as the full action @xmath140 , divided by @xmath77 , of a given classical orbit labelled by @xmath141 and @xmath24 , consisting of the radial part @xmath142 and the angular part @xmath143 . when doing the full double summation over all @xmath141 and @xmath24 and the integration over @xmath111 exactly , yields the ebk spectrum to leading order in @xmath77 . note that the limits of the @xmath111 integral are the cases of zero angular momentum , which corresponds to the diameter orbits , and its maximum value @xmath125 at a given energy , which corresponds to the circle orbits . the latter have the radius @xmath144 at which the effective scaled potential @xmath145 has its minimum ; for this motion the radial action is zero : @xmath146 , since all energy is in the angular motion . all contributions with @xmath147 to the integral correspond to motion which has both radial and angular components . the formula , valid for arbitrary ( but correctly scaled ) spherical potentials @xmath118 with smooth walls , is in principle a trace formula , but it does not yet have its characteristic form . that form is obtained by evaluating the integral over @xmath111 in the semiclassical limit @xmath148 by evaluating its leading contributions from the critical points of the phase function @xmath149 in the exponent of the integrand . to leading order in @xmath77 , these are stationary points as far as they exist . using the standard stationary - phase evaluation of the integral around the stationary points , one obtains contributions to @xmath122 with amplitudes proportional to @xmath150 , as expected for the fully three - fold degenerate orbits in a spherical system ( cf . the discussion at the end of the previous section ) . next to leading order in @xmath77 , one obtains contributions from the end points of the integral ( see , e.g. , @xcite ) , sometimes referred to as `` edge corrections '' . as already announced above and shown explicitly below , these correspond here to the diameter and circular orbits , giving contributions with amplitudes proportional to @xmath151 . a special contribution to @xmath122 comes from @xmath152 . as generally proved by berry and mount @xcite , this must be the thomas - fermi ( tf ) value of the density of states , which is the leading contribution to its smooth part . from we obtain with @xmath152 g_ebk^(0)(e ) = _ 0^l_m(e)dl^2 t_r(e , l ) . [ tfint ] the integral of over @xmath153 is straightforward . note that the contribution from the upper limit @xmath125 in is zero , since the turning points coincide : @xmath154 . the result is g_ebk^(0)(e ) = _ 0^q_2 ^ 0q^2dq = g_tf(e ) . [ gtf ] here @xmath155 is the upper turning point for @xmath136 . ( the lower one is zero : @xmath156 . ) that really is equal to the tf density of states follows from its general definition g_tf(e ) = d^3p d^3r ( e- ( , ) ) = d^3 d^3q ( e- ( , * ) ) . using polar coordinates for @xmath157 and @xmath158 and doing the @xmath158 integration leads to . the analytical expression valid for the potential is given in . in the introduction we have stated that the average part of the density of states is generally given by the etf approximation which contains @xmath77 corrections to the tf limit . for the spherical three - dimensional ho , these corrections are of relative order @xmath42 , as seen in . for the present perturbed ho potential we will see that the tf approximation is sufficient to reproduce the average part of the quantum - mechanical density of states , at least up to the energies for which the quantum spectrum is numerically available . * the stationary condition for the phase function in at a point @xmath159 reads [ ns_r(e , l)+2ml]|_l_nm = n|_l_nm+2m= 0 . [ stat ] due to energy conservation we may write de = d(s_r , l ) = |_l ds_r + |_s_r dl= 0 , so that = -2 , [ dsrdl ] where the frequencies of angular and radial motion are defined as usual by _ ( e , l ) = , _ r(e , l ) = 2 . with this , the stationary condition becomes = = , [ tori ] which is the periodicity condition for the `` rational tori '' as the most degenerate classical orbits @xcite . clearly , @xmath24 and @xmath141 must have the same sign , since frequencies and periods are positive quantities . because of , the lower integration limit @xmath136 in is a stationary point corresponding to the diameter orbit with @xmath160 . however , the stationary - phase integration of for @xmath161 yields a zero contribution due to the factor @xmath111 in the integrand , so that the only semiclassical contribution due to the diameter orbits is the end - point correction for @xmath136 discussed in the next section . as we shall see further below , solutions of with @xmath162 do not exist for all energies @xmath106 and for all pairs @xmath163 of integers . therefore we postpone the contributions of the rational tori to and concentrate first on the diameter and circle orbits . before developing the corresponding trace formula for our present system , valid to all orders in @xmath1 , we want to establish here the connection to the first - order perturbative approach used in sect . [ secpert ] . from the taylor expansions given in appendix [ app3 ] , we obtain for the radial action integral the following first terms : 2s_r(e , l ) = 2(-l ) -(3e^2-^2l^2 ) + o(^2 ) . [ srad1 ] in the second term we recognise precisely the first - order action shift given in . inserting into it @xmath164 and the leading - order expression for @xmath165 given in , we obtain the first - order action shifts of the diameter and circle orbits , respectively , given in . we now insert into the expression for the density of states , neglecting the terms of higher order in @xmath1 and keeping only the zero - order terms in the amplitude of @xmath166 , given in , and of the upper integration limit @xmath165 . noting furthermore that for the unperturbed ho orbits the ratio @xmath167 becomes equal to 2 , so that the resonance condition implies @xmath168 , we assume for the moment that all other combinations of @xmath24 and @xmath141 in this limit may be neglected . writing @xmath169 , we obtain the following result for the density of states up to first order in @xmath1 ( note that the linear terms in @xmath96 cancel in the exponent of the integrand ) : g_ebk^(1)(e ) = _ k=-^(-1)^k e^i2ke/ _ 0^e/ dl^2 e^-ik(3-^2l^2/e^2)]/ , [ gebkp ] where @xmath46 is the quantity defined in . using the substitution @xmath170 , the above integral becomes identically equal to the integral for the perturbative modulation factor , and the result is precisely the perturbed density of states defined in with the oscillating part given in . rather than proving at this stage that all contributions to with @xmath171 either cancel or are negligible to the leading order in @xmath77 , we now go on to derive the full trace formula for the diameter and circle orbits , doing the summations over all @xmath24 and @xmath141 exactly in the semiclassical limit . equipped with the above results , we are now in a position to evaluate the full trace formula for the contribution of diameter and circle orbits for our present hamiltonian . the smooth tf part of the level density , @xmath172 , is given explicitly in the appendix [ app3 ] . the contributions of the rational tori , which bifurcate from the circle orbits only for high enough energy and repetition numbers , will be discussed in section [ sectori ] . we therefore now evaluate the end - point contributions to the integral in in the semiclassical limit @xmath148 . the lower end point @xmath136 in yields the asymptotic contribution g_d(e ) = t_r(e,0 ) _ n=-^ ( -1)^n e^i[ns_r(e,0)/+/2]u_n(e ) , [ gd1 ] where we have left out the contribution from @xmath173 which contributes to the tf part , and defined the quantity u_n(e ) = _ l0 . using the identity @xcite = _ m=-^ , ( zn , n ) [ polesum ] we find u_n(e ) = _ l0 . [ une ] we now exploit the result from which we find = -+ 2a(e ) l + o(l^2 ) . for odd values of @xmath24 , the sin function in gives always a nonzero denominator in the limit @xmath136 , so that @xmath174 becomes zero . for even @xmath175 we get in the limit @xmath176 ( ) = ( -k+2kla(e ) ) = ( -1)^k(2kla(e ) ) ( -1)^k 2kla(e ) , so that we obtain u_2k(e ) = . inserting this into using , leaving out the contribution @xmath38 which is contained in @xmath177 , we obtain the final contribution of the diameter orbits to the density of states : g_d(e ) = -_k=1^ ( ks_d(e)/ ) , [ dgd2 ] where the actions @xmath178 are given explicitly in . in the low - energy limit we obtain with exactly the diameter contribution to the perturbative trace formula . note that the presence of the linear term @xmath179 in the quantity is instrumental in annihilating the contributions with odd @xmath24 to the sum in and hence in establishing the fact that only an even number of radial oscillations yields a physical periodic orbit with angular momentum @xmath164 . the upper end point @xmath180 in yields the asymptotic contribution g_c(e ) = l_m(e)t_r(e , l_m ) _ m=-^(-1)^m e^i[2l_m(e)/-/2 ] _ n=-^. [ gc1 ] using , and noting that @xmath181 , we get |_l_m = - . employing again to perform the @xmath24 summation , writing @xmath182 and omitting the @xmath38 term , we obtain after some manipulations the contribution of the circle orbits to the density of states g_c(e ) = _ k=1^ ( ks_c(e)/ ) , [ dgc2 ] where @xmath183 is given explicitly in , and the quantity @xmath184 is defined by w(e ) = = . [ wofe ] in the limit @xmath185 this quantity is found with the r.h.s . of the analytical results and to go like w(e ) = 2(1+e/4 + ) = 2+e + expanding the denominator in , using the above @xmath184 , up to first order in @xmath1 brings the amplitude exactly into that of the circle orbit contribution to the perturbative result . note that the denominator under the sum in looks exactly like that in the gutzwiller trace formula @xcite for an isolated stable orbit , whereby @xmath186 corresponds to one - half of the stability angle . indeed , the semiclassical amplitude in diverges when @xmath187 becomes an integer @xmath188 . at the corresponding nonzero energies , the circle orbit bifurcates ; the condition for this to happen is exactly the resonance condition for the rational tori with @xmath189 , taken at @xmath190 . ( we come back to this point in section [ sectori ] . ) using and , we find that @xmath184 is restricted to the following range : 2 < w(e ) = < 0 . [ wlim ] the smallest @xmath36 for which we can have @xmath191 with @xmath192 is @xmath193 with @xmath194 , so that @xmath195 , which happens at the scaled energy @xmath196 ( see below ) . this means that the shortest new orbit is the 7:3 torus , bifurcating from the 3rd repetition of the circle orbit . the next bifurcation ( from @xmath197 ) is that of the 9:4 torus at @xmath198 . still longer orbits bifurcate at lower energies , but from higher repetitions of the circle orbit . therefore , we can ignore the bifurcations at low energies and small repetition numbers @xmath36 and concentrate on the contributions of the diameter and circle orbits to the density of states . let us finally write down the trace formula which we have obtained for the oscillating part of the density of states : g(e ) _ k=1^ , [ dguni ] where the semiclassical amplitudes are given by _ k^c(e ) = , _ k^d(e ) = . [ ampcd ] in the following we shall test this formula against the quantum - mechanical density of states . as we have shown above , goes over into the perturbative trace formula in the limit @xmath26 . this confirms the assumption , made at the end of sect . [ secpois ] , that the contributions with @xmath171 to do not contribute in this limit . in this section we compare our semiclassical results with those obtained from the exact quantum - mechanical spectrum ( obtained numerically by solving the radial schrdinger equation on a discrete mesh ) . in order to focus on the _ coarse - grained _ shell structure , we convolute both results over the energy @xmath18 with a normalised gaussian of width @xmath199 . the quantum - mechanical density of states then becomes g_qm(e ) = _ n e^-(e - e_n)^2/^2 . [ gqmgam ] in order to obtain its oscillating part , we subtract from it the tf expression @xmath172 which we can calculate analytically ( see appendix [ app3 ] ) . the semiclassical trace formula becomes , after doing the convolution in stationary - phase approximation ( cf . @xcite , sec . 5.5 ) g_sc(e ) _ po a_po(e ) e^-[t_po(e)/2]^2 [ s_po(e)/-_po/2 ] , [ dgscgam ] so that orbits with longer periods @xmath200 will be exponentially suppressed . in we show a comparison of the results obtained with the uniform trace formula for the case @xmath201 with those obtained from the exact numerical quantum spectrum . the width of the gaussian smoothing was chosen to be @xmath202 ; in the semiclassical result the harmonics @xmath203 then do not contribute noticeably . the agreement is seen to be perfect ; tiny differences can only be noted near the beat minima on the amplified scale below . in we show a similar comparison for @xmath204 , exhibiting much fine structures . we now start to resolve the spectrum with a higher resolution ; note that we still only plot the oscillating part @xmath7 of the density of states . in the top panel , the semiclassical result includes harmonics ( i.e. , repetitions of the two orbits ) with @xmath205 . this is evidently sufficient to reproduce the quantum - mechanical oscillations up to @xmath206 with a high accuracy . in the centre panel , we have added two more harmonics , going up to @xmath207 . here we recognise the appearance of divergences at the scaled energies @xmath208 and @xmath209 which correspond to the bifurcations of the 19:9 and 17:8 tori from the repetitions of the circle orbit with @xmath210 and @xmath211 , respectively ( cf . below ) . the small wiggles close to the divergences , with decreasing amplitudes when going away from them , are due to the missing contributions from those torus orbits in the trace formula ( see the following section ) . = 13.5 cm in order to anticipate the results of a suitable uniform trace formula including the contributions of the tori , we show in the bottom panel of the result of the ebk trace integral obtained by a numerical integration over the variable @xmath111 . hereby the summations over @xmath24 and @xmath141 have been done in the limits @xmath212 , @xmath213 using @xmath214 . [ the period to be used in the gaussian damping factor of is in this case given by @xmath215 . ] now the divergences and all other discrepancies have disappeared ; the difference to the quantum - mechanical result can hardly be recognised . the integral therefore is also a uniform expression for the semiclassical density of states , containing the contributions of all periodic orbits . the numerical evaluation becomes , however , very time consuming for even finer energy resolutions ( i.e. , still smaller values of @xmath199 ) . our main result here is that up to a rather fine resolution , the shell structure of the present perturbed ho system is dominated by the two - fold degenerate families of diameter and circle orbits , although these are not the most degenerate orbits . this is a rather unusual situation , due to the fact that the shortest tori with the highest degeneracy are considerably longer ( by a factor @xmath21610 ) than the primitive diameter and circle orbits . let us finally test the range of validity of the perturbative trace formula . in we compare its results to that of the full trace formula for the two values @xmath217 ( upper panel ) and @xmath201 ( lower panel ) . as expected , the two curves agree in the limit @xmath185 . however , already for @xmath217 the amplitude modulation of the perturbative result deviates so much that the position of the first beat node is predicted by about 10% too low in energy . for @xmath201 , the perturbative result becomes so bad that it predicts the second beat node approximately at the position of the correct first one . these results underline the fact that in @xcite , the quantum results for weak interactions could be qualitatively reproduced by the perturbative trace formula , but not quantitatively with correct positions of the beat nodes . a more detailed analysis of the selfconsistent hf results of @xcite using the uniform trace formula is in progress @xcite . as we have seen in the previous section , periodic orbit with both angular and radial motion appear as the rational tori fulfilling the periodicity condition . for the hamiltonian , this condition can only be fulfilled in the limits given in . this means that stationary points @xmath218 of the @xmath96 integral in satisfying and hence only exist if @xmath219 . hereby @xmath220 corresponds to the repetition number @xmath36 of the circle orbit that undergoes a bifurcation exactly when @xmath218 is equal to the maximum angular momentum @xmath221 . in we list some of the lowest energies @xmath222 at which this happens for the smallest integers @xmath189 . each of the @xmath223 tori only exists above its bifurcation energy , i.e. , at energies @xmath224 . [ cols="^,^,<,<,<",options="header " , ] the contributions of the bifurcated tori to the semiclassical trace formula is obtained by a stationary - phase evaluation of the @xmath111 integral in , leading to g_tori(e ) = _ n>2m>0 a^t_nm(e)(-1)^n+m ( + ) . [ dgt ] the summation goes only over those pairs @xmath163 of positive integers for which the stationary condition can be fulfilled . ( note that the contribution of the corresponding pairs of negative integers is included in the amplitudes by a factor 2 . ) the actions of the tori are given by s^t_nm(e ) = ns_r(e , l_nm)+m2l_nm(e ) , [ stori ] and their amplitudes by ^t_nm(e ) = t_r(e , l_nm)l_nm(e ) ( n| |_l = l_nm(e))^-1/2(e - e_n : m ) , [ amptori ] where @xmath114 is the action unit given in . the heaviside step function in is defined as ( x ) = 0 x < 0,(x ) = 1/2 x = 0,(x ) = 1 x > 0,where the value 1/2 for @xmath225 accounts for the fact that one obtains only half a fresnel integral when the stationary point is the upper end point of the integral in . the power @xmath150 in the amplitude @xmath226 is characteristic of the three - fold degenerate orbits in a three - dimensional spherical system @xcite . . corresponds to the standard berry - tabor trace formula @xcite for the most degenerate orbits in an integrable system . here , however , each contribution to is only valid if the energy @xmath106 is sufficiently larger than the corresponding bifurcation energy @xmath222 . in the neighbourhood of the bifurcation energies , we have to replace the sum of the diverging circle orbit contribution and the corresponding torus contribution with @xmath227 by a common uniform contribution . as it is well - known from the semiclassical theory of bifurcations @xcite , one has to include hereby also the contributions of so - called `` ghost orbits '' , which are the imaginary ( or complex ) continuations of the bifurcated orbits ( here : the tori ) to the other side of the bifurcation ( here : @xmath228 ) . it is , however , not the purpose of our present paper to discuss in more detail the uniform approximation suitable for this situation . the present scenario is actually almost identical to that described in @xcite for the bifurcations of tori in the integrable hnon - heiles system . although there the orbit from which the tori bifurcate is an isolated one , while it here is a family of circle orbits , the uniform trace formula given in @xcite can be applied to the present system in a straightforward manner . we refer the interested reader to this paper for all the detailed formulae as well as for a numerical demonstration showing how diverging discrepancies of the type seen in the centre part of disappear when a proper uniform trace formula is used . its result would here be practically indistinguishable from the result shown in the bottom part of , in which we have evaluated the integral in numerically instead of using the stationary phase approximation with end - point corrections . in this section we discuss briefly a purely quartic oscillator potential , i.e. , we start from the hamiltonian h ( , ) = 12p^2 + r^4 = e. [ hr4 ] here both the energy and @xmath1 can be scaled away . after dividing the above equation by @xmath18 and introducing the scaled variables q_i = ( ) ^1/4r_i , = ( e)^1/4t , l = ^1/4e^-3/4l , we obtain the `` unit energy '' equation 1 = 12q^2 + 14q^4 + , where the dot again means derivative with respect to the scaled time @xmath110 . hence the classical equations of motion become independent of energy and @xmath1 . after obtaining all the classical results , we can reintroduce @xmath18 and @xmath1 just by remembering that lengths scale with @xmath229 , momenta with @xmath230 , times ( periods ) with @xmath231 , actions and angular momenta with @xmath232 . we obtain the semiclassical trace formula from the results of the previous sections and of the appendix [ app3 ] simply by taking the limit @xmath30 , which means practically by extracting the leading terms for @xmath233 . in this way , we obtain the following primitive periods and actions for the circle orbit : t_c(e ) = 2 ^ -1/4(4e/3)^-1/4,s_c(e ) = 2l_m(e ) = 2 ^ -1/4(4e/3)^3/4 , [ circr4 ] and for the diameter orbit : t_d(e ) = 2 ^ -1/4k_0e^-1/4,s_d(e ) = 2 ^ -1/4(4/3)k_0e^3/4 . [ diagr4 ] here @xmath234 is the constant elliptic integral k_0=k(_0)=1.8540746773 ,_0 = 1/. furthermore we get t_r(e , l_m ) = ( 3e)^-1/4 . the resonance condition becomes independent of the energy = w(l_nm ) = , [ wofl ] since both periods scale with the same power of the energy . for the frequency ratio we find the limits 2 w(l_nm ) = n : m < 0 l_nm < l_m = ( 4/3)^3/4 = 1.240806 [tori4 ] note that the upper limit @xmath235 never becomes a stationary point , so that no bifurcations of the circle orbits occur in this system . the values of the stationary points @xmath236 ( in scaled dimensionless units ) for the lowest values of @xmath24 and @xmath141 are given in the rightmost column of ; as we just have seen , they do not depend on the energy @xmath18 . the result @xmath237 is a special case of the general result , given in @xcite , that for a potential @xmath94 = @xmath238 one has @xmath239 . for the amplitudes of the circle and diameter orbits in the trace formula , which keeps the same form , we obtain here _ k^c(e ) = , _ k^d(e ) = . [ ampr4 ] the actions @xmath83 and @xmath82 to be used in are those given in and . finally , the tf level density is found to be g_tf(e ) = . the tori , which in the perturbed ho system discussed in the last section occur only at sufficiently high energies after their bifurcations from the repeated circle orbits , exist in the present system at all energies . they affect again only the finer quantum structures of the density of states since they only exist for @xmath240 due to the selection by their resonance condition ( cf . also ) . this is demonstrated in the following two figures . in , we compare the results of the semiclassical trace formula for the diameters and circles using the above quantities to the exact quantum - mechanical density of states , both gaussian - averaged over a width @xmath241 ( in energy units ) . the agreement is again perfect even on the enlarged scale in the lower part of the figure . the reason is that for the chosen energy resolution with @xmath241 , only repetition numbers @xmath242 contribute , for which there exist no tori ( see ) . = 9.2 cm = 4.5 cm shows the same comparison for @xmath243 . here some small discrepancies appear which become more visible with increasing energy . they are due to the missing contributions from the tori with @xmath244 . since there are no bifurcations involved in the present system , the contributions of the tori are given by the berry - tabor trace formula , employing the appropriate expressions for the actions and periods used therein . the analytical expression of the action integral @xmath245 in terms of elliptic integrals is the same as for the perturbed ho potential , as given in eqs . - , except that here the term @xmath246 in the potential and the corresponding contribution @xmath247 must be omitted . even at the rather fine resolution obtained in , our main result is the observation that the circle and diameter orbits dominate the shell structure also for the purely quartic oscillator . motivated by recent hartree - fock ( hf ) calculations of harmonically trapped fermionic gases with a repulsive interaction @xcite , in which pronounced super - shell effects were obtained , we have developed a semiclassical trace formula for the density of states of the anharmonically perturbed three - dimensional isotropic harmonic oscillator ( ho ) defined in . we find that its gross - shell structure is dominated by the periodic orbits of diameter and circle shape , whose interference leads to super - shell beats which explain the findings from the numerical hf results . in a first step , we have used the perturbative approach of creagh @xcite to describe the symmetry breaking u(3 ) @xmath25 so(3 ) for weak anharmonicities @xmath1 , resulting in the perturbative trace formula . it uniformly restores the u(3 ) limit , yielding the exact trace formula of the unperturbed ho system in the limit @xmath26 . for the derivation of the perturbative result , one must in principle integrate over the 4-fold degenerate families of unperturbed ho orbits which cover the manifold @xmath2p@xmath3 . as shown in app . [ app2 ] , however , it turns out to be easier to integrate over the energy shell in phase space , which is a five - sphere s@xmath31 , whereby the integration can be reduced , under exploitation of the so(3 ) symmetry , to the one - dimensional integral given analytically in . an interesting result hereby is the fact that its two contributions , corresponding to the diameter and circle orbits , already have the characteristic form occurring in the general trace formula . usually , this form is obtained from a perturbative trace formula only asymptotically in the semiclassical limit @xmath148 ; this happens , e.g. , also when one investigates the corresponding hamiltonian in two dimensions or other perturbed 2-dimensional systems @xcite . next we have used the standard ebk quantisation and the poisson summation formula to derive a general semiclassical trace formula for arbitrary central potentials in terms of a one - dimensional integral over the angular momentum @xmath96 of the classical orbits , given in . exact integration and summation over all @xmath248 yields , to leading order in @xmath77 , the ebk spectrum . its asymptotic evaluation in the limit @xmath148 yields the standard - type contributions to the trace formula . the end - point contributions yield the diameter orbits with @xmath164 and the circle orbits with maximum angular momentum ; both contributions are of next - to - leading order in @xmath77 , whereas the leading - order contributions corresponding to the typical rational tori of integrable systems according to the berry - tabor theory @xcite come from the stationary points inside the integration interval . one interesting mathematical aspect is the mechanism by which the ratio @xmath249 of radial to angular frequency of the diameter orbit is naturally selected by a property of the general radial action integral @xmath250 as a function of angular momentum @xmath96 which appears to be universal for regular spherical potentials . it holds also for a particle in a spherical box with specular reflection , for which we have re - derived in appendix [ appbb ] the trace formula given by balian and bloch @xcite from our general formula , taking into account the changes of the maslov indices due to hard - wall reflections ) . the coulomb potential , for which @xmath251 , is discussed in appendix [ appcoul ] where we also re - derive the exact trace formula for the rydberg spectrum given in @xcite . for the hamiltonian , we have derived a uniform trace formula for the diameter and circle orbit contributions that is valid for arbitrary @xmath1 . we find that for low energies and low repetition numbers no tori exist , so that the gross - shell structure of this system is dominated by the circle and diameter orbits and their lowest repetitions . the tori bifurcate , at sufficiently high energy , out of the repetitions with @xmath78 of the circle orbits , as was also observed recently for homogeneous power - law potentials in @xcite . their contributions sufficiently high above the respective bifurcation energies are given by the usual berry - tabor type trace formula . we have not discussed here the common uniform treatment of the circle orbits and the tori bifurcating from them , since this scenario is identical to the one discussed in detail in @xcite . the same qualitative results are found also in the limit @xmath252 in which the potential becomes a purely quartic oscillator . although the tori with @xmath80 here exist at all energies , they are so much longer ( by a factor @xmath253 ) than the shortest diameter and circle orbits that they only affect the finer details of the quantum spectrum . we thus have found the interesting and quite atypical situation of a three - dimensional system in which the periodic orbits with highest , i.e. , three - fold degeneracy are only responsible for finer details of the quantum spectrum , whereas its gross - shell structure is to an astonishing degree dominated by the orbits of next - to - leading order in @xmath77 , i.e. , the diameter and circle orbits occurring in two - fold degenerate families . this is totally different from the situation , observed first by balian and bloch @xcite , of a spherical cavity potential ( with ideally reflecting walls ) which has turned out to be a realistic model for large spherical metal clusters . ( we re - derive the semiclassical trace formula of balian and bloch for the spherical cavity in appendix [ appbb ] . ) there the famous super - shells @xcite come from the interference of the shortest three - fold degenerate tori ( the triangle and square orbits ) , whereas the two - fold degenerate diameter orbit family has virtually no effect on the observable shell structure of these systems . our results explain qualitatively the numerical quantum - mechanical hf results @xcite for harmonically trapped fermionic atom gases . a more quantitative comparison , in which the strength @xmath1 of the perturbation in is determined directly from the numerically obtained self - consistent hf potentials , is in progress @xcite . lazzari _ et al . _ @xcite have studied the periodic orbits in a spherical woods - saxon potential in connection with the physics of metal clusters and thereby focused on the contributions of the diameter orbits . we find agreement of their results for the tori and the diameters with our results , and , respectively , if we interpret their function @xmath254 as being _ the radial period @xmath255 at the corresponding stationary value @xmath218 of the angular momentum (= 0 for the diameters ) . they have , however , neglected the contributions from the circle orbits ; furthermore they did not compare their semiclassical results with quantum - mechanical results . we suspect that the neglect of the circle orbits might be justifiable due to the chosen steepness of their woods - saxon potential , for which the triangle - like tori are perhaps more important than the circles . ozorio de almeida _ et al . _ @xcite have discussed the summation of all periodic orbits in integrable systems in connection with spectral determinants . they conclude that edge corrections corresponding to lower - degenerate orbits may be neglected because they are implicitly approximated by the nearest - lying rational tori ( characterised by very large numbers @xmath248 in our notation ) . while their argument applies to the _ full quantisation _ of the exact ( or ebk - approximated ) quantum spectrum obtained by summing over _ all orbits _ , it can not be used when the spectrum is coarse - grained with a finite energy window @xmath199 , such as we have used the periodic - orbit sum in our present investigation ( cf . also the discussion in the appendix of ref . @xcite ) which is more practically oriented . we acknowledge enlightening discussions with s. bechtluft - sachs , s. creagh , and n. sndergaard , and helpful comments by an anonymous referee . we are particularly grateful to k. jnich for providing us with the elegant reduction of the s@xmath31 integration given in appendix [ app2 ] . we thank a. magner for valuable comments on our manuscript and for bringing the refs . @xcite to our attention . this work was financially supported by the swedish foundation for strategic research and the swedish research council . m. . acknowledges financial support from the deutsche forschungsgesellschaft ( graduiertenkolleg 638 : _ nichtlinearitt und nichtgleichgewicht in kondensierter materie _ ) , and m.b . acknowledges the warm hospitality at the lth during several research visits . we reproduce here the parametrisation of @xmath2p@xmath3 given in @xcite . we start with three complex numbers @xmath256 : ( z_1,z_2,z_3 ) = ( n_1,n_2e^i_2,n_3e^i_3 ) , with @xmath257 , @xmath258 , and @xmath259 , @xmath260 , whereby the @xmath261 are restricted to n_1 ^ 2+n_2 ^ 2+n_3 ^ 2=1 . [ ensh ] the @xmath262 define the complex projective space @xmath2p@xmath3 . the @xmath261 are in one - to - one correspondence with the points on the first octant of the 2-sphere @xmath60 , and the @xmath263 form a 2-torus . at the edges of the octant , the torus contracts to a circle or to a point , and one or both of the @xmath263 are not defined . physically , we interpret the real and negative imaginary parts of the @xmath262 as coordinates and momenta , respectively , of a starting point @xmath264 r_(0 ) = e z_,p_(0 ) = - m z _ , in the 6-dimensional phase space @xmath265 of the 3-dimensional isotropic harmonic oscillator . the hamiltonian , which has u(3 ) symmetry @xcite , is given by h ( , ) = 12 _ = 1 ^ 3 = e. the @xmath266-periodic solutions of the equations of motion are then given by r_(t ) = e ( z_e^it ) , p_(t ) = -m ( z_e^it ) . the equation defines the energy shell , so that the six quantities @xmath267 are points on a 5-sphere , s@xmath31 ( see also appendix [ app2 ] below ) . by taking the total energy to be @xmath268 and setting @xmath269 , we fix the radius of the 5-sphere to be unity . taking out the phase factor @xmath270 takes us from s@xmath31 to the projective space @xmath2p@xmath3 . using the fubini - study metric , the authors of @xcite find the squared distance @xmath271 between two points on @xmath2p@xmath3 to be ds^2 = dn_1 ^ 2+dn_2 ^ 2+dn_3 ^ 2 + n_2 ^ 2(1-n_2 ^ 2)d_2 ^ 2+n_3 ^ 2(1-n_3 ^ 2)d_3 ^ 2 - 2n_2 ^ 2n_3 ^ 2d_2d_3 . [ ds2 ] the first three terms correspond to the standard metric on @xmath60 , and the last three to the metric on a flat 2-torus whose shape depends on where we are on the octant . we now wish to re - parametrise the @xmath261 in terms of the two polar angles @xmath272 on @xmath60 by writing n_1 = , n_2 = , n_3 = , [ coord ] with @xmath273 $ ] . collecting our four angles in a vector @xmath274 by defining = ( _ 1,_2,_3,_4 ) = ( , , _ 2,_3 ) , we can write the squared distance @xmath271 as ds^2 = _ i , j=1 ^ 4 g_ijd_id_j , [ ds2a ] and find the elements @xmath275 of the metric tensor to be @xmath276 , @xmath277 , @xmath278 , @xmath279 , @xmath280 , and @xmath281 . from this , we get the integration measure on @xmath2p@xmath3 to be d_^2 = |det(g_ij)|^1/2 d_1d_2d_3d_4 = ^3d dd_2d_3 . the integrated volume of @xmath2p@xmath3 , consistent with a general formula for @xmath2p@xmath282 , is _ ^2 = d_^2 = _ ^n = . here we give a strict proof of eq . of the modulation factor @xmath283 for the anharmonically perturbed ho system . the most straightforward parametrisation of the six - dimensional phase space of the unperturbed ho is given in terms of the coordinate vector @xmath284 and the momentum vector @xmath285 @xmath286 . since a periodic solution is uniquely determined by their initial values at @xmath287 : @xmath288 and @xmath289 , and the total energy @xmath290 is a constant of motion , the six components of @xmath291 and @xmath292 must cover a 5-sphere with radius @xmath293 . we introduce dimensionless coordinate and momentum vectors @xmath294 and @xmath295 , respectively , by = _ 0 , = _ 0 , [ unvec ] so that in these variables the energy shell becomes the unit sphere s@xmath31 : ^2+^2 = 1 . [ ens5 ] from eq . , the first - order action perturbation @xmath296 is given in terms of energy @xmath18 and angular momentum @xmath96 by s_1 = -(3-^2l^2/e^2 ) , [ ds1 ] with @xmath46 defined in . the modulation factor for the first - order perturbed ho system becomes therefore ( k/ ) = d_s^5 e^iks_1/ , [ modfac5 ] where @xmath297 is the five - dimensional solid angle and @xmath298 is the integration measure on s@xmath31 with @xmath299 . one may now express @xmath296 in directly in terms of the five polar angles of six - dimensional hyperspherical coordinates , which becomes a very complicated function , and integrate over all five angles . we have done this numerically and verified that it yields exactly the same result as the numerical 4-dimensional @xmath2p@xmath3 integration of with the phase function . it is , however , not necessary to perform the full five - dimensional integral for . we follow a much more economical route @xcite , exploiting the rotational symmetry of the system and the so(3 ) invariance of the expression for @xmath296 . due to the restriction , we can write the s@xmath31 sphere as s^5 = \{(*e*_,*e * _ ) | ; _ , * e*_s^2 } , [ s5new ] where @xmath300 and @xmath301 are the unit vectors in the directions of @xmath294 and @xmath295 , respectively . the square of the conserved total angular momentum then is l^2 = * l*^2 = ( _ 0_0)^2 = ^2 ^ 2 ( * e*__)^2 , so that the action perturbation becomes , after the substitution @xmath302 , s_1 = - , . [ ds1a ] the integration measure for is d_s^5 = ^2d_s^2(*e * _ ) ^2d_s^2(*e*_)d = 18 ^ 2d d_s^2(*e * _ ) d_s^2(*e * _ ) , and the modulation factor becomes ( k/ ) = _ 0^d^2 d_s^2(*e * _ ) d_s^2(*e * _ ) e^-ik/. [ mf5a ] we now introduce the angle @xmath303 between the unit vectors @xmath300 and @xmath301 , so that ( * e*__)^2 = ^2 , and choose @xmath300 as the direction of the north pole ( @xmath304 ) of polar angles @xmath305 for the vector @xmath301 . the integrand then becomes independent of @xmath300 , so that the corresponding s@xmath3 integral just gives @xmath306 . for the other s@xmath3 integral we have @xmath307 since the integrand does not depend on @xmath308 . we thus obtain , after the standard substitution @xmath309 and an obvious reduction of the integration limits , ( k/ ) = _ 0^/2 d^2 _ 0 ^ 1 du e^-ik/. next , we make the substitution @xmath310 and then a further substitution @xmath311 to obtain ( k/ ) = _ 0^/2d _ 0^ dt e^-ik(3-^2+t^2)/ = _ 0 ^ 1 ds _ 0^ dt e^-ik(2+s^2+t^2)/. since the last integral goes exactly over the first quadrant of a unit disk in the @xmath312 plane , we can use polar coordinates @xmath313 and furthermore @xmath314 to obtain our final result ( k/ ) = _ 0^/2 d _ 0 ^ 1 drre^-ik(2+r^2)/ = _ 0 ^ 1 dze^-ik(3-z)/ , which is identical to that given in or in . in this appendix we derive analytical expressions for the particular scaled potential @xmath118 in . rewriting the radial action integral using the substitution @xmath315 , we obtain s_r(e , l ) = 2_q_1^q_2 dq = _ z_1^z_2 , [ srint ] where g(z ) = 4ez-2z^2-z^3 - 2l^2 = ( z_1-z)(z_2-z)(z_3-z ) . [ gz ] the classical turning points are defined by the roots of the cubic equation @xmath316 , which are always real : the discriminant can easily be shown to be negative except for @xmath180 where it becomes zero due to the double root @xmath317 . selecting the roots such that z_3 < z_1 z_2 , one finds that @xmath318 is always negative , whereas @xmath319 and @xmath320 are positive definite and represent the squares of the real classical turning points @xmath134 and @xmath135 . the integrals in can not be found in most tables . however , after an integration by parts we may write the r.h.s . as s_r(e , l ) = s_r^(1)(e , l)+ s_r^(2)(e , l)-l^2 s_r^(-1)(e , l ) , [ srltot ] where we have defined s_r^(n)(e , l ) = _ z_1^z_2 dz . in byrd and friedman @xcite , these integrals are found to be : s_r^(1)(e , l ) & = & [ ( ^2-^2)k()+^2e ( ) ] , + s_r^(2)(e , l ) & = & [ ( 3 ^ 4 - 6 ^ 2 ^ 2 + 2 ^ 4+^2 ^ 4 ) k()+2 ^ 2(3 ^ 2-^2-^2 ^ 2 ) e ( ) ] , + s_r^(-1)(e , l ) & = & ( ^2 , ) . [ sradn ] here @xmath321 , @xmath322 and @xmath323 are the complete elliptic integrals of first , second and third kind with modulus @xmath324 , and = , ^2 = 1- . [ kap ] note that @xmath325 , which corresponds to the circular case of the elliptic integral @xmath323 @xcite . this function becomes singular like @xmath326 , i.e. , for @xmath327 which happens when @xmath176 . since @xmath319 goes to zero like @xmath328 in this limit , @xmath323 has a first - order pole at @xmath136 . using the laurent expansion of @xmath323 given in eq . 906.04 of @xcite , we find s_r^(-1)(e , l ) = -2a(e ) + o(l ) = -2a(e ) + o(l ) , [ srlm1 ] with a(e ) = - = ( 1-e + ) . [ aofe ] here @xmath329 and @xmath330 are the @xmath136 values of the quantities given in eqs . and below . the last equality in follows because @xmath331 , as seen from . the functions @xmath332 with @xmath333 are both regular in @xmath136 . altogether we obtain the following taylor expansion for the total radial action integral : s_r(e , l ) = s_r(e,0 ) - l + a(e)l^2 + o(l^3 ) , [ srlexp ] the radial period is easily found to be t_r(e , l ) = s_r^(0)(e , l ) = k ( ) . [ trad1 ] in the limit @xmath136 , which is relevant for the diameter orbits and the tf density of states , the above results simplify considerably . we obtain s_r(e,0 ) & = & 23(1 + 4e)^1/4 , [ sr0 ] + t_r(e,0 ) & = & 2(1 + 4e)^-1/4k(_0 ) [ tr0 ] . note that in this limit we get z_1 ^ 0=0,z_2 ^ 0=(q_2 ^ 0)^2=-1,z_3 ^ 0=--1 , [ zi0 ] and the modulus of the elliptic integrals becomes _ 0 = = = ( 1-e + ) . [ kap0 ] the tf expression for the density of states becomes g_tf(e ) & = & ( 1 + 4e)^1/4 [ 2(1 + 3e)e(_0)-(1 + 3e+)k(_0 ) ] . [ gtf0 ] it is of interest to see the first terms in the taylor expansion of the above results around @xmath334 , whose leading terms are those of the pure harmonic oscillator : s_r(e , l ) & = & ( e - l)-18(3e^2-l^2)+ , [ srtay ] + t_r(e , l ) & = & -34e+(7e^2-l^2)+ , [ trtay ] + g_tf(e ) & = & e^2(1 - 5e/4 + ) .[gtftay ] the unscaled leading ho terms are s_r^ho(e , l ) = ( -l ) = , t_r^ho = , g_tf^ho(e ) = . the diameter orbit with @xmath164 makes two full radial oscillations during its primitive period . therefore its primitive action and period are given by s_d(e ) = 2s_r(e,0),t_d(e)=2t_r(e,0 ) . [ stdia ] for the circle orbit , the primitive action integral @xmath183 is just @xmath266 times the maximum value of the angular momentum , @xmath335 , which for the present potential is found to be l_m(e ) = s , s_c(e ) = 2l_m(e ) = ( 1-e/4 + ) . [ scirc ] the primitive period of the circle orbit is given by its energy derivative t_c(e ) = = ( 1-e/2 + ) . [ tcirc ] we also give here the value of the radial period @xmath336 at @xmath180 . the corresponding frequency @xmath337 is easily obtained from the second derivative of the effective potential at its minimum @xmath144 , which for the present potential is given by q_0 ^ 2=z_0=2(-1)/3 , using @xmath338 , and the result becomes t_r(e , l_m ) = = ( 1 + 3e)^-1/4 = ( 1 - 3e/4 + ) . [ trlm ] this agrees with the result obtained from , noting that for the maximum value of @xmath180 , @xmath339 and @xmath340 , so that @xmath324 becomes zero and @xmath341 . in order to illustrate the validity and the usage of our general ebk trace integral , we present in this appendix briefly its application to two popular spherical potentials . we consider a particle with mass @xmath19 in a spherical box with radius @xmath58 and ideal specular reflection from the boundary , i.e. , a three - dimensional spherical billiard . quantum - mechanically , one obtains the spectrum by solving the schrdinger equation with dirichlet boundary conditions . the semiclassical trace formula for this system has been derived by balian and bloch @xcite using a multi - reflection expansion of the green function . to derive it from our formula , we need the radial action integral for arbitrary energy @xmath342 and angular momentum @xmath343 . working throughout with unscaled variables , we find ( with @xmath344 ) s_r(e , l ) = 2_r_1^r dr & = & 2 - 2l(l / r ) + & = & 2r - l + l^2/r + o(l^3 ) , [ srlbox ] thus fulfilling our general relation with a(e ) = 1/r . with this , the stationary condition gives 2n(l / r ) = 2 m with the solutions l_nm = r(m / n ) . as in the potentials discussed in sect . [ secuni ] , the solution with @xmath164 corresponding to n=2 m contributes only via the end - point correction of to the diameter orbit with the primitive action s_d(e ) = 2s_r(e,0 ) = 4r and , according to the lower equation in , with the amplitude of its @xmath36-th repetition _ k^d(e ) = . ( k=|m| ) the upper end - point correction with @xmath345 can , for @xmath346 , only be reached formally in the limit @xmath347 . it corresponds to the `` whispering gallery mode '' with amplitude @xmath348 , cf . below , and does therefore not contribute to the trace formula . the tori with @xmath349 have the actions s^t_nm(e ) = ns_r(e , l_nm)+2 m l_nm = 2nr_nm , _ nm = m / n . [ stbb ] this corresponds exactly to the actions of the polygonal orbits given in @xcite with winding number @xmath350 and @xmath351 corners ( i.e. , reflections from the boundary ) . the case @xmath352 corresponds to the @xmath36-th repetition of the diameter orbit . for the polygons , @xmath353 is half the polar angle covered by one of their segments . the amplitudes of the tori with @xmath354 become , according to , ^t_vk(e ) = ( 2_vk ) . [ ampbb ] before we can write down the trace formula with the correct phases , we have to correct the radial quantisation condition , because the maslov index changes from 1/4 to 1/2 per turning point if a reflection happens there . hence the quantisation condition is @xmath355 for the diameter orbit and @xmath356 for the tori . this changes the phases in and all the results derived from it . taking this into account , we obtain the semiclassical trace formula for the spherical billiard g(e ) = _ v>2k>0^t_vk(e ) - _ k>0 a_k^d(e ) , [ dgbb ] which is identically the result of balian and bloch @xcite . here we derive the trace formula for the coulomb potential as a special central potential which is not regular in @xmath357 : v(r ) = - . [ vcoul ] we need not use scaled variables here , but it is useful to introduce the positive energy @xmath358 since all bound states in the potential have @xmath359 . the maximum angular momentum and the action of the circle orbit are l_m(e ) = , s_c(e)= , from which the period of the circle orbit is found to be t_c(e ) = 2 = . the radial action integral becomes elementary ( cf . @xcite ) : s_r(e , l ) = _ r_1^r_2 dr = 2(-l ) = 2 . [ srcoul ] from it we find the radial period t_r(e ) = . the difference from the other potentials treated in this paper is that the term linear in @xmath96 in here has the coefficient @xmath360 . furthermore , we find from the general relation = -2 _ ( e ) = _ r(e ) , t_r(e ) = t_(e ) . [ trat ] thus , the radial and angular frequencies are identically the same for all values of the angular momentum @xmath96 . this is a special property of kepler s ellipses . [ note that here , in contrast to the spherical ho , the angular momentum @xmath96 is defined with respect to one of the focal points and not to the symmetry centre of the ellipse ! ] inserting the above results into the general ebk trace integral for the density of states , we obtain g_coul(e ) = _ m , n=-^ ( -1)^m+n_0^l_m(e ) ldl e^i2/. [ gcoul ] due to , the general resonance condition implies @xmath361 for all periodic orbits . we therefore only use the terms with @xmath362 in the double sum of ( cf . the discussion at the end of sect . [ secpois ] ) . we then find g_coul(e ) = . writing this in terms of atomic units ( note that we have put @xmath19 ) e_at= = ry , we finally obtain the quantum - mechanically exact trace formula g_coul(e ) = _ n=1^n^2 ( e+e_at / n^2 ) = , [ gcoul1 ] which has been given in @xcite . the eigenenergies e_n = -,(n = n_r++1 ) with degeneracies @xmath363 are , as is well known , also found directly from the ebk quantisation conditions - using and . m. c. gutzwiller : _ chaos in classical and quantum mechanics _ ( springer verlag , new york , 1990 ) ; + h .- j . stckmann : _ quantum chaos : an introduction _ ( cambridge university press , cambridge , uk , 1999 ) ; + f. haake : _ quantum signatures of chaos _ ( springer , 2nd edition , 2001 ) .

cosmological gamma - ray bursts are one of the most interesting and exotic phenomena in astrophysics . in standard burst models ( e.g. see piran 1999 ; mszros 2001 for reviews ) , the rapidly expanding fireball decelerates , converting the internal energy of the hot plasma into kinetic energy of the beamed , relativistically moving ejecta and electron - positron pairs . at the point where the fireball becomes optically thin and the gamma - ray burst ( grb ) we see is emitted , the matter will not emit non - thermal gamma - rays unless some mechanism can efficiently re - convert the kinetic energy back into internal energy , i.e. , unless some particle acceleration process takes place . diffusive shock acceleration is widely believed to be this mechanism ( e.g. , rees & mszros 1992 ; piran 1999 ) for the prompt emission ; it is also likely to be a key element for the underlying physics of x - ray and optical flashes and burst afterglows . an important question concerning burst models is whether or not their prompt emission can be accurately described by a detailed shock acceleration analysis , and if so , how is the pertinent parameter space of models limited by the observations ? tavani ( 1996a , b ) proposed a synchrotron shock emission model that invoked such diffusive acceleration as the means for generating non - thermal distributions . this work has been an interpretative driver in the field , due in part to the impressive model fits of data he obtained . yet , tavani s model and many subsequent works do not treat the critical involvement the acceleration process has on shaping the relationship between the thermal and non - thermal portions of the electron distribution . the normal approach is to simplify the distribution to a truncated or broken power - law , occasionally with an additional thermal component . the viability of the synchrotron model , or emission scenarios that invoke any other radiation mechanism , hinges on such subtleties of the particle distribution , which are inextricably linked to the nature of shock acceleration , diffusive or coherent . in addition , they must provide compatibility with the hard x - ray spectral index , an issue raised by preece et al . ( 1998 ; 2000 ) , who observed that around 1/3 of all batse bursts exhibited spectra that were too flat below the peak to accommodate a synchrotron interpretation . this paper explores the general expectations for burst emission spectra from shock acceleration theory . parametric descriptions for a linear combination of thermal and non - thermal distributions are used to obtain fits to burst spectra for the cases of the synchrotron , inverse compton and synchrotron self - compton ( ssc ) mechanisms , and these distributions are interpreted in the context of known results from acceleration simulations . cases of particles cooling much slower than their acceleration are treated in this exposition ; justification for these are discussed in section [ sec : cooling ] . the focus is on grbs observed by the compton gamma - ray observatory , specifically those with egret detections that provide broad spectral coverage and therefore more constrained fits around the spectral peak . it is found that acceptable fits are only possible with a marked dominance of non - thermal electrons , contrasting the particle distributions obtained in acceleration simulations . this poses a general problem for any burst acceleration paradigm where the non - thermal population is drawn probabilistically from a thermal gas . in addition , fits are found to be more or less equally acceptable for the synchrotron and inverse compton processes ; it is anticipated that swift will provide a number of bright bursts where spectral fitting down into the x - ray band might enable discrimination between such emission mechanisms . however , at this juncture , spectroscopic discrimination against unabsorbed synchrotron self - compton scenarios seems probable , given that the characteristically broad ssc spectra in space are difficult to reconcile with cgro data . in service of this analysis , a compact formalism for ssc emissivities from thermal and truncated power - law electron distributions is present in the appendix . it is anticipated that glast s broad spectral coverage from kev to gev , supplied by the gamma - ray burst monitor in the batse band and the large area telescope in and above the egret band , will render glast a powerful tool in constraining and interpreting burst spectral properties in the future . this section explores the relationship of synchrotron spectra to the electron energy and angular distributions , inferring lorentz factor distributions using data from the bright burst grb 910503 as a representative case study . the formalism for synchrotron radiation in optically thin environs is standard and readily available in many textbooks , including the treatments of bekefi ( 1966 ) , jackson ( 1975 ) and rybicki & lightman ( 1979 ) . throughout this paper , the standard convention for the labelling of the photon polarizations will be adopted , namely that refers to the state with the photon s _ electric _ field vector parallel to the plane containing the magnetic field and the photon s momentum vector , while denotes the photon s electric field vector being normal to this plane . the classical ( photon ) angle - integrated emissivities for the and polarization for monoenergetic electrons of lorentz factor moving at angle to the field are ( e.g. westfold 1959 ; jackson 1975 ; rybicki and lightman 1979 ) @xmath0 in units of , where the are modified bessel functions of the second kind . the generally higher probability of emitting photons than ones reflects the intrinsic dipolar radiative nature of accelerating / oscillating charges ; the polarization is elliptical for photon propagation along the field . the characteristic synchrotron rate factor that scales these emissivities is @xmath1 here , is the fine structure constant . the photon energies that appear in these functions that form the scaling for the energies of emission are [ with the electron compton wavelength over being ] : @xmath2 where is the particle s larmor radius , and gauss is the quantum critical field . note that since , the photon angles to the field form a narrow distribution about . furthermore , the expressions in eq . ( [ eq : synchrate ] ) are appropriate for cases where the particle momenta are not closely aligned with the magnetic field ; situations of such near alignment are discussed in section [ sec : pitch_synch ] below . the total synchrotron emissivity is obtained by integrating over the energy and angular distribution of the particles . for electron distributions in units of , summing over polarizations forms the expression for the emissivity that will be used mostly throughout this paper ( in units of ): @xmath3 this is the differential photon spectrum ; frequently in this work , reference will be made to the so - called spectrum , a spectral energy distribution that is proportional to . physical diagnostics on the grb environment are obtained by comparing the emission spectrum in eq . ( [ eq : synch_emiss ] ) with observations . the obvious database of choice comprises batse observations from the compton gamma - ray observatory ( cgro ) , specifically the batse spectroscopy catalog of preece et al . ( 2000 ) with its pubicly - accessible electronic compilation . yet this data possesses a principal limitation , namely that imposed by the restricted energy band of batse . inferences of particle distributions are generally on a more secure footing when using broader burst spectra , particularly in domains of significant spectra structure , which is the intrinsic nature of the batse pool . thus , the two decades of energy afforded by batse often are insufficient to describe the nature of the non - thermal particle distribution invoked in models . data from the egret experiment on cgro suitably augment this situation and extend the dynamic range by 24 orders of magnitude in energy . accordingly , a focus on egret bursts is insightful and offers more secure determinations of particle distributions . use is therefore made of spectral compilations of batse , comptel and egret data that are offered in schaefer et al . specifically , a case study is made here of grb 910503 , observing that the inferences to be made generally extend to the handful of other egret bursts , and also to a large number of bright batse bursts that possess no confirmed egret detections . note that the focus here is primarily on time - integrated spectra . given the complexity and diversity of burst time profiles , and associated variations in spectral shape and the well - known hard - to - soft evolution , detailed spectral fits will depend on the time window chosen . a classic example for this is provided by the gradual emergence of a hard gamma - ray component in the spectrum of grb 941017 ( gonzalez , et al . the main conclusions of this paper apply to the mean properties of the emitting particle distributions , and are striking enough to indicate their applicability for most or nearly all of the duration of the bursts studied herein . a variety of particle distributions are possible in modeling burst spectra . clearly , all should contain a non - thermal component , since the data are clearly non - thermal : purely isothermal models do not work ( pacynski 1986 ; goodman 1986 ) , though constructed convolutions of thermal spectra of different temperatures are possible . the emphasis here is motivated by physical interpretation , and the leading contender for non - thermal particle energization in bursts is diffusive acceleration at relativistic shocks in the burst outflow . such acceleration is well studied in non - relativistic flows , and is widely believed to originate with thermal particles that are diffusively transported when interacting with field turbulence in the shock environs . such transport effects many crossings of the shock layer ( e.g. see drury 1983 ; blandford & eichler 1987 ; jones & ellison 1991 for reviews ) , that lead to a friction term in the momentum evolution , i.e. acceleration . while other acceleration models such as reconnection and coherent electrodynamic energization exist and are appropriate in a variety of astrophysical environments ( e.g. the solar corona , pulsar magnetospheres ) , diffusive shock acceleration is the most widely invoked in astrophysical models ; it forms the centerpiece of discussion here as well as for many investigations of bursts . tavani ( 1996a , b ) proposed a shock emission model that invoked such diffusive acceleration as the means for generating non - thermal distributions . this work has been an interpretative driver in the field , in no small part due to the impressive model fits of data obtained . the principal purpose of this section is to delve deeper into the physical implications of such data fitting . accordingly , it is appropriate to use an electron distribution similar to tavani s ( 1996a , b ) to add insight to the discussion . here , the quasi - isotropic electron distribution @xmath4 is adopted , where is a step function with for and zero otherwise , and is the system or post - shock temperature . the factor directly relates to the efficiency of acceleration , and is a parameter in the fitting protocol , as are ( assumed much greater then unity ) , the dimensionless , the non - thermal index , and the normalization factor . the ensuing discussion will provide physical grounds justifying this choice of distribution in shocked environs ; other possibilities with different mandates include the broken power - law adopted by lloyd & petrosian ( 2000 ) . note that is independent of for true isotropy . in many cases in this paper , will be chosen for simplicity , since this corresponds to the dominant radiative contribution from the synchrotron mechanism . the parameter is introduced to generalize from tavani s ( 1996a , b ) fits , permitting the non - thermal distribution to emerge perhaps at truly super - thermal energies , rather than exactly in the thermal peak ( tavani s restricted case ) . typical acceleration spectra in non - relativistic shocks assume forms similar in general appearance to eq . ( [ eq : elec_dist ] ) . specifically , they correspond to the portion of parameter space ( at least for ions ) , so that dissipation in the shock layer leads to a heating plus non - thermal acceleration that ensues only at super - thermal energies in the maxwell - boltzmann tail . evidence for such a structured distribution can be found both in heliospheric measurements of accelerated ions near shocks ( e.g. earth s bow shock : see ellison , mbius & paschmann 1990 ; interplanetary shocks : see baring et al . 1997 ) , and also a variety of acceleration simulations and theoretical analyses ( e.g. see scholer , trattner & kucharek 1992 ; giacalone et al . 1993 , baring , ellison & jones 1993 ; ellison , baring & jones 1996 ; kang & jones 1997 ) . the situation for relativistic shocks is less certain , due in no small part to the absence of _ in situ _ spacecraft determinations of particle spectra ; simulation expectations for such environments are discussed below . in this paper , values of appear , being in general only weakly constrained by the fits and largely motivated by the above physical and theoretical considerations . while numerical evaluations of eq . ( [ eq : synch_emiss ] ) given the electron distribution in eq . ( [ eq : elec_dist ] ) were performed to obtain approximations to burst spectra , standard asymptotic limits are readily attainable in analytic form . the non - thermal contribution to the synchrotron spectrum ( from the truncated power - law ) , dominant whenever , assumes the following power - law forms ( derived , for example , from eq . [ [ eq : synchfinal ] ] ) : @xmath5 here , is the component of the field perpendicular to the instantaneous electron motion . the thermal component , sampled when , necessarily declines exponentially at photon energies well above the thermal peak , but possesses the same power - law dependence as in the non - thermal case for low photon energies : @xmath6 this spectral similarity at low energies follows from the narrow nature of the distributions being convolved with the synchrotron emissivity functions in eq . ( [ eq : synchrate ] ) . accordingly , it is not possible to distinguish between emission from thermal or power - law distributions if spectrum measurements exist only at energies below the peak ; some other diagnostic is required . the exponential decline above the thermal peak possesses a somewhat complex two - domain structure ; the reader is referred to pavlov & golenetskii ( 1986 ) , brainerd & petrosian ( 1987 ) and baring ( 1988a ) for the presentation of appropriate asymptotic forms in cases . it must be noted that throughout this paper , photon spectra are generated in the shock rest frame ( srf ) and then boosted in energy to the observer s frame ( of ) by a fixed doppler factor , independent of photon angle in the srf . this approximation , essentially adopted in many models of grb spectra , is expedient but does not accommodate the fact that photon distributions in the srf are inherently anisotropic due both to the instrinsic nature of the synchrotron and inverse compton processes , and also to the fact that relativistic shocks naturally generate anisotropic ions and electrons . notwithstanding , this boost approximation does not dramatically distort the angle - integrated of spectral shape for typical srf angular distributions and suffices to describe the continuum for the purposes of the discussions here . before presentation of results , it is appropriate to address the modeling procedure used in this paper . approximate spectra , as opposed to data fits , were derived visually using both log - log and semi - log representations of spectra to increase confidence in the results . least squares trials were performed in some cases , but with no addition of insight or perceived accuracy . this choice of a visual approach was one of expediency , and did not compromise the scientific conclusions presented here ; the essential points are unambiguously made without the need of statistical analysis . this latitude is afforded by the limited energy range of the data in the batse catalog : slight adjustments in parameters used to fit the extant data for a given burst would almost certainly have been dominated by those needed to accommodate a broader spectral range . for the specific cases of grb 910503 and grb 910601 , where egret data are incorporated in the study , the uncertainty in the mev data points is a major contributor to the determination of model parameters . in bursts where no egret or comptel data were used , the `` fit '' was actually to the band ( 1993 ) model spectral fit presented in the preece et al . ( 2000 ) spectroscopy catalog , not to the actual data . in such cases , the high energy band model index was adopted , and other distribution parameters adjusted . in this sense , there was in implicit second order incorporation of the statistics embodied in the band model fit . it must be emphasized that there is no uniquely preferred method for statistically fitting public data since it is already a convolution of model and instrumental response . the correct approach ( as in lloyd & petrosian 2000 ) would be to fold spectral models here through the batse or egret detector response function , an enterprise that rapidly becomes more ambitious when different instrumental responses must be blended . such complexity seems premature for the explorations of this paper , and is deferred until the more refined data of swift and glast becomes available . results for the spectral approximation of the grb 910503 data are presented in figure [ fig : synch_spec_dist ] , together with the multi - instrument , time - integrated cgro data as provided by schaefer ( private communication ) prior to the public release in schaefer et al . we remark that the spectral deconvolutions published in schaefer et al . ( 1998 ) were made independently for each data set from the different cgro instruments , applying an artificial normalization to reconcile offsets in the different data amplitudes . such a procedure is less accurate than a joint spectral deconvolution of the entire spectral data set ( see for example the analysis of grb 990123 data in briggs et al . notwithstanding , the major conclusions of this paper are robust and are not affected by such subtleties of data analysis . figure [ fig : synch_spec_dist ] displays both synchrotron emission spectra for the case of pitch angles ( the so - called _ quasi - isotropic _ case ) , and the inferred electron distribution that generates the continuum . for the solid curve approximate case , the overall normalization is obviously a free parameter and was offset slightly from the data for visual clarity . the electron distributions were normalized to unit area in space . the solid curve depicts the case that would be required to explain the photon spectrum . the non - thermal particle distribution clearly dominates the thermal component in this `` fit . '' its index differs from the value of 3.8 that tavani ( 1996a ) obtained in his analysis of this burst , a discrepancy that is actually insignificant given the error bars in the egret data . adjustment of the parameter to improve the appearance of the model approximation around and above the spectral peak led to the assumed value of , accurate to around 15% . notwithstanding , there appears to be some spectral structure around the peak that can not be explained by the simple particle distribution in eq . ( [ eq : elec_dist ] ) . the positioning of the peak in energy produced the fit value , where is the lorentz factor associated with the bulk motion of the grb expansion . here , is interpreted as the magnetic field in the burst expansion frame , i.e. , not in the ism or shock rest frames . the parameter describing the relative normalization of the non - thermal and thermal components was found to be ( ) , a strikingly large value that is not addressed explicitly in detail by tavani ( 1996a ) . a discussion of the implications of an inferred particle distribution with such a large follows just below . a similar property is obtained for synchrotron model fits for batse bursts grb 940619 ( trigger number 3035_6 ) and grb 940817 ( trigger number 3128_5 ) from the preece et al . ( 2000 ) catalog , and also for grb 910601 and grb 910814 whose composite spectra are presented in schaefer et al . the spectra for bursts grb 930131 and grb 940217 that possess egret detections also suggest a similar result , though the lack of a composite broad - band multi - instrument spectrum for each in the literature precluded a detailed analysis . the parameter regime of is expected to be pervasive for batse bursts , due to the smoothness of their circumpeak continua . for comparative purposes , another distribution and its resultant synchrotron is illustrated in figure [ fig : synch_spec_dist ] as dashed curves . the distribution was deliberately chosen to possess a somewhat dominant thermal component , motivated by expectations from shock acceleration theory , and clearly yields a spectrum that grossly deviates from the observations both well above and well below the peak . while these deductions are obtained using a time - integrated spectrum , such a conclusion must apply either for most of the burst duration , or at sub - intervals when the burst was at its brightest . for the synchrotron mechanism to be a viable explanation of grb spectra , the underlying particle distribution must be physically realistic . here , expectations of particle distributions from shock acceleration theory are the predominant focus , partly because this was the preferred acceleration mechanism in tavani ( 1996a , b ) , and largely because this is the underlying energization assumed in most burst models . in a nutshell , both theory and observation pertaining to non - relativistic shocks strongly indicate and cases . note that is essentially the parameter that demarcates the relative normalization of the non - thermal and thermal components in equation ( [ eq : elec_dist ] ) . applications of shock acceleration in the heliosphere and at supernova remnant shells more or less mandate these parameter regimes , since the seed pool of particles that can be subjected to acceleration are thermal and moreover plentiful . in heliospheric environs , the source is the cold solar wind , and there is little evidence of significant pre - heating that is not connected to the shock system under study . the _ in situ _ spacecraft measurements exhibit strong heating at shocks and the generation of dominant thermal ions in the downstream regions , corresponding to cases . data taken at the high sonic mach number earth s bow shock ( e.g. see ellison , mbius & paschmann 1990 ; scholer , trattner & kucharek 1992 ) and low sonic and afvnic mach number interplanetary shocks ( e.g. see gosling et al . 1981 ; baring et al . 1997 ) clearly support this assertion . in supernova remnants , direct particle measurements are , or course , impossible , and inferences can only be made from radiative signatures . while electron temperatures and densities can be estimated from x - ray signals and molecular emission in the ir / optical bands , the data can not yet clearly discriminate and situations . from a theoretical perspective , diffusive shock acceleration naturally draws particles from the thermal pool via transport in shock - associated turbulence . whether this be first or second order in momentum transport , thermal and non - thermal particles are subjected to similar diffusion , at least as far as ions are concerned , so that circumstances would require virtually complete suppression of thermalization and a 100% efficiency of acceleration out of the thermal pool . such a scenario would require unusually anomalous spatial diffusion coefficients ( i.e. extremely small ) at thermal momenta , or perhaps strange and non - dissipative electrodynamic properties in the shock layer . note that is more likely to increase at low energies due to the possibility of wave damping by thermal populations ( e.g. see forman , jokipii & owens 1974 , for the relationship between and wave power spectra within the confines of quasi - linear theory ) . without the assumption of anomalous forms for , monte carlo transport simulations ( e.g. see baring , ellison & jones 1993 ; ellison , baring & jones 1996 ) clearly indicate a downstream thermal population that dominates the extended non - thermal distribution in number , corresponding to . the `` injection '' of ions from the thermal pool is extremely efficient , but not incredibly so . hybrid plasma simulations ( e.g. see trattner & scholer 1991 , 1993 ; scholer , trattner & kucharek 1992 ; giacalone et al . 1992 , 1993 ; liewer , goldstein & omidi 1993 ) of wave generation and ion transport near non - relativistic shocks , which treat the electron component as a background fluid and therefore address only some of the electrodynamic aspects of the plasma in the shock layer , produce similar distributions . these indications are strongly suggestive , but not decisive for the gamma - ray burst problem , since they largely do nt focus on electron properties and equally importantly do not explore acceleration at relativistic shocks . in situ measurements of non - thermal electrons at shocks are limited ; examples include recent geotail observations of interplanetary shocks ( shimada , et al . 1999 ; terasawa et al . , 2001 ) . accordingly , discussion is dependent on simulational information , as is the case for relativistic shocks , and the literature is sparse on such a subject . monte carlo simulations of ion acceleration at relativistic shocks ( e.g. ellison , jones & reynolds 1990 ; baring 1999 ; ellison & double 2002 ) generate the same injection properties as do non - relativistic shocks , i.e. . plasma simulations of relativistic shocks are similarly sparsely studied . gallant et al . ( 1992 ) and hoshino et al . ( 1992 ) presented results from one - dimensional ( 1d ) particle - in - cell ( pic ) simulations of electron - positron and electron - positron - proton plasmas in ultrarelativistic flows with perpendicular shocks , where the mean field was orthogonal to the shock normal . such simulations treat the full electrodynamics of such plasmas , but are severely limited by cpu constraints . accordingly , they tend to use unrealistically small proton to electron mass ratios , and model relatively small spatial scales . these constraints tend to mask the production of some wave modes and limit consideration to particles with energies at most a few to ten times thermal . the context for these investigations was pulsar wind flows , and they revealed no acceleration in pure plasmas . with the addition of protons in significant abundances , hoshino et al . ( 1992 ) demonstrated that non - thermal positrons ( and not electrons ) could be created , as the ions generated elliptically - polarized magnetosonic modes that resonantly interacted with the positrons . mace and jones ( 1994 , unpublished ; f. jones , private communication ) explored a 1d pic simulation similar to gallant et al . ( 1992 ) for an ultrarelativistic plasma shock of arbitrary field obliquity , and found that electron / positron acceleration arose in quasi - parallel scenarios , but disappeared when the field obliquity to the shock normal increased above around degrees , concurring with zero acceleration in the earlier results for perpendicular shocks . a scaled version of their parallel shock distribution ( binned in energy ) is depicted in fig . [ fig : synch_spec_dist ] , and is fairly proximate to the , case . a notable feature of the simulation data is the very limited range of energies accessible to the pic technique , imposed by the severe cpu limitations for such complex codes . some interesting recent pic simulations results have been presented by shimada & hoshino ( 2000 ) and hoshino & shimada ( 2002 ) for electron - proton plasmas containing mildly - relativistic shocks , where suprathermal power - law electrons are generated and the distribution closely resembles the pic simulation data depicted in fig . [ fig : synch_spec_dist ] . yet , as with the mace and jones results , the complete distribution reveals that the power - law emerges only well into the maxwellian tail , corresponding to the situation elicited in the monte carlo simulations . it is noted in passing that the one - dimensional nature of these simulations potentially imposes an important absence of critical physics to the acceleration problem : jokipii , kta & giacalone ( 1993 ) and jones , jokipii and baring ( 1998 ) demonstrate that particle diffusion across field lines is eliminated when simulations restrict the spatial dimensions to less than three . the implication of this result for the shocks is a cessation of acceleration when thermal particles can not be transported across fields to return to the shock from the downstream side ; baring , ellison & jones ( 1993 ) demonstrated that for high mach number , non - relativistic ion shocks , a quenching of acceleration through this effect would arise at around , a result that matched the behavior seen in the relativistic pic simulations of mace and jones . moreover this result is consistent with the absence of acceleration in non - relativistic 1d hybrid plasma simulations applied to the quasi - perpendicular heliospheric termination shock ( e.g. liewer , goldstein & omidi 1993 ; kucharek & scholer 1995 ) , a site that is believed to generate the non - thermal anomalous cosmic rays . the essential loss of physics in simulations of restricted dimensionality is clearly crucial to the correct modeling and interpretation of particle acceleration in gamma - ray bursts . the principal conclusion of this discussion is that there is little observational or simulational evidence for quasi - isotropic electron distributions in the environs of shocks that would render a synchrotron spectrum commensurate with the grb910503 data in particular and typical burst spectra in general . an issue that has already been substantially discussed in the literature is whether the synchrotron model is compatible with spectral slopes below the burst peak . for grb 910503 , fig . [ fig : synch_spec_dist ] clearly indicates that there is no conflict between the data and the model . the fixed value of for the low energy index in eqs . [ eq : ntsynch_limits ] and [ eq : thsynch_limits ] suggests that this is not the situation for all bursts , a fact that was demonstrated by preece et al . ( 1998 ; see also preece et al . 2000 ; 2002 ) . synchrotron theory distinctly predicts this canonical index at low energies for any isotropic electron distribution that has a characteristic lorentz factor as its lower scale . correlations between lorentz factors and pitch angles can alter this property , as discussed in section [ sec : pitch_synch ] below . any index greater than this value can not be accommodated by quasi - isotropic distributions ; indices less than this are easily described by a steepening particle distribution above . preece et al.s ( 2000 ) display of the distribution of low energy indices clearly indicated a conflict , as illustrated in fig . [ fig : batse_low_index ] . there the low energy indices for various empirical fitting models used in the batse spectroscopy catalog are binned in a histogram , with the permitted range of clearly marked . the most common fitting model used in the catalog was the popular band ( 1993 ) grb model for the observed flux in photons/@xmath7/sec : @xmath8 most bursts possess values and . in folding the band model , broken power - law models and others through the batse detector response , preece et al . ( 2000 ) found that the band model performed best in roughly 2/3 of the burst sample . note that the index recorded in the catalog is the asymptotic low energy index ( preece et al . 2000 ) , and does not exactly match the index local to the lowest energy kev in the batse window . the inferred values of are subject to instrumental selection effects , particularly when the peak energy is low . hence , there is the possibility that the distribution might move to slightly higher when the detection band moves to lower energy , the situation for swift . note that such an effect was observed by preece et al . ( 1998 ) when applying a correction to accommodate the energy of the batse window . a depiction of the histogram for time - integrated spectra in the batse spectroscopy catalog is given in lloyd & petrosian ( 2000 ) ; it exhibits a similar distribution , as expected . the quasi - isotropic synchrotron model clearly can not account for nearly 25% of sample burst spectra . this is true even for pathological choices of electron energy distributions : from the form in eq . ( [ eq : synch_emiss ] ) , it is easily shown that the spectral index is a monotonically increasing function of energy for any electron distribution , due to the broad and smooth nature of the synchrotron emissivities in eq . ( [ eq : synchrate ] ) . while this is disturbing , it may be more of an indication of problems with assumptions underpinning the bound than a suggestion that synchrotron radiation is not responsible for grb prompt emission . for instance , relaxing an adherence to isotropy or distributions can potentially alter the spectral slope . it is demonstrated in section [ sec : anis_synch ] below that for an independent of , the spectrum below the peak can not be flattened beyond cases , and in fact steepens . however , introducing a -dependent angular distribution can suitably ease the bound ; this is exemplified in the focus on pitch angle synchrotron emission in section [ sec : pitch_synch ] . an alternative recourse is synchrotron self - absorption , i.e. relaxation of an optically thin assumption . given the theoretical expectation of extremely flat spectra below the break ( for power - law electrons and a raleigh - jeans tail for thermal : see , e.g. , rybicki & lightman 1979 ) , different groups have invoked such a mechanism to account for cases in the batse catalog ( e.g. crider & liang 1999 ; granot , piran & sari 2000 ) . pushing the absorption break to gamma - ray energies requires extreme physical conditions ( @xmath9 , gauss ) , as noted by lloyd & petrosian ( 2000 ) , and can be inferred from eq . ( [ eq : selfabs_energy ] ) . these parameter regimes would be more extreme in the absence of the strong doppler boosting associated with grb outflows . equipartition is usually invoked to justify such high fields , however a widely - accepted and well - understood mechanism for the generation of such fields remains elusive . a convenient resolution of this issue is afforded by the proposition of panaitescu & meszaros ( 2000 ) , namely that the prompt emission is an inverse compton image of a self - absorbed synchrotron spectrum . the concomitant moving of the absorption break energy to below the optical band lowers the required values of and to less extraordinary ranges . the x - ray spectral indices in the scenario of panaitescu & meszaros ( 2000 ; see also liang , boettcher & kocevski 2003 ) can comfortably accommodate the domain , as discussed in section [ sec : alphalow_comp ] . an additional concern is that the turnover energy ( in the grb expansion frame ) satisfying is a function of a number of environmental parameters . equation ( 6.53 ) of rybicki & lightman ( 1979 ) for the optical depth for self - absorption can be manipulated to arrive at the equation for . this result can be found in many papers ; here eq . ( 7 ) of baring ( 1988b ) is adapted to accommodate the specific form for the non - thermal portion of the distribution in eq . ( [ eq : elec_dist ] ) : @xmath10 for @xmath11 here , is the thomson optical depth of the emission region , and , as before . note that represents the gamma function in eq . ( [ eq : phidef ] ) . the density and size of the absorbing medium are incorporated in . these parameters possess implicit dependence on the bulk lorentz factor of the flow , with different evolutionary dependences for adiabatic and radiative expansion cases ( e.g. , see dermer , chiang & bttcher 1999 ) . if eq . ( [ eq : selfabs_energy ] ) is interpreted in the frame of the burst outflow , then is in the euv / soft x - ray band for typical burst lorentz factors . given the number of parameters ( five in all ) contributing to the computation of the self - absorption optical depth , it is difficult to comprehend how can be fine - tuned to the batse band in the observer s frame ; such a conspiracy is an issue also for optically thin synchrotron emission , whose peak depends on three parameters , , and . a strong test of the self - absorption proposition could be offered by swift , which should determine whether or not batse was fully capable of determining the `` asymptotic '' values of , especially for those bursts with peaks around 100 kev or lower . the gamma - ray burst monitor ( gbm ) on glast will provide an additional broad - band probe . should the batse distribution be mirrored by the swift and gbm ones , synchrotron self - absorption in the gamma - ray band could then only be operating in a small minority of bursts , rendering it less germane to the spectral shape discussion , except perhaps for synchrotron self - compton models . the foregoing discussion has focused on the case of synchrotron emission from quasi - isotropic or large pitch angle electron distributions . distributing the angles of electrons in a non - uniform manner can alter the spectral index of the emitted radiation . a prime example of this was the flattening incurred in distributing magnetic pair creation turnovers in neutron star environments , probed by baring ( 1990 ) . here , obviously a distribution skewed towards small pitch angles is a potential candidate for modifying . such a distribution is not at all contrived : synchrotron cooling can produce it quite naturally ( e.g. see brainerd & lamb 1987 ) . however , as alluded to above , it is now demonstrated that such distributions , if uncorrelated with lorentz factor , yield only spectral steepening , and so can not resolve the low energy index problem . the spectral behavior can adequately be demonstrated using the non - thermal portion of eq . ( [ eq : elec_dist ] ) , which for the purposes of this discussion will be written in the form for and is zero otherwise . here . if is defined as the dimensionless cyclotron energy , then the differential rate of production of synchrotron photons in eq . ( [ eq : synch_emiss ] ) can be quickly manipulated by reversing the integrations over and . the result is @xmath12 where @xmath13 is the synchrotron emissivity scale , and @xmath14 this form for the synchrotron rate will be used later in the discussions on synchrotron self - compton emission . by adding and subtracting suitable forms like eq . ( [ eq : synchfinal ] ) with different choices of , a general expression for the synchrotron emission from a piece - wise continuous power - law distribution of electrons can be obtained . defining the photon spectral index to be , it is easily shown that is a monotonically increasing function of energy for any electron distribution with a low cutoff , using piecewise continuous power - laws to successively refine approximations to a smooth distribution . ( [ eq : synchfinal ] ) possesses two readily identifiable asymptotic limits , namely when and . these lead to the forms given in eq . ( [ eq : ntsynch_limits ] ) . now consider an angular distribution defined over the range for . the superposition of spectral forms in eq . ( [ eq : ntsynch_limits ] ) then immediately indicates that such an angular distribution must generate a spectrum with asymptotic forms at , and when . the spectrum in between depends on how rapidly varies with angle , and can be either of slope or steeper , but never flatter than and never steeper than . this follows from the convolution over angles of spectra that are monotonically declining in energy necessarily being also monotonically declining in . this assertion holds even for that is strongly peaked towards , and monotonically declining with . hence , angular distributions with uncorrelated with can not ameliorate the low energy index issue due to the spectral shape implied by eq . ( [ eq : synchfinal ] ) . a more viable possibility for flattening the low energy index is when there is a correlation between the pitch angle and the lorentz factor of charged particles . this provides a distinct kinematic difference from the foregoing considerations of synchrotron radiation from quasi - isotropic electron distributions . the archetypical example for consideration here is so - called _ pitch - angle synchrotron _ radiation , a mechanism explored in detail by epstein ( 1973 ; for applications , see epstein & petrosian 1973 ) . this mechanism corresponds to synchrotron emission from particles possessing very small pitch angles . the polarization - averaged emissivity derived by epstein ( 1973 ) in this limit , for electrons in a uniform magnetic field , can be written in the form @xmath15 this form is simply obtained from the flux emissivity in equation ( 10 ) of epstein ( 1973 ) , and is applicable for ; at higher energies , the usual synchrotron formula in eq . ( [ eq : synch_emiss ] ) is operable if . note that this process exhibits strong circular polarization due to the near coalignment of particles with the field , though there is significant linear polarization in narrow bands of photon energy . observe that defines the intrinsic energy scale of pitch angle synchrotron ( pas ) emission . ( [ eq : synch_pas_rate ] ) is virtually independent of for domains , thereby defining an extremely flat spectrum . this property is a direct consequence of the kinematic coupling , yielding a dependence that contrasts the isotropic synchrotron dependence . this weaker dependence on immediately implies an inference of stronger source fields in order to move the characteristic pas photon energy into the batse window . the spectrum drops dramatically above if , otherwise it merges into the synchrotron form if . these shapes imply ( i ) that prompt grb spectra that are flatter than below the peak ( i.e. around 98% of bursts : see figure [ fig : batse_low_index ] ) can be easily accommodated by a pitch angle synchrotron scenario ( just like the inverse compton scattering case considered in section [ sec : icformalism ] below ) , and ( ii ) accordingly that a variety of grb spectral shapes can be accurately modeled due to the characteristically narrow peaking of the pas emissivity in eq . ( [ eq : synch_pas_rate ] ) . these were conclusions of lloyd - ronning and petrosian ( 2002 ) , who explored fits to batse bursts for pitch angle synchrotron models , and addressed correlations between spectral parameters and temporal properties . hence , pitch angle synchrotron is an attractively versatile spectroscopic possibility for bursts . note that the mechanism of `` jitter '' radiation in turbulent magnetic fields ( medvedev 2000 ) possesses similar spectral properties due to the intrinsic kinematic coupling , thereby offering an alternative possibility of similar promise . pitch angle synchrotron might arise when electrons stochastically increase their pitch angle from virtually zero during an acceleration process . for it to provide the major contribution , a prevalence of small pitch angles must be strongly favored , i.e. the pitch angles must not be significantly populated , so that normal synchrotron emission does not dominate . it is not clear how this might come about , though rapid cooling and inefficient diffusion transverse to the field is an interesting possibility ( e.g. see lloyd - ronning and petrosian 2002 ) . in addition , small pitch angles must be prepared in the first instance , which suggests a coherent electrodynamic mode of acceleration as opposed to diffusive one . while pas may afford a means of enhancing the non - thermal radiative signal relative to the thermal component , a detailed exploration of pas spectral shapes and polarization for thermal and non - thermal distributions in the context of diffusive acceleration will be the subject of future work . while detailed exploration of cooling effects is beyond the scope of this paper , a brief discussion is appropriate . cooling is known to strongly influence spectral shape in quasi - isotropic synchrotron scenarios ( e.g. ghisellini , celotti & lazzati 2000 ) , introducing a second `` line of death '' with a value of . this presents problems for models invoking strong cooling in generating spectra like those seen by cgro . such models are most applicable to grb environments where the emission regions are somewhat remote from the acceleration ones . this is because even though synchrotron radiation is a remarkably efficient process , diffusive acceleration at shocks is even more so . accordingly , coincident acceleration and emission zones of the same physical scale can exhibit prolific acceleration up to some maximum lorentz factor , before radiative cooling in the magnetic field causes cessation of acceleration . the spectral turnover that marks this cessation is quasi - exponential in nature , and have not been observed in burst spectra to date . this combination acceleration / radiation scenario is widely envisaged in the contexts of x - ray and gamma - ray emission from supernova remnants and blazars . whenever multiple spatial or temporal scales are sampled , convolution of spectral contributions can blur the turnover into a broad distribution , essentially the case explored by ghisellini , celotti & lazzati ( 2000 ) . the rate of acceleration scales as the inverse gyrofrequency , which has a weaker dependence on electron lorentz factor than the synchrotron cooling rate . specifically , the acceleration timescale ( in seconds ) can be approximated by , where is the energy of the electrons in units of tev , is the shock speed in units of , is the mean magnetic field strength ( measured in gauss ) , and is a factor of order unity that accounts for corrections to the diffusive timescale appropriate for relativistic shocks . the synchrotron cooling timescale ( also in seconds ) is . various versions of these formulae are widely available , though a specific discussion of them in the context of blazars is in baring ( 2002 ) ; they are effectively applicable in the comoving frame of the burst outflow , as opposed to the observer s frame . the electron energy and the magnetic field are coupled by placing the synchrotron peak in the soft gamma - ray band , i.e. setting gauss . here is the bulk lorentz factor of the outflow . these quickly lead to the ratio estimate @xmath16 acceleration is clearly much more rapid than synchrotron cooling for typical , a property that changes only when considering electrons that emit above about 10 gev , i.e. for gauss . in the special case of , the result in eq . ( [ eq : cooling_ratio ] ) corresponds to the well - known synchrotron cooling cutoff property of the crab pulsar wind nebula ( e.g. see de jager et al . , 1996 ) , namely that the spectral turnover occurs at around 50 mev , independent of the magnetic field strength . since significant questions surround the viability of the synchrotron mechanism in explaining burst spectra , consideration of alternative emission processes is warranted . here , the focus is on inverse compton scattering , another very efficient radiative mechanism . for the purposes of the discussion here , it comes in two relevant varieties : scattering off a narrow - band soft photon distribution , and upscattering of synchrotron photons ( synchrotron self - compton , or ssc ) that provide an inherently broad seed spectrum . it shall be seen that the resulting predictions for prompt emission spectra for these two cases are manifestly different . the inverse compton ( ic ) mechanism possesses two obvious advantages over synchrotron radiation : ( i ) high values of can be avoided , since there is no requirement that the synchrotron peak appear in the gamma - ray band , and ( ii ) the low energy index can , in cases to be discerned in sections [ sec : icformalism ] and [ sec : alphalow_comp ] , be as flat as . at the same time , inverse compton scenarios ( particularly the ssc case ) have the disadvantage of requiring that synchrotron radiation or other components be unobservable at optical and lower wavebands during the prompt and early afterglow phases . the parameter constraints thereby imposed are discussed in section [ sec : constraint ] . the well - known criterion for ic to dominate synchrotron as an electron cooling mechanism ( i.e. also in total bolometric luminosity for a spatially uniform medium ) is simply that the energy density of the soft photons exceeds that of the ambient magnetic field , ( e.g. rybicki & lightman , 1979 ) . in general , the value of can vary enormously depending on the particular invocation for soft photon generation . for the specific case of a ssc model , couples to the field and can be determined by integrating in eq . ( [ eq : synch_emiss ] ) over and multiplying by some fiducial photon escape timescale , , that reflects the physical size of the emission region . for monoenergetic electrons of lorentz factor , the familiar result ( e.g. see p. 168 of rybicki & lightman , 1979 ) for the synchrotron energy loss rate can be used to write . hence , the ssc emission can dominate the bolometric luminosity for significant but not particularly large values of the product . it was precisely this property that led gonzalez et al . ( 2003 ) to argue that the intriguing hard gamma - ray component seen by the egret tasc in grb 941017 was not due to ssc emission , but perhaps due to inverse compton scattering seeded by an external soft photon source . the criterion for an ssc component to fall in the soft gamma - ray band is , which can easily accommodate fields in the gauss range for and . a more thorough exploration of ssc parameter spaces for burst prompt and afterglow emission can be found , for example , in mszros , rees , & papathanassiou ( 1994 ) , sari & esin ( 2001 ) and zhang & mszros ( 2001 ) . the formalism for the inverse compton ( ic ) emissivity in optically thin environs is readily available in many astrophysical textbooks , such as rybicki & lightman ( 1979 ) . since the description of ic polarization effects is more subtle than for synchrotron radiation , being dependent on anisotropies in the soft photon population , the developments here are restricted to cases where photon polarizations are summed over . the rate of emission , integrated over the angles of the upscattered photons , is independent of the direction of momentum of the seed photons . hence , no integration over is required , contrasting eq . ( [ eq : synch_emiss ] ) for synchrotron radiation . however , the rate is dependent on the soft photon energy , which must be integrated over , yielding the form ( in units of ) @xmath17 here is the thomson cross section , is the energy distribution of soft photons , and @xmath18\ ; \theta ( z ) \quad , \quad z\ ; = \ ; \dover{\erg}{4\gamma^2\erg_{\rm s}}\;\ ; , \label{eq : fdef}\ ] ] for a heaviside step function when and otherwise . mirroring the treatment of synchrotron radiation , while numerical evaluations of eq . ( [ eq : ic_emiss ] ) were performed using the electron distribution in eq . ( [ eq : elec_dist ] ) to obtain approximations to burst spectra in section [ sec : part_comp ] , asymptotic spectral indices are readily attainable . for monoenergetic soft photons , the non - thermal contribution to the inverse compton spectrum ( from the truncated power - law ) is dominant whenever , and assumes the following power - law forms : @xmath19 the index at high energies , , is identical to that for synchrotron radiation , since both mechanisms emit photons with the same dependence on the electron lorentz factor : . at low energies , the differential spectrum is approximately independent of the upscattered energy of the photon , due to the limit of and the fact that electrons near the cutoff dominate the low energy photon production . this index of has been known since the early days of application of inverse compton to astrophysical problems : see , for example , figure 3 of the comprehensive treatment of jones ( 1968 ) that also illustrates that the compton spectrum flattens further to in the domain of downscattering . as in the case of synchrotron radiation , the thermal component , sampled when , necessarily declines exponentially at photon energies well above the thermal peak , but possesses the same power - law dependence as in the non - thermal case for low photon energies : @xmath20 this follows because the low energy photons are produced by electrons mostly near the peak of the thermal distribution , again sampling const . from the perspective of spectral shapes at and below the gamma - ray peak , it is important to explore the properties of synchrotron self - compton ( ssc ) radiation , to distinguish its character from ic emission that is seeded by quasi - monoenergetic soft photons . here the analysis will first be confined to the case of a purely non - thermal electron distribution , i.e. regimes , where the distribution can be written in the form for and is zero otherwise . hereafter , when making the identification with the non - thermal portion of the distribution function in eq . ( [ eq : elec_dist ] ) , , and one can set and to establish a correspondence . for such a truncated power - law , a compact analytic form for the ssc spectrum was obtained , essentially reducing the quadruple integration to one over a single variable . defining a fiducial photon escape timescale , for a one - zone emission region of size , the average density of synchrotron photons that seed the ssc emission can be written as , so that the form in eq . ( [ eq : synchfinal ] ) can be directly inserted into eq . ( [ eq : ic_emiss ] ) . the analytic reduction is expounded in appendix a , with the result being @xmath21 \label{eq : sscfinal}\\[-5.5pt ] & \times & \biggl\ { \int_{0}^{\chi } dy \ , k_{5/3}(y ) p_{\delta } \bigl(\dover{y}{\chi}\bigr ) + \int_{\chi}^{\infty } dy \ , k_{5/3}(y ) q_{\delta } \bigl(\dover{\chi}{y}\bigr ) \ ; \biggr\}\;\ ; , \nonumber\end{aligned}\ ] ] where @xmath22 is the effective scale for the rate of the synchrotron - self - compton process . here , is the thomson optical depth of the emission region . the emergent photon energy is encapsulated in a single dimensionless parameter @xmath23 that denotes the energy scale of the peak for ssc emission , which putatively could be in the gamma - ray band , and is the only spectral structure in the emissivity . the functions and are relatively compact and are given in eqs . ( [ eq : pdef ] ) and ( [ eq : qdef ] ) in appendix a in terms of elementary functions . ( [ eq : sscfinal ] ) can be routinely integrated numerically . asymptotic energy dependences of the ssc rate in eq . ( [ eq : sscfinal ] ) are readily obtained . in cases where , the term never contributes to the integration since it is exponentially suppressed by the behavior of the bessel function . accordingly , it is quickly deduced that the ssc spectrum approximately obeys at energies , as expected : a convolution of an extended synchrotron spectrum with a truncated power - law electron distribution with index is a power - law of index . there is a slight logarithmic modulation as outlined in eq . ( [ eq : sscnt_highchi ] ) . the limit can also be promptly obtained . the term samples only the asymptotic limit of the bessel function , i.e. the form . from this , it is quickly verified that the integration is proportional to . it is simple to show using the eqs . ( [ eq : pdef ] ) and ( [ eq : qdef ] ) in appendix a that the term possesses an identical dependence . the net result is , as embodied in eq . ( [ eq : sscnt_lowchi ] ) , a property that has been evident in various ssc models of bursts ( e.g. see sari & esin , 2001 ) . these two limits can be summarized via @xmath24 the low energy index images the synchrotron low energy tail , a result that is not surprising since such a tail is steeper ( and therefore its constituent photons are more populous ) than the tail resulting from ic upscattering of monoenergetic soft photons in eq . ( [ eq : ntic_limits ] ) . accordingly , the bound is placed on ssc spectra so that this mechanism can not aid with the low energy spectral index issues discussed in section [ sec : alphalow_synch ] . to facilitate the fitting of grb spectra with ssc spectra , the case of ultra - relativistic thermal particles must also be considered , for which the distribution can be written in the form . when making the identification with the thermal part of the distribution function in eq . ( [ eq : elec_dist ] ) , one can set and to establish a correspondence with correct normalization . in appendix b , an expedient derivation of the emission spectrum is offered , making use of the result already presented in eq . ( [ eq : sscfinal ] ) for a truncated power - law electron distribution . basically , the limit is taken to generate the spectrum for monoenergetic electrons , and this is then convolved with a relativistic maxwell - boltzmann distribution . setting @xmath25 as the effective scale for the rate of the synchrotron - self - compton process for the hot thermal particles , one arrives at @xmath26^{1/4}\bigr)\quad , \label{eq : sscfinal_th_too}\ ] ] where the function is defined in eq . ( [ eq : caljdef ] ) , and is steeply declining as increases . here , . asymptotic forms for the spectrum in eq . ( [ eq : sscfinal_th_too ] ) are routinely derived , being presented in eq . ( [ eq : sscthlimits ] ) . as expected , an form arises at energies below the characteristic energy , and the spectrum is a weakly declining exponential far above this . results for the spectral fitting of the grb 910503 data for the case of inverse compton scattering are presented in figure [ fig : ic_spec_dist ] . the fits were performed using the formalism presented in section [ sec : icformalism ] , specifically with numerical evaluations of eq . ( [ eq : ic_emiss ] ) . again , the multi - instrument , time - integrated cgro data as published in schaefer et al . ( 1998 ) was used . the figure displays both inverse compton emission spectra for the case of an isotropic , monoenergetic soft photon distribution , and the inferred electron distribution that generates the continuum . the solid curve depicts the case that would be required to explain the photon spectrum , with the overall normalization again being a free parameter ; this curve was offset slightly from the data for visual clarity . the electron distributions were normalized to unit area in space . the non - thermal particle distribution clearly dominates the thermal component in this `` fit , '' as is the case for synchrotron emission . the index was controlled by the egret data , and so is identical to that in fig . [ fig : synch_spec_dist ] . adjustment of the parameter to improve the appearance of the model approximation around and above the spectral peak led to the assumed value of , again accurate to around 15% . the positioning of the peak in energy produced the fit value , where is the lorentz factor associated with the bulk motion of the grb expansion . the parameter describing the relative normalization of the non - thermal and thermal components was found to be ( % ) , still a large value though only a tenth of that obtained for the synchrotron fit . the conclusion is essentially identical to that in section [ sec : part_synch ] : the observational data can only be accommodated by an electron distribution that has a preponderance of non - thermal electrons as fig . ( [ fig : ic_spec_dist]b ) . such a determination was also obtained for batse bursts grb 940619 ( trigger number 3035_6 ) and grb 940817 ( trigger number 3128_5 ) from the preece et al . ( 2000 ) catalog . moreover , the fits in figs . [ fig : synch_spec_dist ] and [ fig : ic_spec_dist ] are of comparable quality , at least below 1 mev and above 5 mev ; i.e. there is little potential in the cgro data to discriminate between the two emission mechanisms . there is a small bump / excess in the theoretical spectrum in fig . [ fig : ic_spec_dist ] just above 1 mev that is a consequence of the sharper inverse compton scattering kernel ; its prominence can be muted somewhat by slightly reducing both and the index . such a liberty may not be afforded in the glast era when burst spectral indices will be better constrained because of its broad energy range , sampled by the gamma - ray burst monitor ( gbm ) in the batse band and the large area telescope ( lat ) in the egret band . the last point to note is that the inverse compton fit is tolerated here because the low energy asymptotic index is possibly not fully realized in the batse window . the data possess a slight curvature that may be consistent with a continued flattening of the spectrum below 30 kev , as would be expected for the inverse compton mechanism . yet the data match the synchrotron emissivity also , indicating that batse can not discriminate between these two mechanisms on the basis of the soft gamma - ray portion of the bandwidth . this impasse should be addressed when swift is launched in the coming year . swift possesses the sensitivity in the critical 1 -100 kev band to probe the spectrum below the peak and afford diagnostics on the emission mechanism . furthermore , the spectral coverage afforded by the gbm and the lat on glast will also provide powerful observational constraints . the low energy asymptotic forms in eqs . ( [ eq : ntic_limits ] ) and ( [ eq : thic_limits ] ) raise the possibility that inverse compton emission can eliminate the so - called `` line of death '' problem that is pervasive in synchrotron models . the ic mechanism can accommodate low energy asymptotic indices as flat as , with steeper indices easily being generated by appropriately distributed electrons . the domain of compatibility is indicated in fig . [ fig : batse_low_index ] , where around 98% of batse burst spectra in the spectroscopy catalogue of preece et al . ( 2000 ) can be accounted for by the inverse compton process . this is an enticing property , but is not a definitive statement of the operation of this mechanism in grb prompt emission regions . it must be emphasized that the establishment of an domain is contingent upon the soft photons that seed the upscattering assuming a narrow energy ( i.e. quasi - monoenergetic ) distribution . this caveat is exemplified by the ssc value of being identical to that for synchrotron radiation due to the breadth of the synchrotron seed spectrum . hence invoking a synchrotron self - compton model only dramatically impacts the operating range of and not the spectral shape in the very soft gamma - rays . given the discussion in section [ sec : anis_synch ] pertaining to synchrotron radiation , it is natural to ask whether anisotropies in either the soft photon or electron populations can dramatically alter the spectrum . the inverse compton photon production rate in eq . ( [ eq : ic_emiss ] ) implicitly assumes isotropy of the soft photons constituting . furthermore , it assumes that the angular distribution of electrons is not seriously deficient along the line of sight to the observer ; i.e. that the viewing perspective is not towards the side of an electron jet . calculating ic spectra for anisotropic distributions is quite involved , and moreover sensitive to the choice of geometry . yet spectral behavior depends mostly on the interplay between such geometry and the kinematics of ic scattering . if the electron distribution is highly collimated and the observer s direction lies outside the cone of jet collimation , then the kinematics of scattering requires a smaller angle of scattering in the electron rest frame to send the photon towards the observer . this small yields a lower final photon energy in the observer s frame , so that the emitted inverse compton spectrum is softer outside the jet core . the spectrum then flattens at lower energies ( e.g. see dermer & schlickeiser 1993 ; baring 1994 for blazar contexts ) due to the kinematic depletion of photons in the observer s direction , and can be even slightly flatter than . a problem with this scenario immediately arises : the energetics are dominated by emission in the jet cone so that assuming `` off - axis '' ic emission places more severe constraints on the grb radiative efficiency . note that fluctuations in the jet orientation or structure would lead to the emergent inverse compton spectra being highly variable both in flux ( observed in bursts and therefore not a problem , in principle ) and in spectral shape any structure such as the peak would be expected to vary in time rapidly in energy due to the scattering kinematics . while modest fluctuations ( as opposed to hard - to - soft evolution ) are present in gamma - ray burst prompt emission , detailed modeling would be required to discern whether they could match those produced by anisotropic scattering scenarios . any anisotropy in the soft photon population is not sampled until they become more collimated than the electrons ; only then does the emissivity change appreciably from eq . ( [ eq : ic_emiss ] ) . then , electrons within the photon `` cone '' of beaming experience an almost full range of incident photon angles in their rest frames , so that they contribute an inverse compton spectrum similar to that for isotropic soft photons . however , the majority of electrons lie outside this cone , and for these , collisions with photons are predominantly head - on , leading to a depletion of the softer photons from the overall inverse compton spectrum just as with electron collimation . hence , one expects that such soft photon anisotropies could in principal also flatten the spectrum below the peak . to provide a meaningful comparison with the preceding mechanisms , grb 910503 is again used as a focal point . yet , so as to provide a variation , here a case study for another egret burst , grb 910601 is also presented . results for the spectral fitting for synchrotron self - compton scattering are presented in figure [ fig : ssc_spec_dist ] . the `` fits '' were performed using the formalism presented in section [ sec : sscformalism ] , specifically with numerical evaluations of eqs . ( [ eq : sscfinal ] ) and ( [ eq : sscfinal_th_too ] ) . the figure displays the ssc emission spectra for the case of an isotropic electron distribution , for each burst . the solid curve depicts the case that would most closely explain the photon spectrum , with the overall normalization again being a free parameter . due to the strong dependence of the characteristic energy on electron lorentz factor , the values of required to move the ssc peak into the gamma - ray band lead to inferences of sub - gauss fields if . consider first grb 910503 . the most obvious features are that the `` fit '' is more difficult than for the emission mechanisms previously considered , with poor results for the lower energies much below the peak , and that the thermal contribution must be dramatically suppressed . the cause is easily identified : the ssc spectrum is intrinsically much broader than either the ic or synchrotron spectra . this is due to the characteristic energy for ssc embodied in the dimensionless parameter ( see equation [ [ eq : chidef ] ] ) scaling as ( or for the thermal case ) . consequently , small changes in the electron near the peak of the maxwell - boltzmann distribution incur large variations in the ssc photon energy so that the spectral peak is broad . this effect is manifested in the non - thermal ssc emissivity in eq . ( [ eq : sscfinal ] ) , which leads to the excesses evident around 30 kev in the `` fits '' of figure [ fig : ssc_spec_dist ] , and more profoundly for the thermal ssc contribution , which is extremely broad . from eq . ( [ eq : sscthlimits ] ) a notable conclusion can be drawn : that the thermal ssc spectrum in the depicted representation only starts to fall above around 30 mev , and then only slowly , i.e. it is blueward of the peak from the non - thermal ssc contribution . this drives the requisite dramatic suppression of the normalization of a thermal component to the underlying electron distribution , leading to the same issues ( but in this case more strikingly emphasized ) for acceleration from thermal energies that were discussed above . note that for values of , a comparably satisfactory ssc fit can be obtained with . observe that for such a large , the tail of the thermal maxwellian emerges above that of the non - thermal distribution for a small range of lorentz factors . the situation for grb 910601 is even more constrained , due to the markedly steeper gamma - ray spectrum . the thermal electrons must be even less populous to accommodate the comptel and egret data , and the non - thermal electrons must possess a very steep index , considerably steeper than the required for a synchrotron fit . in fact , both the synchrotron and ic models had much greater success fitting the grb 910601 data , though are not depicted . steepening the index to produces only marginal improvement in the fit . the inevitable conclusion of this analysis is that the breadth of the ssc emissivity for even monoenergetic electrons ( derived in equation ( [ eq : sscfinal_mono ] ) , and closely approximated by the case in figure [ fig : ssc_spec_dist ] ) is large enough to provide significant or severe problems for the ssc interpretation of prompt emission from gamma - ray bursts . this contention is not based solely on the grb 910503 and in particular the grb 910601 data ; the similarity of the spectra of grb 910814 ( schaefer et al . 1998 ) and grb 990123 ( briggs et al . 1999 ) to that of grb 910601 strongly suggests that a spectroscopic incompatibility of the ssc mechanism for prompt emission may be commonplace , specifically for differential burst spectra steeper than around above the peak . bursts such as the superbowl one ( grb 930131 ) with high energy index may be much more compatible with an ssc scenario . while this assessment may achieve closure in the swift / glast era , a caveat is possibly provided if the synchrotron continuum is strongly self - absorbed . this is the scenario envisaged by panaitescu & meszaros ( 2000 ) and liang , boettcher & kocevski ( 2003 ) , with the latter group suggesting that ssc with moderate optical depths ( ) could explain x - ray excesses seen in some burst spectra . with such strong absorption , the supply of soft photons for inverse compton upscattering is more sharply peaked than even the synchrotron continuum from electrons possessing a truncated power - law distribution , thereby approximately mimicking the pure inverse compton spectral shapes explored in section [ sec : part_comp ] . the requirement that self - absorption be effective in narrowing the synchrotron self - compton peak is that the turnover energy exceeds that of the synchrotron peak , which is below the x - ray band . using eq . ( [ eq : selfabs_energy ] ) , one quickly arrives at the requisite condition : @xmath27 for @xmath28 here is the energy of the peak in the prompt emission . for , one deduces that , so that optical depths can be tolerated for typical without pushing too low , i.e. into the mildly - relativistic regime . detailed computations of self - absorbed ssc spectra are deferred to future work . in addition to the soft gamma - ray spectral shape , a probing diagnostic on the emission mechanisms is whether their extrapolation down to the optical band can be accommodated by optical monitoring data for bursts . such broadband constraints turn out not to be profoundly limiting for synchrotron or pure inverse compton emission . however , they do impact ssc models , since these can potentially have a prominent synchrotron component in the optical that might violate upper bounds or , in the case of grb 990123 , prompt optical detections . it is straightforward to assess how optical band data can constrain ssc model parameters using the formalism derived here . first , the focus is on observational data . grb 990123 ( with redshift ) is to date the only burst with a prompt optical detection ( akerlof et al . 1999 ) , peaking at around magnitude 8.8 some 47 seconds after batse trigger and contemporaneous with the batse signal . these prompt observations were made by the robotic optical transient search experiment - i ( rotse - i ) . the recent bright burst grb 030329 ( redshift ) was only detected optically by rotse - iii in the afterglow epoch at visual magnitude 12.5 over an hour after the burst trigger on hete-2 ( smith et al . kehoe et al . ( 2001 ) report upper bounds of around magnitude 1315 to half a dozen bursts , in two cases starting just 15 - 20 seconds after trigger . these low limiting magnitudes can potentially be severely constraining on models with a high optical to gamma - ray flux expectation , though contemporaneous multiwavelength detections are requisite for unambiguous diagnostics . taking magnitude 9 as a benchmark optical signal for the purposes of exploring broadband constraints , a number that must be considered modulo the source distance and intrinsic luminosity , this corresponds to a differential photon flux of the order of photons sec@xmath29 @xmath30 kev@xmath29 at around 2ev . this can be compared with the batse spectrum of grb 910503 , a burst brighter than most of those in the sample of kehoe et al . ( 2001 ) , which has a time - integrated differential flux of photons sec@xmath29 @xmath30 kev@xmath29 at around 100 kev . the interpolation between such an optical magnitude and gamma - ray flux is a roughly spectrum . immediately , one infers that synchrotron and inverse compton models for the gamma - rays can accommodate such an optical magnitude ; only limiting magnitudes of around 19 or higher can provide conflicts for these mechanisms . note that a similar conclusion would apply for pitch angle synchrotron models , which possess very flat low energy asymptotic spectra like inverse compton scenarios . the situation for ssc broadband spectra is very different . the ratio of the energy of the ssc peak to that of the synchrotron peak is simply of the order , which would set the synchrotron peak in the infra - red for lorentz factors of , independent of the bulk flow . lower can move the synchrotron peak into the optical and even the uv . the ratio of emissivities at the synchrotron and ssc peaks is quickly determined from equations ( [ eq : ndotsyn_def ] ) and ( [ eq : ndotsscdef ] ) to be in the notations of this paper , i.e. with . hence , for , the peaks of the two components would lie roughly on an spectrum . moreover , opting for to position a synchrotron peak in the optical window , normalizing to the batse spectrum of grb 910503 would yield a count rate an order of magnitude below the magnitude 9 benchmark posited above . at face value , this would render the ssc scenario compatible with limiting magnitudes of around 11 for this source . yet comptonization of the spectrum is not evident in the gamma - rays , so it is inferred that , thereby increasing the ratio and lowering the permissible limiting magnitude for a viable ssc model . accordingly , for the only source with a prompt optical detection , grb 990123 , for the ssc model not to overestimate the observed optical flux , the synchrotron peak has to move somewhat out of the optical , though not much . this amounts to a modest constraint in phase space that can be quickly deduced , though at this juncture is not particularly enlightening ; such an exercise will be more timely once swift is providing a number of simultaneous hard x - ray and optical detections of bursts . the interpretative modeling presented here provides a strong indication that the synchrotron and inverse compton processes may provide a more viable explanation of prompt burst spectra than the ssc mechanism . the spectroscopic bias against ssc is due to its inherently broad character around the peak , and is only marginally constrained by broad - band considerations , i.e. limiting magnitudes in the optical band . however , strong self - absorption of the synchrotron component can provide an escape clause for ssc scenarios , though this imposes significant fine - tuning of parameter space . inverse compton provides an attractive possibility for the vast majority of bursts , as does pitch angle synchrotron emission , though quasi - isotropic synchrotron radiation may only be reconciled with around 2/3 of bursts at the lowest energies in the batse window . discrimination between these processes may prove possible after swift launches and in the glast era , though polarimetric measurements would enhance the diagnostic capability of observations . in analyzing ssc spectral models , attractively compact analytic forms for the synchrotron self - compton emissivity from thermal and truncated power - law electron distributions were developed ; these will prove useful for applications of ssc spectral models to a variety of astrophysical environments . a distinctive conclusion of this work is that agreeable `` fits '' with either synchrotron or inverse compton emission scenarios are only attainable if the underlying electron distribution is profoundly dominated by a non - thermal component . this result is in contradistinction with standard understanding of acceleration mechanisms such as the diffusive ( fermi ) process at shocks , where the non - thermal particles are drawn probabilistically from the shocked thermal population , and therefore are subdominant in number . this constraint has profound implications for any proposed grb model , providing a challenge for theorists in an era when computational resources are permitting new avenues for exploration using intensive numerical simulations of particle acceleration at shocks . this work was supported by the nsf extragalactic astrophysics program . matthew braby acknowledges the support of a shell foundation scholarship towards this research during his undergraduate degree at rice university . we thank the referee , rob preece , for insightful and constructive suggestions for polishing the paper , and nicole lloyd - ronning , jerry fishman and demos kazanas for reading through the manuscript and providing helpful comments . we also thank brad schaefer for supplying the grb 910503 compilation data prior to its publication in 1998 . in this appendix , developments leading to the form in eq . ( [ eq : sscfinal ] ) for the rate of synchrotron self - compton emission from a truncated power - law distribution of electrons are expounded . the starting point is the insertion of times the synchrotron rate in eq . ( [ eq : synchfinal ] ) , serving as the soft photon density , directly into the expression in eq . ( [ eq : ic_emiss ] ) for the inverse compton emissivity . this presents a triple integral that can be reduced to a compact single integration by appropriate manipulations . the lorentz factor of the electron that upscatters the synchrotron photons proves to be a less convenient integration variable than , which is defined in eq . ( [ eq : fdef ] ) . likewise , to proves expedient to use @xmath31 as an integration variable , rather than the synchrotron photon energy . performing these two changes of variables leads quickly to an ssc emissivity of @xmath32 where is the ssc rate scale factor given in eq . ( [ eq : ndotsscdef ] ) , and is the dimensionless ssc energy variable defined in eq . ( [ eq : chidef ] ) . here , and the maximum value of can be written @xmath33 the -integration over the inverse compton scattering kernel leads to a definition of : @xmath34 \label{eq : gzdef}\\[-5.5pt ] g_{\delta}(z ) & = & \dover{2z}{\delta+3}\ ; \biggl ( 2\log_ez + \dover{\delta-1}{\delta+3 } \biggr ) + \dover{2}{\delta+1 } - \dover{4z^2}{\delta+5}\quad , \nonumber\end{aligned}\ ] ] the integration can be reduced by noting that the definitions of and establish two critical values of , namely and . which delineate distinct dependences of the integrand on . the relative ordering of and depends on the value of assumed : and . this naturally divides the integration into two ranges , , and . the developments are straightforward , and are facilitated using an function defined via @xmath35 \label{eq : hzdef}\\[-5.5pt ] h_{\delta}(z ) & = & \dover{4z}{(\delta+3)^2}\ ; \biggl ( 2\log_ez + \dover{\delta-5}{\delta+3 } \biggr ) + \dover{4}{(\delta+1)^2 } - \dover{8z^2}{(\delta+5)^2}\quad .\nonumber\end{aligned}\ ] ] the case requires an integration by parts in one contributing interval , and one quickly arrives at the compact form @xmath36 i.e. eq . ( [ eq : sscfinal ] ) , using the following functional forms ( with and defined as just above ) : @xmath37\ , g_{\delta}(1 ) + h_{\delta}(1)\,\biggr\ } \;\ ; , \label{eq : pdef}\ ] ] for , and for @xmath38 + h_{\delta}(z)\,\biggr\ } \quad , \nonumber\\[-5.5pt ] \label{eq : qdef}\\[-5.5pt ] & = & - 2z\,\dover { \delta + 1}{(\delta + 3)^2}\biggl [ 2 \ , \log_ez - \dover{\delta + 11}{\delta + 3 } \biggr ] - 4 \dover{\delta -1}{(\delta + 1)^2 } -2 \dover{\log_ez}{\delta + 1 } + 2z^2\ , \dover{\delta + 1}{(\delta + 5)^2}\quad .\nonumber\end{aligned}\ ] ] observe that as , , as demanded by continuity of the spectrum . asymptotic dependences of the ssc rate on the upscattered photon energy , encapsulated in the proportionalities of eq . ( [ eq : ntssc_limits ] ) , can be simply derived . in the limit , both the and terms contribute to eq . ( [ eq : sscfinal_app ] ) , with the integrals being dominated by the domain . the bessel function can be replaced by its power - law asymptotic form , resulting in elementary integrals . for the integration converges to @xmath39 the case is less simple , though the integration over does not contribute significantly since it is exponentially suppressed by the dependence of the bessel function . the integration can be developed using identity 6.561.16 of gradshteyn and ryzhik ( 1980 ) , which can be specialized to the case appropriate for this analysis : @xmath40 the derivative of this with respect to can also be used to expedite the developments , and can be written in the form @xmath41 where is the digamma function . setting , since the upper limit to the can be replaced by infinity , one readily arrives at the result @xmath42 \label{eq : sscnt_highchi}\\[-5.5pt ] & & \hbox{\hskip 0pt } - \dover{1}{2 } \psi \biggl ( \dover{\delta}{4 } + \dover{19}{12}\biggr ) - \dover{1}{2 } \psi \biggl ( \dover{\delta}{4 } - \dover{1}{12}\biggr)\ , \biggr ] \ , g_{\delta}(1)+ h_{\delta}(1 ) \biggr\}\ ; \dover{\ndotssc}{\chi^{(\delta + 1)/2}}\quad , \quad \chi\,\gg\ , 1\;\ ; .\nonumber\end{aligned}\ ] ] the logarithmic term will clearly dominate for extremely large values of . the comparison of synchrotron self - compton ( ssc ) models invoking shock acceleration with burst spectra requires also the ssc rate for relativistically hot thermal particles , which implies using a four - dimensional integral defined by inserting the thermal contribution to eq . ( [ eq : elec_dist ] ) into eq . ( [ eq : synch_emiss ] ) and then the result subsequently into eq . ( [ eq : ic_emiss ] ) . rather than laboriously reducing the integral along the lines of the developments of appendix a , the result in eq . ( [ eq : sscfinal_app ] ) can be used to advantage . taking the limit generates the ssc emissivity for monoenergetic electrons with lorentz factor . the result can then be convolved with the thermal electron distribution to yield the required emissivity . as becomes very large , approaches zero at all positive less than unity , so that the integration over in eq . ( [ eq : sscfinal_app ] ) does not contribute . using the result @xmath43 the integration over the thermal electron distribution produces a double integral that can be evaluated by reversing the order of integrations , yielding @xmath26^{1/4}\bigr)\quad , \label{eq : sscfinal_th}\ ] ] where is given in eq . ( [ eq : ndotsscdef ] ) , , and @xmath44 \label{eq : caljdef}\\[-5.5pt ] & \equiv & 2\dover{e^{-\kappa}}{\kappa } \int_0^{\infty } \dover{e^{-v}s(v)}{\kappa + v}\ , dv \ ; = \ ; 2\dover{e^{-\kappa}}{\kappa^3 } \int_0^{\infty } \dover{e^{-v } \ , v^2\ , s(v)}{\kappa + v}\ , dv \quad , \nonumber\end{aligned}\ ] ] where @xmath45 the first integral defining can be expressed as a sum of incomplete gamma functions and one of their derivatives , using identities 3.381.3 and 4.358.1 of gradshteyn and ryzhik ( 1980 ) , though this does not expedite the evaluation of . however , the integral representation of the incomplete gamma function in 8.353.3 of gradshteyn and ryzhik ( 1980 ) proves useful , leading to the alternative integral forms posited on the second line of eq . ( [ eq : caljdef ] ) . these latter forms are readily amenable to numerical computation in a fashion similar to evaluation of euler s integral form for the gamma function . note that the first form on the second line is expedient for cases , while the second form on this line , which was obtained using the fact that the integrals of and on are identically zero , is advantageous for evaluation when . asymptotic forms for are simply obtained : for , and for . the appropriate asymptotic limits of eq . ( [ eq : sscfinal_th ] ) can then be quickly determined : @xmath46 ^ 2\;\chit^{-2/3}\;\ ; , & $ \quad \chit\ll 1\quad , \vphantom{\biggl(}$\cr \dover{8\pi}{\sqrt{5}}\ ; \dover{2^{9/10}}{\chit^{4/5}}\ ; \exp \biggl\ { -\dover{5}{2^{8/5}}\ ; \chit^{1/5 } \biggr\}\;\ ; , & $ \quad \chit\gg 1\quad , \vphantom{\biggl(}$\cr } \label{eq : sscthlimits}\ ] ] where the low argument power - law dependence for the bessel function is used when , and the method of steepest descents is employed to derive the case , for which the exponential portion of is sampled . note that the ssc emissivity for a monoenergetic electron distribution can be determined by omitting the convolution with the thermal distribution , so that eq . ( [ eq : qlimit ] ) can be used efficaciously to yield @xmath47 as is a monotonically decreasing function of on , for it flattens the ssc spectrum relative to that for synchrotron emission , and so effects a broadening of the peak . this monoenergetic formula can be convolved with arbitrary electron distributions to obtain resultant synchrotron emissivities . by the same token , eq . ( [ eq : sscfinal_th ] ) can be used to facilitate a laplace transform approach for special cases for the electron distribution . akerlof , c. , et al . 1999 , @xmath48398 , 400 . band , d. , et al . 1993 , @xmath48413 , 281 . baring , m. g. 1988a , @xmath48235 , 51 . baring , m. g. 1988b , @xmath48235 , 79 . baring , m. g. 1990 , @xmath48244 , 49 . baring , m. g. 1994 , in _ proc . of the second compton symposium _ , eds . gehrels , n. & norris , j. , ( aip conf . 304 , new york ) p. 639 . baring , m. g. 1995 , proc . of the erice nato asi _ currents in high energy astrophysics , eds . shapiro , m. , wefel , j. and silberberg , r. , ( kluwer academic , dordrecht ) p. 21 . baring , m. g. 1999 , in _ proc . of the 26th international cosmic ray conference _ , vol . iv , p. 5 , held in salt lake city , utah . baring , m. g. 2002 , pub . @xmath4819 , 60 . baring , m. g. , ellison , d. c. & jones , f. c. 1993 , @xmath48409 , 327 . baring , m. g. , ogilvie , k. w. , ellison , d. c. , & forsyth , r. j. 1997 , @xmath48476 , 889 . bekefi , g. 1966 , _ radiation processes in plasmas , _ ( wiley and sons , new york ) blandford , r. d. & eichler , d. 1987 , phys . @xmath48154 , 1 . brainerd , j. j. & lamb , d. q. 1987 , @xmath48313 , 231 . brainerd , j. j. & petrosian , v. 1987 , @xmath48320 , 703 . briggs , m. s. , et al . 1999 , @xmath48524 , 82 . crider , a. & liang , e. p. 1999 , @xmath48138 , 405 . de jager , o. c. , et al . 1996 , @xmath48457 , 253 . dermer , c. d. , chiang , j. & bttcher , m. 1999 , @xmath48513 , 656 . dermer , c. d. & schlickeiser , r. 1993 @xmath48416 , 209 . drury , l. oc . @xmath4846 , 973 . ellison , d. c. , baring , m. g. & jones , f. c. 1996 , @xmath48473 , 1029 . ellison , d. c. & double , g. p. 2002 , @xmath4818 , 213 . ellison , d. c. , jones , f. c. , & reynolds , s. p. 1990 , @xmath48360 , 702 . ellison , d. c. , mbius , e. , & paschmann , g. 1990 , @xmath48352 , 376 . epstein , r. i. 1973 , @xmath48183 , 593 . epstein , r. i. & petrosian , v. 1973 , @xmath48183 , 611 . forman , m. a. , jokipii , j. r. & owens , a. j. 1974 , @xmath48192 , 535 . gallant , y. a. , hoshino , m. , langdon , a. b. , arons , j. & max , c. e. 1992 , @xmath48391 , 73 . ghisellini , g. , celotti , a. & lazzati , d. 2000 , @xmath48313 , l1 . giacalone , j. , burgess , d. , schwartz , s. j. & ellison , d. c. 1992 , @xmath4819 , 433 . giacalone , j. , burgess , d. , schwartz , s. j. & ellison , d. c. 1993 , @xmath48402 , 550 . goodman , j. 1986 , @xmath48308 , l47 . gonzalez , m. m. , et al . 2003 , @xmath48424 , 749 . gosling , j. t. , et al . 1981 , @xmath4886 , 547 . gradshteyn , i. s. and ryzhik , i. m. 1980 , _ table of integrals , series and products _ , ( academic press , new york ) granot , j. , piran , t. & sari , r. 2000 , @xmath48534 , l163 . hoshino , m. , arons , j. , gallant , y. a. & langdon , a. b. 1992 , @xmath48390 , 454 . hoshino , m. & shimada , n. 2002 , @xmath48572 , 880 . jackson , j. d. 1975 , _ classical electrodynamics , _ ( wiley and sons , new york ) jokipii , j. r. , kta , j. , & giacalone , j. 1993 , @xmath4820 , 1759 . jones , f. c. 1968 , @xmath48167 , 1159 . jones , f. c. & ellison , d. c. 1991 , @xmath4858 , 259 . jones , f. c. , jokipii , j. r. & baring , m. g. 1998 , @xmath48509 , 238 . kang , h. & jones , t. w. 1997 , @xmath48476 , 875 . kehoe , r. , et al . 2001 , @xmath48554 , l159 . kucharek , h. , & scholer , m. 1995 , @xmath48100 , 1745 . liang , e. p. , boettcher , m. & kocevski , d. 2003 , in _ gamma - ray burst and afterglow astronomy 2001 _ , eds . g. r. ricker & r. k. vanderspek ( aip conf . 662 ) , p. 295 . liewer , p. c. , goldstein , b. e. & omidi , n. 1993 , @xmath4898 , 15,211 . lloyd , n. m. & petrosian , v. 2000 , @xmath48543 , 722 . medvedev , m. v. 2000 , @xmath48540 , 704 . mszros , p. 2001 , science @xmath48291 , 79 . mszros , p. , rees , m. j. & papathanassiou , h. 1994 , @xmath48432 , 181 . pacyski , b. 1986 , @xmath48308 , l43 . panaitescu , a. & mszros , p. 2000 , @xmath48544 , l17 . pavlov , g. g. & golenetskii , s. v. 1986 , @xmath48128 , 341 . piran , t. 1999 , phys . @xmath48314 , 575 . preece , r. d. , et al . 1998 , @xmath48506 , l23 . preece , r. d. , et al . 2000 , @xmath48126 , 19 . preece , r. d. , et al . 2002 , @xmath48581 , 1248 . rees , m. j. , & mszros 1992 , @xmath48258 , 41 . rybicki , g. b. & lightman , a. p. 1979 , _ radiative processes in astrophysics , _ ( wiley and sons , new york ) sari , r. & esin , a. 2001 , @xmath48548 , 787 . schaefer , b. e. , et al . 1998 , @xmath48492 , 696 . scholer , m. , trattner , k. j. & kucharek , h. 1992 , @xmath48395 , 675 . shimada , n. & hoshino , m. 2000 , @xmath48543 , l67 . shimada , n. , et al . 1999 , @xmath48264 , 481 . smith , d. a. et al . 2003 , @xmath48596 , l151 . tavani , m. 1996a , @xmath4876 , 3478 . tavani , m. 1996b , @xmath48466 , 768 . terasawa , t. , et al . 2001 , in proc . 27th international cosmic ray conference , ( copernicus gesellschaft , germany ) , sh2.3/sh119 [ see _ http://www.copernicus.org/icrc/program.htm_ ] trattner , k. j. & scholer , m. 1991 , @xmath4818 , 1817 . trattner , k. j. & scholer , m. 1993 , ann . geophysicae @xmath4811 , 774 . westfold , k. c. 1959 , @xmath48130 , 241 . zhang , b. & mszros , p. 2001 , @xmath48559 , 110 . _

the objective of this investigation is to study three - dimensional ( 3d ) flows of viscous incompressible fluids which are constructed to exhibit extreme growth of enstrophy . it is motivated by the question whether the solutions to the 3d incompressible navier - stokes system on unbounded or periodic domains corresponding to smooth initial data may develop a singularity in finite time @xcite . by formation of a `` singularity '' we mean the situation when some norms of the solution corresponding to smooth initial data have become unbounded after a finite time . this so - called `` blow - up problem '' is one of the key open questions in mathematical fluid mechanics and , in fact , its importance for mathematics in general has been recognized by the clay mathematics institute as one of its `` millennium problems '' @xcite . questions concerning global - in - time existence of smooth solutions remain open also for a number of other flow models including the 3d euler equations @xcite and some of the `` active scalar '' equations @xcite . while the blow - up problem is fundamentally a question in mathematical analysis , a lot of computational studies have been carried out since the mid-90s in order to shed light on the hydrodynamic mechanisms which might lead to singularity formation in finite time . given that such flows evolving near the edge of regularity involve formation of very small flow structures , these computations typically require the use of state - of - the - art computational resources available at a given time . the computational studies focused on the possibility of finite - time blow - up in the 3d navier - stokes and/or euler system include @xcite , all of which considered problems defined on domains periodic in all three dimensions . recent investigations by @xcite focused on the time evolution of vorticity moments and compared it with the predictions derived from analysis based on rigorous bounds . we also mention the studies by @xcite and @xcite , along with the references found therein , in which various complexified forms of the euler equation were investigated . the idea of this approach is that , since the solutions to complexified equations have singularities in the complex plane , singularity formation in the real - valued problem is manifested by the collapse of the complex - plane singularities onto the real axis . overall , the outcome of these investigations is rather inconclusive : while for the navier - stokes flows most of recent computations do not offer support for finite - time blow - up , the evidence appears split in the case of the euler system . in particular , the recent studies by @xcite and @xcite hinted at the possibility of singularity formation in finite time . in this connection we also mention the recent investigations by @xcite in which blow - up was observed in axisymmetric euler flows in a bounded ( tubular ) domain . a common feature of all of the aforementioned investigations was that the initial data for the navier - stokes or euler system was chosen in an ad - hoc manner , based on some heuristic arguments . on the other hand , in the present study we pursue a fundamentally different approach , proposed originally by @xcite and employed also by @xcite for a range of related problems , in which the initial data leading to the most singular behaviour is sought systematically via solution of a suitable variational optimization problem . we carefully analyze the time evolution induced by the extreme vortex states first identified by @xcite and compare it to the time evolution corresponding to a number of other candidate initial conditions considered in the literature @xcite . we demonstrate that the taylor - green vortex , studied in the context of the blow - up problem by @xcite , is in fact a particular member of the family of extreme vortex states maximizing the instantaneous rate of enstrophy production in the limit of vanishing enstrophy . in addition , based on these findings , we identify the set of initial data , parameterized by its energy and enstrophy , for which one can a priori guarantee global - in - time existence of smooth solutions . this result therefore offers a physically appealing interpretation of an `` abstract '' mathematical theorem concerning global existence of classical solutions corresponding to `` small '' initial data @xcite . we also emphasize that , in order to establish a direct link with the results of the mathematical analysis discussed below , in our investigation we follow a rather different strategy than in most of the studies referenced above . while these earlier studies relied on data from a relatively small number of simulations performed at a high ( at the given time ) resolution , in the present investigation we explore a broad range of cases , each of which is however computed at a more moderate resolution ( or , equivalently , reynolds number ) . with such an approach to the use of available computational resources , we are able to reveal trends resulting from the variation of parameters which otherwise would be hard to detect . systematic computations conducted in this way thus allow us to probe the sharpness of the mathematical analysis relevant to the problem . the question of regularity of solution to the navier - stokes system is usually addressed using `` energy '' methods which rely on finding upper bounds ( with respect to time ) on certain quantities of interest , typically taken as suitable sobolev norms of the solution . a key intermediate step is obtaining bounds on the rate of growth of the quantity of interest , a problem which can be studied with ode methods . while for the navier - stokes system different norms of the velocity gradient or vorticity can be used to study the regularity of solutions , the use of enstrophy @xmath5 ( see equation ) is privileged by the well - known result of @xcite , where it was established that if the uniform bound @xmath6 holds , then the regularity of the solution @xmath7 is guaranteed up to time @xmath8 ( to be precise , the solution remains in a suitable gevrey class ) . from the computational point of view , the enstrophy @xmath9 is thus a convenient indicator of the regularity of solutions , because in the light of , singularity formation must manifest itself by the enstrophy becoming infinite . while characterization of the maximum possible finite - time growth of enstrophy in the 3d navier - stokes flows is the ultimate objective of this research program , analogous questions can also be posed in the context of more tractable problems involving the one - dimensional ( 1d ) burgers equation and the two - dimensional ( 2d ) navier - stokes equation . although global - in - time existence of the classical ( smooth ) solutions is well known for both these problems @xcite , questions concerning the sharpness of the corresponding estimates for the instantaneous and finite - time growth of various quantities are relevant , because these estimates are obtained using essentially the same methods as employed to derive their 3d counterparts . since in 2d flows on unbounded or periodic domains the enstrophy may not increase ( @xmath10 ) , the relevant quantity in this case is the palinstrophy @xmath11 , where @xmath12 is the vorticity ( which reduces to a pseudo - scalar in 2d ) . different questions concerning sharpness of estimates addressed in our research program are summarized together with the results obtained to date in table [ tab : estimates ] . we remark that the best finite - time estimate for the 1d burgers equation was found _ not _ to be sharp using the initial data obtained from both the instantaneous and the finite - time variational optimization problems @xcite . on the other hand , in 2d the bounds on both the _ instantaneous _ and _ finite - time _ growth of palinstrophy were found to be sharp and , somewhat surprisingly , both estimates were realized by the same family of incompressible vector fields parameterized by energy @xmath13 and palinstrophy @xmath14 , obtained as the solution of an _ instantaneous _ optimization problem @xcite . it is worth mentioning that while the estimate for the instantaneous rate of growth of palinstrophy @xmath15 ( see table [ tab : estimates ] ) was found to be sharp with respect to variations in palinstrophy , the estimate is in fact not sharp with respect to the prefactor @xmath16 @xcite , with the correct prefactor being of the form @xmath17 . we add that what distinguishes the 2d problem , in regard to both the instantaneous and finite - time bounds , is that the rhs of these bounds are expressed in terms of two quantities , namely , energy @xmath13 and enstrophy @xmath5 , in contrast to the enstrophy alone appearing in the 1d and 3d estimates . as a result , the 2d instantaneous optimization problem had to be solved subject to _ two _ constraints . in the present investigation we advance the research program summarized in table [ tab : estimates ] by assessing to what extent the finite - time growth of enstrophy predicted by the analytic estimates and can be actually realized by flow evolution starting from different initial conditions , including the extreme vortex states found by @xcite to saturate the instantaneous estimate . the key finding is that , at least for the range of modest enstrophy values we considered , the growth of enstrophy corresponding to this initial data , which has the form of two colliding axisymmetric vortex rings , is rapidly depleted and there is no indication of singularity formation in finite time . thus , should finite - time singularity be possible in the navier - stokes system , it is unlikely to result from initial conditions instantaneously maximizing the rate of growth of enstrophy . we also provide a comprehensive characterization of the extreme vortex states which realize estimate together with the resulting flow evolutions . the structure of the paper is as follows : in the next section we present analytic estimates on the instantaneous and finite - time growth of enstrophy in 3d flows . in [ sec:3d_instopt ] we formulate the variational optimization problems which will be solved to find the vortex states with the largest rate of enstrophy production and in [ sec:3d_instopt_e0to0 ] we provide an asymptotic representation for these optimal states in the limit of vanishing enstrophy . in [ sec:3d_instopt_e ] we present numerically computed extreme vortex states corresponding to intermediate and large enstrophy values , while in [ sec : timeevolution ] we analyze the temporal evolution corresponding to different initial data in order to compare it with the predictions of estimates and . our findings are discussed in [ sec : discuss ] , whereas conclusions and outlook are deferred to [ sec : final ] . l|c|c & estimate + & realizability + & & + & & + & @xmath18 + @xmath19 & + 2d navier - stokes + finite - time & @xmath20 + @xmath21 & + & & + 3d navier - stokes + finite - time & @xmath22 + @xmath23 $ ] & + we consider the incompressible navier - stokes system defined on the 3d unit cube @xmath24 ^ 3 $ ] with periodic boundary conditions [ eq : nse3d ] @xmath25 where the vector @xmath26 $ ] is the velocity field , @xmath27 is the pressure and @xmath28 is the coefficient of kinematic viscosity ( hereafter we will set @xmath29 which is the same value as used in the seminal study by @xcite ) . the velocity gradient @xmath30 is the tensor with components @xmath31_{ij } = \partial_j u_i$ ] , @xmath32 . the fluid density @xmath33 is assumed to be constant and equal to unity ( @xmath34 ) . the relevant properties of solutions to system can be studied using energy methods , with the energy @xmath35 and its rate of growth given by @xmath36 where `` @xmath37 '' means `` equal to by definition '' . the enstrophy @xmath38 and its rate of growth are given by @xmath39 for incompressible flows with periodic boundary conditions we also have the following identity @xcite @xmath40 hence , combining , the energy and enstrophy satisfy the system of ordinary differential equations @xmath41 a standard approach at this point is to try to upper - bound @xmath2 and using standard techniques of functional analysis it is possible to obtain the following well - known estimate in terms of @xmath13 and @xmath5 @xcite @xmath42 for @xmath43 an absolute constant . a related estimate expressed entirely in terms of the enstrophy @xmath5 is given by @xmath44 by simply integrating the differential inequality in with respect to time we obtain the finite - time bound @xmath45 which clearly becomes infinite at time @xmath46 $ ] . thus , based on estimate , it is not possible to establish the boundedness of the enstrophy @xmath47 globally in time and hence the regularity of solutions . therefore , the question about the finite - time singularity formation can be recast in terms of whether or not estimate can be saturated . by this we mean the existence of initial data with enstrophy @xmath48 such that the resulting time evolution realizes the largest growth of enstrophy @xmath47 allowed by the right - hand side ( rhs ) of estimate . a systematic search for such most singular initial data using variational optimization methods is the key theme of this study . although different notions of sharpness of an estimate can be defined , e.g. , sharpness with respect to constants or exponents in the case of estimates in the form of power laws , the precise notion of sharpness considered in this study is the following [ def : notionsharpness ] given a parameter @xmath49 and maps @xmath50 , the estimate @xmath51 is declared sharp in the limit @xmath52 if and only if @xmath53 from this definition , the sharpness of estimates in the form @xmath54 for some @xmath55 and @xmath56 can be addressed in the limit @xmath57 by studying the adequacy of the exponent @xmath58 . the question of sharpness of estimate was addressed in the seminal study by @xcite , see also @xcite , who constructed a family of divergence - free velocity fields saturating this estimate . more precisely , these vector fields were parameterized by their enstrophy and for sufficiently large values of @xmath5 the corresponding rate of growth @xmath1 was found to be proportional to @xmath59 . therefore , in agreement with definition [ def : notionsharpness ] , estimate was declared sharp up to a numerical prefactor . however , the sharpness of the instantaneous estimate alone does not allow us to conclude about the possibility of singularity formation , because for this situation to occur , a sufficiently large enstrophy growth rate would need to be sustained over a _ finite _ time window @xmath60 . in fact , assuming the instantaneous rate of growth of enstrophy in the form @xmath61 for some @xmath62 , any exponent @xmath63 will produce blow - up of @xmath47 in finite time if the rate of growth is sustained . the fact that there is no blow - up for @xmath64 follows from grnwall s lemma and the fact that one factor of @xmath5 in can be bounded in terms of the initial energy using as follows @xmath65 \leq \frac{1}{2\nu } \k(0 ) . \label{eq : kt}\ ] ] this relation also leads to an alternative form of the estimate for the finite - time growth of enstrophy , namely @xmath66 \label{eq : evs_t_fixe}\end{aligned}\ ] ] which is more convenient than from the computational point of view and will be used in the present study . we note , however , that since the rhs of this inequality can not be expressed entirely in terms of properties of the initial data , this is _ not _ in fact an a priori estimate . estimate also allows us to obtain a condition on the size of the initial data , given in terms of its energy @xmath67 and enstrophy @xmath68 , which guarantees that smooth solutions will exist globally in time , namely , @xmath69 from which it follows that @xmath70 thus , flows with energy and enstrophy satisfying inequality are guaranteed to be smooth for all time , in agreement with the regularity results available under the assumption of small initial data @xcite . sharpness of instantaneous estimate , in the sense of definition [ def : notionsharpness ] , can be probed by constructing a family of `` extreme vortex states '' @xmath71 which , for each @xmath72 , have prescribed enstrophy @xmath73 and produce the largest possible rate of growth of enstrophy @xmath74 . given the form of , the fields @xmath71 can be expected to exhibit ( at least piecewise ) smooth dependence on @xmath0 and we will refer to the mapping @xmath75 as a `` maximizing branch '' . thus , information about the sharpness of estimate can be deduced by analyzing the relation @xmath0 versus @xmath74 obtained for a possibly broad range of enstrophy values . a maximizing branch is constructed by finding , for different values of @xmath0 , the extreme vortex states @xmath71 as solutions of a variational optimization problem defined below . hereafter , @xmath76 will denote the sobolev space of functions with square - integrable second derivatives endowed with the inner product @xcite @xmath77 where @xmath78 are parameters with the meaning of length scales ( the reasons for introducing these parameters in the definition of the inner product will become clear below ) . the inner product in the space @xmath79 is obtained from by setting @xmath80 . the notation @xmath81 will refer to the sobolev space @xmath76 of functions with zero mean . for every fixed value @xmath0 of enstrophy we will look for a divergence - free vector field @xmath71 maximizing the objective function @xmath82 defined in . we thus have the following [ pb : maxdedt_e ] given @xmath83 and the objective functional @xmath84 from equation , find @xmath85 which will be solved for enstrophy @xmath0 spanning a broad range of values . this approach was originally proposed and investigated by @xcite . in the present study we extend and generalize these results by first showing how other fields considered in the context of the blow - up problem for both the euler and navier - stokes system , namely the taylor - green vortex , also arise from variational problem [ pb : maxdedt_e ] . we then thoroughly analyze the time evolution corresponding to our extreme vortex states and compare it with the predictions of the finite - time estimates and . as discussed at the end of this section , some important aspects of our approach to solving problem [ pb : maxdedt_e ] are also quite different from the method adopted by @xcite . the smoothness requirement in the statement of problem [ pb : maxdedt_e ] ( @xmath86 ) follows from the definition of the objective functional @xmath84 in equation , where both the viscous term @xmath87 contain derivatives of order up to two . the constraint manifold @xmath88 can be interpreted as an intersection of the manifold ( a subspace ) @xmath89 of divergence - free fields and the manifold @xmath90 of fields with prescribed enstrophy @xmath0 . the structure of these constraint manifolds is reflected in the definition of the corresponding projections @xmath91 ( without a subscript , @xmath92 refers to a generic manifold ) which is given for each of the two constraints as follows : * ( _ div - free_)-constraint : the projection of a field @xmath93 onto the subspace of solenoidal fields @xmath94 is performed using the helmholtz decomposition ; accordingly , every zero - mean vector field @xmath86 can be decomposed uniquely as @xmath95 where @xmath96 and @xmath97 are scalar and vector potentials , respectively ; it follows from the identity @xmath98 , valid for any sufficiently smooth vector field @xmath97 , that the projection @xmath99 is given simply by @xmath100 and is therefore calculated as @xmath101 , \label{eq : phodge}\ ] ] where @xmath102 is the inverse laplacian associated with the periodic boundary conditions ; the operator @xmath103 is also known as the leray - helmholtz projector . * @xmath104-constraint : the projection onto the manifold @xmath105 is calculated by the normalization @xmath106 thus , composing with , the projection onto the manifold @xmath88 defined in problem [ pb : maxdedt_e ] is constructed as @xmath107 this approach , which was already successfully employed by @xcite , allows one to enforce the enstrophy constraint essentially with the machine precision . for a given value of @xmath0 , the maximizer @xmath71 can be found as @xmath108 using the following iterative procedure representing a discretization of a gradient flow projected on @xmath109 @xmath110 where @xmath111 is an approximation of the maximizer obtained at the @xmath112-th iteration , @xmath113 is the initial guess and @xmath114 is the length of the step in the direction of the gradient . it is ensured that the maximizers @xmath71 obtained for different values of @xmath0 lie on the same maximizing branch by using the continuation approach , where the maximizer @xmath71 is employed as the initial guess @xmath113 to compute @xmath115 at the next enstrophy level for some sufficiently small @xmath116 . as will be demonstrated in [ sec:3d_instopt_e0to0 ] , in the limit @xmath117 optimization problem [ pb : maxdedt_e ] admits a discrete family of closed - form solutions and each of these vortex states is the limiting ( initial ) member @xmath118 of the corresponding maximizing branch . as such , these limiting extreme vortex states are used as the initial guesses @xmath113 for the calculation of @xmath119 , i.e. , they serve as `` seeds '' for the calculation of an entire maximizing branch ( as discussed in [ sec : discuss ] , while there exist alternatives to the continuation approach , this technique in fact results in the fastest convergence of iterations and also ensures that all computed extreme vortex states lie on a single branch ) . the procedure outlined above is summarized as algorithm [ alg : optimalg ] , whereas all details are presented below . set @xmath120 set @xmath121 @xmath122 @xmath123 @xmath124 compute @xmath125 compute the @xmath126 gradient @xmath127 , see equation compute the sobolev gradient @xmath128 , see equation compute the step size @xmath114 , see equation set @xmath129 , see equations set @xmath130 compute the ` relative error ` @xmath131 set @xmath132 set @xmath133 a key step of algorithm [ alg : optimalg ] is the evaluation of the gradient @xmath134 of the objective functional @xmath135 , cf . , representing its ( infinite - dimensional ) sensitivity to perturbations of the velocity field @xmath93 , and it is essential that the gradient be characterized by the required regularity , namely , @xmath136 . this is , in fact , guaranteed by the riesz representation theorem @xcite applicable because the gteaux differential @xmath137 , defined as @xmath138 $ ] for some perturbation @xmath139 , is a bounded linear functional on @xmath140 . the gteaux differential can be computed directly to give @xmath141\,d{\mathbf{x}}-2\nu\int_{\omega}{\delta}^2{\mathbf{u}}\cdot{\mathbf{u}}'\,d{\mathbf{x}}\label{eq : dr}\ ] ] from which , by the riesz representation theorem , we obtain @xmath142 with the riesz representers @xmath134 and @xmath143 being the gradients computed with respect to the @xmath144 and @xmath126 topology , respectively , and the inner products defined in . we remark that , while the @xmath144 gradient is used exclusively in the actual computations , cf . , the @xmath126 gradient is computed first as an intermediate step . identifying the gteaux differential with the @xmath126 inner product and performing integration by parts yields @xmath145 similarly , identifying the gteaux differential with the @xmath144 inner product , integrating by parts and using , we obtain the required @xmath144 gradient @xmath146 as a solution of the following elliptic boundary - value problem @xmath147 \nabla\r = \nabla^{l_2 } \r \qquad \text{in } \ \omega , \\ & \text{periodic boundary conditions}. \end{aligned } \label{eq : gradrh2}\ ] ] the gradient fields @xmath143 and @xmath134 can be interpreted as infinite - dimensional sensitivities of the objective function @xmath135 , cf . , with respect to perturbations of the field @xmath93 . while these two gradients may point towards the same local maximizer , they represent distinct `` directions '' , since they are defined with respect to different topologies ( @xmath126 vs. @xmath144 ) . as shown by @xcite , extraction of gradients in spaces of smoother functions such as @xmath76 can be interpreted as low - pass filtering of the @xmath126 gradients with parameters @xmath148 and @xmath149 acting as cut - off length - scales and the choice of their numerical values will be discussed in [ sec:3d_instopt_e ] . the step size @xmath114 in algorithm is computed as @xmath150 \right\}\ ] ] which is done using a suitable derivative - free line - search algorithm @xcite . equation can be interpreted as a modification of a standard line search method where the optimization is performed following an arc ( a geodesic ) lying on the constraint manifold @xmath109 , rather than a straight line . this approach was already successfully employed to solve similar problems in @xcite . it ought to be emphasized here that the approach presented above in which the projections and gradients are obtained based on the infinite - dimensional ( continuous ) formulation to be discretized only at the final stage is fundamentally different from the method employed in the original study by @xcite in which the optimization problem was solved in a fully discrete setting ( the two approaches are referred to as `` optimize - then - discretize '' and `` discretize - then - optimize '' , respectively , cf . a practical advantage of the continuous ( `` optimize - then - discretize '' ) formulation used in the present work is that the expressions representing the sensitivity of the objective functional @xmath84 , i.e. the gradients @xmath151 and @xmath146 , are independent of the specific discretization approach chosen to evaluate them . this should be contrasted with the discrete ( `` discretize - then - optimize '' ) formulation , where a change of the discretization method would require rederivation of the gradient expressions . in addition , the continuous formulation allows us to strictly enforce the regularity of maximizers required in problem [ pb : maxdedt_e ] . finally and perhaps most importantly , the continuous formulation of the maximization problem makes it possible to obtain elegant closed - form solutions of the problem in the limit @xmath117 , which is done in [ sec:3d_instopt_e0to0 ] below . these analytical solutions will then be used in [ sec:3d_instopt_e ] to guide the computation of maximizing branches by numerically solving problem [ pb : maxdedt_e ] for a broad range of @xmath0 , as outlined in algorithm [ alg : optimalg ] . it is possible to find analytic solutions to problem [ pb : maxdedt_e ] in the limit @xmath152 using perturbation methods . to simplify the notation , in this section we will drop the subscript @xmath0 when referring to the optimal field . the euler - lagrange system representing the first - order optimality conditions in optimization problem [ pb : maxdedt_e ] is given by @xcite [ eq : kkt_e ] @xmath153 where @xmath154 and @xmath155 are the lagrange multipliers associated with the constraints defining the manifold @xmath88 , and @xmath156 , given by @xmath157 is the bilinear form from equation . using the formal series expansions with @xmath158 [ eq : series3d ] @xmath159 in and collecting terms proportional to different powers of @xmath160 , it follows from that , at every order @xmath161 in @xmath160 , we have @xmath162 similarly , equation leads to @xmath163 at every order @xmath164 in @xmath160 . it then follows from equation that @xmath165\e_0^{2\alpha } + \ldots \\ & = & \e_0,\end{aligned}\ ] ] which , for @xmath166 , forces @xmath167 . hence , @xmath168 , @xmath169 and @xmath170 . the systems at orders @xmath171 and @xmath172 are given by : [ eq : maxdedt_asympt_1 ] @xmath173 [ eq : maxdedt_asympt_2 ] @xmath174 where the fact that @xmath175 for all @xmath176 has been used . while continuing this process to larger values of @xmath164 may lead to some interesting insights , for the purpose of this investigation it is sufficient to truncate expansions at the order @xmath177 . the corresponding approximation of the objective functional then becomes @xmath178 it is worth noting that , in the light of relation , the maximum rate of growth of enstrophy in the limit of small @xmath0 is in fact negative , meaning that , for sufficiently small @xmath0 , the enstrophy itself is a decreasing function for all times . this observation is consistent with the small - data regularity result discussed in introduction . as regards problem defining the triplet @xmath179 , taking the divergence of equation and using the condition @xmath180 leads to the laplace equation @xmath181 in @xmath182 . since for zero - mean functions defined on @xmath182 , @xmath183 , it follows that @xmath184 and equation is reduced to the eigenvalue problem @xmath185 with @xmath186 satisfying the incompressibility condition . direct calculation using equation and condition leads to an asymptotic expression for the objective functional in the limit of small enstrophy @xmath187 solutions to the eigenvalue problem in equation can be found using the fourier expansion of @xmath186 given as ( with hats denoting fourier coefficients ) @xmath188 where @xmath189 is a set of wavevectors @xmath190 for which @xmath191 . the eigenvalue problem then becomes @xmath192\widehat{{\mathbf{u}}}_1({\mathbf{k } } ) & = & 0 \qquad\forall\,{\mathbf{k}}\in{\mathcal{w } } , \\ \widehat{{\mathbf{u}}}_1({\mathbf{k}})\cdot{\mathbf{k } } & = & 0 \qquad\forall\,{\mathbf{k}}\in{\mathcal{w}},\end{aligned}\ ] ] with solutions obtained by choosing , for any @xmath193 , a set of wavevectors with the following structure @xmath194 and @xmath195 with an appropriate form satisfying the incompressibility condition @xmath196 . for the solutions to equation constructed in such manner it then follows that @xmath197 and the optimal asymptotic value of @xmath84 is given by @xmath198 since the fields @xmath186 are real - valued , their fourier modes must satisfy @xmath199 , where @xmath200 denotes the complex conjugate ( c.c . ) of @xmath201 . depending on the choice of @xmath202 , a number of different solutions of can be constructed and below we focus on the following three most relevant cases characterized by the largest values of @xmath203 : 1 . @xmath204 , where @xmath205 , @xmath206 , is the @xmath207 unit vector of the canonical basis of @xmath208 ; the most general solution can then be constructed as @xmath209 with the complex - valued constant vectors @xmath210 $ ] , @xmath211 $ ] and @xmath212 $ ] suitably chosen so that @xmath170 ; hereafter we will use the values @xmath213 ; it follows that @xmath214 @xmath215 , and the optimal asymptotic value of @xmath84 obtained from equation is given by @xmath216 [ c1 ] 2 . @xmath217 , where @xmath218 denotes the set whose elements are the negatives of the elements of set @xmath219 , for @xmath220 ; the most general solution can be then constructed as @xmath221\cdot{\mathbf{x } } } } + \mathbf{b}{\textrm{e}^{2\pi i[1,-1,0]\cdot{\mathbf{x } } } } + \mathbf{c}{\textrm{e}^{2\pi i[1,0,1]\cdot{\mathbf{x } } } } + \nonumber \\ & & \mathbf{d}{\textrm{e}^{2\pi i[1,0,-1]\cdot{\mathbf{x } } } } + \mathbf{e}{\textrm{e}^{2\pi i[0,1,1]\cdot{\mathbf{x } } } } + \mathbf{f}{\textrm{e}^{2\pi i[0,1,-1]\cdot{\mathbf{x } } } } + \textrm{c.c . } \label{eq : uvec_3d_k2}\end{aligned}\ ] ] with the constants @xmath222 suitably chosen so that @xmath223 = 0 $ ] , @xmath224 = 0 , \ldots,\mathbf{f}\cdot[0,1,-1 ] = 0 $ ] , which ensures that incompressibility condition is satisfied , and that @xmath170 ; in this case , @xmath225 , @xmath226 , and the optimal asymptotic value of @xmath84 is @xmath227 [ c2 ] 3 . @xmath228 for @xmath229 ; the most general solution can then be constructed as @xmath230\cdot{\mathbf{x } } } } + \mathbf{b}{\textrm{e}^{2\pi i[-1,1,1]\cdot{\mathbf{x } } } } + \nonumber \\ & & \mathbf{c}{\textrm{e}^{2\pi i[1,-1,1]\cdot{\mathbf{x } } } } + \mathbf{d}{\textrm{e}^{2\pi i[1,1,-1]\cdot{\mathbf{x } } } } + \textrm{c.c . } \label{eq : uvec_3d_k3}\end{aligned}\ ] ] with the constants @xmath231 suitably chosen so that @xmath232 = 0 $ ] , @xmath233 = 0 $ ] , @xmath234 = 0 $ ] and @xmath235 = 0 $ ] , which ensures that incompressibility condition is satisfied , and that @xmath170 ; in this case , @xmath236 , @xmath237 , and the optimal asymptotic value of @xmath84 is @xmath238 [ c3 ] the three constructions of the extremal field @xmath186 given in , and are all defined up to arbitrary shifts in all three directions , reflections with respect to different planes and rotations by angles which are multiples of @xmath239 about the different axes . as a result of this nonuniqueness , there is some freedom in choosing the constants @xmath240 . given that the optimal asymptotic value of @xmath84 depends exclusively on the wavevector magnitude @xmath241 , cf . , any combination of constants @xmath240 will produce the _ same _ optimal rate of growth of enstrophy . thus , to fix attention , in our analysis we will set @xmath242 in case ( i ) , @xmath243 in case ( [ c2 ] ) and @xmath244 in case ( [ c3 ] ) . with these choices , the contribution from each component of the field @xmath186 to the total enstrophy is the same . the maximum ( i.e. , least negative ) value of @xmath84 can be thus obtained by choosing the smallest possible @xmath245 . this maximum is achieved in case ( [ c1 ] ) with the wavevectors @xmath246 $ ] , @xmath247 $ ] , @xmath248 $ ] , and @xmath249 , @xmath250 and @xmath251 , for which @xmath252 . because of this maximization property , this is the field we will focus on in our analysis in [ sec:3d_instopt_e ] and [ sec : timeevolution ] . the three fields constructed in , and are visualized in figure [ fig : maxdedt_vortexcells ] . this analysis is performed using the level sets @xmath253 defined as @xmath254 for a suitable function @xmath255 . in figures [ fig : maxdedt_vortexcells](a c ) we choose @xmath256 with @xmath257 . to complement this information , in figures [ fig : maxdedt_vortexcells](d f ) we also plot the isosurfaces and cross - sectional distributions of the @xmath258 component of the field @xmath186 . the fields shown in figure [ fig : maxdedt_vortexcells ] reveal interesting patterns involving well - defined `` vortex cells '' . more specifically , we see that in case ( [ c1 ] ) , given by equation and shown in figures [ fig : maxdedt_vortexcells](a , d ) , the vortex cells are staggered with respect to the orientation of the cubic domain @xmath182 in all three planes , whereas in case ( [ c3 ] ) , given by equation and shown in figures [ fig : maxdedt_vortexcells](c , f ) , the vortex cells are aligned with the domain @xmath182 in all three planes . on the other hand , in case ( [ c2 ] ) , given by equation and shown in figures [ fig : maxdedt_vortexcells](b , e ) , the vortex cells are staggered in one plane and aligned in another with the arrangement in the third plane resulting from the arrangement in the first two . these geometric properties are also reflected in the @xmath258-component of the field @xmath186 which is independent of @xmath258 in cases ( [ c1 ] ) and ( [ c2 ] ) , but exhibits , respectively , a staggered and aligned arrangement of the cells in the @xmath259 plane in these two cases . in case ( [ c3 ] ) the cells exhibit an aligned arrangement in all three planes . the geometric properties of the extreme vortex states obtained in the limit @xmath117 are summarized in table [ tab : e0 ] . we remark that an analogous structure of the optimal fields , featuring aligned and staggered arrangements of vortex cells in the limiting case , was also discovered by @xcite in their study of the maximum palinstrophy growth in 2d . while due to a smaller spatial dimension only two optimal solutions were found in that study , the one characterized by the staggered arrangement also lead to a larger ( less negative ) rate of palinstrophy production . l|c|c|c|c case + & formula for the velocity field & arrangement of cells in @xmath259 plane & dependence of @xmath258 component of @xmath186 on @xmath258 & remarks + ( [ c1 ] ) + & & staggered & uniform & staggered abc flow + ( [ c2 ] ) + & & aligned & uniform & aligned abc flow + ( [ c3 ] ) + & & aligned & cell - like & taylor - green vortex + it is also worth mentioning that the initial data for two well - known flows , namely , the arnold - beltrami - childress ( abc ) flow @xcite and the taylor - green flow @xcite , are in fact particular instances of the optimal field @xmath186 corresponding to , respectively , cases ( [ c1 ] ) and ( [ c3 ] ) . following the notation of @xcite , general abc flows are characterized by the following velocity field @xmath260 where @xmath261 , @xmath262 and @xmath263 real constants . the vector field in equation can be obtained from equation by choosing @xmath264 $ ] , @xmath265 $ ] and @xmath266 $ ] . by analogy , we will refer to the state described by as the `` aligned abc flow '' . likewise , the well - known taylor - green vortex can be obtained as a particular instance of the field @xmath186 from equation again using a suitable choice of the constants @xmath267 . traditionally , the velocity field @xmath268 $ ] characterizing the taylor - green vortex is defined as @xcite @xmath269 for @xmath270 . for given values of @xmath271,@xmath272 and @xmath273 in , the corresponding constants @xmath267 in can be found by separating them into their real and imaginary parts denoted , respectively , @xmath274 and @xmath275 . then , after choosing @xmath276,\ ] ] the imaginary parts can be determined by solving the following system of linear equations @xmath277 \left [ \begin{array}{c } { \mathbf{a}_{\textrm{im } } } \\ { \mathbf{b}_{\textrm{im } } } \\ { \mathbf{c}_{\textrm{im } } } \\ { \mathbf{d}_{\textrm{im } } } \end{array } \right ] = \left [ \begin{array}{c } \mathbf{0 } \\ \gamma_1\mathbf{e}_1 \\ \gamma_2\mathbf{e}_2 \\ \gamma_3\mathbf{e}_3 \end{array } \right],\ ] ] where @xmath278 is the @xmath279 identity matrix . the values of @xmath280 are thus given by @xmath281,\ , { \mathbf{b}_{\textrm{im } } } = -\frac{1}{8}\left [ \begin{array}{c}-\gamma_1 \\ \gamma_2 \\ \gamma_3 \end{array } \right],\ , { \mathbf{c}_{\textrm{im } } } = -\frac{1}{8}\left [ \begin{array}{c}\gamma_1 \\ -\gamma_2 \\ \gamma_3 \end{array } \right],\ , { \mathbf{d}_{\textrm{im } } } = -\frac{1}{8}\left [ \begin{array}{c}\gamma_1 \\ \gamma_2 \\ -\gamma_3 \end{array } \right].\ ] ] a typical choice of the parameters used in the numerical studies performed by @xcite and @xcite is @xmath282 and @xmath283 . we remark that the taylor - green vortex has been employed as the initial data in a number of studies aimed at triggering singular behaviour in both the euler and navier - stokes systems @xcite . it is therefore interesting to note that it arises in the limit @xmath117 as one of the extreme vortex states in the variational formulation considered in the present study . it should be emphasized , however , that out of the three optimal states identified above ( see table [ tab : e0 ] ) , the taylor - green vortex is characterized by the smallest ( i.e. , the most negative ) instantaneous rate of enstrophy production @xmath1 . on the other hand , we are not aware of any prior studies involving abc flows in the context of extreme behaviour and potential singularity formation . the time evolution corresponding to these states and some other initial data will be analyzed in detail in [ sec : timeevolution ] . in this section we analyze the optimal vortex states @xmath71 obtained for finite values of the enstrophy in which we extend the results obtained in the seminal study by @xcite . as was also the case in the analogous study in 2d @xcite , there is a distinct branch of extreme states @xmath71 parameterized by the enstrophy @xmath0 and corresponding to each of the three limiting states discussed in [ sec:3d_instopt_e0to0 ] ( cf . figure [ fig : maxdedt_vortexcells ] and table [ tab : e0 ] ) . each of these branches is computed using the continuation approach outlined in algorithm [ alg : optimalg ] . as a key element of the gradient - based maximization technique , the gradient expressions are approximated pseudo - spectrally using standard dealiasing of the nonlinear terms and with resolutions varying from @xmath284 in the low - enstrophy cases to @xmath285 in the high - enstrophy cases , which necessitated a massively parallel implementation using the message passing interface ( mpi ) . as regards the computation of the sobolev @xmath144 gradients , cf . , we set @xmath286 , whereas the second parameter @xmath149 was adjusted during the optimization iterations and was chosen so that @xmath287 $ ] , where @xmath288 is the length scale associated with the spatial resolution @xmath289 used for computations and @xmath290 is the characteristic length scale of the domain @xmath182 , that is , @xmath291 and @xmath292 . we remark that , given the equivalence of the inner products corresponding to different values of @xmath148 and @xmath149 ( as long as @xmath293 ) , these choices do not affect the maximizers found , but only how rapidly they are approached by iterations . for further details concerning the computational approach we refer the reader to the dissertation by @xcite . as was the case in the analogous 2d problem studied by @xcite , the largest instantaneous growth of enstrophy is produced by the states with vortex cells staggered in all planes , cf . case ( [ c1 ] ) in table [ tab : e0 ] . therefore , in our analysis we will focus exclusively on this branch of extreme vortex states which has been computed for @xmath294 $ ] . the optimal instantaneous rate of growth of enstrophy @xmath295 and the energy of the optimal states @xmath296 are shown as functions of @xmath0 for small @xmath0 in figures [ fig : rvse0_fixe_small](a ) and [ fig : rvse0_fixe_small](b ) , respectively . as indicated by the asymptotic form of @xmath84 in and the poincar limit @xmath297 , both of which are marked in these figures , the behaviour of @xmath298 and @xmath296 as @xmath117 is correctly captured by the numerically computed optimal states . in particular , we note that @xmath298 is negative for @xmath299 and exhibits the same trend as predicted in for @xmath117 . for larger values of @xmath0 the rate of growth of enstrophy becomes positive . likewise , the asymptotic behaviour of the energy of the optimal fields does not come as a surprise since , as discussed in [ sec:3d_instopt_e0to0 ] , in the limit @xmath152 the maximizers of @xmath84 are eigenfunctions of the laplacian , which also happen to saturate poincar s inequality . the structure of the optimal vortex states @xmath71 is analyzed next . they are visualized using in which the vortex cores are identified as regions @xmath300 for @xmath301 defined as @xcite @xmath302,\ ] ] where @xmath303 and @xmath304 are the symmetric and anti - symmetric parts of the velocity gradient tensor @xmath305 , that is , @xmath306_{ij } = \frac{1}{2}(\partial_j u_i + \partial_i u_j)$ ] and @xmath307_{ij } = \frac{1}{2}(\partial_j u_i - \partial_i u_j ) $ ] , @xmath32 . the quantity @xmath301 can be interpreted as the local balance between the strain rate and the vorticity magnitude . the isosurfaces @xmath308 representing the optimal states @xmath71 with selected values of @xmath0 are shown in figures [ fig : rvse0_fixe_small](c)-(e ) . for the smallest values of @xmath0 , the optimal fields exhibit a cellular structure already observed in figure [ fig : maxdedt_vortexcells](a ) . for increasing values of @xmath0 this cellular structure transforms into a vortex ring , as seen in figure [ fig : rvse0_fixe_small](e ) . the component of vorticity normal to the plane @xmath309 for @xmath310 $ ] and @xmath311 $ ] is shown in figures [ fig : rvse0_fixe_small](f)-(h ) , where the transition from cellular structures to a localized vortex structure as enstrophy increases is evident . the results corresponding to large values of @xmath0 are shown in figure [ fig : rvse0_fixe_large ] with the maximum rate of growth of enstrophy @xmath298 plotted as a function of @xmath0 in figure [ fig : rvse0_fixe_large](a ) . we observe that , as @xmath0 increases , this relation approaches a power law of the form @xmath312 . in order to determine the prefactor @xmath313 and the exponent @xmath314 we perform a local least - squares fit of the power law to the actual relation @xmath298 versus @xmath0 for increasing values of @xmath0 starting with @xmath315 ( this particular choice the starting value is justified below ) . then , the exponent @xmath314 is computed as the average of the exponents obtained from the local fits with their standard deviation providing the error bars , so that we obtain @xmath316 ( the same approach is also used to determine the exponents in other power - law relations detected in this study ) . we note that the exponent @xmath314 obtained in is in fact very close to 3 which is the exponent in estimate . for the value of the viscosity coefficient used in the computations ( @xmath29 ) , the constant factor @xmath317 in estimate has the value @xmath318 which is approximately seven orders of magnitude larger than @xmath319 given in . to shed more light at the source of this discrepancy , the objective functional @xmath84 from equation can be separated into a negative - definite viscous part @xmath320 and a cubic part @xmath321 defined as @xmath322 so that @xmath323 . the values of @xmath324 are also plotted in figure [ fig : rvse0_fixe_large](a ) and it is observed that this quantity exhibits a power - law behaviour of the form @xmath325 while the value of @xmath326 is slightly larger than the value of @xmath319 in , it is still some six orders of magnitude smaller than the constant factor @xmath317 from estimate . these differences notwithstanding , we may conclude that estimate is sharp in the sense of definition [ def : notionsharpness ] . the power laws from equations and are consistent with the results first presented by @xcite , where the authors reported a power - law with exponent @xmath327 and a constant of proportionality @xmath328 . the energy of the optimal fields @xmath296 for large values of @xmath0 is shown in figure [ fig : rvse0_fixe_large](b ) in which we observe that the energy stops to increase at about @xmath329 . this transition justifies using @xmath330 as the lower bound on the range of @xmath0 where the power laws are determined via least - square fits . figures [ fig : rvse0_fixe_large](c)-(e ) show the isosurfaces @xmath308 representing the optimal fields @xmath71 for selected large values of @xmath0 . the formation of these localized vortex structures featuring two rings as @xmath0 increases is evident in these figures . the formation process of localized vortex structures is also visible in figures [ fig : rvse0_fixe_large](f)(h ) , where the component of vorticity normal to the plane @xmath331 for @xmath332 $ ] and @xmath333 $ ] is shown ( we note that the planes used in figures [ fig : rvse0_fixe_small](c)(e ) and [ fig : rvse0_fixe_large](c)(e ) have different orientations ) . + + + + next we examine the variation of different diagnostics applied to the extreme states @xmath71 as enstrophy @xmath0 increases . the maximum velocity @xmath334 and maximum vorticity @xmath335 of the optimal fields are shown , respectively , in figures [ fig : scalinglaws_fixe](a ) and [ fig : scalinglaws_fixe](b ) as functions of @xmath0 . for each quantity , two distinct power laws are observed in the forms @xmath336 and @xmath337 in order to quantify the variation of the relative size of the vortex structures , we will introduce two characteristic length scales . the first one is based on the energy and enstrophy , and was defined by @xcite as @xmath338^{1/2}.\ ] ] it is therefore equivalent to the taylor microscale @xmath339 used in turbulence research @xcite . another length scale , better suited to the ring - like vortex structures shown in figures [ fig : rvse0_fixe_large](c)-(e ) , is the average radius @xmath340 of one of the vortex rings calculated as @xmath341 and @xmath342 is the characteristic function of the set @xmath343 , \ { \mathbf{x}}_0 = [ 1/2,1/2,1/2 ] \}.\end{aligned}\ ] ] in the above definition of the set @xmath344 , the intersection of the two regions is necessary to restrict the set @xmath340 to only one of the two ring structures visible in figures [ fig : rvse0_fixe_large](c)(e ) . the quantity @xmath345 can be therefore interpreted as the geometric centre of one of the vortex rings . the dependence of @xmath346 and @xmath347 on @xmath0 is shown in figures [ fig : scalinglaws_fixe](c , d ) in which the following power laws can be observed @xmath348 by comparing the error bars in the key power laws and with the error bars in power - law relations , , and , we observe that there is less uncertainty in the first case , indicating that the quantities @xmath334 , @xmath335 , @xmath346 and @xmath347 tend to be more sensitive to approximation errors than @xmath349 . non - negligible error bars may also indicate that , due to modest enstrophy values attained in our computations , the ultimate asymptotic regime corresponding to @xmath350 has not been reached in some power laws . a useful aspect of employing the average ring radius @xmath340 as the characteristic length scale is that its observed scaling with respect to @xmath0 can be used as an approximate indicator of the resolution @xmath351 required to numerically solve problem [ pb : maxdedt_e ] for large values of enstrophy . from the scaling in relation , it is evident that a two - fold increase in the value of @xmath0 will be accompanied by a similar reduction in @xmath347 , thus requiring an eight - fold increase in the resolution ( a two - fold increase in each dimension ) . this is one of the reasons why computation of extreme vortex states @xmath71 for large enstrophy values is a very challenging computational task . in particular , this relation puts a limit on the largest value of @xmath0 for which problem [ pb : maxdedt_e ] can be in principle solved computationally at the present moment : a value of @xmath352 , a mere order of magnitude above the largest value of @xmath0 reported here , would require a resolution of @xmath353 used by some of the largest navier - stokes simulations to date . to summarize , as the enstrophy increases from @xmath354 to @xmath355 , the optimal vortex states change their structure from cellular to ring - like . while with the exception of @xmath74 and @xmath296 , all of the diagnostic quantities behave in a monotonous manner , the corresponding power laws change at about @xmath356 , which approximately marks the transition from the cellular to the ring - like structure ( cf . figure [ fig : rvse0_fixe_small](e ) vs. [ fig : rvse0_fixe_large](c ) ) . this is also the value of the enstrophy beyond which the energy @xmath296 starts to decrease ( figure [ fig : rvse0_fixe_large](b ) ) . this transition also coincides with a change of the symmetry properties of the extreme vortex states @xmath71 while in the limit @xmath117 these fields feature reflection and discrete rotation symmetries ( cf . [ sec:3d_instopt_e0to0 ] ) , for @xmath357 the optimal states are characterized by axial symmetry . the asymptotic ( as @xmath358 ) extreme vortex states on locally maximizing branches corresponding to the aligned abc flow and the taylor - green vortex ( cf . table [ tab : e0 ] ) are similar to the fields shown in figures [ fig : rvse0_fixe_large](c)(h ) , except for a different orientation of their symmetry axes with respect to the periodic domain @xmath182 ( these results are not shown here for brevity ) . the different power laws found here are compared to the corresponding results obtained in 2d in [ sec : discuss ] . it is also worth mentioning that , as shown by @xcite , all power laws discussed in this section , cf . , , , and , can be deduced rigorously using arguments based on dimensional analysis under the assumption of axisymmetry for the optimal fields @xmath71 . + finally , the findings of this section allow us to shed some light on the `` small data '' result which provides the conditions on the size of the initial data @xmath359 , given in terms of its energy @xmath67 and enstrophy @xmath68 , in the navier - stokes system guaranteeing that smooth solutions exist globally in time . the power - law fits and allow us to sharpen condition be replacing the constant on the rhs with either @xmath360 or @xmath361 , so that we obtain @xmath362 the region of the `` phase space '' @xmath363 described by condition is shown in white in figure [ fig : k0e0 ] . the gray region represents the values of @xmath67 and @xmath68 for which long - time existence of smooth solutions can not be a priori guaranteed ( the two shades of gray correspond to the two constants which can be used in ) . solid circles represent the different extreme states found in this section , whereas the thin curves mark the time - dependent trajectories which will be analyzed in [ sec : timeevolution ] . we conclude from figure [ fig : k0e0 ] that the change of the properties of the optimal states @xmath71 discussed above occurs in fact for the values of enstrophy ( @xmath364 ) for which the states @xmath71 are on the boundary of the region of guaranteed long - time regularity . the goal of this section is to analyze the time evolution , governed by the navier - stokes system , with extreme vortex states identified in [ sec:3d_instopt_e ] used as the initial data @xmath359 . in particular , we are interested in the finite - time growth of enstrophy @xmath47 and how it relates to estimates , and . we will compare these results with the growth of enstrophy obtained using other types of initial data which have also been studied in the context of the blow - up problem for both the euler and navier - stokes systems , namely , the taylor - green vortex @xcite , the kida - pelz flow @xcite , colliding lamb - chaplygin dipoles @xcite and perturbed antiparallel vortex tubes @xcite . precise characterization of these different initial conditions is provided in table [ tab : initialconditions ] and , for the sake of completeness , the last three states are also visualized in figure [ fig : summaryic ] . we comment that , with the exception of the taylor - green vortex which was shown in [ sec:3d_instopt_e0to0 ] to be a local maximizer of problem [ pb : maxdedt_e ] in the limit @xmath117 , all these initial conditions were postulated based on rather ad - hoc physical arguments . we also add that , in order to ensure a fair comparison , the different initial conditions listed in table [ tab : initialconditions ] are rescaled to have the same enstrophy @xmath0 , which is different from the enstrophy values used in the original studies where these initial conditions were investigated @xcite . as regards our choices of the initial enstrophy @xmath0 , to illustrate different possible behaviours , we will consider initial data located in the two distinct regions of the phase space @xmath363 shown in figure [ fig : k0e0 ] , corresponding to values of @xmath365 and @xmath0 for which global regularity may or may not be a priori guaranteed according to estimates . .characterization of the different initial data used in time evolution studies in [ sec : timeevolution ] . [ cols="^,^,^",options="header " , ] system is solved numerically with an approach combining a pseudo - spectral approximation of spatial derivatives with a third - order semi - implicit runge - kutta method @xcite used to discretize the problem in time . in the evaluation of the nonlinear term dealiasing was used based on the @xmath366 rule together with the gaussian filtering proposed by @xcite . massively parallel implementation based on mpi and using the fftw routines @xcite to perform fourier transforms allowed us to use resolutions varying from @xmath367 to @xmath368 in the low - enstrophy and high - enstrophy cases , respectively . a number of different diagnostics were checked to ensure that all flows discussed below are well resolved . we refer the reader to the dissertation by @xcite for additional details and a validation of this approach . the time - dependent results will be shown with respect to a normalized time defined as @xmath369 with @xmath370 and @xmath371 ( cf . equation ) playing the roles of the characteristic velocity and length scale . we begin by showing the time evolution of the enstrophy @xmath372 corresponding to the five different initial conditions listed in table [ tab : initialconditions ] with @xmath373 and @xmath374 in figures [ fig : fvs_t_fixe](a ) and [ fig : fvs_t_fixe](b ) , respectively ( because of the faster time - scale , the time axis in the latter figure is scaled logarithmically ) . we see that the maximizers @xmath71 of problem [ pb : maxdedt_e ] are the only initial data which triggers growth of enstrophy for these values of the initial enstrophy and , as expected , this growth is larger for @xmath374 than for @xmath373 . the other initial condition which exhibits some tendency for growth when @xmath374 is the taylor - green vortex . in all cases the enstrophy eventually decays to zero for large times . next we examine whether the flow evolutions starting from the instantaneous maximizers @xmath71 as the initial data saturate the finite - time estimate . we do this by defining functions [ eq : fg ] @xmath375 \label{eq : fgb}\end{aligned}\ ] ] representing , respectively , the left- and right - hand side of the estimate and then plotting them with respect to the normalized time @xmath376 , which is done in figures [ fig : fg](a ) and [ fig : fg](b ) for @xmath373 and @xmath374 , respectively . the constant @xmath62 in the definition of @xmath377 is numerically computed from the power - law fit in . it follows from estimate that @xmath378 pointwise in time . the hypothetical extreme event of a finite - time blow - up can be represented graphically by an intersection of the graph of @xmath379 with the horizontal line @xmath380 , which is also shown in figures [ fig : fg](a)(b ) . the behaviour of @xmath377 , representing the upper bound in estimate , is quite distinct in figures [ fig : fg](a ) and [ fig : fg](b ) reflecting the fact that the initial data @xmath71 in the two cases comes from different regions of the phase diagram in figure [ fig : k0e0 ] . in figure [ fig : fg](a ) , corresponding to @xmath373 , the upper bound @xmath377 never reaches @xmath381 , in agreement with the fact that the finite - time blow - up is a priori ruled out in this case . on the other hand , in figure [ fig : fg](b ) , corresponding to @xmath374 , the upper bound @xmath377 does intersect @xmath381 implying that , in principle , finite - time blow - up might be possible in this case . the sharpness of estimate can be assessed by analyzing how closely the behaviour of @xmath379 matches that of @xmath377 . in both figures [ fig : fg](a ) and [ fig : fg](b ) we observe that for a short period of time @xmath379 exhibits a very similar growth to the upper bound @xmath377 , but then this growth slows down and @xmath379 eventually starts to decrease short of ever approaching the limit @xmath381 . we further characterize the time evolution by showing the maximum enstrophy increase @xmath382 and the time when the maximum is achieved @xmath383 as functions of @xmath0 in figures [ fig : emax_vse0_fixe](a ) and [ fig : emax_vse0_fixe](b ) , respectively . in both cases approximate power laws in the form @xmath384 are detected in the limit @xmath358 ( as regards the second result , we remark that @xmath385 is not equivalent to the time until which the enstrophy grows at the sustained rate proportional to @xmath386 , cf . figure [ fig : fg ] ) . to complete presentation of the results , the dependence of the quantities @xmath387\ ] ] on the initial enstrophy @xmath0 is shown in figures [ fig : emax_vse0_fixe](c ) and [ fig : emax_vse0_fixe](d ) , respectively . it is observed that both quantities approximately exhibit a power - law behaviour of the form @xmath388 . discussion of these results in the context of the estimates recalled in [ sec : intro ] is presented in the next section . in this section we provide some comments about the results reported in [ sec:3d_instopt_e0to0 ] , [ sec:3d_instopt_e ] and [ sec : timeevolution ] . first , we need to mention that our gradient - based approach to the solution of optimization problem [ pb : maxdedt_e ] can only yield local maximizers and , due to nonconvexity of the problem , it is not possible to guarantee a priori that the maximizers found are global . to test for the possible presence of branches other than the ones found using the continuation approach described in [ sec:3d_instopt ] , cf . algorithm [ alg : optimalg ] , we tried to find new maximizers by initializing the gradient iterations with different initial guesses @xmath113 . they were constructed as solenoidal vector fields with prescribed regularity and random structure , which was achieved by defining the fourier coefficients of @xmath113 as @xmath389 with the amplitude @xmath390 and the phases @xmath391 chosen as random numbers uniformly distributed in @xmath392 $ ] . however , in all such tests conducted for @xmath393 the gradient optimization algorithm would always converge to maximizers @xmath71 belonging to one of the branches discussed in [ sec:3d_instopt_e ] ( modulo possible translations in the physical domain ) . while far from settling this issue definitely , these observations lend some credence to the conjecture that the branch identified in [ sec:3d_instopt_e ] corresponds in fact to the global maximizers . these states appear identical to the maximizers found by @xcite and our search has also yielded two additional branches of locally maximizing fields , although we did not capture the lower branch reported by @xcite . however , since that branch does not appear connected to any state in the limit @xmath117 , we speculate that it might be an artifact of the `` discretize - then - optimize '' formulation used by @xcite , in contrast to the `` optimize - then - discretize '' approach employed in our study which provides a more direct control over the analytic properties of the maximizers . we add that the structure of the maximizing branches found here is in fact quite similar to what was discovered by @xcite in an analogous problem in 2d . since the 2d problem is more tractable from the computational point of view , in that case we were able to undertake a much more thorough search for other maximizers which did not however yield any solutions not associated with the main branches . the results reported in [ sec:3d_instopt_e ] and [ sec : timeevolution ] clearly exhibit two distinct behaviours , depending on whether or not global - in - time regularity can be guaranteed a priori based on estimates . these differences are manifested , for example , in the power laws evident in figures [ fig : scalinglaws_fixe ] and [ fig : emax_vse0_fixe ] , as well as in the different behaviours of the rhs of estimate with respect to time in figures [ fig : fg](a ) and [ fig : fg](b ) . however , for the initial data for which global - in - time regularity can not be ensured a priori there is no evidence of sustained growth of enstrophy strong enough to signal formation of singularity in finite time . indeed , in figure [ fig : emax_vse0_fixe](c ) one sees that the quantity @xmath394 behaves as @xmath395 , where @xmath396 , when @xmath0 increases , revealing no tendency to approach @xmath381 which is a necessary precursor of a singular behaviour ( cf . discussion in [ sec : timeevolution ] ) . to further illustrate how the rate of growth of enstrophy achieved initially by the maximizers @xmath71 is depleted in time , in figure [ fig : dedte ] we show the flow evolution corresponding to @xmath71 with @xmath374 as a trajectory in the coordinates @xmath397 . from the discussion in introduction we know that in order for the singularity to occur in finite time , the enstrophy must grow at least at a sustained rate @xmath398 for some @xmath63 . in other words , a `` blow - up trajectory '' will be realized only if the trajectory of the flow , expressed in @xmath399 coordinates , is contained in the region @xmath400 , for some positive constants @xmath401 and @xmath402 . for the flow corresponding to the instantaneous optimizers @xmath71 , the initial direction of a trajectory in @xmath399 coordinates is determined by the vector @xmath403 $ ] and , for initial conditions @xmath359 satisfying @xmath404 , it follows that @xmath405 since the optimal rate of growth is sustained only over a short interval of time , the trajectory of the flow in the @xmath406 coordinates approaches the region @xmath407 only tangentially following the direction of the lower bound @xmath408 , and remains outside @xmath407 for all subsequent times . this behaviour is clearly seen in the inset of figure [ fig : dedte ] . an interpretation of this behaviour can be proposed based on equation from which it is clear that the evolution of the flow energy is closely related to the growth of enstrophy . in particular , if the initial energy @xmath67 is not sufficiently large , then its depletion due to the initial growth of enstrophy may render the flow incapable of sustaining this growth over a longer period of time . this is in fact what seems to be happening in the present problem as evidenced by the data shown in figure [ fig : k0e0 ] . we remark that , for a prescribed enstrophy @xmath68 , the flow energy can not be increased arbitrarily as it is upper - bounded by poincar s inequality @xmath409 . this behaviour can also be understood in terms of the geometry of the extreme vortex states @xmath71 . figure [ fig : ring ] shows a magnification of the pair of vortex rings corresponding to the optimal field @xmath71 with @xmath410 . it is observed that the vorticity field @xmath411 inside the vortex core has an azimuthal component only which exhibits no variation in the azimuthal direction . thus , in the limit @xmath412 the vortex ring shrinks with respect to the domain @xmath182 ( cf . figure [ fig : scalinglaws_fixe](d ) ) and the field @xmath71 ultimately becomes axisymmetric ( i.e. , in this limit boundary effects vanish ) . at the same time , it is known that the 3d navier - stokes problem on an unbounded domain and with axisymmetric initial data is globally well posed @xcite , a results which is a consequence of the celebrated theorem due to @xcite . with @xmath374 in the coordinates @xmath406 . for comparison , in the inset the thin line represents the borderline growth at the rate @xmath413,scaledwidth=90.0% ] . [ fig : dedte ] with @xmath374 . the colour coding of the vortex lines is for identification purposes only.,scaledwidth=60.0% ] we close this section by comparing the different power laws characterizing the maximizers @xmath71 and the corresponding flow evolutions with the results obtained in analogous studies of extreme behaviour in 1d and 2d ( see also table [ tab : estimates ] ) . first , we note that the finite - time growth of enstrophy @xmath414 in 3d , cf . figure [ fig : emax_vse0_fixe](a ) , exhibits the same dependence on the enstrophy @xmath0 of the instantaneously optimal initial data as in 1d , i.e. , is directly proportional to @xmath0 in both cases @xcite . this is also analogous to the maximum growth of palinstrophy @xmath14 in 2d which was found by @xcite to scale with the palinstrophy @xmath415 of the initial data , when the instantaneously optimal initial condition was computed subject to _ one _ constraint only ( on @xmath415 ) . when the instantaneously optimal initial data was determined subject to _ two _ constraints , on @xmath365 and @xmath415 , then the maximum finite - time growth of palinstrophy was found to scale with @xmath416 @xcite . on the other hand , the time @xmath385 when the maximum enstrophy is attained , cf . figure [ fig : emax_vse0_fixe](b ) , scales as @xmath417 , which should be contrasted with the scalings @xmath418 and @xmath419 found in the 1d and 2d cases , respectively . this implies that the time interval during which the extremal growth of enstrophy is sustained in 3d is shorter than the corresponding intervals in 1d and 2d . by constructing the initial data to exhibit the most extreme behaviour allowed for by the mathematically rigorous estimates , this study offers a fundamentally different perspective on the problem of searching for potentially singular solutions from most earlier investigations . indeed , while the corresponding flow evolutions did not reveal any evidence for finite - time singularity formation , the initial data obtained by maximizing @xmath1 produced a significantly larger growth of enstrophy in finite time than any other candidate initial conditions ( cf . table [ tab : initialconditions ] and figure [ fig : fvs_t_fixe ] ) . admittedly , this observation is limited to the initial data with @xmath420 , which corresponds to reynolds numbers @xmath421 lower than the reynolds numbers achieved in other studies concerned with the extreme behaviour in the navier - stokes flows @xcite . given that the definitions of the reynolds numbers applicable to the various flow configurations considered in these studies were not equivalent , it is rather difficult to make a precise comparison in terms of specific numerical values , but it is clear that the largest reynolds numbers attained in these investigations were at least an order of magnitude higher than used in the present study ; for euler flows such a comparison is obviously not possible at all . however , from the mathematical point of view , based on estimates , there is no clear indication that a very large initial enstrophy @xmath0 ( or , equivalently , a high reynolds number ) should be a necessary condition for singularity formation in finite time . in fact , blow - up can not be a priori ruled out as soon as condition is violated , which happens for all initial data lying on the gray region of the phase space in figure [ fig : k0e0 ] . we remark that additional results were obtained ( not reported in this paper ) by studying the time evolution corresponding to the optimal initial data @xmath71 , but using smaller values of the viscosity coefficient @xmath422 , thereby artificially increasing the reynolds number at the price of making the initial data suboptimal . although these attempts did increase the amplification of enstrophy as compared to what was observed in figures [ fig : fg ] and [ fig : emax_vse0_fixe ] , no signature of finite - time singularity formation could be detected either . our study confirmed the earlier findings of @xcite about the sharpness of the instantaneous estimate . we also demonstrated that the finite - time estimate is saturated by the flow evolution corresponding to the optimal initial data @xmath71 , but only for short times , cf . figure [ fig : fg ] , which are not long enough to trigger a singular behaviour . in [ sec : discuss ] we speculated that a relatively small initial energy @xmath67 , cf . figure [ fig : rvse0_fixe_large](b ) , might be the property of the extreme vortex states @xmath71 preventing the resulting flow evolutions from sustaining a significant growth of enstrophy over long times . on the other hand , in introduction we showed that estimate need not be saturated for blow - up to occur in finite time and , in fact , sustained growth at the rate @xmath423 with any @xmath63 will also produce singularity in finite time . thus , another strategy to construct initial data which could lead to a more sustained growth of enstrophy in finite time might be to increase its kinetic energy by allowing for a smaller instantaneous rate of growth ( i.e. , with an exponent @xmath424 instead of @xmath425 ) . this can be achieved by prescribing an additional constraint in the formulation of the variational optimization problem , resulting in [ pb : maxdedt_ke ] @xmath426 it differs from problem [ pb : maxdedt_e ] in that the maximizers are sought at the intersection of the original constraint manifold @xmath88 and the manifold defined by the condition @xmath427 , where @xmath428 is the prescribed energy . while computation of such maximizers is more complicated , robust techniques for the solution of optimization problems of this type have been developed and were successfully used in the 2d setting by @xcite . preliminary results obtained in the present setting by solving problem [ pb : maxdedt_ke ] for @xmath429 are indicated in figure [ fig : k0e0 ] , where we see that the flow evolutions starting from @xmath430 do not in fact produce a significant growth of enstrophy either . an alternative , and arguably more flexible approach , is to formulate this problem in terms of multiobjective optimization @xcite in which the objective function @xmath135 in problem [ pb : maxdedt_e ] would be replaced with @xmath431 where @xmath432 $ ] . solution of such a multiobjective optimization problem has the form of a `` pareto front '' parameterized by @xmath433 . clearly , the limits @xmath434 and @xmath435 correspond , respectively , to the extreme vortex states already found in [ sec:3d_instopt_e0to0 ] and [ sec:3d_instopt_e ] , and to the poincar limit . another interesting possibility is to replace the energy @xmath35 with the helicity @xmath436 in the multiobjective formulation , as this might allow one to obtain extreme vortex states with a more complicated topology ( i.e. , a certain degree of `` knottedness '' ) . we note that all the extreme vortex states found in the present study were `` unknotted '' , i.e. , were characterized by @xmath437 , as the vortex rings were in all cases disjoint ( cf . figure [ fig : ring ] ) . finally , another promising possibility to find initial data producing a larger growth of enstrophy is to solve a _ finite - time _ optimization problem of the type already studied by @xcite in the context of the 1d burgers equation , namely [ pb : maxde ] @xmath438 where @xmath439 is the length of the time interval of interest and @xmath359 the initial data for the navier - stokes system . in contrast to problems [ pb : maxdedt_e ] and [ pb : maxdedt_ke ] , solution of problem [ pb : maxde ] is more complicated as it involves flow evolution . it represents therefore a formidable computational task for the 3d navier - stokes system . however , it does appear within reach given the currently available computational resources and will be studied in the near future . the authors are indebted to charles doering for many enlightening discussions concerning the research problems studied in this work . the authors are grateful to nicholas kevlahan for making his parallel navier - stokes solver available , which was used to obtain the results reported in [ sec : timeevolution ] . anonymous referees provided many insightful comments which helped us improve this work . this research was funded through an early researcher award ( era ) and an nserc discovery grant , whereas the computational time was made available by sharcnet under its dedicated resource allocation program . diego ayala was funded in part by nsf award dms-1515161 at the university of michigan and by the institute for pure and applied mathematics at ucla .

is evolution deterministic , hence predictable , or stochastic , that is unpredictable ? what would happen if one could `` replay the tape of evolution : '' will the outcomes of evolution be completely different or evolution is so constrained that history will be repeated ? arguably , these questions are among the most intriguing and most difficult in evolutionary biology . in other words , the predictability of evolution depends on the fraction of the trajectories on fitness landscapes that are accessible for evolutionary exploration . because direct experimental investigation of fitness landscapes is technically challenging , the available studies only explore a minuscule portion of the landscape for individual enzymes . we therefore sought to investigate the topography of fitness landscapes within the framework of a previously developed model of protein folding and evolution where fitness is equated with robustness to misfolding . we show that model - derived and experimental landscapes are significantly smoother than random landscapes and resemble moderately perturbed additive landscapes ; thus , these landscapes are substantially robust to mutation . the model landscapes show a deficit of suboptimal peaks even compared with noisy additive landscapes with similar overall roughness . thus , the smoothness and substantial deficit of peaks in fitness landscapes of protein evolution could be fundamental consequences of the physics of protein folding . one of the most intriguing questions in evolutionary biology is : to what extent evolution is deterministic and to what extent it is stochastic and hence unpredictable ? in other words , what happens if `` the tape of evolution is replayed : '' are we going to see completely different outcomes or the constraints are so strong that history will be repeated @xcite ? if evolution is envisaged as movement of a population across a fitness landscape , the question can be reworded more specifically : among the numerous trajectories connecting any two points on the landscape , what fraction is accessible to evolution ? until recently , these remained purely theoretical questions as experimental study of fitness landscapes in the actual sequence space was impractical , due both to the technical difficulty of producing and assaying numerous expressed sequence variants and to the more fundamental problem of defining an adequate quantitative measure of fitness . however , recent experimental studies of fitness landscapes could potentially shed light on the problem of evolutionary path predictability . the most thoroughly characterized feature of empirical fitness landscapes is the structure near a peak . in experiments that examine the peak structure , a high fitness sequence is typically subjected to either random mutations or an exhaustive set of mutations at a small number of important sites . the resulting library of mutants is then assayed to measure a proxy of fitness @xcite . significant sign epistasis ( a situation in which the fitness effect of a particular mutation can be either positive or negative depending on the genetic context ) near the peak has been typically observed whereas local deviations from the additive model have been found to be uncorrelated with the genetic context and derived from a nearly normal distribution @xcite . because these studies characterize only a small region of the landscape , they can not be used to address the question of path predictability . another broad class of experiments probes the evolutionary trajectories from low to high fitness . usually , in such experiments , a random peptide is subjected to repeated rounds of random mutagenesis and purifying selection @xcite . during this process fitness grows with each generation and eventually stagnates at a suboptimal plateau . the characteristics of the fitness growth as well as the dependence of the plateau height on the library size can be used to classify landscapes @xcite . a quantitative comparison to the @xmath0 model of random epistatic landscapes ( @xmath1 is the number of sites in an evolving sequence and @xmath2 is the number of sites that affect the fitness contribution of a particular site through epistatic interactions ) can even yield quantitative estimates of @xmath1 and @xmath2 @xcite . the directed evolution studies explore the evolutionarily accessible portion of the landscape and could in principle be used to shed light on the question of path predictability . however , the inaccessible regions of the landscape remain unexplored and the volume of data at this point is insufficient to obtain quantitative conclusions regarding path predictability . a different type of landscapes has been explored in various microarray experiments where protein - dna(rna ) binding affinity serves as the proxy for fitness @xcite . these experiments produce vast , densely sampled landscapes . a comparison with a sophisticated landscape state machine model of a correlated fitness landscapes yields estimates of the model parameters @xcite . the dna binding landscapes , in principle , contain the information required for the analysis of path statistics , and could be a valuable resource for advancing the understanding of evolutionary path predictability . empirical studies that exhaustively sample a region of the fitness landscape allow one to actually assess the accessibility of the entire set of theoretically possible evolutionary trajectories in a particular ( small ) area of the fitness landscape . for example , all mutational paths between two states of an enzyme , e.g. , the transition from an antibiotic - sensitive to an antibiotic resistant form of @xmath3-lactamase @xcite or the transition between different specificities of sesquiterpene synthase @xcite have been explored . the results of these experiments , which out of necessity explore only short mutational paths of @xmath4 amino acid replacements , suggest that there is a substantial deterministic component to protein evolution : only a small fraction of the possible paths are accessible for evolution @xcite . theoretical analyses of available empirical fitness data reveal a tight link between genetic and molecular interactions which are responsible for the landscape ruggedness and ubiquitous sign epistasis @xcite . the emerging quantitative analysis of fitness landscapes can shed light on some of the most fundamental aspects of evolution but the interpretation of the currently available experimental results requires utmost caution as only a minuscule part of the sequence space can be explored , and that only for a few more or less arbitrarily selected experimental systems . here we focus on the question of the predictability of mutational paths which is intimately tied to the ruggedness / smoothness of the fitness landscapes . the study of random landscapes of low dimensionality revealed an intuitively plausible negative correlation between the roughness of a landscape and the availability of pathways of monotonic fitness @xcite . in the same study , carneiro and hartl showed that experimentally characterized landscapes are significantly smoother than their permuted counterparts and exhibit greater peak accessibility @xcite . to gain insights into the structure of the fitness landscapes of protein evolution and in particular the accessibility of mutational paths we used a previously developed simple model of protein folding and evolution @xcite . the key assumption of this model , which is based on the concept of misfolding - driven evolution of proteins @xcite , is that the fitness of model proteins is determined solely by the number of misfolded copies that are produced before the required abundance of the correctly folded protein is reached . we have previously shown that this model accurately reproduces the shape of the universal distribution of the evolutionary rates among orthologous protein - coding genes along with the dependencies of the evolutionary rate on protein abundance and effective population size @xcite . these results appear to suggest that our folding model ( described in detail the methods section ) is sufficiently rich to reproduce some of the salient aspects of evolution . the model is also simple enough to allow exhaustive exploration of the fitness landscapes , which prompted us to directly address the problem of evolutionary path predictability . we build on the efforts of carneiro and hartl @xcite who examined the statistics of evolutionary trajectories . although counting monotonic fitness paths reveals important features of the landscapes , we argue that reliable retrodiction of the evolutionary past is possible ( i.e. , evolution is quasi - deterministic ) only when the available monotonic paths are similar to each other in a quantifiable way . we therefore propose a measure of path divergence to quantify the difference between the available monotonic paths . our aims are to investigate the structure of the fitness landscapes of protein evolution and to elucidate the connection between the roughness of landscapes and the predictability of mutational trajectories . we analyze three classes of fitness landscapes : landscapes in which fitness is derived from the folding robustness of model polymers ; additive random landscapes perturbed by noise ; and experimental landscapes derived from the combinatorial mutation analysis of drug resistance and enzymatic activity . we show that all three classes of landscapes are markedly smoother than their randomly permuted counterparts and all exhibit a similar qualitative connection between roughness and path predictability . however , at the same level of path predictability , the folding landscapes have substantially fewer fitness peaks . given that the statistical properties of the model landscapes can be directly traced to the constraints imposed by the energetics and kinetics of a folding heterpolymer , we hypothesize that the relative smoothness and the suppression of suboptimal peaks in fitness landscapes of protein evolution are fundamental consequences of protein folding physics . carneiro and hartl compared small random landscapes to several empirical fitness landscapes using deviation from additivity as a measure of roughness @xcite . they found that empirical landscapes were significantly smoother than their random counterparts and that the degree of smoothness was correlated with the number of monotonic paths to the main summit . deviation from additivity of a landscape is computed by fitting an additive model in which the fitness of each sequence is different from the peak fitness by the sum of contributions of the substitutions that differentiate it from the peak sequence . the negative fitness contributions of the substitutions to the peak fitness are adjusted to minimize the sum @xmath5 of squares of the differences between the actual fitnesses in the landscape and the fitnesses predicted by the additive model . deviation from additivity is defined as @xmath6 , where @xmath7 is the number of points in the landscape . because roughness of a multidimensional landscape with variable degree connectivity is not an intuitive concept , we introduce three additional quantitative measures to probe alternative facets of the concept of roughness . first , local roughness is the root mean squared difference between the fitness of a point and its neighbors , averaged over the entire landscape . as defined , local roughness conflates the measures of roughness and `` steepness . '' for example , a globally smooth landscape , in which fitness depends only on the distance from the peak , will have a non - zero local roughness . however , because there is a large number of directions that change the distance from the peak by one , the local roughness of a globally smooth landscape will be vanishingly small . in addition , our landscapes tend to be globally flat so that the average decrease in fitness due to a single mutation step away from the main peak is much smaller than the local fitness variability everywhere except a small region around the main peak ( see fig . [ fig:7 ] ) . therefore , the landscape - average local roughness in our case is a true measure of the local fitness variability . second , the fraction of peaks is the number of points with no fitter neighbors divided by the total number of points in the landscape . a strictly additive landscape has a single peak @xcite whereas the peak fraction in landscapes derived from the folding model as well as the corresponding randomized landscapes depends on the method of landscape construction , alphabet size and sequence length . third , the roughness of a landscape can be assessed by identifying its tree component . the tree component is the set of all nodes with no more than one neighbor of higher fitness . thus , the tree component includes peaks and plateaus . monotonic fitness paths along the tree component form a single or several disjoint tree structures without loops . in the limit of high selection pressure , a mutational trajectory that finds itself on the tree component has a single path to the nearest peak or plateau , i.e. evolution on the tree component is completely deterministic . we use the mean distance to the tree component , i.e. the distance to the tree component averaged over the landscape , as a measure of roughness . in a fully additive landscape , only the peak sequence and its immediate neighbors belong to the tree component and therefore the mean distance to the tree component is a measure of the diameter of an additive landscape ( which , for example , could be defined as the maximum pairwise distance between points on the landscape ) . kauffman and levin have shown that in a large class of correlated random landscapes , the mean distance to the tree component grows only logarithmically with the number of points in the landscape @xcite . we utilize two quantitative measures of the predictability of evolutionary trajectories . first is fraction of monotonic paths to the main peak @xmath8 which is computed by counting the number @xmath9 of simple ( without reverse substitutions or multiple substitutions at the same site ) monotonic paths to the main peak from each point @xmath10 on the landscape , dividing it by the total number of simple paths @xmath11 ( where @xmath12 is the hamming distance from point @xmath10 to the peak ) , and averaging over the landscape via @xmath13 where @xmath7 is the number of points in the landscape and the sum excludes the main peak . the monotonic path fraction @xmath8 measures the scarcity of accessible evolutionary paths when selection is strong . when the monotonic path fraction is small , evolution is more constrained . second , the mean path divergence , is a fine - grained measure of evolutionary ( un)predictability . we first define the divergence @xmath14 of a pair of paths @xmath15 and @xmath16 , as the average of the shortest hamming distances from each point on one path to the other path . suppose that we have a way of generating stochastic evolutionary paths . the outcome of a large number of evolutionary dynamics simulations is a collection of paths with their associated probabilities of occurrence . in general , the probability of occurrence of an evolutionary path is proportional to the product of fixation probabilities of its constituent mutation steps . given a bundle of paths with the same starting and ending points , we define its mean path divergence to be @xmath17 where @xmath18 is the probability of occurrence of path @xmath19 in the ensemble . in other words , if two paths were drawn from the bundle at random with probabilities proportional to @xmath18 , their expected divergence would be @xmath20 . alternatively , if we were to fix one path to be the most likely path in the bundle and to select the second path at random with probability proportional to @xmath18 , the divergence would be proportional to @xmath20 as well . in an additive landscape , the mutational trajectory is maximally ambiguous . as every substitution that brings the sequence closer to the peak increases fitness , substitutions can occur in any order and all shortest mutational trajectories to the peak without reverse substitutions or multiple substitutions at the same site are monotonic in fitness . in the strong selection limit of our model defined below , all monotonic trajectories have roughly the same probability of occurrence , so the mutational path can not be predicted . the mean path divergence is a better measure of the predictability of evolutionary trajectories than the number or fraction of accessible paths . even when only a small fraction of paths are monotonic in fitness , these paths could potentially be quite different , perhaps randomly scattered over the landscape . in such a case , prediction of the evolutionary trajectory would be inaccurate despite the scarcity of accessible paths which will be reflected in a high value of path divergence . equation ( [ eq:1 ] ) introduces the mean path divergence of a bundle of paths with the same starting and ending points . the landscape - wide mean path divergence is measured by constructing representative path bundles with all possible [ start , peak ] pairs including suboptimal peaks as trajectory termination points . path divergence is averaged over all bundles with the starting and ending points separated by the same hamming distance . to construct the path bundles , we employed a low mutation rate model in which the attempted substitutions are either eliminated or fixed in the population before the next mutation attempt occurs . we invoke the misfolding - cost hypothesis to assign a fitness to a sequence that folds with probability @xmath21 to a particular structure . to produce an abundance @xmath22 of correctly folded copies , an average of @xmath23 of misfolded copies are produced . the `` fitness '' of a sequence should be a monotonically decreasing function of the cost incurred by the misfolded proteins . previously we showed that qualitative conclusions drawn from the average population dynamics on the fitness landscape did not depend on the precise functional relationship between the number of misfolded copies and fitness @xcite . we use simply the negative of the number of misfolded copies and assign a fitness @xmath24 , to a sequence whose probability of folding to the reference structure is @xmath21 . because the exact population dynamics model is not important , we use diploid population dynamics in the low mutation rate limit . therefore , the probability of fixation of a mutant @xmath25 in the background of @xmath10 is given by @xmath26 where @xmath27 is the effective population size @xcite which in all simulations was fixed at @xmath28 . the required abundance @xmath22 is a measure of the strength of selection . in the limit of large @xmath22 , the probability of fixation of a beneficial mutation is unity whereas neutral and deleterious mutations are never fixed . in this limit , all uphill steps in the fitness landscape are equally likely and all monotonic uphill trajectories have equal evolutionary significance . in the analysis that follows , we study the association between landscape roughness and path predictability for the folding landscapes and their randomized ( also referred to as permuted or scrambled ) versions . in the scrambled landscapes , the topology ( i.e.connectivity ) of the landscape is preserved but the fitness values are randomly shuffled . we also compare the roughness and path predictability characteristics of the model and the experimental landscapes for @xmath3-lactamase @xcite and sesquiterpene synthase @xcite to those for noisy additive landscapes with a continuously tunable amount of roughness . we first establish that the folding and the experimental landscapes are significantly different from their randomly permuted counterparts . the deviation from additivity of the folding landscapes is typically several standard deviations below the mean of their scrambled counterparts . although the additivity hypothesis accounts for less than 40% of the fitness variability ( computed by comparing the sum of the squares of the fitnesses in the landscape to the sum of the squares of the residuals of the additive fitness model fit ) in all but one of the folding landscapes , the deviation from additivity of the permuted landscapes is substantially greater ( fig . [ fig:1]a ) . the experimental landscapes follow the same pattern , in agreement with the earlier findings of carneiro and hartl @xcite . furthermore , both in the folding and in the experimental landscapes , the fraction of monotonic paths to the main peak is several standard deviations greater than in the respective scrambled landscapes ( fig . [ fig:1]b ) . an even more striking disparity exists between the fraction of peaks in the folding landscapes and their permuted versions : the folding landscapes contain at least an order of magnitude fewer peaks than their scrambled counterparts ; the experimental landscapes resemble the folding landscapes more closely than their own randomized versions ( fig . [ fig:1]c ) . to further characterize the deviation of the folding and experimental landscapes from their permuted counterparts , each landscape metric was measured and the mean and standard deviation were computed among 100 randomly permuted landscapes . we then compute the z - score ( deviation from the mean measured in the units of the standard deviation ) of the original non - permuted landscape compared to the ensemble of the permuted landscapes . this z - score shows how much more correlated the original landscape is , as measured by the chosen characteristic , compared to its scrambled counterparts ( figure [ fig:2 ] ) . notably , despite the considerable scatter of the z - score values for the folding landscapes , they all showed extremely large difference ( mean z - score greater than 20 standard deviations ) from the scrambled landscapes for all measures , with the sole exception of the monotonic path fraction ( figure [ fig:2 ] ) . the two experimental landscapes also significantly differed from the scrambled landscapes albeit less so than the folding landscapes , again with the exception of the monotonic path fraction in which case the two classes of landscapes had similar z - scores ( figure [ fig:2 ] ) . aside from the significant correlation ( pearson @xmath29 ) between peak fraction and mean distance to the tree component , there was little or no correlation between the four measures of landscape roughness ( fig . [ fig:3 ] ) . roughness of landscapes of high and variable dimensionality is impossible to capture by a single quantity . therefore , the different measures seam to reveal distinct aspects of landscape architecture . the strong negative correlation between the peak fraction and mean distance to the tree component is due to the fact that each peak spawns a distinct subset of the tree component . the higher the density of peaks on the landscape , the larger fraction of the landscape that is covered by the tree component . therefore the average distance to the tree component declines with the increasing density of peaks . starting from a random non - peak sequence in the landscape , we introduced random mutations and accepted or rejected them according to equation ( [ eq:2 ] ) until the trajectory arrived at a fitness peak . this procedure was repeated a large number of times , and path bundles were constructed for all pairs of starting and ending sequences . then the mean path divergence was computed for each path bundle using equation ( [ eq:1 ] ) and averaged over all bundles for which starting and ending points were separated by the same hamming distance . when selection is weak , all mutations which do not result in a sequence with zero folding probability are accepted . thus , evolution is a random walk on the landscape and the statistical properties of evolutionary trajectories are fully determined by the topology of the landscape ( i.e. the connectivity of each node ) . conversely , in the strong selection limit , only mutations that increase fitness are fixed . the mean path divergence varies smoothly between the two limits ( fig . [ fig:4 ] ) and saturates at high selection pressure . in our analysis , we focus on the strong selection limit plateau . in the weak selection limit , the diversity of trajectories stems solely from the number of neighbors of each point ; by contrast , in the strong selection limit , the statistics of the monotonic trajectories depend on the roughness of the landscape . thus , the weak selection limit probes only the topology of the landscape whereas the strong selection limit also exposes its topography which appears to be critical for assessing predictability of evolution . all four measures of landscape roughness can serve as predictors of path divergence and monotonic path fraction to some degree ( fig.[fig:5 ] ) , in agreement with the notion that each of these measures reflects salient properties of fitness landscapes . the properties of the folding and empirical landscapes are consistent with those of additive landscapes that were perturbed by a moderate amount of noise ( see methods for details ) . a striking exception is the dearth of peaks and monotonic paths in folding landscapes all other characteristics being similar . deviation from additivity and fraction of peaks are negatively correlated with path divergence . this relationship captures the intuitive notion that in rough landscapes there are fewer accessible evolutionary paths than in smooth landscapes , and furthermore , in rough landscapes , even those paths that are accessible show the tendency to aggregate within small areas on the landscape . indeed , in both the folding model - derived landscapes and the experimental landscapes , the mean path divergence for all hamming distances between the starting and ending points was dramatically greater than in scrambled landscapes ( fig . [ fig:6 ] ) . interpreting these findings in terms closer to biology , the fitness landscapes derived from the model as well as experimental landscapes show greater robustness to mutations than random landscapes : a random mutation in a model - derived or experimental fitness landscape is more likely than expected for random landscapes to have no deleterious effect , leading to another monotonic path to the peak . consequently , evolution on the model - derived and experimental landscapes is less predictable ( deterministic ) than it would be on uncorrelated random landscapes . in contrast to deviation from additivity , the mean distance to the tree component is positively correlated with path divergence . when the tree component comprises a large fraction of the landscape , the mean distance to the nearest tree branch is small . consequently , the path divergence is reduced as the paths that reach the tree component do not deviate from each other from that point onward . by the same token , when the tree component is large , there are fewer monotonic paths . the origin of the positive correlation between the local roughness and path divergence ( fig . [ fig:5 ] ) is less obvious . paradoxically , greater noise results in lower mean local roughness of noisy additive landscapes . the lowering of the overall mean fitness with noise and , more importantly , the flattening of the mean fitness dependence on the distance from the peak ( fig . [ fig:7 ] ) appear to provide an explanation for this counter - intuitive result . indeed we found that in noisy additive landscapes there is a characteristic fitness value of approximately 0.2 above which roughness increases with increasing noise and below which roughness declines with increasing noise . given that roughly 75% of the points on the landscape have fitnesses below 0.2 , the landscape - averaged local roughness declines with increasing noise amplitude . here we examined the fraction of monotonic paths and introduced mean path divergence as quantitative measures of the degree to which the starting and ending points determine the path of evolution on fitness landscapes . the lower the mean path divergence value , the more deterministic ( and predictable ) evolution is . global measures of landscape roughness correlate with path divergence in the three analyzed classes of fitness landscapes : additive landscapes perturbed by noise , landscapes derived from our protein folding model and two small empirical landscapes . the folding landscapes are substantially smoother than their permuted counterparts . as a result , although in all analyzed landscapes only a small fraction of the theoretically possible evolutionary trajectories is accessible , this fraction is much greater in the folding and experimental landscapes than it is in randomized landscapes . in addition , the mean path divergence in the randomized landscapes is significantly smaller than in the original landscapes . thus , the model and empirical landscapes possess similar global architectures with many more diverged monotonic paths to the high peaks than uncorrelated landscapes with the same distribution of fitness values . consequently , evolution in fitness landscapes is substantially more robust to random mutations and less deterministic ( less predictable ) than expected by chance . these findings are compatible with the concept that might appear counter - intuitive but is buttressed by results of population genetic modeling , namely , that robustness of evolving biological systems promotes their evolvability @xcite . additionally , the folding landscapes exhibit a substantial deficit of peaks compared to perturbed additive landscapes and experimental landscapes , a property that translates into a substantially greater fraction of paths leading to the main peak . when it comes to the interpretation of the properties of fitness landscapes described here , an inevitable and important question is whether the folding model employed here is sufficiently complex and realistic to yield biologically relevant information . in selecting the complexity of our folding model , we attempted to construct the simplest model which exhibits 1 ) a rich spectrum of low energy conformations across the sequence space , and 2 ) a non - trivial distribution of substitutions effects on the low energy conformations . an important choice is whether the location of monomers is confined to a lattice or can be varied continuously . when the configuration space is continuous , the distribution of energy barriers between energetically optimal conformations can extend to zero . therefore , the subtlety of distinctions between conformations can lead to a richer structure of the fitness landscape . we chose not increase the complexity of the model further and treated monomers as point - like particles in a chain where the distance between nearest neighbors is fixed but the angle between successive links in the chain in unrestricted . our level of abstraction is therefore somewhere between lattice models and all - atom descriptions of proteins @xcite . another important choice is the number of the model monomer types . again , we opted for an intermediate level of abstraction and chose four types of monomers : hydrophobic , hydrophilic , and charged . this choice drastically reduces the size of the sequence space while retaining some of the substitution complexity whereby hydrophilic and charged monomers can be swapped under some conditions without radically altering the native state . the intermediate level of abstraction in our approach has its pros and cons . although the model reproduces key features of protein folding such as the existence of the hydrophobic folding nucleus and two - stage folding kinetics @xcite , compact conformations certainly do not represent proteins . rather , we might think of our monomers as representing structurally grouped regions several ( perhaps up to a dozen ) amino - acids in length . compact conformations in the model might therefore be analogous to tertiary structures of proteins . representing sequence space with only four monomer types and treating mutations without reference to the underlying dna or genetic code does not accurately reflect the natural mutation process . however , our goal was to isolate the features of fitness landscapes which could be traced directly to the constraints imposed by the heteropolymer folding kinetics and energetics . we therefore used a simple sequence space and a homogeneous mutation model to avoid compounding the fitness landscape structure by the complexity derived from the mutation process . most importantly , our folding model has been shown to reproduce the observed universal distribution of the evolutionary rates of protein - coding genes as well as the dependencies of the evolutionary rate on protein abundance and effective population sizes @xcite . therefore , despite its simplicity , the behavior of this model might reflect important aspects of protein evolution . in particular , the conclusions drawn from the analysis of the model landscapes exhaustively explored here could also apply to the fitness landscapes of protein evolution . in the previous work , we concluded that the universal distribution of evolutionary rates and other features of protein evolution follow from the fundamental physics of protein folding @xcite . the results presented here suggest that the ( relative ) smoothness and a substantial deficit of peaks in the fitness landscapes of protein evolution that lead to mutational robustness and the ensuing evolvability could similarly follow from the fact that proteins are heteropolymers that have to fold in three dimensions to perform their functions . the experimental landscapes considered here are decidedly incomplete . due to experimental limitations , only the analysis of binary substitutions at a handful of sites is feasible at this time . the incompleteness of the empirical landscapes analyzed in this work could be the cause of the observed lack of peak suppression . this proposition will be put to test by the study of larger parts of experimental landscapes that are becoming increasingly available . the goal of this study is to explore the relationship between roughness and path divergence in realistic fitness landscapes . our polymer folding model provides a simple way of constructing such landscapes . the model has been described in detail previously @xcite . in brief , the model polymer is a flexible chain of monomers in which the the nearest neighbors interact via a stiff harmonic spring potential with rest length @xmath30 . the angles between the successive links in the chain are unrestricted . there are four types of monomers : hydrophobic , hydrophilic , and charged and . next nearest neighbors @xmath10 and @xmath25 in the chain and beyond interact via a pairwise potential @xmath31 where @xmath32 is the distance between monomers @xmath10 and @xmath25 , @xmath33 is the monomer s charge , @xmath34 is the debye - hckel screening length , and @xmath35 and @xmath36 depend on the pair in question . the interaction parameters are chosen to mimic the essential features of the amino - acid interactions . to emulate the effects of solvent , we assign a stronger attraction to the pair than to the , , and pairs . there is also a long range repulsion between and and even stronger repulsion between and the charged monomers . the values of the parameters are @xmath37 , debye - hckel screening length @xmath38 . the lennard - jones coefficients @xmath35 and @xmath36 are @xmath39 note that a @xmath40 can be substituted by a @xmath41 in the subscripts and the coefficients are symmetric with respect to the interchange of the indices . the energy of the chain is @xmath42 where the first term is the sum of the pairwise energies given by eq . ( [ eq:3 ] ) over non - nearest neighbor pairs , and the second term reflects the springs connecting nearest neighbors . the spring constant is proportional to temperature @xmath43 . the parameters are fixed for all simulation runs at @xmath44 , and the quench temperature @xmath45 . to mimic the observed tendency of the @xmath1 and @xmath46 termini to be in close proximity , we fixed the endpoint monomers of the model sequences to be of @xmath40 and @xmath41 types . dynamics of folding are simulated via overdamped brownian kinetics which are appropriate when inertial and hydrodynamic effects are not important . units are chosen so that each component @xmath47 of the @xmath10th monomer s coordinates @xmath48 is updated according to @xmath49 where @xmath50 is the time step and @xmath51 is a random variable with zero mean , variance @xmath52 , uncorrelated with @xmath53 for other times , monomers and spatial directions . the `` native structure '' of a particular sequence is represented by an equilibrium ensemble of conformations . the ensemble is constructed by identifying the typical folded conformation and measuring the characteristic rmsd @xmath34 due to thermal fluctuations in the folded state . three thousand quenches are then performed and the resulting folded conformations are accumulated . the equilibrium ensemble that represents the native structure is defined as the largest cluster of quenched conformations within rmsd distance @xmath34 from each other . thus , each conformation in the ensemble differs from any other by an amount comparable to the differences introduced by thermal fluctuations alone . the concept of the native structure ensemble allows us to compute the probability that a sequence folds to a particular structure in a natural , physically plausible fashion . given a native structure ensemble we assess its conformation space density by computing the distance @xmath54 between each member @xmath10 of the ensemble and its closest neighbor . given the set @xmath55 of these shortest distances we compute the median @xmath56 and the median absolute deviation ( mad ) @xmath57 . a new conformation is deemed to belong to the ensemble if the shortest distance from this conformation to the members of the ensemble is smaller than @xmath58 . given a native structure ensemble of some sequence @xmath59 we compute the probability @xmath21 that sequence @xmath60 ( which could be @xmath59 itself ) folds to the this structure by accumulating @xmath61 equilibrated quenched conformations of @xmath60 and using the above criterion to determine the fraction @xmath21 that belong to the native structure ensemble of @xmath59 . because @xmath61 sample conformations are computed , the smallest measurable @xmath21 is @xmath62 . the sample size used to measure @xmath63 dictated by the computational demands of the model , introduces a random component to the model fitness landscapes . as we report below , model landscapes turn out to be substantially smoother than random . therefore the underlying global structure of the model landscapes appears to survive the modest amount of randomness introduced by the relatively small sample size used for measuring @xmath21 . robust folders ( sequences with a high probability of correct folding ) tend to have large linear regions stretched by repulsive coulomb interactions . because the linear regions have no contacts with other monomers , we focused our attention on compact conformations with a high monomer contact density . substitutions in these higher complexity conformations were more likely to exhibit non - trivial effects . to find compact robust folders in the vast available sequence space of @xmath64-mers ( the sequences are of length @xmath65 but the endpoint monomer types are fixed ) with @xmath66 monomer types , we implemented a simulated annealing search which optimized the correct folding probability @xmath21 divided by the cube of the native conformation s radius of gyration . the search produced over 800 sequences with @xmath67 and at least two distinct regions of the polymer in mutual contact . we examined each single substitution mutant of a robustly folding sequence and computed the folding probability @xmath21 to the structure of the original sequence . all mutants with @xmath68 were added to the landscape and if @xmath69 their mutants were also examined . this process is repeated until all mutants of the last sequence under consideration have @xmath70 . from our study of complete landscapes we estimate that on average for each sequence with @xmath68 which is included into the landscape , roughly 6 others with @xmath71 need to be examined . since each quench and equilibration takes about 24 seconds , landscape construction takes roughly 30 minutes to an hour per included sequence . thus landscapes larger than 10,000 sequences take months to compile . at the time of submission , 39 complete landscapes have been constructed , the largest comprising 12969 sequences . the organization of the folding fitness landscapes and experimental landscapes were compared with perfectly additive landscapes perturbed by noise constructed as follows . each substitution to the peak fitness sequence was assigned a negative fitness differential drawn at random from an exponential distribution with parameter @xmath72 . the sum over the fitness differentials of a particular set of substitution was modified by either additive of multiplicative noise @xcite . additive noise is drawn from a gaussian distribution with zero mean and standard deviation @xmath73 which was varied between @xmath74 and @xmath75 . the multiplicative perturbation is achieved by multiplying the fitness by a number drawn from a uniform distribution @xmath76 raised to a positive power @xmath77 varied between @xmath78 and @xmath79 when @xmath77 is small , multiplicative factors are close to unity and the perturbation is small as well . if the perturbed fitness was positive , the mutant was included into the landscape . the noise amplitude was varied to obtain a family of landscapes of continuously varying roughness . only the data for the additive landscapes with multiplicative noise were included in this manuscript . landscapes perturbed by other types of noise exhibited essentially the same qualitative behavior . the studies on experimental fitness landscapes typically involve constructing a library of all possible combinations of binary mutations at a small number of sites . the first study included in the present analysis measured the minimum inhibitory concentrations ( mic ) of an antibiotic for a complete spectrum of mutants with modified tem @xmath3-lactamases ; the transition from the antibiotic - sensitive to the antibiotic - resistant form requires five mutation , so the landscape encompassed 120 mutational trajectories between the most distant points on the landscape ( or 32 sequences ) @xcite . the logarithm of mic was used as the proxy for fitness . in the second study , catalytic activity of 419 sesquiterpene synthase mutants that differed by at most 9 substitutions was measured @xcite . we used the catalytic specificity ( propensity for producing a particular reaction product rather than a broad spectrum of products ) of the mutant enzymes as the proxy for fitness . before performing the analysis , the fitnesses in the experimental landscapes are mapped onto the @xmath80 interval to enable meaningful quantitative comparisons of the roughness measures . 10 [ 1]`#1 ` urlstyle [ 1]doi:#1 [ 1 ] [ 2 ] _ _ _ _ _ _ _ _ _ _ _ _ _ _ key : # 1 + annotation : # 2 _ _ _ _ _ _ _ _ _ _ _ _ _ _ gould sj ( 1997 ) full house : the spread of excellence from plato to darwin . new york : three rivers press . conway ms ( 2009 ) the predictability of evolution : glimpses into a post - darwinian world . naturwissenschaften 96 : 13131337 . morris sc ( 2010 ) evolution : like any other science it is predictable . philos trans r soc lond b biol sci 365 : 133145 . koonin ev , wolf yi ( 2010 ) constraints and plasticity in genome and molecular - phenome evolution . nat rev genet 11 : 487498 . elena sf , lenski re ( 1997 ) test of synergistic interactions among deleterious mutations in bacteria . nature 390 : 395398 . parera m , perez - alvarez n , clotet b , martnez ma ( 2009 ) epistasis among deleterious mutations in the hiv-1 protease . mol biol evol 392 : 243250 . domingo - calap p , cuevas jm , sanjun r ( 2009 ) the fitness effects of random mutations in single - stranded dna and rna bacteriophages . plos genet 5 : e1000742 . miller sp , lunzer m , dean am ( 2006 ) direct demonstration of an adaptive constraint . science 314 : 458461 . lunzer m , miller sp , felsheim r , dean am ( 2005 ) the biochemical architecture of an ancient adaptive landscape . science 310 : 499501 . beerenwinkel n , pachter l , sturmfels b , elena sf , lenski re ( 2007 ) analysis of epistatic interactions and fitness landscapes using a new geometric approach . bmc evol biol 7 : 60 . martin g , elena sf , lenormand t ( 2007 ) distributions of epistasis in microbes fit predictions from a fitness landscape model . nat genet 39 : 555560 . lunzer m , golding gb , dean am ( 2010 ) pervasive cryptic epistasis in molecular evolution . genet 6 : e1001162 . weinreich dm , watson ra , chao l ( 2005 ) perspective : sign epistasis and genetic constraint on evolutionary trajectories . evolution 59 : 11651174 . voigt ca , kauffman s , wang zg ( 2000 ) rational evolutionary design : the theory of in vitro protein evolution . adv protein chem 55 : 79160 . bershtein s , segal m , bekerman r , tokuriki n , tawfik ds ( 2006 ) robustness - epistasis link shapes the fitness landscape of a randomly drifting protein . nature 444 : 929932 . tracewell ca , arnold fh ( 2009 ) directed enzyme evolution : climbing fitness peaks one amino acid at a time . curr opin chem biol 13 : 39 . romero pa , arnold fh ( 2009 ) exploring protein fitness landscapes by directed evolution . nat rev mol cell biol 10 : 866876 . kryazhimskiy s , tkacik g , plotkin jb ( 2009 ) the dynamics of adaptation on correlated fitness landscapes . proc natl acad sci u s a 106 : 1863843 . kauffman sa , weinberger ed ( 1989 ) the nk model of rugged fitness landscapes and its application to maturation of the immune - response . j theor biol 141 : 211245 . hayashi y , aita t , toyota h , husimi y , urabe i , et al . ( 2006 ) experimental rugged fitness landscape in protein sequence space . plos one 1 : e96 . carlson cd , warren cl , hauschild ke , ozers ms , qadir n , et al . ( 2010 ) specificity landscapes of dna binding molecules elucidate biological function . proc natl acad sci u s a 107 : 45444549 . knight cg , platt m , rowe w , wedge dc , khan f , et al . ( 2009 ) array - based evolution of dna aptamers allows modelling of an explicit sequence - fitness landscape . nucleic acids res 37 : e6 . rowe w , wedge dc , platt m , kell db , knowles j ( 2010 ) predictive models for population performance on real biological fitness landscapes . bioinformatics 26 : 21452152 . rowe w , platt m , wedge dc , day pj , kell db , et al . ( 2010 ) analysis of a complete dna - protein affinity landscape . j r soc interface 7 : 397408 . weinreich dm , delaney nf , depristo ma , hartl dl ( 2006 ) darwinian evolution can follow only very few mutational paths to fitter proteins . science 312 : 111114 . lozovsky er , chookajorn t , brown km , imwong m , shaw pj , et al . ( 2009 ) stepwise acquisition of pyrimethamine resistance in the malaria parasite . proc natl acad sci u s a 106 : 1202512030 . novais a , comas i , baquero f , cantn r , coque tm , et al . ( 2010 ) evolutionary trajectories of beta - lactamase ctx - m-1 cluster enzymes : predicting antibiotic resistance . plos pathog 6 : e1000735 . omaille pe , malone a , dellas n , jr bah , smentek l , et al . ( 2008 ) quantitative exploration of the catalytic landscape separating divergent plant sesquiterpene synthases . nat chem biol 4 : 617623 . poelwijk fj , kiviet dj , weinreich dm , tans sj ( 2007 ) empirical fitness landscapes reveal accessible evolutionary paths . nature 445 : 383386 . kogenaru m , de vos mg , tans sj ( 2009 ) revealing evolutionary pathways by fitness landscape reconstruction . crit rev biochem mol biol 44 : 169174 . dawid a , kiviet dj , kogenaru m , de vos m , tans sj ( 2010 ) multiple peaks and reciprocal sign epistasis in an empirically determined genotype - phenotype landscape . chaos 20 : 026105 . poelwijk fj , tnase - nicola s , kiviet dj , tans sj ( 2011 ) reciprocal sign epistasis is a necessary condition for multi - peaked fitness landscapes . j theor biol 272 : 141144 . carneiro m , hartl dl ( 2010 ) colloquium papers : adaptive landscapes and protein evolution . proc natl acad sci u s a 107 suppl 1 : 17471751 . lobkovsky ae , wolf yi , koonin ev ( 2010 ) universal distribution of protein evolution rates as a consequence of protein folding physics . proc natl acad sci u s a 107 : 29832988 . drummond da , wilke co ( 2008 ) mistranslation - induced protein misfolding as a dominant constraint on coding - sequence evolution . cell 134 : 341352 . drummond da , wilke co ( 2009 ) the evolutionary consequences of erroneous protein synthesis . nat rev genet 10 : 715724 . wolf yi , gopich iv , lipman dj , koonin ev ( 2010 ) relative contributions of intrinsic structural - functional constraints and translation rate to the evolution of protein - coding genes . genome biol evol 2 : 190199 . kimura m ( 1962 ) on probability of fixation of mutant genes in a population . genetics 47 : 713 . wagner a ( 2008 ) robustness and evolvability : a paradox resolved . proc biol sci 275 : 91100 . draghi ja , parsons tl , wagner gp , plotkin jb ( 2010 ) mutational robustness can facilitate adaptation . nature 463 : 353355 . masel j , trotter mv ( 2010 ) robustness and evolvability . trends genet 26 : 406414 . govindarajan s , goldstein ra ( 1998 ) on the thermodynamic hypothesis of protein folding . proc natl acad sci usa 95 : 55455549 . taverna dm , goldstein ra ( 2002 ) why are proteins so robust to site mutations ? j mol biol 315 : 47984 . goldstein ra ( 2011 ) the evolution and evolutionary consequences of marginal thermostability in proteins . proteins 79 : 1396407 . tiana g , shakhnovich be , dokholyan nv , shakhnovich ei ( 2004 ) imprint of evolution on protein structures . proc natl acad sci u s a 101 : 284651 . shakhnovich e ( 2006 ) protein folding thermodynamics and dynamics : where physics , chemistry , and biology meet . chem rev 106 : 15591588 . zeldovich kb , chen p , shakhnovich ei ( 2007 ) protein stability imposes limits on organism complexity and speed of molecular evolution . proc natl acad sci usa 104 : 1615216157 . zeldovich kb , shakhnovich ei ( 2008 ) understanding protein evolution : from protein physics to darwinian selection . ann rev phys chem 59 : 105127 . bastolla u , roman he , vendruscolo m ( 1999 ) neutral evolution of model proteins : diffusion in sequence space and overdispersion . j theor biol 200 : 4964 . bastolla u , porto m , roman he , vendruscolo m ( 2006 ) a protein evolution model with independent sites that reproduces site - specific amino acid distributions from the protein data bank . bmc evol biol 6 : 43 . zhang j , li w , wang j , qin m , wu l , et al . ( 2009 ) protein folding simulations : from coarse - grained model to all - atom model . iuibmb life 61 : 62743 . gillespie b , plaxco kw ( 2004 ) using protein folding rates to test protein folding theories . annu rev biochem 73 : 837859 . finkelstein av , ivankov dn , garbuzynskiy so , galzitskaya ov ( 2007 ) understanding the folding rates and folding nuclei of globular proteins . curr protein pept sci 8 : 521536 . aita t , uchiyama h , inaoka t , nakajima m , kokubo t , et al . ( 2000 ) analysis of a local fitness landscape with a model of the rough mt . fuji - type landscape : application to prolyl endopeptidase and thermolysin . biopolymers 54 : 6479 .

the next linear collider ( nlc ) will explore @xmath1 processes in the tev energy regime , and probe the standard model of elementary particles to great accuracy . electroweak processes which may reveal new physics , such as supersymmetry , are especially interesting . therefore the accurate calculation of the scattering amplitudes of such high energy processes in the standard model is very important . the main corrections to the born amplitudes at high energies , @xmath2 , are double logarithmic ( dl ) contributions of the form @xmath3 which arise from soft vector boson exchanges . here we derive the dl asymptotics of scattering amplitudes , for a sequence of gauge theories leading up to the standard electroweak model . the dl correction for qed processes dates back many years to the paper by sudakov @xcite . in section 2 we present the original sudakov form factor for qed in a form in which it can be generalized to non - abelian gauge theories , a subject to which we then turn . we start with the processes where all the scalar products @xmath4 of the momenta of participating particles are of the same order . in section 3 we compute the dl corrections to processes governed by massless non - abelian gauge theories , and then , in section 4 , we turn to the interesting case of broken gauge symmetries , discussing , in particular , the dl effects for standard model electroweak processes . in all these cases the dl terms can be resummed in exponential forms which are generalizations of @xmath5 here @xmath6 is the probability of the emission of a soft and quasi - collinear gauge boson from an external particle @xmath7 , and the summation is over all @xmath8 external charged particles participating in the process . the virtual particles are subject to an infrared cut - off @xmath9 on their transverse momenta . we use an equation for the evolution of the scattering amplitudes as a function of the infrared cut - off @xmath9 of the transverse momenta of virtual particles . our approach is based on a gauge invariant dispersive method which is a generalization of a powerful theorem on photon bremsstrahlung due to gribov @xcite . his remarkable theorem , proved by dispersive methods , states that , due to gauge invariance , the region of applicability of well known formulas for accompanying photon bremsstrahlung is considerably extended at high energies . in these formulas the amplitude for a process with the emission of a soft photon is expressed , in a factorized form , in terms of the amplitude for the non - radiative process , with the particles taken _ on - mass - shell_. since the amplitude is taken on - shell , _ gauge invariance _ is guaranteed . the evolution equation approach greatly simplifies the computation of dl effects for electroweak processes , which are mediated by broken symmetry with the photon having components in both the @xmath10(2 ) and @xmath11(1 ) gauge groups . the evolution equations for the amplitude as a function of the infrared cut - off parameter @xmath12 are analogous to the renormalization group equations ( rge ) . basically we start from the domain of very large @xmath12 , where the born amplitude applies , and evolve down to small @xmath12 , matching the solution at the mass thresholds bounding every new kinematic domain . it is therefore not surprising that the structure of the exponentiation of the dl effects has a clear physical interpretation . of course for observable processes it is necessary to consider the dl corrections to both the non - radiative and the radiative processes . for radiative processes we again have exponentiation of the sudakov dl corrections in a form similar to ( [ eq : a1 ] ) . the dependence on the infrared cut - off parameter @xmath9 is canceled in the measurable semi - inclusive process , and is replaced by a dependence on the experimental acceptance cuts on the soft boson emissions . in sections 3.1 , 4.1 and 4.2 we address these issues . as well as dl corrections of the sudakov type , there are dl contributions specific to forward or backward scattering , when a final particle is quasi - collinear to an initial particle and their energies are almost equal . this is the domain of regge kinematics and is the subject of section 5 . here we have the dl corrections related to the exchange of soft pairs of fermions or of bosons in the crossed channel . the infrared evolution equations are now non - linear in the amplitude , and in the @xmath13-plane representation are of ricatti - type form . in simple cases the latter forms can be reduced to a schrdinger - type equation with a harmonic potential . this approach can also be applied to production processes in the multi - regge kinematics . in this case the evolution equations are solved in a sequence of domains starting from the region where the born amplitude applies and evolving down in the cut - off @xmath9 to the region where soft particles are emitted . the vector boson and fermion reggeization can be also easily verified in the dl approximation . moreover , the infrared evolution equation allows the construction of scattering amplitudes with quasi - elastic unitarity . the high energy asymptotics of electromagnetic processes was calculated many years ago within the framework of qed @xcite@xcite . in particular the amplitude for @xmath1 elastic scattering at a fixed angle ( @xmath14 a fictitious plays the role of the infrared cut - off . in physical cross sections the divergence in @xmath15 of the elastic amplitude is canceled with the analogous divergences in processes with soft photon emissions . ] photon mass ) in the dl approximation has the form @xcite @xmath16 where @xmath17 is the born amplitude for @xmath1 scattering and @xmath18 is the sudakov form factor . the dl approximation applies in the energy regime @xmath19 where the qed coupling @xmath20 . thus each charged external particle effectively contributes @xmath21 to the total amplitude . the sudakov form factor appears in the elastic scattering of an electron off an external field @xcite . it is of the form : @xmath22 to specify @xmath23 it is convenient to use the sudakov parametrization of the momentum of the exchanged virtual photon : @xmath24 where @xmath25 and @xmath26 are the initial and final momenta of the scattered electron . @xmath27 can then be written as the integral over @xmath28 and @xmath29 : @xmath30 where @xmath31 . the first two factors in the integrand correspond to the propagators of the virtual fermions which occur in the one - loop triangle sudakov diagram . the @xmath32 - function appears as a result of the integration of the propagator of the photon over its transverse momentum @xmath33 , noting that the main contribution comes from the region near the photon mass shell @xcite : @xmath34 to dl accuracy ( 6 ) gives for @xmath35 : @xmath36 where the result comes equally from two different kinematical regions , @xmath37 and @xmath38 . therefore one can write @xmath39 . we can obtain physical insight by presenting the two equal contributions separately . in the first region , with @xmath40 , the virtual photon is emitted along @xmath25 and the parameter @xmath29 is given by the ratio of energies of the photon and the initial electron . here instead of @xmath28 , it is convenient to use eq . ( [ eq : kp ] ) to replace it by the square of the transverse momentum component of the photon . then integrating over @xmath29 and @xmath41 gives @xmath42 in the dl approximation , which may be evaluated to give half of @xmath23 . the quantity @xmath43 is proportional to the probability @xmath44 of the emission of a soft and almost collinear photon from an external particle with energy @xmath2 and mass @xmath45 , i.e. @xmath46 if several charged particles participate in a process , for example @xmath47 , then analogous contributions appear for each external line , provided all external invariants are large and of the same order . this leads to the general result @xmath48 where @xmath8 is the number of external lines corresponding to charged particles . in summary the soft emissions described by the sudakov form factor is a quasi - classical effect which does not depend on the hard dynamics of the process . in particular there are no quantum mechanical interference effects in the dl sudakov corrections , for large scattering angles . sudakov effects have been widely discussed for non - abelian gauge theories , such as @xmath49 and can be calculated in various ways ( see , for instance , @xcite ) . we consider here the scattering amplitude in the simplest kinematics when all its invariants @xmath50 are large and of the same order @xmath51 . a general method of finding the dl asymptotics ( not only of the sudakov type ) is based on the infrared evolution equations describing the dependence of the amplitudes on the infrared cutoff @xmath9 of the virtual particle transverse momenta @xcite . this cutoff plays the same role as @xmath52 in qed , but , unlike @xmath52 , it is not necessary that it vanishes and it may take an arbitrary value . it can be introduced in a gauge invariant way by working , for instance , in a finite phase space volume in the transverse direction with linear size @xmath53 . instead of calculating asymptotics of particular feynman diagrams and summing these asymptotics for a process with @xmath8 external lines it is convenient to extract the virtual particle with the smallest value of @xmath54 in such a way , that the transverse momenta @xmath55 of the other virtual particles are much bigger @xmath56 for the other particles @xmath57 plays the role of the initial infrared cut - off @xmath58 . in particular , the sudakov dl corrections are related to the exchange of soft gauge bosons , see fig . 1 . for this case the integral over the momentum @xmath59 of the soft ( i.e. @xmath60 ) virtual boson with the smallest @xmath61 can be factored off , which leads to the following infrared evolution equation : @xmath62 where the amplitude @xmath63 on the right hand side is to be taken on the mass shell , but with the substituted infrared cutoff : @xmath64 . the generator @xmath65 acts on the color indices of the particle with momentum @xmath66 . the non - abelian gauge coupling is @xmath67 . in eq . ( [ eq : vem ] ) , and below , @xmath68 denotes the component of the gauge boson momentum @xmath69 transverse to the particle emitting this boson . note that in sudakov dl corrections there are no interference effects , so that we can talk about the emission ( and absorption ) of a gauge boson by a definite ( external ) particle , namely by a particle with momentum almost collinear to @xmath69 . it can be expressed in invariant form as @xmath70 for all @xmath71 . the above factorization is related to a non - abelian generalization of the gribov theorem ) below , together with a description of its essential content . ] for the amplitude of the bremsstrahlung of a photon with small transverse momentum @xmath72 in high energy hadron scattering @xcite . the form in which we present eq . ( [ eq : vem ] ) corresponds to a covariant gauge for the gluon with momentum @xmath59 . formally this expression can be written in a gauge invariant way if we include in the sum the term with @xmath73 ( which does not give a dl contribution ) . indeed , in this case we can substitute @xmath74 by @xmath75 , where the polarization matrices of the boson @xmath76 in the various gauges differ by the terms proportional to @xmath77 or @xmath78 giving a vanishing contribution due to the conservation of the total color charge @xmath79 . thus we have the possibility of choosing appropriate gauges for each kinematical region of quasi - collinearity of @xmath59 and @xmath80 . we can , however , use ( [ eq : vem ] ) as well , noting that in this region for @xmath81 we have @xmath82 , where @xmath83 is the energy of the particle with momentum @xmath80 and @xmath84 the frequency of the emitted gauge boson , so that : @xmath85 where @xmath86 is the eigenvalue of the casimir operator @xmath87 ( @xmath88 for gauge bosons in the adjoint representation of the gauge group @xmath49 and @xmath89 for fermions in the fundamental representation ) . the differential form of the infrared evolution equation follows immediately from ( [ eq : dgvem ] ) : @xmath90 where @xmath91 with @xmath92 as in the abelian case , @xmath93 is the probability to emit a soft and almost collinear gauge boson from the particle @xmath94 with mass @xmath95 , subject to the infrared cut - off @xmath9 on the transverse momentum . note again that the cut - off @xmath9 is not taken to zero . the function @xmath43 is determined by ( 9 ) for arbitrary values of the ratio @xmath96 . to logarithmic accuracy , we obtain from ( [ eq : wl ] ) : @xmath97 the infrared evolution equation ( [ eq : ee ] ) should be solved with an appropriate initial condition . in the case of large scattering angles , if we choose the cut - off to be the large scale @xmath98 then clearly there are no sudakov corrections . the initial condition is therefore @xmath99 and the solution of ( [ eq : ee ] ) is thus given by the product of the born amplitude and the sudakov form factors : @xmath100 therefore we obtain an exactly analogous sudakov exponentiation for the gauge group @xmath49 to that for the abelian case , see ( [ eq : a11 ] ) . theories with semi - simple gauge groups can be considered in a similar way . since ultimately we are interested in measurable cross sections we have to consider the dl corrections to amplitudes of processes with soft emissions , as well as those without . only in inclusive cross sections will the dependence on the infrared cut - off @xmath101 disappear , being replaced by parameters specifying the experimental acceptance . to put it in another way , cross sections of the emission processes receive large ( dl ) contributions from regions where the emitted bosons are soft and the emission angles are small . therefore , to be consistent , we need to calculate cross sections of such processes as well . this is easy to do for qed processes , where the single gauge boson ( the photon ) is neutral and does not possess self - interactions . therefore soft photons are emitted independently according to a poisson distribution . in non - abelian theories , gauge bosons are not neutral and interact with each other . consequently , the soft emission does not follow a poisson distribution @xcite . we again consider the simplest situation , when the additional soft gauge boson is emitted in the process with all invariants @xmath102 large . of course , for the emission of a boson almost collinear to the particle the direction of the particle with momentum @xmath103 , the invariant @xmath104 is small in comparison with @xmath98 . in the case of non - abelian gauge theories the corresponding amplitude for the emission of a soft gauge boson with small @xmath105 has , according to the gribov theorem , the following form : @xmath106 the possible corrections to this factorized expression are of the order of @xmath107 . however , to dl accuracy , we can substitute @xmath58 in the arguments of the scattering amplitudes by its boundary value @xmath57 . notice that the amplitude on the r.h.s . of ( [ eq : rem ] ) is taken on - the - mass shell , which guarantees its gauge invariance . the result ( [ eq : rem ] ) is highly non - trivial in the feynman diagram approach . it means , that the region of applicability of the classical formulas for the bremsstrahlung amplitudes is significantly enlarged at high energies . v. gribov proved this result by using dispersion relations in the variables @xmath108 , and demonstrating that for small @xmath109 the pole terms in these invariants are much larger than the corresponding cut contributions due to the gauge invariance of the theory @xcite . the region of applicability of ( [ eq : rem ] ) corresponds to the situation when the momentum of the emitted soft boson does not spoil the kinematics of the non - radiative process . this implies that the frequency @xmath110 of the boson emitted off the particle with momentum @xmath80 should be much smaller than the energy @xmath83 , and that the emission angle latexmath:[$\vartheta _ { l}\simeq the angle between @xmath80 and all other momenta @xmath112 ( otherwise one needs to include interference effects ) . for the physical coulomb gauge ( @xmath113 ) in the kinematical region where the gauge boson is emitted at a small angle @xmath114 with respect to the particle with momentum @xmath103 in ( [ eq : rem ] ) only the term with @xmath115 contributes . the method based on the infrared evolution equation allows us to calculate , in the dl approximation , the amplitudes of the hard processes accompanied by the emission of any number of soft gauge bosons @xcite . let us consider the amplitude for the emission of one soft gauge boson in the region of quasi - collinearity of its momentum with momentum @xmath103 ( i.e. , the emission of a soft boson by a particle with momentum @xmath103 ) . when the transverse momentum @xmath109 of this boson is much less than the infrared cut - off @xmath9 , used for the virtual particles , the amplitude is given by the term with @xmath115 in the sum in ( [ eq : rem ] ) . but we need to know the emission amplitude in the opposite case , @xmath116 . it can be found from the evolution equation in this region using expression ( [ eq : rem ] ) at @xmath117 as the initial condition . the kernel of the evolution equation in this region differs from the corresponding kernel in the region @xmath118 ( that is the kernel ( 16 ) of the evolution equation for the amplitude without emission ) by a term connected with the emission of a virtual boson from the real gauge boson with momentum @xmath59 : @xmath119 where @xmath120 is given by ( [ eq : wl ] ) with @xmath88 and @xmath121 , see fig . it is clear , that this new term in the kernel for evolution from @xmath122 to @xmath58 leads to an additional term @xmath123 in the sudakov exponential . thus , the amplitude for the emission of one gauge boson with small transverse momentum from the hard scattering process is of the form : @xmath124 @xmath125 we note again that here @xmath122 means the square of the component of the three - dimensional momentum transverse to the momentum of the emitting particle , say for example , @xmath80 . we can write @xmath122 in the invariant form @xmath126 with @xmath81 , which does not depend on @xmath13 . again we see that we have the exponentiation of the sudakov dl corrections . note that in the abelian qed case we have @xmath127 and the exponent for the photon bremsstrahlung amplitude remains the same as that for the process without photon emission . it is related to the poisson distribution for soft photon production . the exponentiation of virtual dl corrections holds for multiple emission processes as well . in qed it is trivial , since the soft photons are emitted independently . in non - abelian gauge theories it is not so simple . the main complexity is connected with the nontrivial structure of the amplitudes for multiple emission of real soft gauge bosons , arising from their self - interaction . but in the dl approximation these amplitudes can be calculated . due to the coherence effect , the branching cascade develops only in the region of sequentially shrinking angular cones @xcite . in this region the born amplitudes for multiple emission processes have a factorized form and the virtual corrections exponentiate @xcite . it is proved by solving the infrared evolution equation in a series of regions where the infrared cut - off @xmath9 is bounded between a sequence of decreasing transverse momenta of the emitted gluons , using in each region the solution of the previous region as the initial condition . the final result is @xmath128 @xmath129 where @xmath130 are the momenta of the emitted gluons with strongly ordered energies and @xmath131 are their components transverse to the momenta of the emitting jets . the same method , based on the infrared evolution equation , is also applicable to broken gauge theories . let us consider for definiteness the standard electroweak theory , where the physical gauge bosons are a massless photon ( described by the field @xmath132 ) and massive @xmath133 and @xmath134 bosons ( described correspondingly by fields @xmath135 and @xmath136 ) . to dl accuracy , all masses can be set equal : @xmath137 and the energy considered to be much larger , @xmath138 . in the unbroken phase the corresponding abelian and yang - mills fields are denoted by @xmath139 and @xmath140 , with @xmath141 . the physical fields are linear combinations of the fields in the unbroken theory with coefficients depending on the weinberg angle @xmath142 . the left and right handed fermions are correspondingly doublets ( @xmath143 ) and singlets ( @xmath144 ) of the @xmath10(2 ) weak isospin group and have hypercharge @xmath145 related to the electric charge @xmath146 , measured in units of the proton charge , by the gell - mann - nishijima formula @xmath147 . in the evolution equation in the dl approximation the value of the infrared cutoff @xmath12 can be chosen in two different ranges : 1 ) @xmath148 and 2 ) @xmath149 . the second case is universal in the sense that it does not depend on details of the electroweak theory . it will be discussed below . in the first region we can neglect spontaneous symmetry breaking effects , in particular gauge boson masses , and consider the evolution equation in the unbroken phase with effectively massless particles @xmath150 and @xmath151 . of course one could calculate everything also in terms of the physical fields @xmath152 and @xmath153 . in the unbroken phase this is equivalent to the description in terms of the original fields @xmath154 and @xmath155 and leads to the same final result , but the intermediate steps will be more complicated because there are cancellations between non - exponentiating terms from diagrams with @xmath134 and @xmath156 exchanges . the separation of these contributions is not gauge invariant and if we would consider the diagrams without virtual photons we would violate @xmath157 symmetry . taking into account only such diagrams leads to nonexponentiating dl effects in an axial gauge @xcite . 3 illustrates this at the two loop level . the loss of gauge invariance is related to the fact that the photon field contains the component @xmath158 of the non - abelian field @xmath159 , and so omitting the virtual photons would violate the conservation of the weak isospin current ( in the unbroken theory ) . [ fig : zg ] + + + + in region 1 ) the infrared evolution equation , written in terms of the unbroken fields , is of a form analogous to ( [ eq : ee ] ) if we assume , for simplicity , that all the charged particles have masses @xmath160@xmath161 here @xmath162 is the total weak isospin of particle @xmath7 , @xmath163 is its weak hypercharge , and @xmath67 and @xmath164 are the couplings of the @xmath165 and @xmath166 gauge groups , respectively . the sum in ( [ eq : ubee ] ) is to be performed over all @xmath8 external particles . as before , the initial condition is given by the requirement that for the infrared cut - off @xmath167 we obtain the born amplitude . the solution of ( [ eq : ubee ] ) is thus given by @xmath168 . \label{eq : ubs } \end{aligned}\ ] ] the expression in the brackets in the exponential can be written in terms of the parameters of the broken theory as follows : @xmath169 where the three terms on the r.h.s . correspond to the contributions of the soft photon ( interacting with the electric charge @xmath170 ) , the @xmath133 and the @xmath134 bosons , respectively . although we may rewrite solution ( [ eq : ubs ] ) in terms of the parameters of the broken theory in the form of a product of three exponents corresponding to the exchanges of photons , @xmath133 and @xmath134 bosons , it would be wrong to identify the contributions of the diagrams without virtual photons with this expression for the particular case @xmath171 . this becomes evident when we note that if we were to omit photon lines then the result would depend on the choice of gauge , and therefore be unphysical . only for @xmath172 , where the photon coincides with the @xmath150 gauge boson , would the identification of the @xmath173 term with the contribution of the diagrams with photons be correct . choosing the cutoff @xmath12 in the second region , @xmath149 , the infrared evolution equation takes the following form : @xmath174 evidently , only the photon contribution remains in this region . now the appropriate initial condition is given by ( [ eq : ubs ] ) evaluated at the matching point @xmath175 . the solution is thus @xmath176 \nonumber \\ & & \times \exp \left [ - \frac{1}{2 } \sum_{i=1}^{n } q^2_i \left ( w_i(s,\mu^2 ) - w_i(s , m^2 ) \right ) \right ] \nonumber \\ & = & { \cal m}_{\rm born}(p_{1}, ... ,p_{n } ) \nonumber \\ & & \times \exp \left [ -\frac{g^{2}}{2(4\pi ) ^{2}}\sum_{i=1}^{n}\left ( \left\ { t_{i}(t_{i}+1)+\tan ^{2}\theta _ { w}\left ( \frac{y_{i}}{2}\right ) ^{2}\right\ } \log ^{2}\frac{s}{m^{2}}\right ) \right ] \nonumber \\ & & \times \exp \left [ -\sum_{i=1}^{n } \ ! \left ( \frac{e_{i}^{2}}{(4\pi ) ^{2}}\ ! \left ( \log \frac{s}{m_{i}\,m}\log \frac{m^{2}}{m_{i}^{2}}+\log \frac{s}{m_{i}^{2}}\log \frac{m_{i}^{2}}{\mu ^{2 } } \ ! \right ) \ ! \right ) \ ! \right ] , \label{eq : ubsa } \end{aligned}\ ] ] where the last equality holds for @xmath177 and @xmath178 from the respective expansions of @xmath179 in ( [ eq : a10 ] ) . let us stress that ( [ eq : ubs ] ) and ( [ eq : ubsa ] ) are applicable for processes involving chiral fermions as well as gauge bosons , provided that all the invariants are large ( @xmath180 ) compared to @xmath181 . note , that in the case , when quarks or gluons participate in the reaction , we should multiply these expressions by the sudakov factors corresponding to the virtual gluons emitted by these colored particles . the infrared evolution equations ( [ eq : ubee ] ) and ( [ eq : ubeea ] ) have a clear physical meaning analogous to the renormalization group equations and therefore it is natural to expect that the next - to - leading corrections to the kernels can be calculated . for physical observables soft real photon emission must be taken into account in an inclusive way and effectively the parameter @xmath101 in ( [ eq : ubsa ] ) will be replaced by parameters depending on the experimental requirements . the calculation of amplitudes for processes with the emission of a gauge boson in the kinematical region which gives dl contributions to the cross sections ( i.e. the region of soft quasi - collinear emission ) is similar to the analogous calculation for unbroken gauge theories , with complications of the type that we discussed above . one is that we have to consider separately two regions of @xmath109 of the emitted boson : first @xmath182 and second @xmath183 . at high energies the cross sections of the emission processes receive dl contributions from both of these regions . of course , @xmath133 and @xmath134 boson emissions contribute in the second region only . therefore consideration of the first case is very simple . for values of the infrared cut - off @xmath184 the amplitude for the emission of a soft photon by a particle with momenta @xmath80 ( i.e. emission within the cone along @xmath185 not containing the momenta of the other particles ) has , in the physical ( coulomb ) gauge , a factorized form @xmath186 according to the gribov theorem of ( [ eq : rem ] ) . however , if @xmath187 , the kernel of the infrared evolution equation does not change when the cut - off @xmath58 changes from the domain @xmath188 to @xmath189 ( since the @xmath133 and @xmath134 bosons do not contribute in this first region ) . therefore , ( [ eq : g1 ] ) remains valid at arbitrary values of the cut - off @xmath9 . in the second region , @xmath190 , the result is more involved . we need to start from @xmath191 , where we can use a generalization of gribov s theorem . consider again the case when a gauge boson is emitted by a particle with momentum @xmath80 . we have @xmath192 where @xmath193 is given by ( [ eq : ubs ] ) and @xmath194 with @xmath195 . then we have to solve the evolution equation in the region @xmath196 with the initial condition given by ( [ eq : g2 ] ) at the matching point @xmath197 . in fact , it is more appropriate to work in terms of the fields @xmath140 and @xmath139 . the kernel of the evolution equation remains unchanged for the emission of the @xmath150-particle due to its abelian nature . on the other hand the emission of the @xmath198 -particle leads to the same additional contribution as in the unbroken theory ( see ( [ eq : f1 ] ) ) . therefore in the cut - off region @xmath199 , we obtain @xmath200 with the same amplitude @xmath193 as in ( [ eq : ubs ] ) and @xmath201 finally we study the region of the infrared cut - off @xmath202 . in this region the kernel of the evolution equation is determined by the electromagnetic interaction only ; therefore , the only contribution related to the emitted particles is that for @xmath133emission . this contribution is given by @xmath203 where @xmath204 is defined by ( [ eq : a10 ] ) and ( 9 ) with @xmath205 . consequently , for values of the infrared cut - off @xmath202 we obtain : @xmath206 where @xmath193 is given by ( [ eq : ubsa ] ) and @xmath207 the important point here is the difference of the dl exponent for a nonradiative process and the process with photon emission , which leads to a violation of the poisson distribution for photons in the dl approximation at high energies . it is a direct consequence of the fact that the photon has a non - abelian component . measurable cross sections have an inclusive nature ( at least with respect to photons , since only cross sections with an infinite number of emitted soft photons are observable ) . let us consider the dl corrections to such cross sections . it is clear that for the emission of real gauge bosons the same cut - off @xmath208 must be used as for virtual ones . therefore , to calculate an experimentally measured cross section we have to take the cut - off @xmath58 less than the lower bound @xmath209 of @xmath210 of those photons emitted in processes which are not included in the cross section . the calculation of the cross section is simple if the experimental conditions are such that only processes with emission of photons with @xmath211 are allowed . in this case the non - abelian component of the photon is not essential , so that photon emissions obey a poisson distribution . therefore for the cross section with an arbitrary number of emitted photons with momenta lying inside regions @xmath212 of the momentum space around the emitting particles with momenta @xmath103 , we obtain : @xmath213 where @xmath214 is the cross section of the non - radiative process and @xmath215 is the probability of the emission of photons with @xmath216 inside the allowed region @xmath217 since the upper bound on @xmath57 of the photons which are allowed to be radiated is less than @xmath218 , we must use the cut - off @xmath219 and , consequently , ( [ eq : ubsa ] ) for the matrix element of the non - radiative process . therefore , we obtain @xmath220 \nonumber \\ & & \times \exp \left [ -\sum_{i=1}^{n}q_{i}^{2}\left ( w_{i}(s,\mu ^{2})-w_{i}(s , m^{2})\right ) + w_{\exp}^{\gamma } \right ] \ , , \label{eq : h3 } \end{aligned}\ ] ] where @xmath221 is given by ( 10 ) and ( 9 ) . evidently , the dependence on @xmath9 in @xmath222 and @xmath215 cancels in the exponential . if we include in the observed cross section the emission of gauge bosons with transverse momenta larger than @xmath223 , then the problem becomes much more complicated because of the non - poisson distribution of soft emission in non - abelian gauge theories @xcite . let us consider here the simplest example of the cross section completely inclusive of photons emission in the two - loop approximation . then the cross section can be written as @xmath224 where @xmath225 is the one - loop virtual correction , @xmath226 comes from one - photon emission taken in the born approximation and @xmath227 from the one - loop correction to one - photon emission . due to exponentiation of the dl terms of the sudakov - type in virtual corrections , the term @xmath228 in ( [ eq : h4 ] ) gives the two - loop virtual correction . the term @xmath229 gives the correction from two - photon emission in the born approximation , since in this approximation the two photons are emitted independently . in the considered case of the cross section completely inclusive of photon emission the cut - off parameter @xmath58 can be taken as large as @xmath218 ( but not greater , because the cross section does not include @xmath133 and @xmath134 emission ) . each of the corrections considered above depends on @xmath230 , but their sum in ( [ eq : h4 ] ) does not ; therefore , we can take the most suitable value of the cut - off to calculate the cross section . it is easy to see that the most convenient choice is @xmath231 . in this case from ( 26 ) we have : @xmath232 the correction due to one - photon emission taken in the born approximation is @xmath233 where @xmath234 is given by ( [ eq : h2 ] ) with @xmath231 , @xmath235 and the region @xmath212 defined by inequality @xmath236 . it gives @xmath237 the one - loop contribution to this correction @xmath227 can be found with help of ( 26 ) , ( 31 ) , ( 32 ) and ( 33 ) : @xmath238\ ] ] @xmath239 \ , . \label{eq : h7}\ ] ] the important point here is that the virtual correction to the one - photon emission does not coincide with the corresponding correction to the non - radiative process ( which means violation of the poisson distribution ) and depends on the momentum of the emitted photon . it is a consequence of the fact that the photon has a non - abelian component . one can check that using an arbitrary cut - off @xmath219 gives the same result , but the calculation is more complicated . for two particle scattering in either the forward @xmath240 or the backward @xmath241 directions , corresponding to the regge regime , there are situations , where the amplitudes have dl contributions different from those of the sudakov type . these logarithms appear from feynman diagrams in which the soft particles with the minimal transverse momenta are pairs of virtual fermions or bosons exchanged in the crossed @xmath242 or @xmath28 channels . for qed and qcd the infrared evolution equations for such regge processes are known [ 4 , 5 , 7 ] . the new features presented by the standard model of weak interactions are that the gauge group is semi - simple , @xmath243 , and that there is a large difference in the particle mass scales . the general strategy of finding dl asymptotics in this case is to first solve the evolution equation in the region @xmath244 , where @xmath245 is the largest particle mass in the theory ; then to solve it in the region @xmath246 , where @xmath247 is the next largest particle mass , using the solution of the previous equation at @xmath248 as the initial condition , and to proceed with these steps until the intermediate particles are light quarks , electrons , neutrinos and photons . for simplicity we consider only the evolution equation in the region where the cut - off parameter @xmath9 is much larger than all particle masses . in this case it is natural to calculate the amplitudes in terms of the massless particles of the unbroken theory , that is the leptons and quarks , the higgs , @xmath150 and @xmath198 bosons . by solving the generalized infrared evolution equation in the dl approximation we sum up contributions of ladder - type diagrams in the crossed channel , with all possible insertions of the gauge bosons , leading to the sudakov double logarithms . because the transverse momenta of the virtual particles are strongly ordered , the integral kernels in the ladder diagrams are given by the splitting kernels of the dglap evolution equ ation @xcite which describes all possible transitions among fermions and bosons in the standard model . these splitting kernels can be simplified since the regge kinematic domain corresponds to a strong ordering of the bjorken variables @xmath249 along the ladder . for definiteness we study the simple case of the backward lepton ( @xmath94 ) - antilepton ( @xmath250 ) scattering for @xmath251 . in the born approximation , the contribution of the feynman diagrams with the higgs , photon , @xmath150 and @xmath198 boson exchanges depend on the helicities @xmath252 ( with @xmath253 ) of the initial and final fermions , which have momenta @xmath254 we consider the scattering of the leptons @xmath255 and @xmath256 belonging to the first generation with the smallest masses . because the coupling of the higgs particle to leptons is proportional to their mass , we can neglect the higgs contribution . the dirac spinors describing the @xmath94 and @xmath257 states with the definite helicities @xmath52 are @xmath258 where @xmath259 with the use of these expressions we can calculate the matrix elements of the @xmath260-matrix structures which appear in the born approximation for backward lepton - antilepton scattering , @xmath261 , @xmath262 it is important , to note that both the helicity structures of @xmath263 and @xmath264 are proportional to @xmath265 , which means , that the helicities of the two fermions in the crossed @xmath28 channel are opposite in sign . therefore the born diagram with a virtual @xmath198 boson gives a negligible contribution to the backward scattering . because the @xmath150 boson interacts with hypercharge @xmath145 , its contribution to the born amplitude is also zero for the right - handed neutrinos ( if they were to exist ) . below we consider only the non - trivial case , when only @xmath266 , @xmath267 and @xmath268 can participate in the reaction . because the hypercharge @xmath269 for left - handed fermions and @xmath270 for the right - handed electron , in this case the born amplitude is given by @xmath271 in the dl approximation we also obtain a significant simplification of the helicity structure of the scattering amplitude : @xmath272 where the functions @xmath273 describe contributions with positive and negative signature : @xmath274 the amplitudes @xmath275 are assumed to satisfy dispersion relations of the form @xmath276 where in the born approximation @xmath277 is different from zero only in the region where @xmath278 is of the order of unity . in particular , from the known behavior of the born contribution @xmath279 at large @xmath98 and its degeneracy in the signature we obtain @xmath280 the absence of degeneracy of the amplitude with respect to signature is a result of the presence of the non - planar diagrams which arise from the virtual soft @xmath150-bosons in the infrared evolution equations ( cf . the qcd case @xcite ) : @xmath281 @xmath282 @xmath283 @xmath284 where we introduced the notation @xmath285 the linear terms in @xmath275 , on the right - hand side of the evolution equations , are the sudakov contributions which arise from soft @xmath198 and @xmath286 virtual bosons coupling to external lines with different momenta @xmath287 and @xmath288 . the non - linear terms appear due to soft pairs of leptons exchanged in the crossed @xmath28 channel . note that , in the above equations , the term related to the @xmath98-channel sudakov vertex with @xmath198 exchange is proportional to the casimir operator @xmath289 for the weak isospin group @xmath165 , and the total @xmath150 boson contribution from the four sudakov vertices having singularities in the @xmath98 and @xmath242 channels is proportional to @xmath290 . the corresponding contribution of these vertices to the terms mixing the amplitudes with different signatures is proportional to @xmath291 for @xmath198 exchange and @xmath292 for @xmath150 exchange . non - linear terms take into account the contribution of the ladder diagrams for two leptons interacting through @xmath150 exchanges . the above equations can be simplified if we use the mellin representation for the amplitudes : @xmath293 where @xmath294 are the signature factors @xmath295 and where the integration contour @xmath296 is situated along the imaginary axes to the right of all singularities of the @xmath242-channel partial waves @xmath297 . in the @xmath13-representation , taking into account the degeneracy of the partial waves with different signatures in the born approximation , the evolution equations can be written in the form of ricatti equations [ 7 ] : @xmath298 if we take into account only the terms with effective dl variables @xmath299 and neglect other terms . furthermore , using the following definition for the @xmath242-channel partial waves @xmath300 the equations can be written as follows @xmath301 both equations can be reduced to linear form after introducing new functions @xmath302 and @xmath303 according to the definitions @xmath304 they have the form of schrdinger equations @xmath305 the first is an equation for the harmonic oscillator . therefore taking into account the boundary condition @xmath306 so as to match onto perturbation theory , we obtain the solution @xmath307 where @xmath308 is the parabolic cylinder function : @xmath309 therefore for the scattering amplitude with positive signature we have , to dl accuracy , @xmath310 with @xmath311 . the equation for negative signatures can be easily solved for @xmath312 ( corresponding to complex values of the weinberg angle ) : @xmath313 where @xmath314 is the confluent hypergeometric function : @xmath315 therefore for @xmath312 , the amplitude with negative signature , in the dl approximation , is @xmath316 with @xmath311 . for general values of the parameter @xmath317 the solution can be obtained as a perturbation series in @xmath318 . this perturbative expansion is convergent for all values of the dl parameter @xmath319 . because the function @xmath308 does not have any singularities in the complex plane , the high energy asymptotics of the scattering amplitude with positive signature @xmath320 is determined by the position of its zeroes , @xmath321 , in the left - half plane , re@xmath322 . the function @xmath323 has a singularity @xmath324 as @xmath325 , and therefore the scattering amplitude with negative signature behaves as @xmath326 at high energies . such behavior of @xmath327 is valid also for other values of the parameter @xmath317 . note , that similar dl asymptotics were obtained in qed for @xmath1 backward scattering amplitudes . for energies smaller than the masses of @xmath198 and @xmath134 bosons , only diagrams with virtual photons give rise to a dl contribution . the photons interact with electric charge @xmath328 , which is the same for left- and right - handed electrons . it is interesting , that at very large energies , the behavior of @xmath1 backward scattering amplitudes is different from the qed prediction even in the case of the abelian standard model with @xmath329 the dl asymptotics of lepton - antilepton forward scattering amplitudes , which is related to @xmath242-channel exchange of two fermions and an arbitrary number of @xmath198 and @xmath150 bosons , can be obtained in a similar way . in this case one should take into account the diagram with a virtual @xmath198 boson even in the born approximation . the reggeization of the fermions and @xmath198 boson can also be verified in the dl approximation with the use of the infrared evolution equation . moreover , one can construct amplitudes with quasi - elastic unitarity @xcite , which is important for the theory of the bfkl pomeron @xcite . some other applications of the dl asymptotics in qcd are given in @xcite , @xcite . we have calculated the double logarithmic ( dl ) corrections to the amplitudes of a number of different high energy processes , in particular , of electroweak processes in the standard model . these sudakov - type corrections , which are found to exponentiate , are crucial for the high precision studies planned at the next linear @xmath330 collider . our approach is based on the use of an evolution equation in the infrared cut - off parameter @xmath9 , which in turn is based on a generalization of a gauge invariant dispersive method for photon bremsstrahlung originally due to gribov . we have assumed that the energy of the process is much larger than all the masses in the theory . however new physics could appear in the tev region and it will modify our dl asymptotic forms . because our approach is based on the infrared evolution equation , we need simply to change the initial conditions in this region to incorporate the new particles and/or new interactions of the existing particles . the infrared evolution equation has the form of a renormalization group equation in the two - dimensional impact parameter subspace . the usual bfkl equation can be also considered as a simplified version of the renormalization group equation but in two - dimensional longitudinal subspace . in this case , in the leading logarithmic approximation @xmath331 , instead of their dependence from the infrared cut - off @xmath12 , we have conformal invariance of the @xmath242-channel partial waves in impact parameter space . however , in the next - to - leading approximation , the effect of the running coupling leads to violation of conformal invariance @xcite . after the breaking of conformal invariance the equation takes the form of a quantized callan - symanzik equation @xcite . therefore , even in the case of the bfkl equation , the renormalization group has its peculiarities . it is related to the fact , that the equation determines not only the anomalous dimensions of the operators of different twists , but also their relative contributions . the expressions for the dl asymptotics that we have obtained in the standard model can be used in perturbation theory to verify the first order , and to predict the higher order , expressions for scattering amplitudes , as has been done for the dl asymptotics in qcd @xcite , @xcite . finally , we reiterate that the accurate calculation of scattering amplitudes is important because the effects of new physics can be rather small . after this work was completed a paper by p. ciafaloni and d. comelli @xcite appeared , in which the electroweak dl corrections are considered in the particular case of massless @xmath332 production by a source which is a singlet with respect to the @xmath243 gauge group . the results of this paper ( in particular , nonexponentiating dl corrections ) are not in agreement with our corresponding results . the difference appears due to the fact that the method in @xcite consists of factorizing off the virtual gauge boson with the softest ( i.e. with smallest frequency ) momentum @xmath333 by computing external line insertions only , and in iterating this procedure by setting the virtual momentum @xmath334 in the left - over diagram . we believe that this approach , which is in disagreement with the method based on the infrared evolution equation , is not valid . we thank intas ( 95 - 0311 ) and the royal society for support . we also thank ms . sharon fairless for help in preparing the manuscript . v. n. gribov , yad . fiz . 5 ( 1967 ) 399 ( sov . . phys . * 5 * ( 1967 ) 280 ) ; sov . * 12 * ( 1971 ) 543 ; l. n. lipatov , nucl . phys . * b307 * ( 1988 ) 705 ; v. del duca , nucl . phys . * b345 * ( 1990 ) 369 . v. n. gribov , g. v. frolov , v. g. gorshkov and l. n. lipatov , phys . lett . * 22 * ( 1966 ) 671 ; yad . fiz . * 6 * ( 1967 ) 129 , 361 ( sov . j. nucl * 6 * ( 1967 ) 95 ) ; v. g. gorshkov , sov . * 6 * ( 1968 ) 95 ; v. g. gorshkov , l. n. lipatov , m. m. nesterov , yad . * 9 * ( 1969 ) 1221 ( sov . * 9 * ( 1969 ) 714 ) . l. n. lipatov , yad . fiz . * 6 * ( 1967 ) 564 ( sov . * 6 * ( 1967 ) 411 ) ; yad * 14 * ( 1971 ) 160 , 396 ( sov . * 14 * ( 1971 ) 90 , 223 ) ; v. g. gorshkov and l. n. lipatov , yad * 9 * ( 1969 ) 818 ( sov . * 9 * ( 1969 ) 480 ) . j. j. carazzone , e. c. poggio and h. r. quinn , phys . rev . * d11 * ( 1975 ) 2286 ; j. m. cornwall and g. tiktopolous , phys . * 35 * ( 1975 ) 338 ; phys . rev . * d13 * ( 1976 ) 3370 ; j. frenkel and j. c. taylor , nucl . phys . * b116 * ( 1976 ) 185 ; v. v. belokurov and n. i. usyukina , phys . lett . * b94 * ( 1980 ) 251 ; theor . * 44 * ( 1980 ) 657 ; * 45 * ( 1980 ) 957 . e. a. kuraev and v. s. fadin , yad . fiz . * 27 * ( 1978 ) 1107 ( sov . * 27 * ( 1978 ) 587 ) ; b. i. ermolaev and v. s. fadin , jetp lett . * 33 * ( 1981 ) 269 ; a. h. mueller , phys . lett . * b108 * ( 1982 ) 355 ; b. i. ermolaev , l. n. lipatov and v. s. fadin , yad . fiz . * 45 * ( 1987 ) 817 ( sov . * 45 * ( 1987 ) 508 ) ; b. i. ermolaev , l. n. lipatov , int . j. mod . phys . * a4 * ( 1989 ) 3147 . v. n. gribov and l. n. lipatov , sov . 15 * ( 1972 ) 438 , 675 ; l. n. lipatov , yad . * 20 * ( 1974 ) 181 ( sov . * 20 * ( 1974 ) 94 ) ; g. altarelli and g. parisi , nucl . phys . * b126 * ( 1977 ) 298 ; yu . i. dokshitzer , sov . jetp * 46 * ( 1977 ) 641 . lipatov , sov . * 23 * ( 1976 ) 642 ; v.s . fadin , e.a . kuraev and l.n . lipatov , phys . lett . * b60 * ( 1975 ) 50 ; v.s . fadin , e.a . kuraev and l.n . lipatov , sov . jetp * 45 * ( 1977 ) 199 ; ya.ya . balitsky and l.n . lipatov , sov . * 28 * ( 1978 ) 822 . r. kirschner and l. n. lipatov , nucl . * b213 * ( 1983 ) 122 ; b. i. ermolaev , s. i. manaenkov and m. g. ryskin , z. phys . * c69 * ( 1996 ) 259 ; j. bartels , b. i. ermolaev and m. g. ryskin , z. phys . * c70 * ( 1996 ) 273 ; z. phys . * c72 * ( 1996 ) 627 ; s. i. manaenkov , z. phys . * c75 * ( 1997 ) 685 ; y. kiyo and j. kodaira , z. phys . * c74 * ( 1997 ) 631 ; b. badalek and j. kwiecinski , phys . lett . * b418 * ( 1998 ) 229 ; j. bartels , t. gehrmann and m. g. ryskin , desy 99 - 65 , hep - ph/9906204 ; b. i. ermolaev , m. greco and s. i. troyan , cern - th/99 - 155 .

the quantum mechanical spin degree of freedom is now widely exploited to control current transport in electronic devices . for instance , the readout of magnetic hard disks is based on the spin - valve effect , i.e. the tunability of a conductance through the relative orientation of some ferromagnetic polarizations @xcite . however , realizing spin injection in mesoscopic conductors would allow to implement further functionalities , like e.g. a gate control of the spin valve effect @xcite . importantly , electronic interaction effects can occur in mesoscopic structures , due to the electronic confinement . this raises the fundamental question of the interplay between spin - dependent transport and electronic interactions . upon scattering on the interface between a ferromagnet ( f ) and a non - magnetic material , electrons with spin parallel or antiparallel to the magnetization of f can pick up different phase shifts , because they are affected by different scattering potentials . this _ spin - dependence _ of interfacial phase shifts ( sdips ) can modify significantly the behavior of mesoscopic circuits . first , when a mesoscopic conductor is connected to several f leads with non collinear polarizations , the sdips produces an interfacial precession of spins which can modify current transport in the device @xcite . secondly , in collinear configurations , precession effects are not relevant , but the sdips can modify mesoscopic coherence effects . for instance , in superconducting / ferromagnetic hybrid circuits , the sdips introduces a phase shift between electron and holes correlated by andreev reflection @xcite . references and have identified signatures of this effect in the experiments of refs . and , respectively . in principle , normal systems in collinear configurations can also be affected by the sdips . indeed , from ref . , the sdips should produce a spin - splitting of the resonant states in a ballistic interactionless wire contacted with collinearly polarized ferromagnetic leads . however , this has not been confirmed experimentally yet @xcite . and @xmath5 . the magnetic polarizations @xmath6 and @xmath7 of leads @xmath8 and @xmath5 can be parallel ( configuration p ) or antiparallel ( configuration ap ) . the element m is capacitively coupled to a gate voltage source @xmath1 . ] recently , ref . has reported current measurements in a single wall carbon nanotube ( swnt ) connected to two ferromagnetic leads with collinear polarizations . the asymmetries observed in the magnetoresistance ( @xmath0 ) of the swnt versus gate voltage are strikingly similar to those predicted by ref . for an interactionless wire subject to the sdips@xcite . however , the swnt of ref . showed a quantum dot behavior with strong coulomb blockade effects , as demonstrated in a great number of experiments with non - magnetic leads ( see e.g. @xcite ) . therefore , one important question is how interaction effects modify the scheme proposed by ref . . the problem of the effects of interactions on the transport properties of a central region connected to ferromagnetic contacts has already been considered in various regimes , like e.g. the coulomb blockade regime@xcite , the kondo regime@xcite , the luttinger liquid regime@xcite and the marginal fermi liquid regime@xcite . this article develops an approach suitable for the limit of ref . and studies , for the first time , the effect of the sdips on a quantum dot . we consider a quantum dot coupled to metallic leads through spin - active interfaces . * * * * we use an anderson model to study the @xmath0 of the circuit above the kondo temperature , but beyond the sequential tunneling limit . the sdips is taken into account through an effective spin - splitting of the dot energy levels . this splitting makes easier negative @xmath0 effects . when it is strong enough , it can even lead to a giant @xmath0 with a sign oscillating with the dot gate voltage @xmath1 , similarly to what has been found in the non - interacting case . in the non - interacting case , assuming that the properties of the contact are constant with energy and that the sdips is too weak to split the conductance peaks , one finds that the @xmath4 pattern is similar for all conductance peaks . in contrast , the effect of the sdips depends on the occupation of the dot in the interacting case . this is in apparent contradiction with the data of ref . because , in the @xmath3range presented in this ref . , the @xmath4 oscillations are regular . using a two - orbitals model , which takes into account the @xmath9 orbital degeneracy commonly observed for swnts ( see e.g. refs . ) , one can solve this discrepancy . in this framework , we expect non - regular @xmath4 oscillations if the experiment is repeated on a larger @xmath3range . this article is organized as follows : we start with summarizing the results found for the non - interacting case in section [ sec : non - interacting ] . then we introduce a model for the interacting case in section [ sec : interacting ] . section [ sec : single - level ] addresses the case of a one - orbital quantum dot circuit , and section [ sec : two - level ] the case of a two - degenerate - orbitals quantum dot circuit . finally section [ sec : conclusion ] concludes . we consider a mesoscopic element m connected to ferromagnetic leads @xmath8 and @xmath5 ( fig . [ figure1 ] ) . the chemical potential of m can be shifted by @xmath10 using the gate voltage @xmath1 , with @xmath11 the ratio between the gate capacitance and the total capacitance of @xmath12 . the magnetic polarizations @xmath6 and @xmath7 of leads @xmath8 and @xmath5 can be parallel ( configuration @xmath13 ) or antiparallel ( configuration @xmath14 @xmath15 ) . before introducing the interacting model investigated in this article , it is useful to reconsider the results obtained by ref . for the case in which m is a non - interacting single - channel ballistic wire of length @xmath8 . in a scattering approach@xcite , the conductance of the circuit depends on the transmission probability @xmath16 for electrons with spin @xmath17 through contact @xmath18 , and on the reflection phase @xmath19 for electrons with spin @xmath20 coming from the wire towards contact @xmath21 . the index @xmath22 $ ] denotes the parallel [ antiparallel ] leads configuration . a spin dependence of @xmath19 can occur due to the magnetic properties of the contact materials used to engineer lead @xmath21 . due to size quantization , the conductance @xmath23 of the circuit in configuration @xmath24 presents fabry - perot like resonances for @xmath25 , with @xmath26 a resonant energy , @xmath27 , @xmath28 the wire fermi energy , @xmath29 the density of orbitals states at the fermi level in the wire and @xmath30 the spin direction opposite to @xmath20 ( we have used @xmath31 ) . from this eq . , in configuration @xmath24 , the sdips produces a spin - splitting @xmath32 of the resonant energies . when the effective field @xmath33 is strong enough to produce a spin - splitting of the conductance peaks , the circuit can display a giant @xmath0 effect with a sign oscillating with @xmath1 , due to the strong shift of the conductance peaks from the @xmath34 to the @xmath15 configurations . in the opposite case , @xmath0 remains smaller , but the sdips can still be detected through characteristic asymmetries in the oscillations of @xmath0 versus @xmath1 ( see fig . 2-right of ref . ) importantly , assuming that @xmath16 and @xmath19 are constant with @xmath1 , one has @xmath35)$ ] . this implies that when @xmath33 is not strong enough to produce a spin - splitting of the conductance peaks , the @xmath4 pattern is similar for all the peaks displayed by @xmath36 . we now assume the presence of strong coulomb interactions inside m , such that we have a quantum dot connected to ferromagnetic leads . such a system can be realized for instance by using granular films@xcite , nanoparticles@xcite , carbon nanotubes@xcite , or @xmath37 molecules@xcite . in the non - interacting case of section [ sec : non - interacting ] , we have considered that the spin - dependent confinement potential felt by electrons causes the sdips , which leads to the spin - splitting of the resonant states . in the interacting case , the scattering approach is not suitable anymore . however , the energy of the quasi - bound single particle states in quantum dot m can depend on spin due to the spin - dependent confinement potential . on this ground , we adopt the effective anderson hamiltonian @xmath38 with @xmath39 here , @xmath40 refers to the energy of the dot orbital state @xmath41 for spin @xmath20 , @xmath42 to the energy of lead state @xmath43 for spin @xmath20 and @xmath44 is an hoping matrix element ( we assume that the spin @xmath20 is preserved upon tunneling like in section [ sec : non - interacting ] ) . the index @xmath43 runs over the electronic states of lead @xmath8 and @xmath5 . coulomb interactions are taken into account through the term in @xmath45 , with @xmath46 and @xmath47 the total capacitance of the quantum dot m. * * * * by construction of the model ( see above ) , for @xmath48 , each orbital level @xmath40 corresponds to a resonant level @xmath49 of section [ sec : non - interacting ] , with @xmath50 . we can therefore regard the effective zeeman splitting @xmath33 in model ( [ h ] ) as a generalization of the sdips concept to the interacting case@xcite . the specificity of this effective field , with respect to an ordinary external field , is that it depends on the configuration @xmath24 of the ferromagnetic electrodes . for instance , in the case of symmetric ferromagnetic contacts , symmetry considerations lead to @xmath51 and @xmath52 . in the following , we calculate the zero - bias conductance of the circuit using @xcite @xmath53\text { .}\nonumber\end{aligned}\ ] ] the above equation involves the retarded green s function @xmath54 with @xmath55 . we also use the fermi distribution @xmath56)^{-1}$ ] and the tunnel transition rates @xmath57 with @xmath58 . note that @xmath59 , @xmath40 and @xmath60 depend on the configuration @xmath61 considered but for simplicity we omit the index @xmath24 in those quantities . we want to study current transport in the limit studied in ref . , i.e. the width of conductance peaks displayed by the circuit is determined not only by temperature but also by the tunnel rates ( @xmath62 @xmath63 ) . this requires to go beyond the sequential tunneling description , i.e. to take into account high - order quantum tunneling processes . for this purpose , we will calculate @xmath59 using the equation of motion ( e.o.m . ) technique@xcite , which is valid for temperatures larger than the kondo temperature @xmath64 of the system@xcite . for simplicity , we first take into account a single orbital level @xmath41 of the dot . we follow the lines of ref . . the e.o.m . technique leads to @xmath65 with @xmath66 \label{article::eq:1}%\ ] ] the average occupation of orbital @xmath41 by electrons with spin @xmath20 . we define @xmath67^{-1}\ , , \label{article::eq:2}%\]]@xmath68^{-1}\ , , \label{article::eq:3}%\]]@xmath69 and , for @xmath70 , @xmath71 here , one has @xmath72 , @xmath73 and @xmath74 ( we anticipate on the next paragraphs by defining @xmath75 for @xmath76 and an arbitrary dot state @xmath77 , but only @xmath78 and @xmath79 are needed for the present one - orbital case ) . we assume that the coupling to the leads is energy independent ( broad band approximation ) , which gives e.g. @xmath80 . the term @xmath81 , which is due to the tunneling of electrons with spin @xmath20 , already occurred in the non - interacting case@xcite . in the parallel configuration ( red full lines ) and conductance @xmath82 in the antiparallel configuration ( black dotted lines ) as a function of the gate voltage @xmath1 , for the circuit shown in fig . [ figure1 ] , with m a 1-orbital quantum dot . we have used @xmath83 , @xmath84 , @xmath85 , @xmath86 and @xmath52 . panels b , d , and f : magnetoresistance @xmath87 ( pink curves ) corresponding to the left conductance plots . the results are shown for @xmath88 ( panels a and b ) , @xmath89 ( panels c and d ) and @xmath90 ( panels e and f ) . ] ( blue lines ) and @xmath91 ( black lines ) of level @xmath41 by spins @xmath92 and @xmath93 as a function of @xmath1 , for the 1-orbital quantum dot circuit of fig . [ figure1 ] . the results are shown for the same parameters as in fig . [ figure2 ] and lead polarizations in the parallel configuration ( @xmath13 ) , with @xmath94 ( panel a ) , @xmath95 , ( panel b ) and @xmath96 ( panel c ) . for @xmath94 , @xmath97 and @xmath91 remain very close , simply showing two steps corresponding to the two conductances peaks visible in fig . [ figure2]-a . in the case of a finite @xmath98 , @xmath99 rises more strongly than @xmath100 at the first conductance peak , revealing that current transport is due in majority to @xmath92 spins for this first peak . then , @xmath97 and @xmath100 become closer when both @xmath101 and @xmath102 are below the fermi level . at the second conductance peak @xmath91 rises more strongly than @xmath97 because current transport is now dominated by down spins . the asymmetry between the behaviors of spins @xmath92 and @xmath93 increases with @xmath98 ( from left to right panels ) . ] in the interacting case , @xmath103 also involves @xmath104 terms related to the tunneling of electrons with spin @xmath30 . the average occupation @xmath105 can be calculated from eqs . ( [ gtot ] ) and ( [ article::eq:1 ] ) as @xmath106 with , for @xmath107 , @xmath108 figure [ figure2 ] shows the conductance @xmath109 in configuration @xmath61 ( panels a , c and e ) and the magnetoresistance @xmath87 ( panels b , d and f ) calculated for different values of @xmath33 , using @xmath110}% ^{l}=\gamma_{l}(1\pm p_{l})$ ] for @xmath18 . we have used parameters consistent with ref . , i.e. @xmath86 , and tunnel rates @xmath111 leading to the proper width and height for the conductance peaks . we have also used relatively low values for @xmath112 because usual ferromagnetic contact materials are not fully polarized @xcite . the conductance peak corresponding to level @xmath41 is split due to coulomb interactions ( see eq . ( [ gtot ] ) , figs . [ figure2]-a , [ figure2]-c and [ figure2]-e ) . at low temperatures @xmath113 , kondo effect is expected in the valley between the two resulting peaks . we have checked that the hypothesis @xmath114 and hence the e.o.m . technique are valid for the parameters of fig . [ figure2 ] ( see refs . ) . for @xmath115 , we already note a strong qualitative difference with the non - interacting case : although the two conductance peaks displayed by @xmath36 are very similar , the @xmath0 variations corresponding to these two peaks have different shapes@xcite . more precisely , for the low values of polarization considered here , @xmath4 is approximately mirror symmetric from one conductance peak to the other . note that in fig . [ figure2 ] , we have used specific parameters such that @xmath0 remains positive for any value of @xmath1 when there is no sdips . nevertheless , it is possible to have @xmath116 for @xmath117 , for instance by increasing @xmath112 ( not shown ) . we now address the effect of a finite effective field @xmath33 . this field produces a shift of the conductance peaks from the @xmath34 to the @xmath15 configurations . for instance , in fig . 2-c and fig . 2-e , plotted for @xmath51 and @xmath52 , the left [ right ] conductance peak is shifted to the right [ left ] from @xmath34 to @xmath15 because it mainly comes from the transport of up [ down ] spins in the @xmath34 case ( this can be seen from the average occupations of the levels versus @xmath1 in fig . [ figure1n ] ) . as a consequence , in fig . [ figure2 ] , @xmath0 becomes negative for certain values of gate voltage . the effective field @xmath33 thus enhances negative @xmath0 effects . if @xmath33 is strong enough , it can even produce a giant @xmath0 effect with its sign tunable with @xmath1 ( fig . [ figure2]-f ) . moreover , because of the opposite shifts of the two consecutive conductance peaks for @xmath13 with respect to those for @xmath118 , the positive - and - then - negative profile of mr corresponding to one conductance peak is generally followed by the negative - and - then - positive profile near the next conductance peak ( approximately mirror symmetric ) . note that a sign change @xmath119 will not modify this behavior for the low values of polarizations considered here , because the spin with lower[higher ] orbital energy will dominate in the left [ right ] peak of @xmath36 . we now compare the results of this section with the experimental data of ref . . like many coulomb blockade devices , the circuit studied in this experiment suffered from low frequency @xmath1-noise , which can be attributed to charge fluctuators located in the vicinity of the device . a strong gate voltage offset jump occurred at @xmath120 , and the data before and after this jump do not necessarily correspond to the filling of consecutive levels . therefore , we will focus on the data taken for @xmath121@xmath122 shown @xcite with black squares in fig . [ figure4 ] . these data display almost 2 regular @xmath4 oscillations , which can not be understood with the 1-orbital model . indeed , as explained above , in this model , the two conductance peaks of @xmath123 are shifted in opposite directions by @xmath33 . as a consequence , the @xmath4 variations corresponding to these two peaks can not be similar for parameters consistent with the experiment . we have shown here curves for @xmath52 , but a finite @xmath124 would not modify this result . using values of @xmath112 larger than in fig . [ figure2 ] would not help either . for simplicity , we have considered in this section the one - orbital case . in reality , there are more than one orbital levels on a quantum dot . as long as these levels are sufficiently well separated from each other ( roughly , by an orbital energy difference larger than the hund - rule exchange energy ) , the two conductance peaks associated to a given level will occur consecutively in @xmath23 and will thus be described qualitatively like above@xcite . in particular , the two peaks will be shifted in opposite directions by @xmath125 ; the first peak to lower values of @xmath1 and the second peak to higher values . therefore , this limit should not allow to obtain two consecutive conductance peaks with analogue @xmath126 patterns . on the contrary , if two ( or more ) levels are nearly degenerate , it is possible that the orbital levels of the quantum dot are not filled one by one while increasing @xmath1 . therefore , consecutive conductance peaks may exhibit a qualitatively different behavior compared with the one - orbital case . to examine this effect , we will consider in next section the extreme case of a quantum dot with a twofold orbital degeneracy . we will see that the discrepancy between the theory and the data can be resolved by using this model . before concluding this section , we make a remark on another possible contribution to the spin splitting of the conductance peaks . even though so far we have mainly considered the contribution from the sdips , in principle , virtual particle exchange processes with the spin - polarized leads can also renormalize the energy levels through the @xmath127 terms of eq . ( [ b ] ) @xcite . indeed , the @xmath104 terms are not negligible in general . for example , in fig . [ figure2]-a , they globally shift the position of the conductance peaks in @xmath36 by about @xmath128 of @xmath129 . nevertheless , for the low values of polarizations @xmath112 and the temperatures used here , the level spin - splitting produced by the @xmath104 terms is much weaker than this global shift and can not compete with the finite values of @xmath33 considered in this article . in order to improve the understanding of ref . , we now take into account the @xmath130-@xmath131 orbital degeneracy commonly observed@xcite in swnts , by considering a two - orbitals model i.e. hamiltonian ( [ h ] ) with @xmath132 and @xmath133 . interestingly , su(4 ) kondo effect involving the orbital and spin degrees of freedom was observed in swnts with the @xmath130-@xmath131 degeneracy @xcite . this suggests that , in this system , the orbital quantum number is conserved during higher order tunnel events , probably because the electrons of the nanotube quantum dot are coupled to the nanotube section underneath the contacts , where they dwell for some time before moving into the metal . for simplicity , we will also assume such a situation here and disregard high - order quantum processes which couple the @xmath130 and @xmath131 orbitals @xcite . in order to calculate the conductance of the system from eq . ( [ cond ] ) , one needs to calculate the retarded green s function @xmath134 for @xmath135 . for this purpose , we again use the eom technique . since it is not possible to obtain a simple analytical expression for @xmath134 in the two - orbitals case , we show below the system of equations of motion calculated by neglecting electronic correlations between the dot and the leads ( @xmath114 ) . using @xmath136 , @xmath137 , @xmath138 and @xmath139 to denote four different dot states in the ensemble @xmath140 , we obtain in the parallel configuration ( red full lines ) and conductance @xmath82 in the antiparallel configuration ( black dotted lines ) , for the circuit of fig . [ figure1 ] , with m a two - orbitals quantum dot . we have used identical tunnel rates to the two orbitals , i.e. @xmath141 , @xmath142 , and @xmath143 . we have also used @xmath86 and @xmath52 . panels b , d and f : magnetoresistance @xmath0 ( pink full lines ) corresponding to the left conductance plots . the results are shown for @xmath88 ( panels a and b ) , @xmath144 ( panels c and d ) and @xmath145 ( panels e and f ) . ] @xmath146 @xmath147 @xmath148 @xmath149 due to interaction @xmath150 , the green s function @xmath151 is coupled to other green s functions @xmath152 with @xmath153 and @xmath154 for @xmath155 . this means that the dynamics of electrons in state @xmath136 is modified by the presence of other electrons on the dot [ in the one orbital case , @xmath156 was coupled to @xmath157 only , which lead to simple expression ( [ gtot ] ) ] . the term @xmath158 is the tunneling self energy for a non - interacting quantum dot , already introduced in previous section . the equations of motion also involve terms @xmath104 , which are defined by eq.([b ] ) , and terms defined by @xmath159 these terms take into account the tunneling of electrons between the leads and a dot state @xmath160 different from @xmath161 . the average level occupations occuring in eqs . ( [ a1])([a4 ] ) are given by @xmath162 $ ] . the green s functions and the level occupations can be calculated numerically from the above equations . up to now , the e.o.m . technique for multilevel systems had been implemented only by neglecting @xmath104 and @xmath163 terms @xcite . like in the one - orbital case , these terms are not negligible in the context of our study . ( blue lines ) and @xmath164 ( black lines ) of levels @xmath130 and @xmath165 by spins @xmath92 and @xmath93 as a function of @xmath1 , for a two - orbitals quantum dot circuit with the same parameters as in fig . [ figure3 ] . the results are shown for lead polarizations in the parallel configuration ( @xmath13 ) , with @xmath94 ( panel a ) , @xmath144 ( panel b ) , and @xmath166 ( panel c ) . for @xmath94 , @xmath167 and @xmath168 remains very close , showing four steps corresponding to the four conductances peaks visible in fig . [ figure3]-a . in the case of a finite @xmath98 , @xmath167 rises more strongly than @xmath169 for the two first conductance peaks , which shows that current transport is due in majority to up spins for these two peaks . the opposite situation occurs for the two last conductance peaks . the asymmetry between the behaviors of spins @xmath92 and @xmath93 increases with @xmath98 ( from left to right panels ) . ] figure [ figure3 ] shows the conductance ( panels a , c and e ) and @xmath0 curves ( panels b , d and f ) calculated from eqs . ( [ cond ] ) and ( [ a1]-[a4 ] ) , for different values of @xmath33 . for simplicity , we have assumed that the coupling to the leads is identical for the two orbitals , i.e. @xmath170}^{l}=\gamma_{k^{^{\prime}% } \uparrow\lbrack\downarrow]}^{l}=\gamma_{l}(1\pm p_{l})$ ] for @xmath18 . we have again used parameters consistent with ref . i.e. @xmath86 , relatively low polarizations @xmath171 and values of @xmath111 leading to the proper width and height for the conductance peaks . we have checked that these parameters are compatible with the hypothesis @xmath114 , with @xmath64 the kondo temperature associated to the su(4 ) kondo effect expected in this system @xcite . in most cases , the curves @xmath23 show 4 resonances , the first two associated with a single occupation of @xmath130 and @xmath165 , and the other two to double occupation ( see e.g. fig . [ figure3]-a ) . for @xmath172 and the parameters used here , @xmath0 remains positive for any value of @xmath1 ( fig . [ figure3]-b ) . like in the 1-orbital case , a finite @xmath33 makes easier negative @xmath0 effects and can even lead to a giant @xmath0 effect with a sign tunable with @xmath1 ( figs 4-d and 4-f ) . importantly , the effect of @xmath33 depends on the occupation of the dot . for instance , in fig . [ figure3]-e plotted for @xmath173 larger than the linewidth of the conductance peaks , the first two conductance peaks of @xmath174 ( peaks 1 and 2 ) are strongly shifted to the left by @xmath98 because they are due in majority to up spins , as can be seen from the average occupation of the levels in fig . [ figure2n]-c . this allows to get a @xmath0 pattern approximately similar for these two peaks , i.e. a transition from positive to negative values of @xmath175 ( fig . [ figure3]-f ) . on the contrary , peak 4 corresponds to a transition from negative to positive values of @xmath0 because the associated conductance peak is due in majority to down spins . in fig . [ figure3]-e , the shape of the @xmath4 pattern associated to peak 3 is more particular ( positive / negative / positive ) because , for the values of parameters considered here , coulomb blockade does not entirely suppress the up spins contribution in peak 3 , which is therefore spin - split @xcite . remarkably , this allows to obtain , at the left of fig . [ figure3]-f , three positive @xmath0 maxima which differ in amplitude but have rather similar shapes . in the case of @xmath176 finite but smaller than the linewidth of the conductance peaks ( fig . [ figure3]-c ) , the amplitude of the @xmath0 signal is much smaller than in the previous case but its shape remains comparable . in the parallel configuration ( top panel ) and the corresponding magnetoresistance @xmath177 ( bottom panel ) . the theory is shown for parameters consistent with the experiment i.e. @xmath178 , @xmath86 and @xmath179 . we also use relatively low values of polarization @xmath143 because usual ferromagnetic contact materials are not fully polarized @xcite . assuming identical tunnel couplings for the two orbitals , the values of tunnels rates @xmath180 and @xmath181 are imposed by the width and height of the conductance peaks . then , @xmath182}$ ] are the only free fitting parameters which remain for interpreting the @xmath0 curve . we have assumed @xmath183 for all the theoretical curves shown in this figure . we have plot the @xmath184 curves of bottom panel for @xmath52 ( pink curve , corresponding to fig . [ figure3]-d ) , @xmath185 ( green dashed curve ) and @xmath186 ( blue dot - dashed curve ) . note that in this figure , we show @xmath187 instead of @xmath188 @xmath189 in order to be consistent with ref . . a strong gate voltage offset jump occurred at @xmath120v , therefore , we show the data at the left / right of this jump with grey / black symbols . ] we now reconsider the experimental data of ref . . even the two - orbital model can not not provide a reasonable fit to the data if we assume @xmath94 and @xmath52 . in contrast , the two - orbital model exhibits a good agreement with the experimental data for @xmath190 , @xmath52 , @xmath191 , @xmath192 , @xmath193 , and parameters @xmath194 , @xmath195 , and @xmath179 given by the experiment ( see fig . [ figure4 ] , red and pink full curves ) . we now discuss the value of @xmath196 found for the above fit . this corresponds to a magnetic field of about @xmath197 , which is too strong to be attributed to stray fields from the ferromagnetic electrodes ( see e.g. ref . ) . this is in favor of generalizing the sdips concept to swnts quantum dots circuits , i.e. considering that the energy levels of the dot are spin - split because the confinement potential created by the ferromagnetic electrodes is spin - dependent . for comparison , we have estimated @xmath98 in the non - interacting theory@xcite , using realistic parameters i.e leads with a fermi energy @xmath198 and a density of states polarized by @xmath199 , and a nanotube with fermi wavevector @xmath200 @xmath201 , fermi velocity@xcite @xmath202 @xmath203 , length @xmath204 like in ref . , and density of states @xmath205 . we have modeled the interfaces between the nanotube and the leads with dirac potential barriers@xcite , with a height which is spin - polarized by @xmath199 and an average value which corresponds to@xcite @xmath206 @xmath207 @xmath208(for comparison the fitting parameters used in fig . [ figure4 ] correspond to @xmath209 @xmath208 and @xmath210 @xmath211 ) . we obtain @xmath212 , which is consistent with the above analysis . in the above discussion , we have assumed @xmath52 for simplicity . the height of the conductance peaks in the data imposes to use a strong assymetry @xmath213 between the left and right tunnel rates . thus , the two tunnel barriers are not symmetric , and there is no fundamental reason to assume @xmath52 . figure [ figure4 ] shows examples of @xmath184 curves plotted for a finite @xmath124 . using @xmath186 ( green dashed curve ) enhances the fit of the @xmath184 at peak 2 whereas @xmath185 ( blue dot - dashed curve ) enhances the fit of the @xmath184 at peak 3 . interestingly , with the non - interacting model , assuming the most simple situation in which @xmath214 has the same sign for the two leads , one finds @xmath215 , which is in agreement with the values used here . the fact that the best fit for the @xmath184 patterns at peaks 2 and 3 correspond to different values of @xmath216 might be due to a gate dependence of the sdips . this is indeed possible since the potential profile of the interfaces between the wire and the leads can vary with @xmath1 . we now comment briefly on the data taken for @xmath217v . it is not sure that the data shown before and after @xmath120v correspond to the filling of consecutive levels because of the gate voltage jump which occured at this value of @xmath1 . nevertheless , the shape of the mr curve corresponding to @xmath217v is rather consistent with the theory shown in fig . this suggests that these data really correspond to peak 1 . at this stage , it is important to point out that other orbital levels not taken into account in our calculation should slightly modify the conductance peaks 1 and 4 . the discrepancy between the theory and the data for @xmath217v could be explained by the effect of the other orbitals . for the data at @xmath218v , our fit is more quantitative since we have used peaks 2 and 3 of the theory . in principle , the modelisation of the orbital levels in swnts can be refined by taking into account an exchange energy @xmath219 which favors spin alignment , an excess coulomb energy @xmath220 related to the double occupation of the same orbital , and a subband mismatch @xmath221 ( see ref . ) . in practice , @xmath220 is rather small but @xmath219 and @xmath222 can be of the same order as @xmath150 . two different regimes of parameters can occur in practice . if @xmath223 , two electrons with opposite spins will fill consecutively the same energy level while @xmath1 increases ( see refs . ) , and the behavior of the device should thus be analogue to the non - degenerate multi - orbital case evoked at the end of section [ sec : single - level ] . nevertheless , if @xmath224 , peaks 1 and 2 [ 3 and 4 ] will correspond in majority to the same spin direction , as observed experimentally by . in this case , the effect of @xmath225 should be qualitatively the same as described in the present section . we expect that the weights of k and k in peaks 1 and 2 differ due to @xmath226 , but this should not change the way in which @xmath33 shifts the conductance peaks from @xmath34 to @xmath15 . in future experiments , it would be interesting to obtain continuous data on a larger @xmath1-range , in order to check that the shape of the @xmath126 pattern depends on the occupation of the dot . this would also allow to study the gate voltage dependence of @xmath33 . it would also be interesting to engineer contacts with ferromagnetic insulators or highly polarized ferromagnets in order to observe the sdips - induced giant @xmath184 effect . note that although we have considered here the limit @xmath227 , a strong enough sdips should also affect the behavior of the quantum dot in the sequential tunneling limit @xmath228 , through an analogous mechanism . using an anderson model , we have studied the behavior of a quantum dot connected to ferromagnetic leads through spin - active interfaces . the spin activity of the interfaces makes easier negative magnetoresistance ( @xmath0 ) effects and can even lead to a giant @xmath0 with a sign oscillating with the gate voltage of the dot . due to coulomb blockade , the @xmath0 versus gate voltage pattern can not be identical for all conductance peaks . it is nevertheless possible to account for the @xmath0 data measured by ref . in single - wall carbon nanotubes by taking into account the @xmath9 orbital degeneracy commonly observed in those systems . a.c . acknowledges discussions with b. douot , t. kontos , g. montambaux and i. safi . this work was supported by grants from rgion ile - de - france , the src / erc program ( r11 - 2000 - 071 ) , the krf grant ( krf-2005 - 070-c00055 ) , and the sk fund . a. brataas , y.v . nazarov , and g. e. w. bauer , phys . lett . * 84 * , 2481 ( 2000 ) ; d.h . hernando , y.v . nazarov , a. brataas , and g.e.w . bauer , phys . b * 62 * , 5700 ( 2000 ) ; a. brataas , y.v . nazarov and g.e.w . bauer , eur . j. b * 22 * , 99 ( 2001 ) . a. millis , d. rainer , and j. a. sauls , phys . b * 38 * , 4504 ( 1988 ) ; m. fogelstrm , _ ibid . _ * 62 * , 11812 ( 2000 ) ; j.c . cuevas and m. fogelstrm , _ ibid . _ * 64 * , 104502 ( 2001 ) ; n.m . chtchelkatchev , w. belzig , y.v . nazarov , and c. bruder , jetp lett . * 74 * , 323 ( 2001 ) ; d. huertas - hernando , y.v . nazarov , and w. belzig , phys . lett . * 88 * , 047003 ( 2002 ) ; j. kopu , m. eschrig , j. c. cuevas , and m. fogelstrm , phys . rev . b * 69 * , 094501 ( 2004 ) ; e. zhao , t. lfwander , and j. a. sauls , _ ibid . _ * 70 * , 134510 ( 2004 ) . very recently , et al_. @xcite have reported a spin - dependent transport experiment through a swnt connected to collinearly - polarized ferromagnetic leads in the resonant tunneling regime . the data of this ref . are interpreted following the lines in ref . . in this particular experiment , the sdips has been found to be vanishing . this is probably due to the fact that , considering the strong values of @xmath229 in this experiment , the effects of sdips on the @xmath4 curves are too weak to be resolved in the actual experiment . j. barnas and a. fert , phys . lett . * 80 * , 1058 ( 1998 ) ; a. braatas , yu . v. nazarov , j. inoue and g. e. w. bauer , phys . b * 59 * , 93 ( 1999 ) ; h. imamura , s. takahashi and s. maekawa , phys . rev . b * 59 * , 6017 ( 1999 ) ; b. r. bulka , phys . rev . b * 62 * , 1186 ( 2000 ) ; a. cottet , w. belzig and c. bruder , phys . rev . lett . * 92 * , 206801 ( 2004 ) , phys . rev . b * 70 * , 115315 ( 2004 ) ; h .- f . mu , g. su , and q .- r . zheng , phys . b * 73 * , 054414 ( 2006 ) . n. sergueev , q .- f . sun , h. guo , b. g. wang , and j. wang , phys . b * 65 * , 165303 ( 2002 ) ; j. martinek , m. sindel , l. borda , j. barnas , j. knig , g. schn , and j. von delft , phys . rev . lett . * 91 * , 247202 ( 2003 ) ; m .- s . choi , d. sanchez and r. lopez , phys . lett . * 92 * , 056601 ( 2004 ) ; y. utsumi , j. martinek , g. schn , h. imamura and s. maekawa , phys . b * 71 * , 245116 ( 2005 ) ; j. martinek , m. sindel , l. borda , j. barnas , r. bulla , j. knig , g. schn , s. maekawa and j. von delft , phys . rev . b * 72 * , 121302(r ) ( 2005 ) ; r. swirkowicz , m. wilczynski , m. wawrzyniak and j. barnas , phys . rev . b * 73 * , 193312 ( 2006 ) . l. f. schelp , a. fert , f. fettar , p. holody , s. f. lee , j. l. maurice , f. petroff , and a. vaures , phys . b * 56 * , r5747 ( 1997 ) ; k. yakushiji , s. mitani , k.takanashi , s. takahashi , s. maekawa , h. imamura , and h. fujimori appl . phys . lett.*78 * , 515 ( 2001 ) ; l. zhang , c. wang , y. wei , x. liu , and d. davidovic , phys . rev.b * 72 * , 155445 ( 2005 ) . note that ref . has also introduced an effective exchange field to take into account the sdips in an interacting single electron transistor connected to ferromagnetic leads . these authors have studied precession effects produced by the sdips in non - collinear configurations . they did not consider spin - dependent resonance effects because this is not relevant in the diffusive limit considered in this ref . note that in the limit considered in this article , the spin accumulation concept of ref . simply corresponds to having a finite average spin on the quantum dot . this feature is intrinsically taken into account in our treatment . for @xmath48 , the conductance given by eqs . ( [ cond ] ) and ( [ gtot ] ) can be perfectly mapped onto the interactionless conductance found in ref . for @xmath230 and a level @xmath231 close to resonance , using @xmath232 and @xmath233 . we point out that this splitting effect is unrelated to @xmath33 . this is particularly clear in the limit of no tunneling @xmath237 , in which the renormalization of the levels by the @xmath104 terms vanishes whereas @xmath33 persists . the function @xmath103 is resonant at @xmath238 $ ] and @xmath239 $ ] . interestingly , for @xmath240 , @xmath241 , @xmath242 and @xmath243 , a simplified expression of the resonance spin - splitting caused by the @xmath127 terms can be obtained by evaluating @xmath244 $ ] at @xmath245 and @xmath246 $ ] at @xmath247 . this leads to @xmath248 $ ] where the prime represents cauchy s principal value . interestingly , this expression perfectly matches with eq . ( 3.9 ) of ref . . studies the precession effect produced by @xmath249 in a quantum dot with ferromagnetic leads polarized in non - collinear directions , in the limit @xmath250 . the system is described with eqs . which neglect @xmath249 in the collinear case . note that it is possible to decrease the up spins contribution in peak 3 and to have a @xmath126 pattern similar for peaks 3 and 4 by using values of @xmath112 much smaller than in fig.[figure3 ] ( e.g. @xmath251 ) .

in the context of spacetimes that admit spatially compact spacelike cauchy hypersurfaces , hypersurfaces of constant mean curvature ( cmc ) and slicings by such hypersurfaces have attracted a lot of attention . one major motivation for studying such slicings rests on the fact that cmc foliations give rise to a geometrically distinguished time - function on the spacetime . cmc slicings thus appear prominently whenever cosmological spacetimes are discussed , when the strong cosmic censorship conjecture is investigated , and they also play an important role in numerical studies . there exist several well - established properties of cmc hypersurfaces in cosmological spacetimes satisfying the timelike convergence condition ( which is identical to the strong energy condition and states that the contraction of the ricci tensor with any timelike vector is non - negative ) . in particular , uniqueness of cmc hypersurfaces is well - understood : in a cosmological spacetime satisfying the timelike convergence condition , for any given constant @xmath0 , there exists at most one compact cmc hypersurface with mean curvature @xmath1 ; the proof in @xcite is based on the techniques of @xcite . modulo some subtleties , an analogous result holds for maximal hypersurfaces ( where @xmath2 ) . a consequence of the uniqueness result is that any killing vector the spacetime possesses must be tangential to the cmc hypersurface . the existence of compact cmc hypersurfaces ( and foliations thereof ) is a more delicate issue . ( if there exists one compact cmc hypersurface , then there exists a foliation of a neighborhood of that hypersurface by compact cmc hypersurfaces . this follows from standard arguments if @xmath3 ; if the initial hypersurface is maximal , see @xcite . ) if the spacetime possesses barriers , i.e. , compact hypersurfaces @xmath4 and @xmath5 , @xmath5 in the chronological future of @xmath4 , with mean curvature @xmath6 and @xmath7 such that @xmath8 , then there exists a foliation of compact cmc hypersurfaces between @xmath4 and @xmath5 where the mean curvature assumes each value between @xmath9 and @xmath10 , see @xcite . a spacetime with crushing singularities is globally foliated by compact cmc hypersurfaces @xcite . it has been conjectured that every maximal globally hyperbolic vacuum cosmological spacetime ( satisfying the timelike convergence condition ) with compact cmc hypersurface admits a global cmc foliation , where the mean curvature ranges over @xmath11 or @xmath12 depending on the yamabe class of the cmc slice @xcite . in certain special cases the conjecture has been shown to be true , e.g. , for locally spatially homogeneous spacetimes , and for spherically symmetric spacetimes with spatial topology @xmath13 , see ( * ? ? ? 2 ) for a review and @xcite and references therein . however , the conjecture is not true in general . there exist examples of spacetimes with a cmc foliation that is not global @xcite and spacetimes that do not possess compact cmc hypersurfaces at all @xcite . the failure of the conjecture can not be attributed to the presence of matter , since there exist vacuum cosmological spacetimes without compact cmc hypersurfaces as well @xcite . without the assumption of the timelike convergence condition , fewer results are available : uniqueness of a globally defined cmc time function with compact level sets still holds @xcite ( and the associated cmc foliation is the unique cmc foliation ) . in @xcite another uniqueness result is proved under the assumption that there exists @xmath14 such that @xmath15 @xmath16 : for any given constant @xmath17 , there exists at most one compact cmc hypersurface with mean curvature @xmath1 . existence of a cmc foliation can be deduced from the existence of barriers @xcite : if the spacetime possesses a future crushing singularity ( future mean curvature barrier ) , then a future end of the spacetime admits a cmc foliation , where the range of the mean curvature is a subset of the interval @xmath18 . ( the proofs are based on the crucial fact that @xmath19 for hypersurfaces with @xmath20 . the expression @xmath21 appears in the lapse equation , see . ) since it is as yet unclear which general properties are to be expected of cmc foliations in spacetimes violating the timelike convergence condition , the detailed study of cmc hypersurfaces in explicit classes of spacetimes comes into focus . in @xcite , a class of spatially compact kottler - schwarschild - de sitter spacetimes has been investigated in detail ; these are vacuum solutions of the einstein equations with positive cosmological constant @xmath22 , which causes the timelike convergence condition to be violated , see section [ ksksds ] for the definition . it is found that the spacetimes of this class do not contain compact cmc hypersurfaces with @xmath23 . each spacetime contains a unique family of slicings of compact cmc hypersurfaces , where the mean curvature ranges between @xmath24 . however , the slicings do not cover the entire spacetime , and , in general , the slicings are foliations only during some time of their evolution . in this paper we investigate the properties of cmc slicings in vacuum _ kantowski - sachs _ models with cosmological constant @xmath25 , which we introduce in some detail in section [ ksksds ] . let us summarize the main results we prove in this work : there are two ( of eight ) classes of kantowski - sachs spacetimes for which there exists a global cmc time function and thus a unique global cmc foliation . for the remaining classes of kantowski - sachs models this is not true in general . we show that , on the one hand , there are kantowski - sachs spacetimes with a unique global cmc foliation ( which does , however , not define a time function ) ; on the other hand , there are kantowski - sachs spacetimes containing ( arbitrarily ) many families of cmc slicings , where the range of @xmath26 is @xmath27 ( with ) . in the future ` end ' of the spacetime , these slicings are foliations . note that , in this paper , a _ slicing _ denotes a smooth family of smooth ( spacelike ) hypersurfaces . a parametrization of a slicing is a smooth map @xmath28 , where @xmath29 is a and @xmath30 , such that for all @xmath31 , @xmath32 is an embedding . we require that @xmath33 is a hypersurface of the slicing for all @xmath31 , i.e. , by the parametrization of the slicing the hypersurfaces are represented as level sets @xmath34 . note that a slicing is a _ foliation _ iff the map @xmath35 is a diffeomorphism onto its image . section [ ksksds ] of the paper is perhaps a ( short ) paper in its own right . it contains a concise discussion of the ( vacuum , @xmath36 ) kantowski - sachs models with a focus on the geometric interpretation of these models in terms of the kottler - schwarzschild - de sitter metric . in section [ data ] we analyze spherically symmetric cmc initial data sets , which are embedded into kantowski - sachs ( and kottler - schwarzschild - de sitter ) spacetimes in the subsequent section . finally , sections [ sec : cdst ] and [ properties ] contain the analysis and the results on cmc slicings and foliations in kantowski - sachs spacetimes . the kantowski - sachs models are a class of locally rotationally symmetric ( lrs ) and spatially homogeneous ( sh ) spacetimes @xcite . the defining property of these models is the existence of a four - dimensional isometry group , whose orbits are three - dimensional spacelike hypersurfaces , where , however , there does not exist any three - dimensional subgroup that acts simply transitively on these orbits . the metric of a kantowski - sachs spacetime is @xmath37}}\:,\ ] ] where @xmath38}}$ ] is the standard metric on the @xmath39-sphere . assuming that the coordinate @xmath40 ranges in @xmath41 , the spatial topology is @xmath42 and thus compact . in this paper we consider kantowski - sachs spacetimes @xmath43 that satisfy the einstein equations with a positive cosmological constant @xmath22 . the qualitative dynamics of these models has been partly investigated in @xcite ; the paper @xcite contains a fairly complete picture . this subsection is a concise compendium . every ( vacuum ) kantowski - sachs model ( with @xmath36 ) is described by a metric @xmath44}}\ : , \tag{\ref{ksmetric}${}^\prime$}\ ] ] whose components @xmath45 , @xmath46 , and @xmath47 are determined by the equations @xmath48 where @xmath49 is a ` seed solution ' , i.e. , a solution of the system of equations [ reddynsys ] @xmath50 \partial_\tau \sigma_+ & = - ( 1-\sigma_+^2 ) ( 1 + 3 h_d \sigma_+ - h_d^2 ) \:,\end{aligned}\ ] ] which is a dynamical system on the relatively compact state space @xmath51 . we refer to @xcite or @xcite for a derivation of the ` reduced dynamical system ' . ( we use the intuitive notation of @xcite ; to compare with @xcite one sets @xmath52 , @xmath53 . ) let us merely note that @xmath54 and @xmath55 are normalized versions of the hubble expansion @xmath56 and the shear variable @xmath57 , [ hdsig+ ] @xmath58 \sigma_+ & = \sigma_+/d & & = \sigma_+\ , \big(h^2 + { { \textstyle{\frac{1}{3}}}}\,g_{2 2}^{-1}\big)^{-1/2 } & & = -{{\textstyle{\frac{1}{3 } } } } \ , ( \tensor{k}{^1_1 } - \tensor{k}{^2_2})\ , \big({{\textstyle{\frac{1}{9}}}}\ , ( \operatorname{tr}k)^2 + { { \textstyle{\frac{1}{3}}}}\,g_{2 2}^{-1}\big)^{-1/2}\:.\end{aligned}\ ] ] in , @xmath59 is the extrinsic curvature of the hypersurfaces of homogeneity ( where we use the convention @xmath60 ) and @xmath61 . the dynamical system encodes the entire dynamics of ( vacuum , @xmath36 ) kantowski - sachs models . we omit the analysis of and content ourselves with the crucial facts : first , the system is smoothly extendible to @xmath62 \times [ -1,1]$ ] ; second , the straight line is an invariant set ; third , the ellipse @xmath63 , in combination with the line @xmath64 , defines future and past invariant subsets of the state space . the result of the analysis is a complete and rigorous picture of the flow of the dynamical system . this picture is given in fig . [ ksflow ] . \times [ -1,1 ] \ni ( h_d , \sigma_+)$ ] . the labeling of the fixed points refers to the asymptotic behavior of solutions that converge to these points . the figure on the r.h . side shows a partition of the state space into eight invariant domains of positive measure : bh ( black hole ) , wh ( white hole ) , fc@xmath65 ( future cosmological region , positive mass ) , fc@xmath66 ( future cosmological region , negative mass ) , pc@xmath65 ( past cosmological region , positive mass ) , pc@xmath66 ( past cosmological region , negative mass ) , sh@xmath67 ( spatially homogeneous , mainly expanding ) , sh@xmath68 ( spatially homogeneous , mainly contracting ) . this nomenclature will become clear in the context of the discussion of the kottler - schwarzschild - de sitter spacetimes.,scaledwidth=100.0% ] a solution that converges to a fixed point labeled ` taub ' in fig . [ ksflow ] exhibits ` taub asymptotics ' , which means @xmath69 and @xmath70 as @xmath71 or @xmath72 , respectively , where @xmath73 represents an initial / final singularity ( in cosmological time @xmath74 ) . the label ` non - flat ' refers to asymptotic behavior governed by the non - flat lrs kasner solution ; for a solution converging to one of these points we have @xmath75 and @xmath76 as @xmath71 or @xmath72 , respectively . finally , de sitter asymptotics corresponds to @xmath77 and @xmath78 as @xmath79 . the orbits converging to the two interior fixed points form a net of separatrices , which yields a partition of the state space into six invariant subsets . in addition , there are two special orbits of , which are depicted by dashed lines in fig . [ ksflow ] . these orbits correspond to the models described in @xcite , which are the unique kantowski - sachs models that are isotropic @xcite . we thereby obtain a partition of the state space into eight domains of positive measure , whose significance will be discussed in the context of the kottler - schwarzschild - de sitter spacetimes . we conclude this section with a brief discussion of the expansion of kantowski - sachs models . from and we have @xmath80 which makes it straightforward to obtain the expansion of models from fig . [ ksflow ] . it is interesting to note that there exist models ( in sh@xmath67 ) that exhibit an intermediate phase of contraction . furthermore , analysis of the equation @xmath81 shows that the time - derivative of @xmath56 can change sign at most once . all of the five classes of initially expanding models ( bh , wh , fc@xmath82 , sh@xmath67 ) are initially decelerating ; however , for three classes ( fc@xmath82 , sh@xmath67 ) , initial deceleration is followed by late - time acceleration ( associated with de sitter asymptotics ) . the kottler - schwarzschild - de sitter metric @xcite reads @xmath83}}\,\quad\text{with}\quad v= v(r ) = 1-\frac{2 m}{r } -\frac{\lambda r^2}{3}\:.\ ] ] we assume a cosmological constant @xmath14 . if the constant @xmath84 satisfies @xmath85 , then there exists a non - trivial interval @xmath86 such that @xmath87 and @xmath88 in @xmath86 . the region @xmath89 is a static region of the spacetime with killing vector @xmath90 . it is straightforward to see that @xmath91 . the static region of the spacetime has an analytic extension reminiscent of the kruskal extension of the schwarzschild spacetime and the de sitter spacetime . on the one hand , the metric is extended beyond the ` black hole horizon ' @xmath92 to arbitrarily small values of @xmath93 , where @xmath94 represents a curvature singularity , on the other hand , the metric is extended beyond the ` cosmological horizon ' @xmath95 to arbitrarily large values of @xmath93 . the extended spacetime is commonly depicted using two charts which cover the regions @xmath96 and @xmath97 , respectively ; for the conformal compactification of these charts see , e.g. , @xcite . in the conformally compactified picture , @xmath98 represents conformal infinity ( which is spacelike ) . as is well - known , the constructed spacetime corresponding to the union of the two regions @xmath96 and @xmath97 can be smoothly ( in fact analytically ) extended in a periodic fashion . thereby one obtains an inextendible , globally hyperbolic spacetime of topology @xmath99 , the kottler - schwarzschild - de sitter spacetime ksds , see fig . [ ksdsfig ] . are separated from the future / past cosmological regions @xmath100 by the static regions @xmath89 . solid lines represent hypersurfaces @xmath101 , dashed lines are hypersurfaces @xmath102 . note that the two charts that yield the conformal compactification of the two regions @xmath96 and @xmath97 do not agree in the static domains @xmath89 ; hence , e.g. , a dashed line represents @xmath103 in one chart , but @xmath104 in the other . ] on ksds there exists an isometric action of @xmath105 . the dashed lines in figs . [ ksdsfig ] are orbits under the ` killing flow ' , i.e. , under the static killing vector @xmath106 . the killing vector @xmath107 is globally defined ; it is null on the killing horizons @xmath108 and @xmath109 which emanate from the bifurcation 2-spheres at which @xmath107 vanishes . the solid lines in fig . [ ksdsfig ] represent hypersurfaces @xmath101 , which are totally geodesic as fixed point sets of the discrete isometries ( ` reflections ' ) @xmath110 . the timelike @xmath101 hypersurfaces we call @xmath101 ` cylinders ' . by a ` cosmological spacetime ' we mean a globally hyperbolic spacetime that is spatially compact , i.e. , contains a spacelike compact cauchy hypersurface . ( then any smooth compact spacelike hypersurface is cauchy @xcite ) . note that we use the term ` cosmological spacetime ' in a less restrictive manner than @xcite , where the timelike convergence condition ( strong energy condition ) is assumed in addition . of course , this condition is violated in the presence of a positive cosmological constant , i.e. , @xmath111 in the following we will construct a number of classes of ` cosmological spacetimes ' from ksds : consider a _ black hole _ region @xmath112 in ksds . by identifying points mapped to each other by a discrete subgroup of the action under @xmath107 , i.e. , by identifying the cylinders @xmath113 and @xmath114 , we obtain a smooth cosmological spacetime of topology @xmath115 , which we denote by bh[t ] , see fig . similarly , identifying @xmath113 and @xmath114 in a _ white hole _ region of ksds , we obtain a spacetime which we denote by wh[t ] ( which is related to bh[t ] by time reversal ) . analogously , consider a _ future cosmological _ region @xmath116 in ksds . identifying @xmath113 and @xmath114 we obtain a spacetime which we call fc@xmath65[t ] , see fig . past cosmological _ region gives rise to pc@xmath65[t ] ( which is related to fc@xmath65[t ] by time reversal ) . the spacetime fc@xmath65[t]/pc@xmath65[t ] is future / past geodesically complete , but geodesically incomplete to the past / future : it possesses a past / future crushing singularity ( in the standard terminology which is due to @xcite ) corresponding to a past / future mean curvature barrier ( in the terminology of , e.g. , @xcite ) ; this will become obvious immediately . bh[t]/wh[t ] possesses a past and future crushing singularity , where the future / past singularity is a curvature singularity . the constructed spacetimes are maximal globally hyperbolic , i.e. , they agree with the maximal cauchy development of initial data defined on any compact spacelike hypersurface . however , there are non - trivial extensions reminiscent of the extensions of the misner universe ( see , e.g. , @xcite ) . that arises from the black hole region in the kottler - schwarzschild - de sitter spacetime ksds . analogously , fc@xmath65[t ] is defined by an identification in the future cosmological region , see the r.h.s . the spacetimes wh[t ] and pc@xmath65[t ] are not depicted ; these are associated with the white hole and the past cosmological region . the constructed spacetimes of the kantowski - sachs type.,scaledwidth=96.0% ] to construct from ksds a spacetime that is a model for a cosmological spacetime containing a black hole , we choose @xmath117 and identify points of equal radius @xmath93 on a @xmath118 cylinder and a @xmath119 cylinder in an adjacent copy of the region @xmath120 on the r.h . thereby we obtain an inextendible cosmological spacetime of topology @xmath121 , which we call ksds[t ] . ksds[t ] is a smooth , in fact analytic , spacetime , see ( * ? ? ? 1 ) for details . in ksds[t ] two types of asymptotic behavior coexist : inextendible timelike geodesics either end in the curvature singularity after finite proper time , or go to scri . in the kottler - schwarzschild - de sitter metric the mass @xmath122 need not necessarily be positive ( as we have assumed so far ) . the kottler - schwarzschild - de sitter spacetime with negative mass , @xmath123 , has a simpler structure than ksds : there is a static region ( whose boundary is a naked singularity ) and a past and a future cosmological region @xmath100 , see fig . [ negm ] . from this spacetime we can construct cosmological spacetimes : in the _ future cosmological _ region @xmath116 we identify @xmath113 and @xmath114 to obtain fc@xmath66[t ] , see fig . [ negm ] . the analogous construction in the _ past cosmological _ region gives rise to pc@xmath66[t ] . ( the subscripts @xmath82 distinguishing the spacetimes fc@xmath82[t ] and pc@xmath82[t ] thus refer to the positivity / negativity of the mass @xmath122 . ) . the cosmological spacetime fc@xmath66[t ] arises from an identification in the future cosmological region ; pc@xmath66[t ] is constructed in the past cosmological region.,scaledwidth=45.0% ] if @xmath84 and @xmath124 in the kottler - schwarzschild - de sitter metric , the spacetime does not have a static region and there are no horizons . we obtain spatially homogeneous cosmological models with topology @xmath115 which we call sh@xmath67[t ] and sh@xmath68[t ] . it remains to systematically describe the relationship between the cosmological spacetimes contructed from the kottler - schwarzschild - de sitter metric and the kantowski - sachs spacetimes . it is immediate that the eight classes of spacetimes bh[t ] , wh[t ] , fc@xmath65[t ] , fc@xmath66[t ] , pc@xmath65[t ] , pc@xmath66[t ] , sh@xmath67[t ] , and sh@xmath68[t ] , are _ kantowski - sachs models _ ; we have @xmath125}}\:.\ ] ] since the metric is explicit , it is straightforward to compute the dynamic quantities @xmath54 and @xmath55 using . we find that the spacetimes bh[t ] of fig . [ bhfc ] give rise to a one - parameter family of orbits , which is parametrized by the ( positive ) mass @xmath122 , that corresponds to the family of orbits of region bh in fig . [ ksflow ] . for small values of @xmath122 we obtain orbits close to the boundary orbit connecting the taub point and the non - flat lrs point . for values of @xmath122 such that @xmath126 is close to @xmath127 we obtain orbits that come close to the separatrix orbits and the interior fixed point . the case is analogous for the spacetimes wh[t ] , which fill the region wh of fig . [ ksflow ] . the spacetimes fc@xmath65[t ] of fig . [ bhfc ] correspond to the orbits of region fc@xmath65 of fig . [ ksflow ] . values of @xmath122 such that @xmath126 is close to @xmath127 correspond to orbits that come close to the separatrix orbits and the interior fixed point . for @xmath128 we obtain orbits in fc@xmath65 close to the special isotropic solution ( dashed line ) . the special orbit itself corresponds to a cosmological spacetime , which may be denoted by fc@xmath129[t ] , that is obtained by an identification in the future cosmological region of the de sitter spacetime . the family fc@xmath66[t ] of fig . [ negm ] fills the region fc@xmath66 of fig . [ ksflow ] . for large negative values of @xmath122 we obtain orbits that are close to the boundary of the state space . the case of the spacetimes pc@xmath82[t ] is completely analogous . the spacetimes sh@xmath67[t ] and sh@xmath68[t ] correspond to positive values of @xmath122 with @xmath130 . for values of @xmath122 such that @xmath126 is only slightly greater than @xmath127 , we obtain orbits in sh@xmath67 and sh@xmath68 that are close to the separatrix orbits and the interior fixed points . for large values of @xmath122 we obtain orbits close to the boundary . let us make some final comments . first , we note that the flow of fig . [ ksflow ] is independent of @xmath14 . it is possible to interpret @xmath131 as a function on the state space @xmath132 which is invariant along the orbits . the net of separatrix orbits ( including the two interior fixed points ) is characterized by @xmath133 . of course , the separatrix orbits admit an interpretation in terms of the kottler - schwarzschild - de sitter metric . in the case @xmath133 the spacetimes defined from this metric are the _ nariai spacetimes_. we refrain from giving a discussion , but refer to @xcite for a recent study . ( note that the kantowski - sachs models associated with the nariai spacetimes form a set of measure zero among all kantowski - sachs models . ) second , a comment in a similar vein . on the boundaries of the state space of fig . [ ksflow ] , where we have to exclude the fixed points , the function @xmath131 takes the values @xmath134 or @xmath135 , respectively . the solutions on the ( ` vacuum ' ) boundaries @xmath136 ( where @xmath137 ) admit a straightforward interpretation : these are the kantowski - sachs models that solve the einstein vacuum equations with @xmath138 ; an explicit representation of these models is based on performing an identification in the black hole / white hole region of the schwarzschild spacetime . third , let us comment on the parameter t , which characterizes the length of the @xmath41 component of the spacetime . this parameter does not enter the picture in fig . [ ksflow ] . this is because it corresponds to the free constant of integration in . finally , the statements on past / future crushing singularities of the kantowski - sachs models bh[t ] , wh[t ] , fc@xmath65[t ] , fc@xmath66[t ] , pc@xmath65[t ] , and pc@xmath66[t ] , which have been made above , are now immediate from fig . [ ksflow ] in combination with . in this section we briefly discuss the parameter space(s ) of compact spherically symmetric cmc data sets . for our purposes , a spherically symmetric cmc initial data set is a triple , where @xmath139 is a one - dimensional manifold , @xmath140 a spherically symmetric riemannian metric , and @xmath59 a spherically symmetric symmetric tensor with constant trace , subject to the einstein vacuum constraints with cosmological constant @xmath25 . we find , cf . @xcite , that these data sets are parametrized by @xmath141 ( ` mean curvature ' ) and a second constant , @xmath142 , through [ gijkij ] @xmath143}}\ : , \\\label{kijinkc } k_{i j } \ , d x^i d x^j & = \frac{k}{3}\ : { g}_{i j}\ , d x^i d x^j + c\ : \big ( \frac{2}{r(l)^3 } \,d l^2 - \frac{1}{r(l ) } \ , { \mathbf{g}_{[s^2]}}\big ) \ : , \ ] ] where @xmath144 is the coordinate on @xmath139 and @xmath38}}$ ] the standard metric on @xmath145 . ( the tensor in brackets in is the standard spherically symmetric transverse traceless tensor . ) the positive function @xmath146 in and is required to be a solution of the differential equation @xmath147 where @xmath122 is an additional constant . the ( local ) development of the data leads to and @xmath122 emerges as the mass @xcite . a spherically symmetric cmc initial data set is _ compact _ , if @xmath148 . this is the case if @xmath149 and @xmath142 are such that @xmath150 has two positive ( simple ) zeros , @xmath151 and @xmath152 , and @xmath153 in the interval @xmath154 ; see fig . [ deltaborderline](a ) . this leads to a periodic solution @xmath146 of that oscillates between @xmath151 and @xmath152 ; we denote the period by @xmath155 , @xmath156 w.l.o.g . we assume that @xmath157 , so that @xmath158 ; it follows that @xmath146 is even . by the natural identification of @xmath159 and @xmath160 , the domain of the function @xmath146 becomes @xmath41 and the cmc initial data set @xmath161 is compact . there are ` degenerate ' cases : if @xmath149 and @xmath142 are such that @xmath150 possesses a root @xmath162 of multiplicity two ( or three ) , see figs . [ deltaborderline](b ) and [ deltaborderline](c ) , then @xmath163 is a constant solution of . clearly , these solutions generate compact cmc initial data sets @xmath161 where the ` length ' @xmath155 of the @xmath41 component is arbitrary . for parameters @xmath164 that induce compact cmc data . ( a ) corresponds to generic cmc data , subfigs . ( b , c ) are borderline cases and generate @xmath165 initial data . ] compact cmc initial data sets with @xmath167 are obtained from as solutions of the algebraic equations @xmath168 , @xmath169 , cf . [ deltaborderline](b ) and [ deltaborderline](c ) . the parameters @xmath149 and @xmath142 associated with these data sets are @xmath170 where @xmath171 is the function of . the admissible range of @xmath162 depends on the values of @xmath22 ( @xmath172 ) and @xmath122 which determine @xmath171 . in the following we concentrate on the case @xmath84 , @xmath173 ( which is the most intricate one ) . accordingly , the admissible range of @xmath162 is @xmath174 . the cmc data sets with @xmath167 are associated with the spacelike hypersurfaces @xmath175 in ksds , @xmath176 . the upper [ lower ] sign in refers to hypersurfaces in the black [ white ] hole and the past [ future ] cosmological region of ksds . equation describes two pairs of curves , @xmath177 , one pair corresponding to @xmath178 and another pair with @xmath179 ; these curves are depicted in fig . [ cbtdu ] . ( it is straightforward to prove the monotonicity properties and the asymptotic behavior of these curves suggested in this figure . ) a conspicuous feature is the cusp of the curve , which means that there exists a unique @xmath180 such that and ; the value of @xmath181 is determined by the equation @xmath182 the value of @xmath183 at @xmath181 is the smallest possible value of the ( modulus of the ) mean curvature within the class @xmath179 ; the same is true for the value of @xmath184 ; we denote these by @xmath185 and @xmath186 , respectively . it is not difficult to show that @xmath187 . in terms of the function @xmath150 , @xmath181 is determined by the conditions @xmath188 , @xmath189 , and @xmath190 . hence , for @xmath191 , and only for these values , the function @xmath150 has a zero of third order , which corresponds to the borderline case between figs . [ deltaborderline](b ) and [ deltaborderline](c ) . cmc initial data sets ( in the case , ) . the mirror image of the ` wedge ' in the upper right corner , which is obtained by reflection at the origin , is not depicted.,scaledwidth=85.0% ] we define the parameter space @xmath192 of compact cmc data to be the interior of the set of pairs @xmath164 that generate compact cmc data . for @xmath164 to be in @xmath193 , @xmath150 must have the form depicted in fig . [ deltaborderline](a ) ; then @xmath146 is an even function that oscillates between @xmath151 and @xmath152 , where the period @xmath155 is given by . the parameter space @xmath193 is the disjoint union of three open connected domains , @xmath194 which are defined by the relations [ kcdef ] @xmath195 \label{kc1def } & ( k , c)\in \mathscr{kc}_{\pm 1 } \ : & & \longleftrightarrow\ : \,r_{\mathrm{c } } < r_{\min } < r_{\max } \:.\end{aligned}\ ] ] the domain @xmath196 arises from @xmath197 by the inversion at the origin , while @xmath198 is invariant under the inversion . let us elaborate on . since the function @xmath150 is non - negative in @xmath86 , see , the roots @xmath151 and @xmath152 either satisfy @xmath199 , cf . , or @xmath200 , or @xmath201 . to obtain we need to show that the subcase @xmath202 and the entire third scenario @xmath201 are excluded : first , if we assume that @xmath203 , then @xmath204 , and hence @xmath205 ; however , this contradicts the definition of @xmath151 , which includes the condition @xmath206 . second , we note that @xmath150 can possess at most three zeros , of which @xmath152 is the largest . this follows from an analysis of the critical points of @xmath150 , see ( * ? ? ? * appendix a ) . accordingly , @xmath150 must be negative for all @xmath207 . this being the case , the assumption @xmath201 ( @xmath208 ) immediately leads to a contradiction , since @xmath209 . this establishes the implicit claims of . the parameter spaces @xmath198 and @xmath210 are depicted in fig . [ kc01 ] . among the boundaries of these spaces are the curves of fig . [ cbtdu ] that represent the pairs @xmath164 that generate @xmath165 initial data . for these ` critical ' values of @xmath164 the profile of the function @xmath150 is a borderline case , see fig . [ deltaborderline ] . there are arbitrarily small perturbations of fig . [ deltaborderline](b ) and [ deltaborderline](c ) that lead to the profile [ deltaborderline](a ) and thus to compact cmc data , but there are other perturbations that lead to a profile of @xmath150 that is not associated with compact cmc data . the remaining boundary components of @xmath198 and @xmath210 are the straight lines @xmath211 . for , we find @xmath212 , cf . [ deltaborderline](a ) , because the function @xmath150 is positive for large @xmath93 , if @xmath213 . and @xmath197 of compact cmc data are bounded by the curves of fig . [ cbtdu ] that represent @xmath214 initial data and by the straight line(s ) @xmath215.,scaledwidth=90.0% ] the analysis of and the parameter space of compact cmc data is simpler in the remaining cases . in the case @xmath123 we find @xmath216 , where the space @xmath210 is a ` wedge ' whose properties are analogous to those sketched in fig . the case @xmath84 , @xmath130 is similar ; however , we refrain from going into the details . in this section we investigate whether compact cmc initial data sets are embeddable in the kantowski - sachs spacetimes and/or the cosmological kottler - schwarzschild - de sitter spacetime ksds[t ] . consider a compact cmc initial data set @xmath217 represented by @xmath22 , @xmath122 , and a pair . the universal covering @xmath218 of the data is embeddable as a cmc hypersurface @xmath219 in ksds through @xmath220 see @xcite ; the integral is understood in the principal value sense . the embedding is not unique : @xmath221 , @xmath222 defines a one - parameter family of embeddings of the data ; this reflects the continuous isometry of the spacetime generated by the killing vector @xmath223 . the hypersurface @xmath219 represented by is reflection symmetric , i.e. , invariant under the discrete isometry @xmath224 of the spacetime . the requirement of reflection symmetry is a convenient way of fixing a representative within the one - parameter family of embeddings @xmath221 , @xmath222 : @xmath219 is the unique reflection symmetric embedding of the data . the ( future pointing ) unit normal of @xmath219 in ksds is given by @xmath225 where the prime denotes the derivative w.r.t . it is straightforward to check that @xmath226 . the killing vector @xmath90 of ksds admits a decomposition into a lapse function @xmath227 and a shift vector @xmath228 : from and we obtain @xmath229 and @xmath230 . we now make a fundamental definition . let @xmath231 and @xmath232 be the ( half-)period characterizing the compact cmc initial data set . we define @xmath233 the value of the quantity @xmath234 is a property of the initial data set . it is possible ( and useful ) to regard @xmath234 as a function of @xmath164 on @xmath193 . in ( * ? 3 ) the following statement has been proved : a compact cmc initial data set associated with a pair @xmath235 is embeddable as a smooth cmc hypersurface @xmath236 in ksds[t ] if and only if @xmath237 . the hypersurface @xmath236 is then a cauchy hypersurface and the embedding is unique modulo the killing flow . we now prove an analogous statement for @xmath238 . [ kc1embedprop ] let @xmath123 , or @xmath84 with @xmath173 . consider a compact cmc initial data set @xmath239 associated with @xmath240 . this data set is embeddable as a smooth cmc hypersurface @xmath236 ( @xmath241 ) in the kantowski - sachs spacetime fc@xmath66[t ] or fc@xmath65[t ] , depending on the sign of @xmath122 , if and only if @xmath242 . the @xmath243-fold covering of the data is embeddable as a smooth cmc hypersurface @xmath244 in fc@xmath82[t ] if and only if @xmath245 . the embedded hypersurfaces are cauchy hypersurfaces , and the embeddings are unique modulo the killing flow . consider the universal covering of the data and the embedded hypersurface @xmath219 in ksds as given by . since @xmath146 oscillates between @xmath151 and @xmath152 where @xmath246 , see , @xmath219 is entirely contained in a cosmological region of ksds . the assumption @xmath240 implies @xmath247 and we infer from that @xmath219 is contained in a future cosmological region of ksds , see fig . [ corrugatedsheet ] . the hypersurface @xmath219 is invariant under the reflections @xmath248 for all @xmath249 ; this is because @xmath146 is even about @xmath250 @xmath251 . since the unit normal vectors of @xmath219 at @xmath252 must be invariant under these reflections as well , they must be tangential to the fixed point sets @xmath253 of these discrete isometries . we conclude that @xmath254 at @xmath250 . therefore , the identification of the cylinders @xmath255 and @xmath256 ( @xmath257 ) in ksds turns @xmath219 into a smooth cmc hypersurface @xmath244 in fc@xmath82[@xmath258 ( where the sign is determined by the sign of @xmath122 ) . hence , since @xmath244 is isometric to the @xmath243-fold covering of the initial data , we have found that the @xmath243-fold covering of the data is embeddable as a smooth cmc hypersurface @xmath244 in fc@xmath82[t ] , @xmath259 . in spacetimes fc@xmath82[t ] with @xmath260 @xmath261 , however , the identification of @xmath113 with @xmath262 is inconsistent with the identifications of @xmath263 with @xmath264 on the level of the data , which is an obstruction to ( smooth ) embeddability . the proof of the remaining claims is trivial . ' '' '' ( and @xmath265 , @xmath173 ) appears as a `` corrugated sheet '' confined between the @xmath102 hypersurfaces @xmath266 and @xmath267 . in fc@xmath65[t ] with @xmath237 , it is a smooth cauchy hypersurface.,scaledwidth=90.0% ] the statement of the proposition holds analogously for @xmath268 if we replace fc@xmath82[t ] ( future cosmological ) by pc@xmath82[t ] ( past cosmological ) . the case @xmath269 is completely analogous as well . the ( @xmath243-fold covering of the ) data is embeddable in the cosmological spacetime fc@xmath129[t ] obtained from the de sitter spacetime , cf . section [ ksdssubsec ] , if @xmath259 . we do not consider the case @xmath84 , @xmath130 ; however , we expect a result that is analogous to proposition [ kc1embedprop ] , where fc@xmath82[t ] and pc@xmath82[t ] are replaced by sh@xmath67[t ] and sh@xmath68[t ] . the results of section [ embeddings ] leave open the question which of the kantowski - sachs spacetimes fc@xmath82[t ] and pc@xmath82[t ] actually contain cmc hypersurfaces . another issue that will be discussed in this section is the question of whether the embedded cmc hypersurfaces evolve to form slicings or foliations . the interesting case is that of the kantowski - sachs spacetimes fc@xmath82[t ] and pc@xmath82[t ] . let us begin , however , by investigating the kantowski - sachs spacetimes of the classes bh[t ] and wh[t ] . the results of sections [ data ] and [ embeddings ] show that there do not exist ( spherically symmetric ) cmc hypersurfaces in these spacetimes other than the @xmath103 hypersurfaces ; see and in particular . the @xmath103 hypersurfaces are associated with the @xmath165 initial data sets which are represented by pairs @xmath164 determined by with @xmath270 , see fig . [ cbtdu ] . the @xmath103 hypersurfaces form a global foliation of bh[t]/wh[t ] , and the mean curvature is a monotone function along this foliation , which we infer from fig . [ ksflow ] in combination with or from fig . [ cbtdu ] . in other words , _ the kantowski - sachs spacetimes * bh[t ] * and * wh[t ] * possess a global cmc time function _ and thus a unique global cmc foliation . the kantowski - sachs spacetimes of the classes fc@xmath82[t ] and pc@xmath82[t ] exhibit a richer structure of cmc slicings . let us concentrate on the spacetimes fc@xmath82[t ] . the @xmath271 hypersurfaces , @xmath272 , form a global cmc foliation of fc@xmath82[t ] . this foliation is reflection symmetric and invariant under the killing flow . however , this global cmc foliation does not define a cmc time function . note that the mean curvature is not monotone and attains a minimal value @xmath273 along the foliation , where @xmath274 is given by and ; it satisfies @xmath275 . more specifically , from fig . [ ksflow ] in combination with or fig . [ cbtdu ] we see that @xmath149 ( @xmath276 ) decreases monotonically from @xmath277 to @xmath273 as @xmath162 increases from @xmath278 to @xmath181 ; at @xmath279 it attains its minimal value @xmath273 ; as @xmath162 increases further , @xmath149 is monotonically increasing as well and converges to @xmath280 in the limit @xmath281 . in some of the kantowski - sachs spacetimes fc@xmath82[t ] , this global cmc foliation is the unique cmc foliation : [ cdsthm1 ] consider a kantowski - sachs spacetime fc@xmath82[t ] or pc@xmath82[t ] with @xmath282 then the unique ( spherically symmetric ) cmc foliation is the global foliation of hypersurfaces , where @xmath272 . recall that @xmath283 is a measure of the length of the @xmath41 factor of the kantowski - sachs spacetime . the situation is more intricate and more interesting for kantowski - sachs spacetimes with larger @xmath283 . [ cdsthm ] consider a kantowski - sachs spacetime fc@xmath82[t ] or pc@xmath82[t ] with @xmath284 and let @xmath285 be the greatest integer strictly less than @xmath286 . then the spacetime contains the global cmc foliation of @xmath271 hypersurfaces and , in addition , @xmath287 distinct families of ( maximally extended ) cmc slicings . the existence of ` families ' of cmc slicing is due to the fact that the embedded cmc hypersurfaces are not unique . the one - parameter family of isometries associated with the killing vector @xmath90 turns each cmc hypersurfaces into a one - parameter family of cmc hypersurfaces ( where of course each hypersurface of the family is characterized by the same parameters @xmath149 and @xmath142 ) . requiring reflection symmetry about the @xmath288 cylinder is a convenient way of fixing a unique representative of the family , see section [ embeddings ] . therefore , theorem [ cdsthm ] can be restated : under the given assumptions , _ fc@xmath82[t ] ( or pc@xmath82[t ] ) contains the global cmc foliation of @xmath271 hypersurfaces and @xmath287 slicings of reflection symmetric compact cmc hypersurfaces_. the remaining compact cmc slicings arise from these reflection symmetric slicings by an appropriate admixture of the killing flow . in addition to the statement of theorem [ cdsthm ] we will see that along the nontrivial slicings the mean curvature is a strictly monotone function where @xmath149 ranges over a subinterval of @xmath289 . the remainder of this section is concerned with the _ proof _ of the theorems and the additional claims . the first part of the proof is essentially identical to the proof of theorem 4.1 of @xcite . let us thus merely sketch those arguments : the starting point is proposition [ kc1embedprop ] , which establishes the existence of ( reflection symmetric ) compact cmc hypersurfaces @xmath244 in ( certain ) fc@xmath82[t ] spacetimes . the proof that each of these hypersurfaces evolves into a unique ( reflection symmetric ) local slicing @xmath290 , where @xmath291 is a parameter of the slicing such that @xmath292 , is based on an analysis of the lapse equation , see , e.g. , @xcite , @xmath293 where @xmath294 is the laplacian of the induced metric and @xmath295 ; the @xmath291-dependencies are suppressed . in ( 4.1 ) it is shown that possesses a unique even solution @xmath296 provided that @xmath297 , which implies the existence ( and uniqueness ) of ( reflection symmetric ) local cmc slicings . if the hypersurface @xmath244 corresponds to the point @xmath298 in the parameter space @xmath210 , see fig . [ kc01 ] , then the constructed local slicing @xmath290 corresponds to a ( piece of a ) smooth curve in @xmath210 , which passes through that point . smoothness follows from smoothness of @xmath59 and @xmath299 along @xmath290 , because @xmath300 to determine these curves in @xmath210 which represent cmc slicings , we differentiate and use the evolution and the lapse equation on the l.h.s.to obtain @xmath301 in this equation , @xmath302 is the ( unique ) even solution of and the prime denotes the derivative w.r.t . evaluation of at @xmath303 , which corresponds to @xmath304 , and use of yields @xmath305 see @xcite for details . we infer that the tangent vectors of the curves @xmath306 in @xmath210 which represent cmc slicings are determined by . summarizing , compact cmc slicings are represented in @xmath210 by the integral curves of the ` oriented direction field ' @xmath307 given by , see fig . [ kc1flown ] . maximal extension of the integral curves of the oriented direction field on @xmath210 corresponds to maximal extension of the cmc slicing in the spacetime . for the case @xmath308 , @xmath309 . the integral curves of this direction field are equipotential lines of the function @xmath234 on @xmath210 and represent compact cmc slicings in kantowski - sachs spacetimes fc@xmath82[t ] , @xmath259 , @xmath310.,scaledwidth=80.0% ] consider an integral curve @xmath306 of the oriented direction field . by construction , the quantity @xmath234 of is constant along this curve and the associated cmc slicing @xmath311 is a slicing in fc@xmath82[@xmath234 ] . however , every point @xmath240 represents the original cmc data set as well as its ( finite ) coverings . therefore , the integral curve represents not only a cmc slicing @xmath311 ( @xmath313 ) in fc@xmath82[@xmath234 ] , but also a cmc slicing @xmath290 in fc@xmath82[@xmath258 for all @xmath310 ; each hypersurface of the slicing @xmath290 corresponds to the @xmath243-fold covering of the original data . to establish the claims of the theorems ( and to obtain additional information on the properties of the cmc slicings ) it is necessary to understand the behavior of @xmath234 , viewed as a function of @xmath240 , see . first , we investigate the behavior of @xmath234 on the boundary ; the relevant statements are proved in appendix [ tonboundary ] : consider a curve in @xmath210 converging to a point on the boundary , which is either of following : a point on the right boundary with @xmath314 ; a point on the lower boundary ( including the cusp ) , which is uniquely represented by a value of @xmath162 in an interval @xmath315 $ ] ( where @xmath316 is known explicitly , see appendix [ tonboundary ] ) ; or a point on the left boundary which is uniquely represented by a value of @xmath162 in the interval @xmath317 ; see fig . [ cbtdu ] . we find that @xmath318 if the curve converges to a point on the right boundary ( i.e. , @xmath314 ) or on the lower boundary ( including the cusp ) . however , if the curve converges to a point represented by @xmath319 on the left boundary , then @xmath234 converges along this curve according to @xmath320 where the values of @xmath149 and @xmath142 that enter @xmath150 are @xmath321 and @xmath322 , cf . ; see appendix [ tonboundary ] for the proofs . the function @xmath323 in is strictly monotonically decreasing in . moreover , @xmath324 as @xmath325 and @xmath326 for the proof we refer to the appendix . based on and we are able to complete the proof of the theorems . consider an integral curve @xmath306 of the oriented direction field on @xmath210 , see fig . [ kc1flown ] ; we have @xmath327 along the curve . since @xmath328 , the @xmath302-limit set of the curve must be contained in the left boundary and/or the lower boundary of @xmath210 . however , @xmath329 on the lower boundary ( including the cusp ) , hence the curve must emerge from the left boundary . since @xmath323 is strictly monotone , the curve must emerge from one point on the left boundary , which is determined by solving the equation @xmath330 for @xmath331 , cf . ; the associated values of @xmath332 and @xmath333 are then given through . since , conversely , each point on the left boundary gives rise to exactly one integral curve of the oriented direction field on @xmath210 , we find a one - to - one correspondence between integral curves of the direction field and points on the left boundary . as a consequence , the range of @xmath234 on @xmath210 coincides with the range of @xmath334 on the left boundary . taking into account we therefore conclude that @xmath335 on @xmath210 and thus on every integral curve of the direction field . hence , invoking proposition [ kc1embedprop ] we have established theorem [ cdsthm1 ] . from each point on the left boundary an integral curve emerges . the monotonicity of @xmath336 along this boundary in combination with implies that , for a given @xmath337 , the equation @xmath338 distinguishes a unique integral curve of the direction field in @xmath210 . consider a kantowski - sachs spacetime fc@xmath82[t ] with @xmath337 . the integral curve @xmath338 represents a unique ( reflection symmetric ) slicing @xmath311 ( @xmath313 ) of compact cmc hypersurfaces in fc@xmath82[t ] . there is more , however . the equation @xmath339 defines a unique integral curve in @xmath210 for each @xmath340 , cf . theorem [ cdsthm ] . each of these integral curves corresponds to a slicing @xmath290 in fc@xmath82[t ] , whose hypersurfaces are isometric to the @xmath243-fold covering of the original data sets generated by @xmath306 , cf . [ kc1embedprop ] . accordingly , there exist @xmath243 slicings @xmath290 , @xmath341 , in fc@xmath82[t ] . this proves the remaining claim of theorem [ cdsthm ] . ' '' '' theorem [ cdsthm ] establishes the existence of @xmath342 compact cmc slicings @xmath290 , different from the @xmath271 slicing , in kantowski - sachs spacetimes fc@xmath82[t ] or pc@xmath82[t ] with sufficiently large @xmath283 . in the proof of theorem [ cdsthm ] in section [ sec : cdst ] we have shown that @xmath290 is represented in @xmath210 by the curve @xmath343 determined by the equation @xmath339 ( which is a maximally extended integral curve of the oriented direction field of fig . [ kc1flown ] ) . the _ asymptotic properties _ of the slicing @xmath290 are the following : toward the future , @xmath344 , i.e. , the mean curvature increases monotonically and converges to @xmath345 ; simultaneously , @xmath346 . these results follow straightforwardly from the discussion of section [ sec : cdst ] . in particular , we find that @xmath347 along @xmath290 as @xmath348 ( or , if we use a different time gauge , as @xmath291 approaches the finite supremum of its range ) , which follows from the fact that @xmath349 as @xmath350 in @xmath210 . toward the past , the slicing @xmath290 converges to a limit hypersurface . let @xmath162 be the solution of @xmath351 , i.e. , @xmath162 is the solution of @xmath352 cf . and . then @xmath353 and @xmath354 as @xmath355 ( where @xmath356 denotes the infimum of the maximal range of @xmath291 ) . these statements follow straightforwardly from the discussion of section [ sec : cdst ] . in particular , since @xmath357 and @xmath358 , see appendix [ tonboundary ] , we find that the slicing @xmath290 converges to the hypersurface of constant @xmath359 as @xmath355 . a natural question in connection with the slicing @xmath290 concerns its limit hypersurface . let us first note that for @xmath290 we have @xmath360 as @xmath361 , where we assume a gauge of @xmath291 such that @xmath302 remains bounded in the limit . to see this , consider the defining equation for the direction field . since @xmath362 converges to zero as @xmath363 , because , @xmath364 in this limit . an additional argument then shows that @xmath365 and @xmath366 separately as @xmath361 . ( suppose @xmath367 ; then @xmath368 and @xmath369 as @xmath361 . this implies that @xmath370 is tangential to the left boundary of @xmath210 , since @xmath371 by . this is false , however ; the oriented direction field is transversal to the left boundary , which can be straightforwardly deduced from the behavior of the function @xmath234 on the boundary and in a small neighborhood thereof . we simply recall that @xmath234 ( @xmath372 ) is strictly monotone along the boundary , while @xmath373 along the integral curves of the direction field . ) let us now consider the lapse equation and its solutions on the hypersurface @xmath374 , i.e. , for @xmath375 . first , recall that the metric on @xmath374 is given by @xmath376}}$ ] ; the variable @xmath144 is related to @xmath377 by @xmath378 . ( this follows directly from or from with . ) since @xmath374 is regarded as a hypersurface in fc@xmath82[t ] or pc@xmath82[t ] , @xmath159 is identified with @xmath160 , where @xmath379 . the lapse equation @xmath380 specializes to @xmath381 where @xmath382 . standard algebraic manipulations show that @xmath383 for @xmath375 , for all @xmath319 , and @xmath384 as @xmath385 ( i.e. , @xmath386 ) . in the case @xmath387 the general even solution of is @xmath388 with @xmath389 . ( in the case @xmath390 we obtain @xmath391 . ) intuitively speaking , the lapse function connects @xmath374 with a infinitesimally neighboring cmc hypersurface in fc@xmath82[t ] or pc@xmath82[t ] , if and only if @xmath392 for some @xmath310 , i.e. , if . using that @xmath393 , it is straightforward to see that this condition is equivalent to @xmath394 , cf . . this is in prefect accord with the previous results and we conclude that the slicing @xmath290 , @xmath395 , can be smoothly extended to a slicing @xmath396 , which also includes @xmath374 . in addition , the slicing @xmath290 , @xmath396 , can be joined continuously to the slicing of @xmath102 hypersurfaces at @xmath374 . however , the junction is not smooth , because for the slicing of @xmath102 hypersurfaces we have @xmath297 and the lapse function is @xmath397 , while @xmath398 and holds for @xmath290 at @xmath399 . it remains to investigate whether the slicings @xmath290 are in fact foliations . in analogy to we may define @xmath400 as the set of all @xmath401 such that the lapse function @xmath302 is positive . ( then the slicing @xmath290 is a foliation for those values of @xmath291 for which the integral curve @xmath306 associated with @xmath290 is in @xmath400 . ) recall the parameter space @xmath210 from figs . [ cbtdu ] and [ kc01 ] . we can prove that there exists a neighborhood @xmath402 in @xmath403 of the left boundary and a neighborhood @xmath404 of the lower boundary of @xmath210 such that @xmath405 and @xmath406 . the former statement follows immediately from , the proof of the latter statement is analogous to the proof of ( * ? ? ? furthermore , there exists a neighborhood @xmath407 in @xmath403 of the straight line @xmath408 , which is the right boundary of @xmath210 , such that @xmath409 . ( to see this we exploit the fact that @xmath410 as @xmath411 , which is intimately connected with the fact that the cmc hypersurfaces with @xmath314 , which are not compact , form a foliation ; for the details we refer to upcoming work . ) . numerical investigations suggest that @xmath400 is a connected set , whose minimal value of @xmath149 is strictly greater than @xmath273 . for large @xmath142 , the boundary @xmath412 approximates the left boundary of @xmath210 ; this leads to the fact that every integral curve lies in @xmath400 for sufficiently large @xmath291 . expressed in terms of the associated slicing in fc@xmath82[t ] : @xmath290 is indeed a foliation for sufficiently large @xmath291 . a concluding remark : compact cmc hypersurfaces and slicings in spacetimes that satisfy the timelike convergence condition are a ( comparatively ) straightforward matter . spacetimes that do not satisfy the timelike convergence condition , on the other hand , may exhibit a rich structure of compact cmc hypersurfaces and cmc slicings . the vacuum kantowski - sachs spacetimes with positive cosmological constant are a fine example to illustrate this fact and to provide indications about what to expect of cmc slicings in more general spacetimes . this appendix is concerned with an analysis of the function @xmath234 on boundary of @xmath210 and the proof of the statements of section [ sec : cdst ] in connection with , , and , which are essential for the proof of theorem [ cdsthm1 ] and theorem [ cdsthm ] . the boundary of the parameter space @xmath210 consists of ( a piece of ) the straight line @xmath314 , the curve @xmath415 , @xmath416 $ ] , which corresponds to the lower boudary and includes the cusp , and the curve @xmath415 , @xmath417 , which is the left boundary , see and figs . [ cbtdu ] and [ kc01 ] . in this context we have @xmath418 which follows by setting @xmath419 in and using that @xmath420 . let us begin by proving . consider a curve @xmath421 in @xmath210 that converges , as @xmath422 , to a point @xmath423 on the lower boundary that is represented by @xmath424 . let @xmath425 be the function defined in with @xmath426 and @xmath427 ; its profile is that of fig . [ deltaborderline](a ) . as @xmath428 , however , @xmath429 , where @xmath430 is of the form depicted in fig , [ deltaborderline](b ) ( since @xmath431 because of the assumption @xmath432 ) . from we have @xmath433 where @xmath434 and @xmath435 are the relevant zeros of @xmath436 , see fig . [ deltaborderline](a ) . as @xmath437 , we have @xmath438 and @xmath439 , where @xmath440 is a simple zero of @xmath430 . let @xmath441 ; we split into an integral over the interval @xmath442 and another over the interval @xmath443 . the latter converges to some constant as @xmath437 ; the former , however , and thus @xmath444 itself , will be shown to diverge as @xmath437 . to that end , we make the expansion @xmath445 as @xmath446 , where we note that @xmath447 and @xmath448 as @xmath437 . in this context , a prime denotes the derivative w.r.t . the argument , i.e. , @xmath93 . we then find @xmath449 as @xmath446 , where @xmath450 is independent of @xmath451 . in addition , @xmath452 then integration yields @xmath453 where @xmath454 is independent of @xmath451 . letting @xmath437 while keeping @xmath455 fixed , we obtain divergence , because @xmath447 . accordingly , @xmath456 as @xmath437 . likewise , we find divergence of @xmath234 along every curve in @xmath210 that converges to a point on the right boundary @xmath314 . to see that , consider a curve @xmath421 in @xmath210 that converges , as @xmath422 , to a point @xmath457 on @xmath458 . the profile of @xmath425 is that of fig . [ deltaborderline](a ) . as @xmath428 , we have @xmath429 , where @xmath459 which entails that the zero @xmath434 of @xmath425 converges to @xmath460 , which is a zero of @xmath430 ; however , @xmath461 as @xmath428 . for sufficiently small @xmath451 , there exists a uniform bound for @xmath425 on @xmath462 $ ] ; furthermore , @xmath463 is positive on @xmath403 and thus uniformly bounded from below . therefore , @xmath464 and , since @xmath461 as @xmath428 , we obtain @xmath465 in this limit . this concludes the proof of . let us proceed by proving . consider a curve @xmath421 in @xmath210 that converges , as @xmath422 , to a point @xmath423 on the left boundary that is represented by . as @xmath428 , @xmath429 , where @xmath430 is of the form depicted in fig . [ deltaborderline](c ) . let @xmath466 be defined as the maximum of @xmath425 , see fig . [ deltaborderline](a ) ; obviously , the three values @xmath467 converge to @xmath162 as @xmath437 . set @xmath468 ( for purely aesthetic reasons ) and consider the function @xmath469 whose zeros we denote by @xmath470 and @xmath471 . obviously , there exists a diffeomorphism @xmath472 such that @xmath473 in the interval @xmath474 $ ] ; we have @xmath475 , , and @xmath476 . we find @xmath477 it is straightforward to calculate the integral on the r.h . side : we obtain @xmath478 . from this result , the claim , i.e. , @xmath479 follows immediately by noting that @xmath480 which is because @xmath481 . it remains to prove the claims in connection with . let @xmath150 be the function of associated with @xmath482 . first , to establish strict monotonicity of @xmath323 , we consider @xmath483 and compute @xmath484 = 8 { r_{\mathrm{m}}}^{-3}\ , \left(1- \frac{3 m}{{r_{\mathrm{m}}}}\right)^2\ : > 0 \:;\ ] ] strict inequality holds because @xmath485 ( @xmath486 ) . ( note that the computation involves derivatives of w.r.t . @xmath162 , since @xmath332 and @xmath333 in @xmath150 . ) second , @xmath324 as @xmath325 , since @xmath487 as @xmath162 approaches the ` cusp ' @xmath181 . third , to show we compute @xmath488 \big ) \:,\ ] ] hence @xmath489 as @xmath281 , and is proved . l. andersson . he global existence problem in general relativity . in p.t . chruciel and h. friedrich , editors , _ the einstein equations and the large scale behaviour of gravitational field _ , pages 71120 . birkhuser , 2004 . y. choquet - bruhat , a.e . fischer , and j.e . maximal hypersurfaces and positivity of mass . in j. ehlers , editor , _ isolated gravitating systems in general relativity _ , pages 396456 . north - holland publishing , 1979 .

cellular automata ( ca ) models for solar flares are successful in explaining solar flare statistics ( peak flux , total flux , and duration distributions ; lu & hamilton 1991 ( hereafter lh91 ) ; lu et al . 1993 ; vlahos et al . 1995 ; georgoulis & vlahos 1996 , 1998 ; galsgaard 1996 ) . they simplify strongly the details of the involved physical processes , and achieve in this way to model large volumes with complex field topologies and a large number of events . on the other hand , mhd simulations give insight into the details of the local processes , they are limited , however , to modeling relatively small fractions of active regions , due to the lack of computing power , yielding thus poor statistics and difficulties in comparing results to observations ( e.g. mikic et al . 1989 ; strauss 1993 ; longcope & sudan 1994 ; einaudi et al . 1996 ; galsgaard & nordlund 1996 ; hendrix & van hoven 1996 ; dmitruk & gomez 1998 ; galtier & pouquet 1998 ; georgoulis et al . 1998 ; karpen et al . 1998 ; einaudi & velli 1999 ) . the _ global _ mhd flare models are still in the state of rather qualitative flare scenarios . the mhd and the ca approach to solar flares seem to have very little in common : the former are a set of partial differential equations , based on fluid - theory and maxwell s equations , whereas the latter are a set of abstract evolution rules , based ( in the case of solar flares ) on the analogy to critical phenomena in ( theoretical ) sand - piles . the scope of this paper is to bridge the gap in - between these two approaches : the solar flare ca models are re - interpreted and extended so as ( i ) to make these models completely compatible with mhd and with maxwell s equations , and so that ( ii ) all relevant mhd variables are made available ( e.g. the current and the electric field , which so far were not available in ca models ) . in an earlier paper ( isliker et al . 1998 ) , we have analyzed the existing solar flare ca models for their soundness with mhd . we asked the question whether the fields in these ca models and the evolution rules can be interpreted in terms of mhd . it turned out that these models can indeed be interpreted as a particular way of implementing numerically the mhd equations . this fact is not trivial , since these models had been derived in quite close analogy to the sand - pile ca model of bak et al . ( 1987 and 1988 ) , with vague association of the model s variables with physical quantities . for instance , some authors ( lu et al . 1993 ) explicitly discuss the question whether their basic grid variable is the magnetic field or not , without reaching to a definite conclusion . isliker et al . ( 1998 ) brought forth not only how the existing ca models are related to mhd and what simplifications are hidden , but also where they differ from or even violate the laws of mhd and maxwell s equation . important is the fact that though the existing ca models can be considered as a strongly simplified numerical solution of the ( simplified ) mhd equations , _ they do not represent the discretized mhd equations _ : the time - step and the spacing between two grid sites are not small ( in a physical sense ) , but finite ; they are a typical temporal and spatial scale of the diffusive processes involved ( see isliker et al . 1998 ) . from the point of view of mhd , the main short - comings of the existing ca models are ( isliker et al . 1998 ) : ( 1 ) there is no control over consistency with maxwell s equations . interpreting , for instance , the vector - field in the ca models as the magnetic field leads to the problem that the gradient of the field ( @xmath0 ) can not be controlled . ( 2 ) secondary quantities , such as currents , are not available , and they can not be introduced in the straightforward way by replacing differential expressions by difference - expressions , since , as mentioned , the grid - size must be considered finite ( see also app . this lack of knowing how to calculate derivatives made it also useless to interpret the primary vector - field in the ca models as the vector potential ( to avoid the @xmath0-problem ) , since @xmath1 could not be derived . the physical interpretation of these ca models remained so far problematic . there are two basically different ways of developing ca models for flares further : ( i ) either one considers ca models _ per se _ , tries to change the existing models further or invent new ones , with the only aim of adjusting them to reproduce still better the observations , i.e. one makes them a tool the results of which explain and maybe predict observed properties of flares . in this approach , one has not to care about possible inconsistencies with mhd or even maxwell s equations , the various components of the model are purely instrumentalistic . ( ii ) on the other hand , one may care about the physical identification and interpretation of the various components of the model , not just of its results , and one may want the ca model to become consistent with the other approach to solar flares , namely mhd . in the approach ( ii ) , some of the freedom one has in constructing ca models will possibly be reduced , since there are more boundary conditions to be fulfilled in the construction of the model : the observations must be reproduced and consistency with mhd has to be reached . ( trials to construct new ca models which are based on mhd and not on the sand - pile analogy were recently made by einaudi & velli ( 1999 ) , macpherson & mackinnon ( 1999 ) , longcope and noonan ( 2000 ) , and isliker et al . ( 2000a ) . ) our aim is in - between these two alternatives : we construct a set - up which can be superimposed onto each classical solar flare ca model , and which makes the latter interpretable in a mhd - consistent way ( by _ classical _ ca models we mean the models of lh91 , lu et al . 1993 , vlahos et al . 1995 , georgoulis & vlahos 1996 , 1998 , galsgaard 1996 , and their modifications , which are based on the sand - pile analogy ) . the set - up thus specifies the physical interpretation of the grid - variables and allows the derivation of quantities such as currents etc . it does not interfere with the dynamics of the ca ( unless wished ) : loading , redistributing ( bursting ) , and the appearance of avalanches and self - organized criticality ( soc ) , if the latter are implied by the evolution rules , remain unchanged . the result is therefore still a ca model , with all the advantages of ca , namely that they are fast , that they model large spatial regions ( and large events ) , and therewith that they yield good statistics . since the set - up introduces all the relevant physical variables into the context of the ca models , it automatically leads to a better physical understanding of the ca models . it reveals which relevant plasma processes and in what form are actually implemented , and what the global flare scenario is the ca models imply . all this was more or less hidden so far in the abstract evolution rules . it leads also to the possibility to change the ca models ( the rules ) at the guide - line of mhd , if this should become desirable . not least , the set - up opens a way for further comparison of the ca models to observations . in sec . 2 , we introduce our set - up . applying it to several ca models ( sec . 3 ) , we will demonstrate the usefulness and some of the benefits such extended models ( i.e. classical models extended with our set - up ) provide over the classical ca models , and we will reveal basic physical features of the ca models . the potential of the extended models to explain more observational facts than the classical ca models is , among others , outlined in the conclusions ( sec . 4 ) . the set - up we propose can be superimposed onto solar flare ca models which use a 3d grid and a basic 3d vector grid - variable , say @xmath2 . the corresponding set of evolution rules is not changed . ( with a few modifications , the set - up can also be superimposed onto ca models which use a scalar field in a planar grid , which our set - up necessarily interprets as slab geometry , as will become clear later . ) we introduce our model on the example of the solar flare ca model of lh91 , which we summarize here in order to make the subsequent presentation more concrete : in the lh91 model , to each grid - site @xmath3 of a 3d cubic grid a 3d vector @xmath4 is assigned . initially , @xmath4 is set to @xmath5 , everywhere . the system is then loaded with the repeated dropping of increments at randomly chosen sites @xmath3 ( one per time - step ) @xmath6 where @xmath7 has all its components as random numbers uniformly distributed in @xmath8 $ ] . after each loading event , the system is checked for whether the local stress , defined as @xmath9 where the sum goes over the six nearest neighbours of the central point @xmath3 , exceeds a threshold @xmath10 , i.e. whether @xmath11 where @xmath12 is used . if this is the case , the field in the neighbourhood of the critical site is redistributed according to @xmath13 for the central point , and @xmath14 for its six nearest neighbours . in such a redistribution event ( burst ) , energy amounting to @xmath15 is assumed to be released . the grid is scanned again and again to search for second , third etc . generation bursts , until the system is nowhere critical anymore and returns to the loading phase ( the details we apply concerning the temporal evolution of the model are given in app . a ; they are not explicitly stated in lh91 ) . the field outside the grid is held constant and assumed to be zero . we turn now to introducing our set - up , starting with a specification : we interpret the vector @xmath4 at the grid sites @xmath3 to denote the local vector - field , @xmath16 . note that this was not specified in the classical ca models . lu et al . ( 1993 ) for instance discuss this point : it might also have been thought of as a mean local field , i.e. the average over an elementary cell in the grid . guided by the idea that we want to assure @xmath17 for the magnetic field @xmath1 , which is most easily achieved by having the vector - potential @xmath2 as the primary variable and letting @xmath1 be the corresponding derivative of @xmath2 ( @xmath18 ) , we furthermore assume that the grid variable @xmath2 of the ca model is identical with the vector - potential . the remaining and actually most basic problem then is to find an adequate way to calculate derivatives in the grid . in general , ca models assume that the grid - spacing is finite , which also holds for the ca model of lh91 ( as shown in detail by isliker et al . 1998 ) , so that the most straightforward way of replacing differential expressions with difference expressions is not adequate ( see the detailed discussion in app . b.1 , below ; vassiliadis et al . ( 1998 ) suggested to interpret ca models as the straightforwardly discretized ( simplified ) mhd equations , which we find problematic for the reasons given in app . b.1 , and we therefore do not follow this approach , here ) . consequently , one has to find a way of continuing the vector - field into the space in - between the grid - sites , which will allow to calculate derivatives . there is , of course , an infinite number of possibilities to do so , and the problem can not have a unique solution . adequate possibilities definitely include : a ) continuation of @xmath2 with the help of an equation ( e.g. demanding the resulting @xmath1-field to be potential or force - free ) ; b ) interpolation , either locally ( in a neighbourhood ) , or globally ( through the whole grid ) . trying several methods , we concluded that 3d cubic spline interpolation is particularly adequate to the problem since it has remarkable advantages over other methods ( e.g. it does not introduce oscillations in - between grid - sites , which would strongly influence the values of the derivatives , and it well reproduces the derivatives of analytically prescribed primary fields ) . the process of evaluating different continuation methods we went through , as well as the comparison of spline interpolation to other continuation - methods are described in app . b. the 3d interpolation is performed as three subsequent 1d interpolations in the three spatial directions ( press et al . 1992 ) . for the 1d splines , we assume natural boundaries ( the second derivatives are zero at the boundaries ) . moreover , since in the ca model of lh91 it is assumed that around the grid there is a zero field which is held constant ( see sec . 2.1 ) , we enlarge the grid by one grid point in all directions to include this constant zero - layer explicitly , using it however only for the interpolation . in the interpolation , the derivatives at the grid - points are immediately given by the analytically differentiated interpolating polynomials . with the help of this interpolation , the magnetic field @xmath1 and the current @xmath19 are calculated as derivatives of @xmath2 , according to the mhd prescription : @xmath20 @xmath21 to determine the electric field @xmath22 , we make the assumption that under coronal conditions the mhd approach is in general valid , and that @xmath22 is reasonably well approximated by ohm s law in its simple form , @xmath23 , with @xmath24 the diffusivity and @xmath25 the fluid velocity . since the classical ca models use no velocity - field , our set - up can yield only the resistive part , @xmath26 in applications such as to solar flares , where the interest is in current dissipation events , i.e. in events where @xmath24 and @xmath19 are strongly increased , eq . ( 9 ) can be expected to be a good approximation to the electric field . theoretically , the convective term in ohm s law would in general yield just a low - intensity , background electric field . ( 9 ) needs to be supplemented with a specification of the diffusivity @xmath24 : isliker et al . ( 1998 ) have shown that in the classical ca models the diffusivity adopts the values @xmath27 at the unstable ( bursting ) sites , and @xmath28 everywhere else . this specifies eq . ( 9 ) completely . _ remark 1 : _ it is worthwhile noting that , since spline interpolation has the property to be the least curved of all twice differentiable interpolating functions , the grid - size is a typical smallest - possible length - scale of field structures , or , if we would think the ca to model mhd turbulence , it is something like an average smallest possible eddy - size . _ remark 2 : _ with our set - up , field lines are made available : the interpolation was introduced to define derivatives at the grid - sites , but it can as well be used to determine the vector - potential , and therewith the magnetic field and the current , in between the grid sites . however , it is important to note that this field in between the grid sites is not used for the time - evolution of the model , it merely allows to visualize the evolution in the standard way through field - lines , if wished so ( second order derivatives in - between the grid - sites , if needed , would have to be done numerically , since else their numerical values would depend on the order in which the three 1d interpolations are done ) . _ remark 3 : _ in this paper , we do not change the rules of the classical ca models to which we apply our set - up except for the definition of energy release ( sec . our aim here is to show the usefulness of the set - up and to give some results and reveal some aspects by extending published , classical ca models . while the redistribution rules have in detail been shown to represent diffusion events and fit nicely into an mhd scenario ( isliker et al . 1998 ) , the loading process is strongly simplified and poorly follows a reasonable flare scenario : for instance , the loading process acts independently everywhere in the simulation box , whereas according to a realistic flare scenario ( see e.g. parker 1993 ) disturbances should appear independently only on one boundary ( the photosphere , due to random foot - point motions or newly appearing flux ) , and propagate then into the interior of the simulation box along the magnetic field lines . to translate such a realistic loading scenario into the language of ca models has not been undertaken , so - far . we just note that it would be quite straightforward to introduce a velocity field into the ca models : e.g. isliker et al . ( 2000a ) propose a ca model which uses a velocity field for the loading phase , but this model does not fall into the category of classical ca models since it does not follow the sand - pile analogy and uses different , mhd based , evolution rules . we leave the problem of introducing a velocity field and a more physical loading process into the classical ca models for a future study . in isliker . 2000b , we will among others analyze in details what this simplified loading process physically represents . our first application is to the ca model of lh91 ( see sec . the lh91 model has a fairly long transient phase and reaches finally a stationary state , the so - called soc ( self - organized criticality ) state , in which spatially spreading series of bursts ( avalanches ) appear , alternating with quiet loading phases . the lh91 model gives basically three results concerning flare statistics , namely the distributions of total energy , peak - flux and durations , which are all power - laws with slopes that are in good agreement with the observations ( lu et al . 1993 , bromund et al . 1995 ) . superimposing our mhd - frame onto the lh91 model such as it stands does not change anyone of the three results , since at this first stage we are not interfering with the dynamics ( i.e. the evolution rules ) . the set - up allows , however , to address several questions in mhd language : our main aim in the subsequent applications is to demonstrate that the set - up indeed yields a new and consistent interpretation of ca - models , to illustrate the behaviour of the secondary variables ( currents , magnetic fields ) , and to reveal major features of them . ( in the subsequent runs , we use a grid of size @xmath29 , as lh91 did to derive their main results . ) first , we turn to the question what the global fields ( vector - potential , magnetic field , current ) look like during the soc state . thereto , the temporal evolution of the model is stopped at an arbitrary time during soc state ( in a phase where there are no bursts , i.e. during loading ) , and the magnitude of the fields at a cut with fixed @xmath30-coordinate are shown as a function of the @xmath31- and @xmath32-coordinates in fig . 1 . @xmath33 obviously exhibits a large - scale organization over the whole grid , it forms a global convex surface ( fig . this convex surface has a slight random distortion over - lying , which visually can not be discerned in fig . 1(a ) , but becomes visible in the plot of @xmath34 ( fig . 1(b ) ) , the curl of @xmath2 , which still exhibits large - scale organization all over the grid , but is clearly wiggled . finally , @xmath35 shows no left - overs of a large scale organization anymore , it reflects the random disturbances of the convexity of @xmath36 ( fig . 1(c ) ) . the large - scale structures shown in fig . 1 are always maintained during the soc state , neither loading nor bursting ( and avalanches ) destroy them , they just tremble a little when such events occur . soc state in the extended lh91 model thus implies large - scale organization of the vector - potential and the magnetic field , in the characteristic form of fig . 1 . the large - scale organization of @xmath2 is not an artificial result of our superimposed set - up , but already inherent in the classical lh91 model : in the classical lh91 ca model , there is only one variable , the one we call here @xmath2 , whose values are not affected by the interpolation we perform since it is the primary grid variable , so that fig . 1(a ) is true also for the classical , non - extended lh91 model . the large scale structure for the primary grid - variable @xmath37 is the result of a combined effect : the preferred directionality of the loading increments ( see sec . 2.1 ) tries to increase @xmath37 throughout the grid . the redistribution events , which already in bak et al . ( 1987 ; 1988 ) were termed diffusive events , and which in isliker et al . ( 1998 ) were analytically shown to represent local , one - time - step diffusion processes , smooth out any too strong spatial unevenness of @xmath2 , and they root the @xmath2-field down to the zero level at the open boundaries . the result is the convex surface of fig . 1(a ) , blown - up from below through loading , tied to the zero - level at the edges , and forced to a maximum curvature which is limited by the local , threshold dependent diffusion events . as the soc state , so is the large - scale structure of @xmath37 independent of the concrete kind of loading , provided it fulfills the conditions that the loading increments exhibit a preferred directionality and are much smaller than the threshold ( with symmetric loading , the soc state is actually never reached , see lh91 and lu et al . ( 1993 ) ) . to make sure of the importance of the boundaries , we performed runs of the model with closed boundaries , and we found that neither a large - scale structure was developed in @xmath37 , nor the soc state was reached . to illustrate the role of the current at unstable sites and during bursts , we plot in fig . 2 the magnitude of the current before and after a typical burst : obviously , the current at the burst site @xmath38 has high intensity before the burst ( fig . 2(a ) ) , and is relaxed after the burst ( fig . inspecting a number of other bursts , we found that , generally , at sites where the lh91 instability criterion is fulfilled , the current is increased , too , and that bursts dissipate the currents . this is a first hint that classical ca models can be interpreted as models for energy release through current - dissipation . after the burst , at the neighbouring site @xmath39 , the intensity of the current is increased , and indeed the presented burst gives rise to subsequent bursts , it is one event during an avalanche . the magnetic field at the bursting site is reshaped , in a way which is difficult to interpret when using only the magnitude of it ( @xmath34 ) for visualization . may - be field line plots would help visualization , but we leave this for a future study . we now turn to the question what relation the energy release formula of lh91 ( eq . 6 ) has to the respective mhd relations : in parallel to using the formula of lh91 , we determine the released energy in the following ways , closer to mhd : first , we assume it to be proportional to @xmath40 ( with the diffusivity @xmath41 at unstable sites , see sec . 2.2 ) , which we linearly interpolate between the two states before and after the burst . this is done in two ways , ( i ) summing over the local neighbourhood , @xmath42 and ( ii ) without summing , but just taking into account the current at the central point , @xmath43 and finally , we monitor the change in magnetic energy due to a burst using the difference in magnetic energy in the local neighbourhood , @xmath44 ( in eqs . ( 10 ) , ( 11 ) , ( 12 ) , we assume @xmath45 and @xmath46 for the grid - spacing @xmath47 and the time - step @xmath48 , since , according to isliker et al . ( 1998 ) , in the classical ca models both values are not specified and set to one . ) the corresponding distributions of total energy and peak - flux are shown in fig . 3 , together with the distributions yielded by the energy - release formula of lh91 , eq . ( 6 ) ( the duration distribution remains the same as in the classical lh91 model , namely a power - law , and is not shown ) . obviously , the four ways of defining the released energy give basically similar results , with larger deviations only at the low and high energy ends ( note that the energy in fig . 3 is in arbitrary units ) . using the formula of ohmic dissipation does thus not change the results of the classical lh91 model . with an estimate of the numerical value of the anomalous resistivity and of the typical size of a diffusive region or the typical diffusive time , it would be possible to introduce physical units . we did not undertake this , since all three parameters are still known only with large observational and theoretical uncertainties . from the similarity of the distributions of the extended model with the ones of the classical lh91 model ( fig . 3 ) , and from fig . 2 , where it was seen that an instability is accompanied by an enhanced current , we are led to ask directly for the relation of @xmath49 to @xmath19 , which we plot as a function of each other in fig . 4 . obviously , the two quantities are related to each other : above @xmath50 , the current is an approximate linear function of the stress , around @xmath50 the current is zero , and below there is again an approximate linear relationship , with negative slope , however ( above @xmath50 the current @xmath19 is actually preferably along @xmath51 , whereas below it is preferably along @xmath52 , i.e. @xmath19 is an approximatly linear function of @xmath49 in the whole range , it merely changes its directivity at @xmath50 with respect to @xmath49 ) . in appendix c , we show analytically why with our set - up a more or less close relation between @xmath49 and @xmath19 has to be expected . of particular interest in fig . 4 is that if @xmath53 is above the threshold @xmath12 , then @xmath54 is also reaching high values : obviously , large values of @xmath53 imply large values of @xmath54 . this confirms the statement made above : the extended ca models can be considered as models for energy release through current dissipation . it also explains why the energy distributions remain very similar when the lh91 formula for the amount of energy released in a burst ( @xmath55 , eq . ( 6 ) ) is replaced by ohmic dissipation ( @xmath56 , sec . 3.1.3 ) : bursts occur only for large stresses @xmath53 , where @xmath54 is also large and an approximate linear function of @xmath53 , so that the distributions of @xmath57 and @xmath58 can be expected to be the same in shape . georgoulis and vlahos ( 1996 , 1998 ) introduced power - law distributed increments for the loading . the main result of such a way of driving the system is that the power - law indices of the energy - distributions depend on the power - law index of the distribution of the loading increments , explaining thus the observed variability of the indices through the variability of the intensity of the driving . we generalize their way of power - law loading , which is for a scalar primary field , to a vector field in the following way : the anisotropic directivity of the loading increment @xmath59 is kept ( see sec . 2.1 ) , but @xmath60 is now distributed according to @xmath61 with @xmath62 $ ] and @xmath63 , the power - law index , a free parameter . simulations were performed for @xmath64 and @xmath65 . interested in global features implied by the ca model , our concern here is the structure of the magnetic field . it turns out that the magnetic field exhibits still a large scale organization , which is very similar to the one of the @xmath1-field of the ( extended with our set - up ) lh91 model ( fig . 1(b ) ) : for @xmath66 , the respective plots are visually indiscernible , and for @xmath67 the overall shape is still roughly the same , it merely seems slightly more distorted . thus , though the statistical results depend on @xmath63 , the strength and variability of the loading , the structure of the magnetic field remains approximately the same as in the case of the extended model of lh91 . large - scale organization ( in the characteristic form of fig . 1 ) must consequently be considered as an inherent property of soc state , through the mechanism explained in sec . 3.1.1 . vlahos et al . ( 1995 ) introduced anisotropic bursts for solar flare ca models , which lead only to small events , but yield a steep distribution at small energies , predicting thus a significant over - abundance of small events with a significant contribution to coronal heating . we have first to generalize the anisotropic evolution rules , which are again for a scalar primary field , to the case of a primary vector field . a natural generalization would be to apply the anisotropic rules to the absolute magnitude of @xmath2 , but it turns out that this causes the algorithm to get trapped in infinite loops ( two neighbouring grid - sites trigger each other mutually for ever ) . the same holds if we apply the anisotropic rules to the absolute magnitudes of the three components of @xmath2 independently . we finally applied the anisotropic rules to the three components of @xmath2 directly , not using absolute magnitudes , as also vlahos et al . ( 1995 ) did not use absolute magnitudes , and this turned out to lead to a stationary asymptotic state : the anisotropic stress in the @xmath31-component is thus defined as @xmath68 where @xmath69 stands for one of the six nearest neighbours . the instability criterion is @xmath70 and the redistribution rules become @xmath71 for the central point and @xmath72 for those nearest neighbours which fulfill the instability criterion ( eq . 15 ) , where the primed sum is over those neighbours for which eq . ( 15 ) holds . the rules for @xmath73 are completely analogous ( so that actually there are 18 possibilities to exceed the threshold ( eq . 15 ) at a given site ) . the released energy is assumed to amount to @xmath74 we performed a run where only the anisotropic burst - rules were applied , in order to isolate their effect , although the anisotropic burst - rules are used always together with the isotropic ones by vlahos et al . ( 1995 ) , since alone they can not explain the complete energy distributions of flares . in fig . 5 , the magnitude of the magnetic field at a cut through the grid is shown ( fixed @xmath30 ) , for an arbitrary time ( in the loading phase ) during the asymptotic stationary state of the model . clearly , there is no overall large scale structure anymore , except that the magnetic field along the boundaries is increased . the magnetic field topology is thus nearer to the concept of a random , relatively unstructured magnetic field than the magnetic field topology yielded by the isotropic models in soc state . the anisotropic burst rules do not yield large - scale structures , as they are , when used alone , also not able to lead the system to soc state : this is obvious from the energy distributions they yield , which are much smaller in extent than the ones given by the isotropic rules ( see vlahos et al . 1995 ) , and confirmed by the result of lu et al . ( 1993 ) that isotropy of the redistribution rules at least on the average is a prerequisite to reach soc state , at all . the anisotropic bursts occur independently in all directions and are in this way not able to organize the field in a neighbourhood systematically , and , as a consequence , also not in the entire grid . the inquiry of the relation of the energy release formula eq . ( 18 ) , which is different from the isotropic formula ( eq . 6 ) , to mhd based formulae we leave for a future study . we just note that the distributions the anisotropic model in our vector - field version yields are at lower energies , smaller in extent , and steeper than the ones of the isotropic models . we have introduced a new set - up for classical solar flare ca models which yields , among others , consistency with maxwell s equations ( e.g. divergence - free magnetic field ) , and availability of secondary variables such as currents and electric fields in accordance with mhd . both are new for solar flare ca models . the set - up specifies the so far open physical interpretation of the ca models . this specification is to some extent unavoidably arbitrary , and it would definitely be interesting to see what alternative interpretations would yield if they can be derived consistently . we can claim , however , that the interpretation we chose is reasonable , it is well - behaved in the sense that the derivatives of analytically prescribed vector - potentials are reproduced and that the abstract stress - measure of the ca models is related to the current , due to general properties of spline interpolation . the central problem which was to solve is how to calculate derivatives in a ca model , i.e. how to continue the primary grid - variable in - between the grid sites , since the notion of derivatives is alien in the context of ca models quite in general . in this article , our main aim with the introduced set - up was to demonstrate that the set - up truly extends the classical ca models and makes them richer in the sense that they contain much more information , now . the main features we revealed about the ca models , extended with our set - up , are : * 1 . large - scale organization of the vector - potential and the magnetic field : * the field topology during soc state is bound to characteristic large - scale structures which span the whole grid , very pronounced for the primary grid variable , the vector - potential , but also for the magnetic field . bursts and flares are just slight disturbances propagating over the large - scale structures , which are always maintained , also in the largest events . the magnitude of the current , as a second order derivative of the primary field , does not show any obvious large - scale structure anymore , it reflects more or less only the random fluctuations of the large - scale organized magnetic field . it is worthwhile noting that the large - scale structure of the primary grid - variable is not an artificial result of our set - up , but a natural consequence of the soc state in which the system finds itself . the appearance of large - scale structures for the primary grid variable was shown here for the first time . it may have been known to different authors , but it never has explicitly been shown : soc models for flares are derived in analogy to sand - pile dynamics , and the paradigm of a pile reappears in the field topologies of the solar flare ca models . * 2 . increased current at unstable grid - sites : * unstable sites are characterized by an enhanced current , which is reduced after a burst has taken place , as a result of which the current at a grid - site in the neighbourhood may be increased . * availability of the electric field : * the electric field is approximated with the resistive part of ohm s law in its simple form , which can in general be expected to be a good approximation in coronal applications and where the interest is in current - dissipation events , e.g. in the case of solar flares . energy release in terms of ohmic dissipation : * we replaced the some - what _ ad hoc _ formula in the ca models to estimate the energy released in a burst with the expression for ohmic dissipation in terms of the current . the distributions yielded in this way are very similar to the ones based on the ad hoc formula , so that the results of the ca models remain basically unchanged . * ca as models for current dissipations : * as a consequence of point 2 and 4 in this list , and of the fact that there is an approximate linear relation between the current and the stress measure of the ca , we can conclude that the _ extended _ ca models can be considered as models for energy release through current dissipation . our set - up is to be contrasted to the recently suggested mhd - derived ( not based on the sand - pile analogy ) ca models of einaudi & velli ( 1999 ) , macpherson & mackinnon ( 1999 ) , longcope and noonan ( 2000 ) , and isliker et al . ( 2000a ) . they all suggest new evolution rules , derived from mhd , and all in different ways ( they actually focus on different processes , namely the microscopic , macroscopic , and mesoscopic physics , respectively , in active regions ) . our set - up , on the other hand , uses existing ca models , does not interfere ( if not wished ) with their evolution rules , does also not change their main results , as shown , but reinterprets them , extends them essentially , and makes them compatible with mhd . the set - up we introduced allows different future applications and posing questions which could not be asked so far in the frame of ca models . in preparation is a study ( isliker et al . 2000b ) to reveal in detail what physical flare scenario the extended ca models imply . we will address the questions : ( 1 ) how to interpret the small scale processes of the models ( loading and bursting ) in terms of mhd ; ( 2 ) what the _ global _ flare - scenario implied by the models is ; ( 3 ) whether the global magnetic field topology of the models can be considered to represent observed magnetic topologies in active regions ; ( 4 ) what spatio - temporal evolution of the electric field during flares is yielded by the models . a different future application we plan with ca models extended with our set - up is the introduction of particles into the models , with the aim to study thermal emission , particle acceleration , and non - thermal emission . this will allow a much deeper comparison of the ca models to observations than was possible so far , and this is actually the most important benefit of the set - up we introduced . such comparisons will allow a new judgment of the adequateness or not of classical ca models ( in their current form ) to the problem of solar flares , beyond the three statistical distributions of the primarily released energy . solar flare ca models which include particle acceleration would represent the first global and complete model for solar flares . we thank k. tsiganis and m. georgoulis for many helpful discussions on several issues . we also thank g. einaudi for stimulating discussions on mhd aspects of flares , and the referee a.l . mackinnon for useful comments . the work of h. isliker was partly supported by a grant of the swiss national science foundation ( nf grant nr . 8220 - 046504 ) . the temporal evolution of the ca models presented in this article is governed by the following rules : * initializing * loading * scanning : create a list of the unstable sites ; if there are none , return to loading ( 1 ) * scanning and bursting : redistribute the fields at the unstable sites which are in the list created in the scannings 2 or 4 * scanning : create a list of the unstable sites . if there are any , go to bursting ( 3 ) , else return to loading ( 1 ) the extra scannings 2 and 4 are needed for causality : if a site becomes unstable through a burst in the neighbourhood , then it should be redistributed in the subsequent scan , and not in the same as the primary unstable site . the same is true for the scanning 4 , since in the next bursting phase ( if any ) only those sites should burst who had become unstable through a burst in their neighbourhood during the foregoing time - step . as a time - step is considered one scanning of the grid , point 3 . the released energy per time - step is the sum of all the energy released by bursts in this time - step ( a burst is considered a single redistribution event in 3 ) . we term a flare or avalanche the loop 3,4 , from the occurring of the first burst in 3 until the activity has died out and one returns via the scanning 4 to loading ( 1 ) . the duration of the flare is the number of time - steps it lasted , the total flare energy is the sum of all the energies released in the duration of the flare , and the peak flux or peak energy is the maximum of the energies of all the time - steps of the flare . we mentioned in sec . 2.2 that other possibilities for continuation of the vector - potential besides spline interpolation would be : a ) continuation of @xmath2 with the help of an equation ; b ) other kinds of interpolation , either locally ( in a neighbourhood ) , or globally ( through the whole grid ) . possibility a ) implies that an equation has to be solved in each time - step ( after each loading and after each burst ) , in the worst case numerically , with open boundary condition and the @xmath4 given at the grid - sites . this computational effort might slow down the algorithm of the model considerably ( and bring it near to the computational effort of mhd equation integration ) . besides that , the problem is what equation to use : to make the magnetic field always a potential field ( i.e. using a corresponding equation for the vector - potential @xmath2 ) implies that , from the point of view of mhd , at all times a very well - behaved magnetic field resides in the ca , with no tendency towards instabilities , which makes it difficult to understand why bursts should occur at all , since critical quantities such as currents do not become excited . a better candidate could be expected to be force - freeness , except that , possibly , one may be confronted with incompatibility of the boundary conditions with the vector - potential values given at the grid - sites , i.e. existence - problems for solutions eventually arise . though definitely possibility a ) can not be ruled out on solid grounds , we found it more promising to proceed with possibility b ) , interpolation . a guide - line for choosing a particular interpolation method is the reasonable demand that the interpolation should not introduce wild oscillations in - between grid - sites , for we want to assure that the derivatives at the grid sites , which are very sensitive to such oscillations , are not random values solely due to the interpolation , but that they reflect more or less directly the change of the primary grid - variable from grid - site to grid - site . this calls for interpolating functions which are as little curved as possible . the easiest and fastest way of interpolating would be to perform local interpolations around a point and its nearest neighbours ( e.g. using low - order polynomials or trigonometric functions of different degrees ) . this interpolation leads , however , to ambiguities for the derivatives : the derivatives , say at a point @xmath3 , are not the same , if the used interpolation is centered at @xmath3 , with the ones calculated with an interpolation centered at e.g. @xmath75 . in this sense , local interpolation is not self - consistent , the derivatives at a grid - site depend on where the used interpolation is centered . finally , we are left with global interpolation through the whole grid . among the candidates are , besides more exotic interpolating functions , polynomials of degree equal to the grid size , trigonometric functions ( also in the form of fourier - transforms ) , low - order smooth polynomials ( e.g. splines ) . the first candidate , polynomials of a high degree @xmath76 ( with @xmath76 the number of grid points in one direction ) , we reject immediately since it is notorious for its strong oscillations in - between grid - sites , mainly towards the edges of the grid . we tried the second candidate , trigonometric interpolation , in the form of discrete fourier transform . testing this by prescribing analytic functions for @xmath77 and comparing the numerical derivatives with the analytic ones , it turned out that there arise problems with representing structures in @xmath2 as large as the entire grid ( the wave - number spectrum is too limited ) , and with structures as short as roughly the grid - spacing ( different prescribed short structures are taken for the same ) . trying cubic spline - interpolation , we found that it does not suffer from the problems stated for the other types of interpolation : neither does it introduce wild oscillations , unmotivated by the values at the grid - sites , nor does spline interpolation have problems with describing large or small scale structures ( if a functional form of @xmath2 is prescribed , then the analytic derivatives and the derivatives yielded by the interpolation give very close values , in general ) . moreover , based on results of sec . 3 , app . c , and isliker et al . ( 1998 ) , there is another reason why spline - interpolation is particularly adequate to our problem : it relates the quantity @xmath49 ( eq . ( 2 ) ) , which measures the stress at a site in the ca model , closely to @xmath78 , the laplacian of @xmath2 ( see app . the latter is related to the current ( @xmath79 ) , which , from the point of view of mhd , can be considered as a measure of stress in the magnetic field configuration . if this relation would not hold , then the redistribution rules ( eqs . ( 4 ) and ( 5 ) ) of the ca would not be interpretable as the diffusion process revealed by isliker et al . ( 1998 ) , and the instability criterion ( eq . 3 ) would not be so closely related to the current ( see sec . 3 and app . we had rejected above ( sec . 1 , sec . 2.2 ) the use of difference expressions to calculate derivatives , stating that differencing is not in the spirit of ca models quite in general , since the nature of ca is truly discrete . we think it worthwhile to make this argument more concrete and to show what problems arise if differencing were used : isliker et al . ( 1998 ) have shown that the classical solar flare ca are not just the discretized form of a differential equations . instead , they describe the time - evolution of a system by rules which express the direct transition from a given initial to a final state which is the asymptotic solution of a simple diffusion equation . the time - step corresponds therewith to the average time needed for smallest scale structures ( structures as large as a neighbourhood ) to diffuse , and the grid - size corresponds to the size of these smallest occurring structures . assuming that the ca models were just discretized differential equations would lead to severe mathematical and physical contradictions and inconsistencies ( continuity for @xmath80 is violated ( with @xmath47 the grid - size ) , and negative diffusivities appear ) . therewith , in order to be consistent with the evolution rules , which assume a finite grid - size , one can not assume for the purpose of differentiating this same grid - size to be approximately infinitesimal . there are several equivalent ways to define numerical derivatives with the use of difference expressions : there are e.g. the backward difference @xmath81 , and the forward difference @xmath82 . both should give comparable values in a given application , else , in the context of differential equation integration , one would have to make the resolution higher . in the case of ca - models , we find that the two difference expressions yield values which differ substantially from each other : e.g. for an initial loading of the grid with independent random values for the @xmath2-field , the difference between the backward and the forward difference expression can be as large as the field itself . such an initial condition would of course not make sense in the context of partial differential equations , in the context of ca , however , it is a reasonable starting configuration , and the evolution is unaffected by such an initial loading . moreover , when the ca models we discuss in this article have reached the soc state , then the differences between e.g. the backward- and forward - difference expressions can be as large as 400% . there is no way to reduce this discrepancy , since grid - refinement is principally impossible for ca : the evolution is governed by a set of rules , and making the grid spacing smaller by introducing new grid - points in - between the old ones would actually just mean to make the grid larger , since the evolution rules remain the same , there are no rules for half the grid - spacing . the stress measure of lh91 , @xmath84 , can be related to continuous expressions by representing the values of @xmath85 as taylor - series expansions around @xmath3 , setting the spatial differences to @xmath86 . it turns out that e.g. @xmath87 and so on for the other two components . _ in general , _ it is therefore _ not _ adequate to consider @xmath49 to be a good 4th order approximation to @xmath88 , since higher order corrections can be large , they depend on the way the vector potential is continued in - between grid - sites . if we had , for instance , chosen global polynomial interpolation instead of spline - interpolation , the higher order terms would not be negligible , above all towards the edges of the grid , since polynomial interpolation is known for introducing fluctuations near the edges of the grid . consequently , @xmath49 would be a bad approximation to @xmath88 . in order @xmath49 to be a good approximation to @xmath88 , interpolation with , for example , 3rd order polynomials would be an optimum choice ( @xmath49 would be an exact approximation to @xmath88 ) . thus , 3rd order polynomials would be the choice for local interpolation , which , however , is not applicable , since it introduces discontinuities in @xmath1 and @xmath19 ( see app . the way out of the dilemma we suggested in this article is the use of cubic splines , which provide global interpolation with 3rd order polynomials , with @xmath1 and @xmath19 continuous , and only third order derivatives are discontinuous ( this is the price of the compromise ) . for splines then , eq . ( c.1 ) writes as @xmath89,\end{aligned}\ ] ] due to the discontinuities in the 3rd order derivatives ( the superscripts @xmath90 and @xmath91 refer to the right and left derivative , respectively ) . thus , in case where the third order right and left derivatives are not too different , @xmath49 is a good approximation to @xmath88 . isliker , h. , anastasiadis , a. , vlahos , l. , 2000a , a solar flare model in between mhd and cellular automaton , in the fourth astronomical conference of the hellenic astronomical society , eds . seimenis , j. et al .

lif is a ionic solid that is particularly transparent to short wavelength radiation due to its large band gap and hence is commonly used in optics for high - pressure and temperature experiments , such as those related to the development of pulsed power @xcite . lif is also a component in molten salts frequently employed as high - temperature thermal fluids . estimates of the transport properties of lif are important to both these applications . specifically , in dynamic high - pressure experiments , a lif window maintains the pressure at the sample interface where velocimetry measurements are typically made . due to the extreme conditions , the necessary transmission properties are difficult to measure directly . in these experiments , a shock or a near - shock ramp compression with pressures up to 800 gpa @xcite is generated by a variety of means and the material response is measured using velocity interferometry , see , _ e.g. _ , refs .. there are many efforts concentrating on estimating the optical properties of lif crucial to this measurement , see , _ e.g. _ , refs .. due to the short but finite timescale of dynamic material experiments , the thermal conductivity of lif windows can significantly affect the temperature measured at the sample interface . in this work , we focus on calculating this thermal conductivity at the extreme conditions relevant to these experiments with the goal of understanding the nonequilibrium energy transfer that governs their behavior . the material properties of lif have been explored in experiments and simulation primarily nearer to ambient conditions . for instance , thacher @xcite measured the sound velocity and thermal conductivity of lif at temperatures less than 100 k. at ambient pressure , [ thacher1967effect ] shows that the thermal conductivity of lif peaks at about 2 w / m - k near 20 k , where the quantum increase of heat capacity begins to be dominated by the decrease in conductivity due to umklapp scattering . andersson and bckstrm @xcite were able to measure the heat capacity and thermal conductivity of lif at room temperature up to pressures of 1 gpa . they showed a linear dependence of thermal conductivity on pressure and measured a conductivity value of 16.3 w / mk at 1 gpa . phase and other transitions can complicate measurements at higher temperatures and pressures . given that the full phase diagram for lif is not known , smirnov @xcite computed an _ ab initio _ phase diagram of lif over pressures ranging from 0 to 500 gpa and temperatures ranging from 0 to 12000 k together with the elastic properties and associated debye temperatures . smirov @xcite predicted that the structure of lif is the nacl - like arrangement b1 at low temperature and pressure but transitions to the cscl - like arrangement b2 at higher pressures and temperatures . smirov correlated his results with experimental data by kormer @xcite and boehler _ et al . _ ( see root _ et al . _ @xcite for a similar study of mgo where _ ab initio _ molecular dynamics and quantum monte carlo methods were also employed to predict a phase diagram at extreme conditions . ) boehler _ et al . _ @xcite studied the high pressure melting regime of lif with diamond anvil experiments and classical molecular dynamics ( md ) fitted to properties at standard temperature and pressure . et al . _ [ clerouin2005ab ] used _ ab initio _ dynamics to estimate the optical properties of lif along the shock hugoniot where it transitions from a transparent solid to a reflective plasma . et al . _ @xcite investigated lif melting with md using a potential tuned with _ ab initio _ data and compare to existing diamond anvil and shock experimental data . in particular , belonoshko _ et al . _ showed that density as a function of pressure and the radial distribution function computed with their potential compares well with trusted data . they also make clear the distinction between thermal instability and melting especially for small systems at high pressures using a phase coexistence method . also using classical md , young @xcite studied ion damage of lif crystals . related to thermal properties of lif , nsslein and schrder @xcite calculated the dispersion and phonon density of states ( phdos ) via polarizable model of the inter - atomic interactions of lif at 0 k. dolling _ et al . _ @xcite also calculated the phdos of lif via lattice dynamics and compared it to dispersion data derived from slow neutron inelastic scattering . in their work , the crystal has phonon content up to 20 thz with most of the low frequency content attributed to the f ion . recently , stegailov @xcite calculated the phdos with density functional theory with the generalized gradient approximation and showed the onset of mechanical instability , which may lead to defect formation or melting , due to hot electrons when the electron temperature reaches 37,000 k. following this body of work , in this paper we use md together with the green - kubo ( gk ) formalism @xcite to estimate the thermal conductivity of lif at stresses on par with the elastic moduli and temperatures in excess of the melt temperature at ambient conditions . in particular , we investigate both volumetric and uniaxial deformation modes similar to ( non - hugniot ) ramp compression experiments . ( [ spataru2015ab ] makes a corresponding study of the optical properties of lif . ) generally speaking , classical md is well - suited to simulating ionic solids since essentially all heat propagates via phonons . the related publications , refs . , are the few examples of estimating the thermal conductivity of alkalis with classical md and typically focus on the thermal properties of molten salts with applications to high temperature thermal transfer fluids . the strong ionic character of lif leads to the usual complications due to long range coulomb interaction , requiring dipole corrections and large cell sizes ; but the main issue is that md is highly reliant on empirical potentials . there has been some work on suitable potentials for ionic solids like lif , typically of the buckingham @xcite or born / tosi - fumi @xcite forms . a high quality potential parameterization for lif of another form was developed by ishii _ et al . _ @xcite but focussed on the properties of molten mixtures , see also refs .. on the other - hand , belonoshko _ et al . _ @xcite carefully constructed a tosi - fumi / born - mayer - huggins potential to suit high - pressure and temperature conditions by dropping the unstable terms in the full tosi - fumi form , and compared its behavior to literature and their own density functional theory ( dft ) results . given the findings in [ belonoshko2000born ] , it was evident that the md potential may not transition to the most stable phase with changes in pressure and temperature but instead becomes stuck in a metastable state . we used this fact together with a phase diagram independently calculated with dft to estimate thermal conductivity over pressures ranging from 100 to 400 gpa and temperatures ranging from 1000 to 4000 k. to compute the phase diagram , we follow smirnov s work @xcite and others @xcite and use plane augmented wave ( paw ) dft with the local density approximation ( lda ) instead of the linear muffin tin orbital method smirnov employed to estimate the zero temperature enthalpy and entropy of the phonon population . from a dynamical matrix calculated with dft , we are able to estimate the entropy component of the free energy with a quasi - harmonic model . the range of the free energy estimated with the quasi - harmonic model limited by the mechanical stability which we also estimated with the _ ab initio _ bond stiffnesses governing the phonon propagation . in addition , we use the _ ab initio _ phdos to validate and recalibrate the belonoshko parameterization for thermal conductivity estimates . given a definition of the heat flux @xmath0 , the thermal conductivity tensor @xmath1 can be obtained from the green - kubo formula dependent on the time - correlation of @xmath0 with itself : @xmath2 where @xmath3 is the system volume , @xmath4 is the temperature , @xmath5 is the boltzmann constant . the bracket @xmath6 denotes the appropriate ensemble average , where it is important to note that @xmath7 in equilibrium . a microscopic formula @xcite for the heat flux @xmath0 is @xmath8 where the per - atom energy @xmath9 is formed from the kinetic energy of the atom @xmath10 and a reasonable partition of the total potential energy comprised of short - range bonds and long - range coulomb interactions to individual atoms @xcite , and the virial stress @xmath11 for atom @xmath10 in terms of fundamental quantities ( which is given in app.[app : virial ] ) . classical molecular dynamics ( md ) provides the necessary positions @xmath12 , velocities @xmath13 , and forces @xmath14 , from the trajectories obtained by integrating newton s equations of motion , @xmath15 , given an initial configuration @xmath16 and atomic masses @xmath17 . the total force @xmath18 on an atom @xmath10 is the sum of interatomic forces derived from an empirical potential @xmath19 . for ionic solids like lif , explicit charges @xmath20 are typically constant and located at ion cores . a widely - used potential for ionic materials is the tosi - fumi / born - mayer - huggins ( tf / bmh ) potential @xcite,@xcite . it is a combination of long - range coulomb and short - range @xmath21 ( repulsion ) pair - wise interactions @xmath22 for species @xmath23,@xmath24 with associated groups of atoms @xmath25,@xmath26 ; inter - atomic distance @xmath27 ; and charge @xmath28 for @xmath29 . here , @xmath30 is the ( vacuum ) permittivity and @xmath31 is the elementary charge . of the empirical parameters : @xmath32 , @xmath33 , @xmath34 and @xmath35 , the last two are related to dipole interactions and are set to zero for high pressure stability considerations by belonoshko _ et al . _ @xcite in their model of lif . the periodic images participating in the coulomb forces on the atoms in the system extend well beyond the explicitly represented periodic box . for efficiency , the energy @xmath36 is decomposed into long ( _ reciprocal _ @xmath37-space ) and short ( real @xmath38-space ) components . this decomposition is the essence of ewald summation and the particle - particle particle - mesh ( pppm ) method @xcite ( see app.[app : virial ] for more details ) . to validate an empirical potential for calculation of thermal conductivity ideally all the properties related to the phonon population and propagation would be compared with experimental or _ ab initio _ data . in lieu of a full comparison of the dispersion relationship for harmonic waves and related properties for anaharmonic interactions , we follow others in comparing the elastic constants and phonon density of states related to the phonon dispersion and wave speeds . as derived from the dispersion relationship of the material , the phonon density of states ( phdos ) is linked to the thermal conductivity of the material . the dispersion relationship is determined by the matrix of bond stiffnesses @xmath39 , which is composed of sub - matrices of linearized force constants : @xmath40_{\alpha\beta } = \left . \frac{\partial^2 \phi}{\partial { \mathbf{x}}_\alpha \partial { \mathbf{x}}_\beta } \right|_{{\mathbf{x}}_\alpha={\mathbf{y}}_\alpha } = - \left . \frac{\partial { \mathbf{f}}_\alpha}{\partial { \mathbf{x}}_\beta } \right|_{{\mathbf{x}}_\alpha={\mathbf{y}}_\alpha}\ ] ] referenced to a given lattice configuration @xmath41 @xcite . the dynamical matrix , a fourier transform of @xmath39 , results from applying a plane wave ansatz for the motion of the atoms about lattice positions @xmath41 : @xmath42_{\alpha\beta } = \sum_{\boldsymbol{\ell } } \frac{1}{\sqrt{m_\alpha m_\beta } } { \mathbb{k}}_{\alpha\beta } \exp\left(-\imath { \mathbf{k}}\cdot \left ( { \mathbf{x}}_{\alpha } - { \mathbf{x}}_{\beta } - \boldsymbol{\ell } \right ) \right ) \ , \ ] ] where @xmath43 ranges across all periodic images of the unit cell including the original one . the dynamical matrix determines the eigenvalues @xmath44 for a given propagation direction ( wave vector ) @xmath37 and polarization . the phdos is constructed by sampling the eigenvalues @xmath44 of eq.([eq : dmat ] ) throughout the brillouin zone . the same procedure can be used in the context of an _ ab initio _ density functional model of the material where the forces @xmath14 are the hellman - feynman forces . the dynamical matrix also determines the ( linear ) phonon and long - wavelength elastic stability . the elastic moduli tensor @xmath45_{iajb } = \left[\frac{\partial^2 \phi}{\partial { { \mathbf{f } } } \partial { { \mathbf{f } } } } \right]_{iajb } = \frac{1}{{\mathrm{v}}_0 } \left [ \sum_{\alpha,\beta } \frac{\partial^2 \phi } { \partial { { \mathbf{x}}_\alpha } \partial { { \mathbf{x}}_\beta } } : \frac{\partial { \mathbf{x}}_\alpha}{\partial { \mathbf{f } } } \frac{\partial { \mathbf{x}}_\beta}{\partial { \mathbf{f } } } \right]_{iajb } = \frac{1}{{\mathrm{v}}_0 } \sum_{\alpha,\beta } \left [ \left [ { \mathbb{k}}\right]_{\alpha \beta } \right]_{ij } \left [ { \mathbf{x}}_{\alpha } \right]_a \left [ { \mathbf{x}}_{\beta } \right]_b\ ] ] is related to the tensor of bond stiffnesses @xmath39 and determines the stability in the continuum limit . here , @xmath46 is the deformation gradient , @xmath47 are the stress - free lattice sites , and @xmath48 . see app.[app : stability ] for more details . lif can change phase over a range of temperatures and pressures . to determine the relatively stable phase as a function of temperature @xmath4 and stress @xmath49 , estimates of the gibbs free energy @xmath50 : @xmath51 a legendre transform of helmholtz free energy @xmath52 , needs to be calculated for both b1 and b2 structures . here , @xmath53 is the first piola - kirchhoff stress . assuming positive frequencies @xmath54 , the helmholtz free energy @xmath52 is commonly estimated with a quasi - harmonic ( qh ) model : @xmath55 which is based on the partition function of independent oscillators ( see _ e.g._[dove1993introduction](chap . this model is composed of two zero temperature components : ( a ) the cohesive energy ( referenced to an infinitely dispersed state ) which can be equated with @xmath56 , and ( b ) the ( non - classical ) zero point / ground state energy of the phonons @xmath57 ; together with a third term : the temperature - entropy product approximated by the harmonic oscillator model . given the wide band - gap of lif , we neglect the thermal electron contribution in this approximate model . clearly , an equation of state ( eos ) @xmath58 is necessary to transform the helmholtz free energy @xmath52 to the gibbs free energy @xmath50 . at zero temperature the data needed to construct an accurate eos can be calculated with dft . the change of stress with temperature can be estimated with the qh model eq.([eq : qh ] ) @xcite or from md simulations . assuming a first order dependence of stress on temperature , we can use @xmath59 to form the necessary inverse @xmath60 . since the systems of interest are cubic , the thermal expansion tensor @xmath61 , and hence only one coefficient @xmath62 needs to be determined to effect thermal expansion . as discussed in the introduction , we have based this study on the potential by belonoshko _ et al . _ this potential was specifically parameterized for high pressure states where the @xmath34 and @xmath35 parameters of the tf / bmh potential in eq.([eq : born ] ) , which cause instability , are set to zero . since belonoshko _ et al . _ @xcite were primarily concerned with investigating phase diagram and mechanical properties and we are employing the potential to estimate thermal conductivity , we compared the phdos resulting from the tf / bmh potential with the belonoshko _ et al . _ parameters to that from an _ ab initio _ calculation as a measure of the validity of phonon transmission . for the dft calculations , we employed the local density approximation ( lda ) , a plane augmented wave basis with cutoff 800 ev with standard pseudo - potentials @xcite , and 20@xmath6320@xmath6320 k - point monkhorst - pack grid , which were arrived at via convergence studies for the dynamical matrix and elastic moduli . fig.[fig : pdos ] shows that the phdos ( calculated via the dft code vasp @xcite and the phonopy package @xcite ) is quite sensitive to compression and hence pressure . the presence of negative frequencies in the phdos of compressed b2 structures also indicates that the b2 phase is unstable for lattice constant @xmath64 2.1 ( and number density @xmath65 0.215 @xmath66 ) . we compared the phdos for simulations with 2@xmath632@xmath632 , 3@xmath633@xmath633 , and 4@xmath634@xmath634 unit cells with the correction based on born effective charges @xcite and found results essentially indistinguishable and hence we employed systems with 2@xmath632@xmath632 for the remainder of the calculations . for the comparison of the phdos derived from the belonoshko potential and that from dft , we picked the compressed b1 configuration with @xmath23= 3.285 ( corresponding to 2000 k , 200 gpa lattice constant based on the belonoshko parameterization ) as representative of our pressure - temperature region of interest . given the poor match shown in fig.[fig : pdos_comparison ] , we re - tuned the potential to achieve a better correspondence , particularly of the peaks in the phdos , which is also shown in fig.[fig : pdos_comparison ] . note that only changing li - f well depth resulted in stable modifications of the crystal that maintained a reasonably representative lattice constant : b1 : 4.051 ( original ) _ vs. _ 4.206 ( modified ) , and b2 : 2.514 ( original ) _ vs. _ 2.588 ( modified ) , at zero temperature . the resulting and original parameters are given in table[tab : parameters ] . l | r r | r r & a [ ev ] & b [ @xmath67 & a [ ev ] & b [ @xmath67 + li - li & 98.933 & 3.3445 & `` & '' + li - f & 401.319 & 3.6900 & 521.714 & `` + li - li & 420.463 & 3.3445 & '' & " + regarding the possibility that lif can have b1 , b2 , and liquid phases over the pressure temperature range of interest and these structural changes can influence the thermal conductivity , we adopted the approach to : ( a ) use dft to predict the appropriate phase for given stress and temperatures conditions , and ( b ) use this phase to initialize the md simulation which generally stays in this phase even if it is only meta - stable with respect to the empirical potential . we did observe some deviations from this assumption , including defect formation and melting , which are noted in the results section . to this end , the qh model derived from the same dynamical matrix used to generate phdos was employed to estimate the relative free energy @xmath68 and thus the thermodynamic stability of the b2 phase relative to b1 . this model is built upon direct _ ab initio _ estimates of the zero temperature enthalpy and limited in its range of validity by the mechanical stability of the phonon population at each particular deformation state . to construct the b1b2 phase diagram , first we constructed an equation of state . we interpolated the function @xmath69 for b1 and b2 directly from dft data and estimated the thermal expansion coefficient @xmath62 from md data ( as opposed to from the qh model ) due to its full representation of anharmonic effects and its good correlation with measured values . fig.[fig : pressure_density ] shows the relevant pressure versus density curves for a range of temperatures . clearly , the finite - temperature md pressure curves are offset from the zero - temperature dft data so that positive thermal expansion coefficients are obtained and the bulk modulus for the md model are effectively the same as for the dft ; however , the zero - temperature equilibrium lattice constants do differ slightly . our estimates of the expansion coefficient @xmath62 employed in eq.([eq : thermal_expansion ] ) are : b1 : 0.008056 gpa / k , b2 : 0.01164 gpa / k , for the original belonoshko parameterization , and b1 : 0.007411 gpa / k , b2 : 0.01082 gpa / k , for the modified parameterization . for reference , the measured coefficient of thermal expansion ( @xmath62 divided by the bulk modulus ) is 37 @xmath63 10@xmath70 /k @xcite at ambient conditions which corresponds to our estimate , 30 @xmath63 10@xmath70 /k , for the unmodified b1 potential . also apparent is the fact that the b1 and b2 phase have similar mechanical responses with b2 being slightly , but distinctly softer than b1 at the same ( number ) density @xmath71 . next , we ascertained the mechanical stability of the b1 and b2 phases through the _ ab initio _ estimates of the elastic moduli and phdos . fig.[fig : stability ] shows the pressure and elastic moduli calculated from the dft data , and the derived stability moduli ( see app.[app : stability ] for details ) . the results : ( a ) b1 is stable over the high pressure range we consider , and ( b ) b2 is only conditionally stable ( @xmath72 2.15 , @xmath73 0.2 atoms / @xmath66 ) based on linearized , long - wavelength elastic stability considerations , are comparable to the findings in [ smirnov2011abinitio](fig . 2 ) . coincide with those derived in app.[app : stability ] . ] examining phonon spectrum corresponding to the phdos data in fig.[fig : pdos ] gives a more detailed account of stability since each mode can be examined independently ( see app.[app : stability ] for a discussion of the connection between the two stability criteria ) . from the phdos data , b2 is apparently stable for @xmath72 2.1 , which corresponds approximately to pressure @xmath74 120 - 130 gpa for the temperatures we consider . ( coincidently , belonoshko @xcite speculates that a b1b2 transition occurs at approximately 130 gpa , which is in the neighborhood of @xmath71 = 0.2 atoms / @xmath66 given fig.[fig : pressure_density ] . ) finally , we evaluated the free energy difference . the zero - temperature energy ( enthalpy ) differences between b2 and b1 shown in fig.[fig : enthalpy ] display trends similar those shown in [ smirnov2011abinitio](fig.1 ) . using the qh model eq.([eq : qh ] ) , we calculate the zero point energy difference @xmath75 , omitted in [ smirnov2011abinitio ] , to be nearly uniformly 0.02 ev / atom over the pressure range we examined , and , hence , has no significant effect on the resulting b1-b2 phase separator . in fact , the change in zero - temperature enthalpy difference between the two phases over the given pressure range is dominated by the pressure - volume work . unlike [ smirnov2011abinitio](fig.6 ) which shows a transition to b2 at temperatures and pressures as low as 1500 k and 150 gpa , we estimate that @xmath76 0.1 ev over the given @xmath4 and @xmath77 range so b1 is always relatively thermodynamically stable . the contours of @xmath78 resemble the slope of b1-b2 separator in [ smirnov2011abinitio](fig.6 ) , and the elastic moduli as a function of pressure and phase are similiar . the qh model of the free - energy difference is arguably better than debye model tuned by linear muffin tin data employed in [ smirnov2011abinitio ] since the qh model does not make assumptions about form of the dispersion relation ; however , its validity at these temperatures is suspect . as fig.[fig : thermal_displacement ] shows , the thermal displacements predicted by the qh model are large , but still much smaller than those given by md with similar elastic properties . this data gives credence to the notion of thermal stabilization of apparently mechanically unstable phases . in this form of non - linear stability , at high temperatures atoms primarily reside outside the zero - temperature minimum state which may be mechanically unstable and in nearby regions of the energy surface with positive curvature @xcite . after these validation and phase determination procedures , we thermalized and pressurized lif lattices with a nos - hoover thermo - barostat ( using the classical md code lammps @xcite ) in order to obtain the equilibrium flux correlations necessary to estimate thermal conductivity . after equilibration , we used 10 replica systems with initial conditions selected from the constant temperature - pressure equilibration simulations of constant energy dynamics to compute the average flux correlation . after transients due to the relaxation from constant temperature dynamics subsided , samples of the correlations contributing to the average @xmath79 were collected every ten 0.5 fs time - steps as a last preliminary , given the spatial decomposition employed by pppm , we checked for finite size effects in the flux correlations . fig.[fig : size ] shows that they are negligible with respect to the inherent noise even for periodic systems as small as 4@xmath634@xmath634 unit cells ; hence , in the following studies we used 4@xmath634@xmath634 systems and a real space/@xmath37-space decomposition cutoff @xmath80 nm for the pppm electrostatic solver . ) . this extrapolation is parallel but not coincident with the dft data ( black ) which implies that the md and dft models of lif have similar elastic elastic properties but different zero - temperature equilibrium lattice constants . the inset shows corresponding data for the modified md potential . for reference , the rose - vinet fits to the dft data are : b1 : @xmath81= 83.93 gpa , @xmath82= 4.594 gpa , @xmath23= 3.905 , and b2 : @xmath81= 78.42 gpa , @xmath82= 4.818 gpa , @xmath23= 2.464 , where @xmath81 is the zero - temperature bulk modulus , and @xmath23 is the corresponding lattice constant . for reference , the measured , ambient lattice constant is 4.03 @xcite ; , title="fig : " ] using the green - kubo ( gk ) method described in sec . [ sec : theory ] and the preliminaries given in sec . [ sec : method ] , we compute the thermal conductivity for compressed states in two studies for different deformations of the lattice : ( a ) volumetric compression over a range of pressures @xmath77 = 100400 gpa and temperatures @xmath4= 10004000 k , and ( b ) uniaxial compression with normal stress 150 gpa to simulate conditions at the initiation of a ramp experiment . since , at ambient pressure the measured debye temperature for lif is 732 k ( * ? ? ? * table 12.1 ) and the melt is temperature 1121 k @xcite , our conditions are well within the classical regime and some of the states may melt . in preliminary studies , we found the difference in the estimated conductivity with the modified _ vs. _ the original belonoshko parameters was at most 10% over the range of interest and typically only 3% . since these differences were comparable to our error estimated from 10 replicas , we report conductivities derived from the original parameters . we attribute these small differences between parameterizations with distinct phdos , and hence dispersion characteristics , to the observation that low frequency / long wavelength phonons carry most of the heat and in that range the phdos of the two parameterizations agree fairly well . in fact , g. chen and co - authors refs . showed that 90% of the heat in si at ambient conditions is carried by phonons with frequencies less than about 2 thz ( estimated from the reported 2 - 5 nm wavelength and the given elastic moduli ) . first , we compared the conductivity estimated with md gk and a method directly employing _ ab initio _ data for b1 lif at constant volume ( @xmath23= 3.2 ) over the temperature range @xmath4=10004000 k. specifically , in the second method the boltzmann transport equation ( bte ) was parameterized with _ ab initio _ second and third order force constants derived from 2@xmath632@xmath632 , and 4@xmath634@xmath634 unit cell systems respectively and solved in the single mode relaxation time approximation . see refs . for details , and the similar approach in [ li2014shengbte ] . given the differences in the methods , the results shown in fig.[fig : abinitio_comparison ] are comparable . the md estimates are uniformly lower than those of the bte , which is consistent with fact that the md has a complete , albeit less exact , hamiltonian with no truncation of the phonon scattering interactions and the temperature is high enough for higher order and non - linear mechanisms beyond those captured by a single relaxation time to be significant . also noteworthy , the thermal conductivity derived from the bte model displays perfect @xmath83 scaling , whereas the md estimates show slightly stronger decay with temperature . fig.[fig : vol_cond ] shows the thermal conductivity estimated with md gk for pressures in the range 1400 gpa and temperatures 10004000 k , and table[tab : kappa ] gives the corresponding data for both the b1 and b2 phases . in fig.[fig : vol_cond ] the phase of the samples used to create the contour plot are marked and a few of the high temperature , relatively low pressure systems melted . as can be seen in table[tab : kappa ] , the estimated thermal conductivity for the b1 and b2 phases have comparable values and same trends . this is plausible given that the elastic properties of the two phases are similar and a simple kinetic model of thermal transport indicates that the resulting comparable sound speeds should lead to similar conductivities . the same basic kinetic interpretation is consistent with the observations that the thermal conductivity increases with increased pressure due to higher wave speeds and with lower temperature due to relatively less scattering and longer phonon mean free path . these trends are monotonic and have decreasing effect on the thermal conductivity . although the belonoshko potential was tuned to high pressure conditions , we also calculated the thermal conductivity nearer ambient conditions . the values we obtain , _ e.g._2.8@xmath840.2 w / mk at 1 gpa , 1000 k and 1.8@xmath840.2 w / mk at 1 atm , 1200 k ( melt , refer to table[tab : kappa ] ) are roughly comparable to the value 1.5 w / mk at 1 atm , 1150 k ( melt ) given by ishi _ et al . _ @xcite using a different empirical potential and the experimental measurements : 15.7 w / mk at 0.1 gpa , 16.3 w / mk at 1.0 gpa , 300 k @xcite , and 4.0 w / mk at 1 atm , 314 k @xcite of solid lif . for this study , we compressed one direction of a b1 lif crystal while fixing the lateral dimension to a 4.02 lattice spacing to mimic initial phases of ramp compression with inertial confinement and examine the resulting differences in the thermal conductivity resulting from unequal principal strains . the compression direction was chosen to be @xmath85 , since this direction lacks surface polarization . the lateral directions were @xmath86 ( equivalent to @xmath85 ) and @xmath87 , respectively . the compressions @xmath88 examined corresponded to normal stresses 150 gpa in the compressed dimension ( note that 75 gpa md crystal was unstable and a dislocation formed ) and temperature range 10003000 k. it was not possible to predict which phases were thermodynamically stable over this deformation temperature range since we predicted that b2 has unstable phonons over the range we examined . the phdos for b1 , fig.[fig : uniaxial_pdos ] , shows that the compressed direction becomes stiffer ( the sound speed is roughly inversely proportional to slope of phdos ) and higher frequency content is added to the phonon spectrum . fig.[fig : uniaxial_anisotropy ] shows stress response to these conditions and corresponding anisotropy of the thermal conductivity of the b1 structure . note that the initial lattice constant is not equilibrium at the given temperatures which immediately induces the anisotropy shown . also the lateral stresses become nearly equal but distinct from the normal stress in the compressed direction as the structure loses perfect crystallinity . fig.[fig : uniaxial_conductivity ] shows that the normal conductivity follows similar trends with temperature and pressure as in the volumetric compression case , namely in this state the highest conductivities are at the highest pressures and lowest temperatures . the data for this study is tabulated in table[tab : uniaxial_kappa ] . in summary , we found that the thermal conductivity of lif at high temperatures and pressures is only marginally dependent on phase and ranged from about 5 w / mk to 70 w / mk over the range 10004000 k and 100400 gpa . for our purposes , the fact that the two expected phases ( b1 and b2 ) have similar conductivity offsets the difficulties in determining their mechanical and thermodynamic stability . our estimates are corroborated by the limited experimental data available as well as direct _ ab initio _ estimates of thermal conductivity . we also found that the uniaxial deformation expected to result from inertia confinement of the targeted ramp compression experiments may lead to significant anisotropy in the thermal conductivity . more rigorous treatment of the relative phase stability exist in the literature than the method we selected , notably [ monserrat2013anharmonic ] which focussed on the influence of the anaharmonic phonon energy and [ desjarlais2013first ] which adapts the phase coexistence technique of [ lin2003two ] to finite temperature dft calculations , which may shed light on phase transitions from b1 at high temperatures and pressures . since our findings indicate that these transitions are unlikely over our temperature and pressure range of interest , whereas the formation of defects appear at relatively low uniaxial compression we intend to pursue investigation of the influence of defects on the thermal conductivity of solid lif next . we thank luke shulenberger , catalin spataru and thomas mattsson for helpful guidance and appreciate the use of lammps @xcite , vasp @xcite , and phonopy @xcite . this work was supported by the nnsa advanced simulation and computing - physics and engineering models program at sandia national laboratories . sandia is a multiprogram laboratory managed and operated by sandia corporation , a wholly owned subsidiary of lockheed martin corporation , for the u.s . department of energy s national nuclear security administration under contract no . de - ac04 - 94al85000 . 10 sandia national laboratories , http://www.sandia.gov / z - machine/. , 2015 . de fratanduono , tr boehly , ma barrios , dd meyerhofer , jh eggert , rf smith , dg hicks , pm celliers , dg braun , and gw collins . refractive index of lithium fluoride ramp compressed to 800 gpa . , 109(12):123521 , 2011 . pa rigg and ym gupta . real - time x - ray diffraction to examine elastic plastic deformation in shocked lithium fluoride crystals . , 73(12):16551657 , 1998 . md knudson , dl hanson , je bailey , ca hall , jr asay , and c deeney . principal hugoniot , reverberating wave , and mechanical reshock measurements of liquid deuterium to 400 gpa using plate impact techniques . , 69(14):144209 , 2004 . bm lalone , ov fatanov , jr asay , and ym gupta . velocity correction and refractive index changes for [ 100 ] lithium fluoride optical windows under shock compression , recompression , and unloading . , 103(9):093505 , 2008 . dg hicks , tr boehly , pm celliers , jh eggert , sj moon , dd meyerhofer , and gw collins . laser - driven single shock compression of fluid deuterium from 45 to 220 gpa . , 79(1):014112 , 2009 . bj jensen , db holtkamp , pa rigg , and dh dolan . accuracy limits and window corrections for photon doppler velocimetry . , 101(1):013523 , 2007 . pa rigg , md knudson , rj scharff , and rs hixson . determining the refractive index of shocked [ 100 ] lithium fluoride to the limit of transmissibility . , 116(3):033515 , 2014 . catalin d spataru , luke shulenburger , and lorin x benedict . ab initio many - body green s function calculations of optical properties of lif at high pressures . , 92(24):245117 , 2015 . philip d thacher . effect of boundaries and isotopes on the thermal conductivity of lif . , 156(3):975 , 1967 . s andersson and g backstrom . thermal conductivity and heat capacity of single - crystal lif and caf@xmath89 under hydrostatic pressure . , 20(35):5951 , 1987 . n. a. smirnov . ab initio _ calculations of the thermodynamic properties of lif crystal . , 83:014109 , 2011 . sb kormer . optical study of the characteristics of shock - compressed condensed dielectrics . , 11(2):229254 , 1968 . reinhard boehler , marvin ross , and david b boercker . melting of lif and nacl to 1 mbar : systematics of ionic solids at extreme conditions . , 78(24):4589 , 1997 . seth root , luke shulenburger , raymond w lemke , daniel h dolan , thomas r mattsson , and michael p desjarlais . shock response and phase transitions of mgo at planetary impact conditions . , 115(19):198501 , 2015 . j clerouin , y laudernet , v recoules , and s mazevet . study of the optical properties of shocked lif . , 72(15):155122 , 2005 . a. b. belonoshko , r. ahuja , and b. johansson . molecular dynamics of lif melting . , 61(18):1192811935 , 2000 . d. a. young . molecular dynamics simulation of swift ion damage in lithium fluoride . , 225:231240 , 2004 . v nsslein and u schrder . calculations of dispersion curves and specific heat for lif and nacl using the breathing shell model . , 21(1):309314 , 1967 . g dolling , hg smith , rm nicklow , pr r vijayaraghavan , and mk wilkinson . lattice dynamics of lithium fluoride . , 168(3):970 , 1968 . v. v. stegailov . stability of lif crystal in the warm dense matter state . , 50(1):3134 , 2010 . l. onsager . reciprocal relations in irreversible processes . i. , 37(4):405426 , 1931 . l. onsager . reciprocal relations in irreversible processes . ii . , 38(12):22652279 , 1931 . m. s. green . markoff random processes and the statistical mechanics of time - dependent phenomena . 2 . irreversible processes in fluids . , 22:398413 , 1954 . r. kubo . statistical - mechanical theory of irreversible processes . i. general theory and simple applications to magnetic and conduction problems . , 12:570586 , 1957 . r. kubo , m. yokota , and s. nakajima . statistical - mechanical theory of irreversible processes . ii . response to thermal disturbance . , 12:12031211 , 1957 . r. zwanzig . elementary derivations of time - correlation formulas for transport coefficients . , 40:25272533 , 1964 . g ciccotti , g jacucci , and ir mcdonald . transport properties of molten alkali halides . , 13(1):426 , 1976 . p. j. d. lindan and m. j. gillan . a molecular dynamics study of the thermal conductivity of caf@xmath89 and uo@xmath89 . , 3:39293939 , 1991 . n. galamaba and c. a. nieto de castro . thermal conductivity of molten alkali halides from equilibrium molecular dynamics simulations . , 120(18):86768682 , 2004 . y. ishii , k. sato , m. salanne , p. a. madden , and n. ohtori . thermal conductivity of molten alkali metal fluorides ( lif , naf , kf ) and their mixtures . , 118:33853391 , 2014 . norikazu ohtori , mathieu salanne , and paul a madden . calculations of the thermal conductivities of ionic materials by simulation with polarizable interaction potentials . , 130(10):104507 , 2009 . mathieu salanne , dario marrocchelli , cline merlet , norikazu ohtori , and paul a madden . thermal conductivity of ionic systems from equilibrium molecular dynamics . , 23(10):102101 , 2011 . yoshiki ishii , keisuke sato , mathieu salanne , paul a madden , and norikazu ohtori . thermal conductivity of molten alkali metal fluorides ( lif , naf , kf ) and their mixtures . 118(12):33853391 , 2014 . y. cherednikov , n. a. inogamov , and h. m. urbassek . influence of defects on extreme ultraviolet laser ablation of lif . , 88:134109 , 2013 . m. tosi . cohesive energy of the alkali halide crystals . , 24(7):965967 , 1963 . mp tosi and fg fumi . ionic sizes and born repulsive parameters in the nacl - type alkali halides - ii : the generalized huggins - mayer form . , 25(1):4552 , 1964 . g jacucci , ir mcdonald , and a rahman . effects of polarization on equilibrium and dynamic properties of ionic systems . , 13(4):1581 , 1976 . alfredo a correa , lorin x benedict , david a young , eric schwegler , and stanimir a bonev . first - principles multiphase equation of state of carbon under extreme conditions . , 78:024101 , 2008 . alfredo a correa , stanimir a bonev , and giulia galli . carbon under extreme conditions : phase boundaries and electronic properties from first - principles theory . , 103(5):12041208 , 2006 . md knudson , mp desjarlais , and dh dolan . shock - wave exploration of the high - pressure phases of carbon . , 322(5909):18221825 , 2008 . j. h. irving and j. g. kirkwood . the statistical mechanical theory of transport processes . iv . the equations of hydrodynamics . , 18:817829 , 1950 . k. k. mandadapu , r. e. jones , and p. papadopoulos . generalization of the homogeneous non - equilibrium molecular dynamics method for calculating thermal conductivity to multi - body potentials . , 80(4):047702 , 2009 . p. k. schelling , s. r. phillpot , and p. keblinski . comparison of atomic - level simulation methods for computing thermal conductivity . , 65(14):14430612 , 2002 . m. l. huggins and j. e. mayer . interatomic distances in crystals of the alkali halides . , 1(9):643646 , 1933 . m. l. huggins . lattice energies , equilibrium distances , compressibilities and characteristic frequencies of alkali halide crystals . , 5(2):143148 , 1937 . m. l. huggins . errata : lattice energies , equilibrium distances , compressibilities and characteristic frequencies of alkali halide crystals . , 15(4):212 , 1947 . paul peter ewald . die berechnung optischer und elektrostatischer gitterpotentiale . , 369(3):253287 , 1921 . roger w hockney and james w eastwood . . crc press , 2010 . naoki karasawa and william a goddard iii . acceleration of convergence for lattice sums . , 93(21):73207327 , 1989 . david m heyes . pressure tensor of partial - charge and point - dipole lattices with bulk and surface geometries . , 49(2):755 , 1994 . timothy w sirk , stan moore , and eugene f brown . characteristics of thermal conductivity in classical water models . , 138(6):064505 , 2013 . paolo giannozzi , stefano de gironcoli , pasquale pavone , and stefano baroni . ab initio calculation of phonon dispersions in semiconductors . , 43(9):7231 , 1991 . g kresse , j furthmller , and j hafner . ab initio force constant approach to phonon dispersion relations of diamond and graphite . , 32(9):729 , 1995 . xavier gonze and changyol lee . dynamical matrices , born effective charges , dielectric permittivity tensors , and interatomic force constants from density - functional perturbation theory . 55(16):10355 , 1997 . martin t dove . , volume 4 . cambridge university press , 1993 . cj kimmer and re jones . continuum constitutive models from analytical free energies . , 19(32):326207 , 2007 . georg kresse and d joubert . from ultrasoft pseudopotentials to the projector augmented - wave method . , 59(3):1758 , 1999 . university of vienna , http://www.vasp.at . , 2015 . a togo and i tanaka . first principles phonon calculations in materials science . , 108:15 , 2015 . a. togo , kyoto university , http://atztogo.github.io/phonopy . , 2015 . lewis s combes , stanley s ballard , and kathryn a mccarthy . mechanical and thermal properties of certain optical crystalline materials . , 41(4):215221 , 1951 . smirnov @xcite apparently omits the pressure dependence despite deriving the pressure dependence of the stability moduli in an earlier co - authored publication @xcite . nevertheless , the shear stability criteria used in ref.+ coincide with those derived in app.+[app : stability ] . we also calculated the enthalpy difference with the generalized gradient approximation ( gga ) but it had a downward trend with pressure which we attributed to gga tendency to over - binding . nikolas antolin , oscar d restrepo , and wolfgang windl . fast free - energy calculations for unstable high - temperature phases . , 86(5):054119 , 2012 . bartomeu monserrat , nd drummond , and rj needs . anharmonic vibrational properties in periodic systems : energy , electron - phonon coupling , and stress . , 87(14):144302 , 2013 . sandia national laboratories , http://lammps.sandia.gov . , 2015 . lc carrison and cb sclar . the compressibility and thermal expansion of lif to 60 kbar and 600 c as determined by x - ray diffraction : report of progress . , 13:167 , 1971 . a pirth and jen slyom . , volume 1 . springer science & business media , 2007 . thomas b douglas and james l dever . lithium fluoride : heat content from 0 to 900@xmath90 , the melting point and heat of fusion . , 76(19):48264829 , 1954 . c dames and g chen . thermal conductivity of nanostructured thermoelectric materials . in d. m. rowe , editor , _ thermoelectrics handbook : macro to nano _ , page 42 . crc press , boca raton , fl , 2006 . keivan esfarjani , gang chen , and harold t stokes . heat transport in silicon from first - principles calculations . , 84(8):085204 , 2011 . atsushi togo , laurent chaput , and isao tanaka . distributions of phonon lifetimes in brillouin zones . , 91(9):094306 , 2015 . a. togo , kyoto university , http://atztogo.github.io/phono3y . , 2015 . wu li , jess carrete , nebil a. katcho , and natalio mingo . a solver of the boltzmann transport equation for phonons . , 185:17471758 , 2014 . michael p desjarlais . first - principles calculation of entropy for liquid metals . , 88(6):062145 , 2013 . shiang - tai lin , mario blanco , and william a goddard iii . the two - phase model for calculating thermodynamic properties of liquids from molecular dynamics : validation for the phase diagram of lennard - jones fluids . , 119(22):1179211805 , 2003 . jacques hadamard . . courier corporation , 1922 . max born and kun huang . . oxford university press , 1998 . robin john knops and ew wilkes . theory of elastic stability(liapunov functions application to stability analysis of dynamic systems and elastic bodies , considering eigenfunction method , maximum principle and energy criterion ) . , pages 125302 , 1973 . frederick milstein and rodney hill . theoretical properties of cubic crystals at arbitrary pressure iii . stability . , 27(3):255279 , 1979 . frederick milstein and rodney hill . divergences among the born and classical stability criteria for cubic crystals under hydrostatic loading . , 43(19):1411 , 1979 . jinghan wang , ju li , sidney yip , simon phillpot , and dieter wolf . mechanical instabilities of homogeneous crystals . , 52:1262712635 , nov 1995 . duane c wallace . . courier corporation , 1998 . jw morris jr and cr krenn . the internal stability of an elastic solid . , 80(12):28272840 , 2000 . dm clatterbuck , cr krenn , marvin l cohen , and jw morris jr . phonon instabilities and the ideal strength of aluminum . , 91(13):135501 , 2003 . krystyn j. van vliet , ju li , ting zhu , sidney yip , and subra suresh . quantifying the early stages of plasticity through nanoscale experiments and simulations . , 67:104105 , mar 2003 . e weinan and pingbing ming . cauchy born rule and the stability of crystalline solids : static problems . , 183(2):241297 , 2007 . ronald e miller and david rodney . on the nonlocal nature of dislocation nucleation during nanoindentation . , 56(4):12031223 , 2008 . tj delph , ja zimmerman , jm rickman , and jm kunz . a local instability criterion for solid - state defects . , 57(1):6775 , 2009 . tj delph and ja zimmerman . prediction of instabilities at the atomic scale . , 18(4):045008 , 2010 . flix mouhat and franois - xavier coudert . necessary and sufficient elastic stability conditions in various crystal systems . , 90(22):224104 , 2014 . jerome harris weiner . . courier corporation , 2012 . a cemal eringen and erdogan s suhubi . . academic press , 1974 . javier bonet and richard d wood . . cambridge university press , 1997 . g. v. sinko and n. a. smirnov . calculations of elastic constants and thermodynamic properties of bcc , fcc , and hcp al crystals under pressure . , 14(29):6989 , 2002 . although we employ the pppm method in sec . [ sec : results ] the gist of how the virial stress and , hence , the heat flux is obtained is easier to explain in the context of the ewald sum ( * ? ? ? 7 ) : @xmath91 where the error function , @xmath92 , and its complement , @xmath93 , play the role of a blending / cutoff function with parameter @xmath94{v } $ ] , and @xmath95 is a relative position vector . note we have used the fourier transforms @xmath96 = \sum_\alpha \exp \imath { \mathbf{k}}\cdot { \mathbf{x}}_\alpha$ ] and @xmath97 = \frac{1}{\|{\mathbf{k}}\|^2 } \exp\left ( -\frac{1}{4 } \|{\mathbf{k}}\|^2 \ell^2 \right ) $ ] . it follows , after dropping the species subscripts @xmath98 for clarity , that the per - atom energy for pair potentials is @xmath99 _ cf . _ [ heyes1994pressure](eq . 8) . thus , the expression for the per - atom virial stress @xmath11 @xcite is : @xmath100 \frac{1}{r_{\alpha\beta}^3 } { \mathbf{x}}_{\alpha\beta } \otimes { \mathbf{x}}_{\alpha\beta } \\ & - \frac{2\pi}{\epsilon v } \sum_{{\mathbf{k}}\neq\mathbf{0 } } \left| \sum_\alpha q_\alpha \exp \imath { \mathbf{k}}\cdot{\mathbf{x}}_\alpha \right|^2 \frac{1}{\| { \mathbf{k}}\|^2}{\exp\left ( -\frac{1}{4}{\| { \mathbf{k}}\|^2 \ell^2}\right ) } \left [ { \mathbf{i}}- \left ( \frac{2}{\| { \mathbf{k}}\|^2 } + \frac{1}{2 } \ell^2 \right ) { \mathbf{k}}\otimes { \mathbf{k}}\right ] \end{split}\ ] ] where @xmath101 is the identity tensor and @xmath102 , _ cf . _ [ heyes1994pressure](eq . many versions of the elastic moduli tensor exist at finite deformations like those investigated in this study and the elastic stability of crystal lattices and elastic materials has been well studied , see , _ e.g. _ , refs .. to connect continuum , elastic stability to atomic , phonon stability , we will assume the current positions @xmath12 are given by small time - varying displacements @xmath103 due to phonon modes superposed on large , static deformations characterized by a homogeneous deformation of the zero temperature , equilibrium lattice @xmath104 @xmath105 since a homogeneous deformation maintains equilibrium , @xmath106 , and hence the ( linearized ) newton equation governing the phonon modes is @xmath107 likewise , in the continuum limit , such that @xmath108 , the linearized balance of momentum @xmath109 governs the long - wavelength elastic waves . here , @xmath110 is the mass density in reference configuration @xmath111 . since background stress @xmath112 is homogeneous the system is also in equilibrium at the continuum level . the elasticity tensor @xmath113 of the first piola - kirchoff stress @xmath49 with respect to the deformation gradient has an atomic - level definition @xmath114_{ij } { \mathbf{e}}_i \otimes { \mathbf{x}}_\alpha \otimes { \mathbf{e}}_j \otimes { \mathbf{x}}_\beta = \frac{1}{v_0 } \sum_{\alpha,\beta } \left [ { \mathbb{k}}_{\alpha\beta } \right]_{ij } { \mathbf{e}}_i \otimes { \mathbf{x}}_\alpha \otimes { \mathbf{e}}_j \otimes { \mathbf{x}}_\beta \\ \end{split}\ ] ] where @xmath115 . further manipulation leads to @xmath116 \\ = & \frac{1}{v_0 } \sum_{a , b , c , d , i , j=1}^3 \left [ { \mathbb{c}}_{abcd } { \mathbf{f}}_{ia } { \mathbf{f}}_{jc } + { \mathbf{s}}_{bd } \delta_{ij } \right ] { \mathbf{e}}_i \otimes { \mathbf{e}}_b \otimes { \mathbf{e}}_j \otimes { \mathbf{e}}_d \end{split}\ ] ] relating @xmath113 to the more familiar elasticity tensor @xmath117 of the symmetric second piola - kirchoff stress @xmath118 with respect to the lagrange strain @xmath119 _ cf . _ [ weiner2012statistical](eq . 4.6.11 ) . using the chain rule @xmath120 , eq.([eq : bmom0 ] ) can be written as : @xmath121 based on the push - forward of @xmath113 ( * ? ? ? 4.2.34 ) : @xmath122_{ijkl } = \frac{1}{\det({\mathbf{f } } ) } \sum_{j , l } \left [ { \mathbb{b}}\right]_{ijkl } \left[{\mathbf{f}}\right]_{jj } \left[{\mathbf{f}}\right]_{ll } = \left [ { { \mathbbold{c}}}\right]_{ijkl } + \left [ { \boldsymbol{\sigma}}\right]_{ik } \delta_{jl}\ ] ] where @xmath123 is the push - forward of @xmath124 by the deformation gradient : @xmath125_{ijkl } = \frac{1}{\det({\mathbf{f } } ) } \sum_{i , j , k , l } \left [ { \mathbb{c}}\right]_{ijkl } \left[{\mathbf{f}}\right]_{ii } \left[{\mathbf{f}}\right]_{jj } \left[{\mathbf{f}}\right]_{kk } \left[{\mathbf{f}}\right]_{ll}\ ] ] the legendre - hadamard criterion for dynamic stability requires that all infinitesimal plane waves @xmath126 have real - valued wave speeds . here , @xmath23 and @xmath127 are the amplitude and polarization of the displacement ( such that @xmath128 ) , and @xmath129 and @xmath130 are the wave number and propagation direction ( such that @xmath131 ) . this leads to an eigenvalue problem for the dyad @xmath132 and the strong ellipticity condition @xmath133_{ijkl } n_i p_j n_k p_l > 0\ ] ] this condition is satisfied when all the eigenvalues of the square matrix @xmath134 are all real and positive . the moduli that vasp and other codes calculate are derivatives of the current , cauchy stress with respect to small strains about a given configuration , which is not @xmath123 . to connect @xmath135 to the moduli obtained from perturbing the system about a given ( not necessarily stress - free reference ) configuration , we start with the derivative of the cauchy stress with respect to a displacement @xmath136 about a deformed state @xmath137 @xmath138 \bar{{\mathbf{f } } } \bar{{\mathbf{s } } } \bar{{\mathbf{f}}}^t + \frac{1}{\det({\mathbf{f } } ) } \left [ \partial_{\mathbf{x}}{\mathbf{u}}\right ] \bar{{\mathbf{f } } } \bar{{\mathbf{s } } } \bar{{\mathbf{f}}}^t \\ & + \frac{1}{\det({\mathbf{f } } ) } \bar{{\mathbf{f } } } \ , \partial_{\mathbf{e}}\bar{{\mathbf{s } } } \bar{{\mathbf{f}}}^t \ , \frac{1}{2 } \left [ \partial_{\mathbf{x}}^t { \mathbf{u}}+ \partial_{\mathbf{x}}^t { \mathbf{u}}\right ] \bar{{\mathbf{f } } } \bar{{\mathbf{f}}}^t + \frac{1}{\det({\mathbf{f } } ) } \bar{{\mathbf{f } } } \bar{{\mathbf{s } } } \bar{{\mathbf{f}}}^t ( \partial_{\mathbf{x}}^t { \mathbf{u } } ) \nonumber\end{aligned}\ ] ] formed from the basic gteaux derivatives in [ bonetwood](eqs . 3.69 , 3.71 , and 3.76 ) . we recognize that the third term on the right - hand side is @xmath123 and @xmath139 in the other terms , so that we can form the frchet derivative of the cauchy stress with respect to the small strain measure @xmath140 as : @xmath141_{ijkl } = - \bar{{\boldsymbol{\sigma}}}_{ij } \delta_{kl } + \delta_{il } \bar{{\boldsymbol{\sigma}}}_{jk } + \delta_{jk } \bar{{\boldsymbol{\sigma}}}_{il } + { { \mathbbold{c}}}_{ijkl}\ ] ] which is the typical moduli calculated by finite differences or perturbation in terms of the current stress @xmath139 and push - forward of the tensor of the traditional elasticities @xmath124 to the current state . eq.([eq : frechet ] ) is identical in form to corresponding equations in the often cited [ wang1995mechanical ] , and in the independently derived [ sinko2002ab ] , but differs in the interpretation as moduli about a deformed state finitely far from the relaxed , stress - free material . when eq.([eq : frechet ] ) is combined with eq.([eq : smallbb ] ) , the stability requirement can be applied to : @xmath142_{ijkl}\ ] ] for our purposes it suffices to find the stability conditions for an orthotropic modulus tensor @xmath143 and a diagonal stress tensor @xmath144 . following [ mouhat2014necessary ] , we obtain : @xmath145 where @xmath146 , @xmath147 ; @xmath148 , @xmath149 , @xmath150 ; and we have used @xmath151 to denote the components of @xmath152 $ ] using traditional voigt notation . this reduces to @xmath153 for cubic symmetry and a hydrostatic pressure @xmath154 . these stability criteria differ from those in [ wang1995mechanical ] and [ sinko2002ab ] in that the shear conditions are unaffected by the pressure and the volumetric instability criterion on the bulk modulus @xmath155 is offset by the pressure only .

while standard wisdom has long held that some significant component of the cosmic ray spectrum is the result of particle acceleration by shocks in snrs , only recently has compelling evidence of such a process been observed . studies carried out with the _ advanced satellite for cosmology and astrophysics _ ( asca ) have clearly established the nonthermal nature of the emission from the limb - brightened regions of the remnant of sn 1006 ( koyama et al . 1995 ) . both the spectrum and the morphology of the observed x - rays can be explained by models in which synchrotron radiation is produced from electrons with energies up to 100 tev accelerated by the remnant blast wave ( reynolds 1996 ) , a scenario which is bolstered by the apparent detection of tev @xmath1-rays from sn 1006 ( tanimori et al . 1998 ) . x - ray spectral results similar to those for sn 1006 are observed for the remnant g347.3@xmath00.5 ( figure 1 ) , a bright snr discovered in the rosat all - sky survey ( rass ; pfeffermann & aschenbach 1996 the rosat catalog source name for the remnant is rx j1713.7@xmath03946 ) . koyama et al . ( 1997 ) serendipitously observed the northwest shell of the remnant in a portion of a galactic plane survey being conducted with asca and found that the emission from this region appears nonthermal ; the spectrum is featureless , extends to high energies , and is well described by a power law . here we present our new x - ray and radio observations of g347.3@xmath00.5 that cover the entire snr . in 2 below , we describe the several data sets we have for the remnant . in 3 we discuss the remnant s distance implied by our co data and the spectral results from the x - ray observations . section 4 contains a discussion of the overall picture the data suggest for g347.3@xmath00.5 , and the conclusions are summarized in 5 . nonthermal emission in supernova remnants is usually associated with the presence of a central pulsar - powered synchrotron nebula . sn 1006 appears to present a glaring exception to this scenario , but cas a also possesses a hard power - law tail ( allen et al . 1997 ) , and ic443 harbors a concentration of hard emission ( keohane et al . 1997 ) that appears to be associated with a molecular cloud interaction site . with synchrotron spectra extending to 10 kev and beyond , this implies the presence of electrons with energies as high as @xmath2 tev for typical interstellar magnetic fields . the origin of this emission is of considerable significance both in terms of the long - surmised source of cosmic rays and because it raises questions about what special conditions are responsible for such production in such a limited number of snrs . like sn 1006 ( but unlike cas a or ic443 ) , the nonthermal emission from g347.3@xmath00.5 _ dominates _ the x - ray flux . this snr thus presents significant evidence for particle acceleration which may lead to a confirmation of the long - suspected connection with the origin of cosmic rays . however , in contrast to sn 1006 , whose age , distance , and progenitor type are known , g347.3@xmath00.5 is an unfamiliar beast . to understand how and why ( or even if ) it is acting as a cosmic ray accelerator , we need to establish the basic fundamentals of its character . ( 0,10.5)(18.5,21 ) ( 0.4,20.7)=8.5 cm ( 9.75,20.7)=8.0 cm ( 0,12.7 ) = 3.5 mm fig . 1. rosat pspc image of g347.3@xmath00.5 . two point - like sources are seen in the central regions . the eastern - most source is associated with a star while the central source is unidentified . asca observations reveal strong nonthermal emission from the two bright regions along the western limb as well as weak nonthermal emission from the rest of the interior . contour levels start at @xmath3 and increase inward in steps of @xmath4 . ( 9.7,12.7 ) = 3.5 mm fig . 2. asca gis image of g347.3@xmath00.5 with pspc contours . regions used for spectral studies are indicated . boxes along the western limb were used for the nw and sw regions ( see text ) . two circles from the eastern half of the remnant were used for the east region and a circle around the central point source was used for the ptsrc region . for deriving upper limits to the thermal emission for the entire snr , the four black circles were used with the white circle excluded . g347.3@xmath00.5 was first discovered in the rass data ( pfeffermann & aschenbach 1996 ) . a follow - up 3 ks pointed observation was then carried out with the position sensitive proportional counter ( pspc ) , providing a good map of the @xmath5 arcmin diameter remnant ( figure 1 ) . the spectral results were inconclusive , however , being unable to distinguish between a thermal or nonthermal origin for the soft x - ray emission . the pspc also reveals two point - like sources in the field of the remnant . the first , 1wgaj1714.4@xmath03945 , has a very soft spectrum and appears to be associated with a star ( pfeffermann & aschenbach 1996 ) . we will not discuss it further . the second source , 1wgaj1713.4@xmath03949 , which is located near the center of the remnant at position 17@xmath6 13@xmath7 28@xmath8 , @xmath039@xmath9 49@xmath10 46@xmath11 ( epoch j2000 here and henceforth ) , has no reported counterpart . the brightest cataloged object near the x - ray source is @xmath12 arcsec away and has an optical magnitude @xmath13 ( @xmath14 ) . this separation is consistent with the expected position errors for the pspc source given the point spread function of the detector as well as aspect reconstruction errors . the large x - ray to optical flux ratio is consistent with what might be expected for a neutron star . a stellar counterpart is virtually ruled out for such a ratio ( stocke et al . 1991 ) , but a background extragalactic source remains a possibility . we observed g347.3@xmath00.5 with asca on 1997 march 25 for a total of 50 ks using three pointings of 10 ks duration and one ( in the fainter southeast region ) of 20 ks duration , which resulted in complete coverage by the gas - imaging spectrometer ( gis ) . complimentary chips on the ccds were used to image the entire solid - state imaging spectrometer ( sis ) field of view ; nevertheless , complete coverage of the remnant was not possible with the sis . risetime information for the gis was sacrificed in order to provide improved temporal resolution so that timing analysis could be carried out on the central source . after standard screening procedures for the gis and sis data , we performed spectral fits for several distinct spatial regions defined as follows : nw - a box surrounding the bright region in the northwest which was first studied by koyama et al . ( 1997 ) ; sw - a box surrounding the bright region in the southwest ; ptsrc - a 3 arcmin radius circle centered on the central source ; and east - combined data from two circles in the northeast and southeast regions of the remnant . these regions are shown schematically in figure 2 where we present the gis image . ( also shown are additional circles used in combination with east to establish limits on the total thermal emission ; see 3.2.2 . ) the corresponding spectra are shown in figure 3 . ( 9.5,-1)(18.5,10.3 ) ( 9.0,10.3)=8.8 cm ( 9.0,1.5 ) = 3.5 mm fig . 3. asca spectra from distinct regions of g347.3@xmath00.5 ( see figure 2 ) along with best - fit models . top panel : gis ( upper ) and sis ( lower ) from nw shell . spectrum is well described by a power law ( shown as histogram ) and is indistinguishable from spectrum of sw shell ( not shown ) . middle panel : gis spectrum from eastern region of remnant along with best - fit power law model . lower panel : gis spectrum of central point - like source along with best - fit blackbody model . table 1 g347.3@xmath00.5 spectral parameters [ cols="<,<,<,<",options="header " , ] unlike the line - dominated x - ray spectra typical of snrs , the spectra for g347.3@xmath00.5 are clearly featureless and extend to at least 10 kev , where the asca effective area becomes very small . for each region , we performed joint fits to data from each gis detector as well as to the associated pspc data from the same spatial region of the remnant . a power law spectrum provides an excellent fit to each spectral region of the remnant . the nw and sw regions display virtually identical spectra while that for the eastern region appears to have a flatter spectral index and somewhat lower absorption . the best - fit spectral values are listed in table 1 . addition of a thermal component ( raymond & smith 1977 ) to the spectral model does not result in a significant improvement in the fit ; there is no significant evidence for thermal emission from any portion of the snr . we note that a pure thermal bremsstrahlung model ( e.g. , an exponential with gaunt factor ) yields a considerably larger value of @xmath15 than does the power law model ( e.g. , for the sw region , @xmath16 for 293 degrees of freedom compared with @xmath17 for the power law model ) . the sis data provide higher sensitivity for detection of line features in the spectra due to its better spectral resolution compared to the gis . however , due to the extended nature of the emission , the number of counts in any particular chip is small . further , background measurements were possible for only two of the chips and these were used as representative background for analysis of data from other chips . variations in the spectral characteristics of each chip , plus possible spatial variations in the cosmic and galactic x - ray background emission , thus introduce larger uncertainties than would be obtained with independently measured backgrounds . we have carried out fits to the sis data from the nw region , the sw region , and for the northeast region . in each case , the sis fits show no evidence of spectral lines and the power - law fit parameters are in reasonable agreement with those obtained from the joint gis / pspc fits . the nonthermal nature of the emission from all regions of the remnant is quite peculiar . unlike sn 1006 , where thermal emission is found in the central regions , any such component from g347.3@xmath00.5 is apparently overwhelmed by the nonthermal emission . we discuss this surprising result in 3 . ( 0,-0.5)(8.5,4 ) ( -0.5,3.5 ) = 3.5 mm table 2 blackbody model for 1wgaj1713.4 - 3949 llparameter & value + @xmath18 & @xmath19 + @xmath20 & @xmath21 kev + @xmath22 & @xmath23 + @xmath24 & @xmath25 + + + the spectrum from the unidentified central source 1wgaj1713.4@xmath03949 is not well constrained by the data . joint fits to the sis / gis / pspc data yield acceptable fits for power - law , raymond - smith , bremsstrahlung , and blackbody models . the blackbody model yields the lowest overall @xmath15 fit ( though only marginally lower than the other models ) , and we summarize these fit parameters in table 2 for use in discussing a neutron star scenario for this source in 3.2.3 . however , we note that additional observations are required to obtain a higher fidelity characterization of the spectral properties . ( 0,11)(18.5,21 ) ( 0,12.7 ) = 3.5 mm fig . 4. iras 60 @xmath26m/100 @xmath26 m ratio image with pspc contours overlaid . a marginal enhancement in the ratio appears over the interior of the remnant , but significant emission is seen outside the western limb possibly indicating the presence of adjacent cool material . ( 9.7,12.7 ) = 3.5 mm fig . 5. most image of radio emission at 843 mhz , with pspc contours . emission from the bulk of the snr rim can be seen with particular enhancements along the west / northwest regions where bright nonthermal x - ray emission is seen . bright emission outside of the western limb corresponds well with the observed infrared emission from this region . we have also investigated the infrared emission from the region with the infrared astronomical satellite ( iras ) 60@xmath26m/100@xmath26 m ratio ( figure 4 ) . an enhancement of this ratio , indicative of hot dust , has been shown to be a powerful discriminator of remnant emission against the background emission of the galactic plane ( saken , fesen , & shull 1992 ; slane , vancura , & hughes 1996 ) . the infrared emission in the inner galactic disk is intense and complicated , but as figure 4 shows , there is a clear enhancement of the 60 mm/100 mm ratio coincident with the remnant . the strongly enhanced ratio to the northwest of the remnant is due to the region g347.61 + 0.20 ( lockman 1979 ) , which may be associated with the same molecular cloud as the remnant ( see 3.1 ) . radio maps from the parkes - mit - nrao ( pmn ) 6 cm survey ( condon , griffith , & wright 1993 ) reveal complex emission from the region surrounding the remnant , but we find very good morphological correspondence with the infrared emission , particularly for the regions directly west and northwest of the remnant . to better delineate any radio emission corresponding to the x - ray remnant , we have carried out 843 mhz observations using the molonglo observatory synthesis telescope ( most ; robertson 1994 ) . the synthesis was carried out on 1997 may 27 , with the telescope in its wide field mode of operation ( large et al . the resulting map ( figure 5 ) reveals regions of faint emission which extend along most of the snr perimeter , with the most distinct emission corresponding to the bright x - ray regions in the northwest and southwest . considerable contributions from sidelobes , baseline errors , and grating rings from surrounding sources result in a rather high noise level of @xmath27 mjy beam@xmath28 ; the contrast has been adjusted considerably to show the faint shell emission . the brightest regions of emission within the confines of the x - ray contours are found along two arcs in the northwest . these are concentrated along the edges of the brightest x - ray emission . arc 1 is centered at approximately 17@xmath6 12@xmath7 46@xmath8 , @xmath039@xmath9 25@xmath10 49@xmath11 . we estimate a surface brightness range of @xmath29 ( @xmath30 ) with @xmath31 uncertainty in these values . arc 2 , centered at 17@xmath6 11@xmath7 27@xmath8 , @xmath039@xmath9 32@xmath10 36@xmath11 has a surface brightness range @xmath32 ( @xmath33 ) with similar uncertainty . this arc appears to be most closely correlated with the bright x - ray emission and , although its east - west elongation is curiously askew with respect to the perceived snr shell at this location , the radio filament does overlay a filament of x - ray emission at this position . an upper limit to any radio counterpart to the unidentified x - ray point source 1wga j1713.4@xmath03949 is 15 mjy ( @xmath34 ) . the intense emission beyond the nw limb of the remnant corresponds to the same region , g347.61 + 0.20 , that was evident in the iras map ( figure 4 ) . additional radio observations of g347.3@xmath005 with higher resolution and sensitivity have been carried out with the australia telescope compact array . analysis of these data is in progress and will be reported in a future publication . to further constrain our picture of the environment surrounding the remnant , we have used observations of the _ j _ = @xmath35 rotational transition of co ( at 115 ghz ) , the best large - scale tracer of interstellar molecular gas , from the cfa - chile 1.2 m telescope ( bronfman et al . figure 6 is a map of total molecular column density over a wide section of the inner galaxy centered on the snr . it is noteworthy that the snr coincides very closely with the largest and deepest hole in the molecular column density near the galactic plane in the inner fourth quadrant . since the co emission in the direction of the remnant is complex and extends over @xmath36 in radial velocity , it is difficult to determine which individual molecular cloud or clouds , if any , along the line of sight might be associated . the most likely candidates based on the morphology of the co and x - ray emission are a pair of clouds at @xmath37 and a third at @xmath38 ( figure 7 ) . given that giant molecular complexes with active star formation sometimes have total velocity extents of 30 km s@xmath28 ( e.g. , w44 ; dame et al . 1986 ) , it is possible that all three of these clouds are part of the same molecular complex that gave birth to the region g347.61 + 0.20 and the snr , with one or more of the clouds significantly perturbed in velocity . the cloud at @xmath38 abuts the region of strongest x - ray and radio enhancement , suggesting that the remnant may be expanding into this cloud . it is worth noting that a compact region of anomalous - velocity emission extending to @xmath39 is seen in both 21 cm ( burton 1985 ) and co ( bronfman et al . 1989 ) emission along the line of sight to the remnant . ( -0.5,11.7)(8.5,24 ) ( -0.75,24)=8.8 cm ( -0.75,14.5 ) = 3.5 mm fig . 7. contours of co intensity integrated over the velocity range @xmath40 to @xmath41 , from the survey of bronfman et al . the contours are spaced every @xmath42 , starting at @xmath43 . the rosat pspc image is shown in grayscale . the molecular clouds that seem most likely to be associated with g347.3@xmath00.5 are labeled with their lsr velocities . the region g347.61 + 0.20 is marked by a `` + '' sign . given their proximity both in projected position on the sky and in velocity , the pair of clouds at @xmath37 are almost certainly associated with each other , and with the region g347.61 + 0.20 , which has a recombination - line velocity of @xmath44 ( lockman 1979 ) . an association between the snr and the pair of clouds at @xmath37 is suggested by the presence of massive star formation in one of them , as evidenced by the region , and by an enhancement of the co(21)/co(10 ) line ratio in the other . studies of molecular clouds in the galaxy find values generally below 1 for the ratio @xmath45 . sakamoto et al.(1995 ) find a mean value of @xmath46 with a variation from 0.5 to 0.8 that has some dependence on galactocentric distance . chiar et al . ( 1994 ) find a value of @xmath47 from their study of molecular clouds in the scutum arm of the galaxy . on the other hand molecular clouds that appear to be undergoing interactions with snrs ( e.g. , w44 and ic443 ; seta et al . 1998 ) , show an enhanced @xmath48 ratio ( @xmath491.2 ) that is indicative of shocked material in the molecular gas . we obtained co(21 ) spectra over the range @xmath50 to 348.5 , @xmath51 to 0.5 , on a 0.25 degree grid ( handa & hasegawa 1999 , private communication ) from the university of tokyo 0.6 m telescope on la silla , chile , which has the same angular resolution at the co(21 ) line as the cfa - chile 1.2 m at the co(10 ) line . over the range for which co(21 ) spectra were available , the mean @xmath48 value was determined to be @xmath52 . however , the largest values of @xmath48 were associated with the cloud at @xmath37 : in particular , toward @xmath53 , @xmath54 , a value of @xmath55 is found near @xmath56 , in the wing of a strong ( 4 k ) and well - defined co(10 ) line near @xmath57 . such an enhancement in the wing of the co line is suggestive of an interaction with the snr shock wave ( seta et al . we note that no enhanced @xmath48 value is observed for the cloud at @xmath58 which appears along the line of sight to the brightest x - ray limb . however , the co(21 ) data are coarsely spaced ( every 1/4 degree ) ; more sensitive and closely spaced spectra in this region may reveal such an enhancement and are of considerable interest . koyama et al . ( 1997 ) compare their measurement of the column density toward g347.3@xmath00.5 ( which is in good agreement with the values we report here ) with the total @xmath18 in the direction of the galactic center to estimate a distance of 1 kpc to the remnant . we argue that a distance as small as this is unlikely , based on several lines of evidence . like koyama et al . ( 1997 ) we first consider the distance estimate based on the x - ray derived column density . the total line - of - sight column density in the direction of g347.3@xmath00.5 is @xmath59 based upon observations ( dickey & lockman 1990 ) and confirmed by co measurements ( bronfman et al.1989 ) . this relatively low value is consistent with the presence of a hole in the molecular column density in this direction mentioned earlier ( see also fig 6 ) . the @xmath18 value we measure for g347.3@xmath00.5 ( @xmath60@xmath61 , see table 1 ) is a significant fraction of the total column through the galaxy and so the distance to the remnant must be considerably more than 1 kpc . we have just argued in the preceeding section that the remnant , the region , and the complex of three molecular clouds in this direction are all likely to be physically associated . this provides an opportunity to obtain a quantitative estimate for the distance to the remnant . from the rotation curve of the galaxy ( burton 1988 ) , the pair of clouds at @xmath37 lie at a distance of @xmath62 kpc or @xmath63 kpc . whether the near or far kinematic distance is appropriate can be resolved in favor of the former by noting that molecular absorption measurements place the associated region at a distance of 6.3 kpc ( lockman 1979 , adjusted to the burton 1988 rotation curve and @xmath64 kpc ) . the cloud at @xmath65 has a near kinematic distance of 5.4 kpc , assuming that its radial velocity is purely due to galactic rotation . therefore we adopt a distance of 6 kpc for the distance to g347.3@xmath00.5 , with the warning that this is uncertain by at least @xmath661 kpc . in the following we present results in terms of @xmath67 . the nonthermal x - ray emission from g347.3@xmath00.5 strongly indicates that the snr shock is accelerating particles to very high energies . the electron energy associated with synchrotron photons radiated at energy @xmath68 is @xmath69 with observed photon energies up to of 10 kev , the spectrum for g347.3@xmath00.5 thus indicates the presence of electrons with energies as high as 300 tev for typical magnetic fields of @xmath70 g . if protons are accelerated to similar energies , it would appear that g347.3@xmath00.5 may be accelerating cosmic rays up to energies near the `` knee '' in the cosmic ray spectrum at @xmath71 ev . the total energy associated with the x - ray producing electrons is of order @xmath72 erg which is a small fraction of the mechanical energy released in the supernova explosion . as noted in table 1 , the spectral index for the emission from the bright regions along the western limb of g347.3@xmath00.5 appears to be steeper than that for the rest of the diffuse emission from the snr . the column density from the spectral fits also differs between the two regions . when @xmath18 is fixed to the value determined for the nw region , the photon index derived for the eastern region is @xmath73 , the same as that for the regions along the western limb . however , the @xmath15 for the fit is considerably higher and the residuals clearly indicate difficulty with such a high absorption value for the eastern region . a variation in column density is not hard to reconcile ; many snrs exhibit such spatial variations in the foreground absorbing column ( e.g. , roberts et al . 1993 ) . for g347.3@xmath00.5 this interpretation is particularly reasonable , since the regions indicating higher column density ( i.e. , the nw and sw enhancement regions ) lie closest to the galactic plane ( see figure 7 ) . variations in spectral index are also well - understood for synchrotron nebulae where the finite radiating lifetimes of electrons accelerated by a central pulsar result in a steepening of the spectral index with distance from the electron injection point ( e.g. , kennel & coroniti 1984 ; slane , bandiera , & torii 1998 ) . for synchrotron nebulae the steepening of the spectrum goes along with a decrease in brightness . for g347.3@xmath00.5 , however , no obvious injection point is evident , and the steeper index appears to be concentrated along the bright western limb . a synchrotron lifetime argument thus appears problematic in that the brightest synchrotron region ( i.e. , the western limb ) would also have to be `` older . '' for completeness we note that the nonthermal x - ray emission from sn 1006 , with a photon index @xmath74 , is steeper than spectra measured here for g347.3@xmath00.5 . although the bright rim regions of sn 1006 are dominated by synchrotron emission , asca studies of the interior reveal the line - dominated x - ray thermal component expected from the hot shocked gas characteristic of snrs ( koyama et al . our asca observations of g347.3@xmath00.5 reveal no such component . the lack of measurable thermal emission sets limits on the amount of hot material , and thus on the density of the region in which the snr evolved . this is illustrated in figure 8 where we have plotted upper limits to the thermal emission measure as a function of temperature based upon our spectral analysis . for the curve plotted as the solid histogram in figure 8 , we have combined data from the gis detectors using regions that cover most of the remnant ( but not the bright regions along the western limb see figure 2 ) and determined 3-sigma upper limits to the amount of thermal emission , assuming normal solar composition , that could be added to the model . the resulting normalization for this thermal component was then scaled by a factor of two to account for the incomplete coverage of the remnant that our spectral extraction regions gave . as indicated in figure 2 , the spectral regions actually cover more than 50% of the area of the snr , so the upper limits shown are quite conservative . under some assumptions about the three - dimensional geometry of the remnant , the upper limit on the emission measure can be converted into a limit on the mean density of the ambient medium . for g347.3@xmath00.5 we assume a spherical shell geometry with a measured angular radius of @xmath75 and an estimated shell thickness equal to @xmath76 . ( note that the factor 23 was chosen for numerical agreement with the integrated emission measure for a sedov solution . ) for the range of allowed emission measures , we then calculate that the hydrogen number density of the ambient preshock medium in the strong shock limit ( i.e. , assuming the density jump across the shock front is a factor of 4 ) is in the range @xmath77 = 0.0140.28 @xmath78 @xmath79 . the lower bound on the ambient density corresponds to temperatures of @xmath600.75 kev , while the upper bound corresponds to the lowest temperature bin in figure 8 . ( 9.5,8)(18.5,20 ) ( 9.0,20)=8.8 cm ( 9.0,11.5 ) = 3.5 mm fig . 8. upper limit to derived emission measure associated with a thermal emission component plotted as a function of temperature . the two histograms correspond to limits derived for the entire snr ( solid ) based upon scaling results which exclude the bright regions ( see text ) , and for the bright northwest region ( dashed ) . also shown are curves for models derived assuming a sedov solution . solid ( dotted ) curves correspond to a distance of 6 kpc ( 2 kpc ) with @xmath80 varying from 0.2 ( lower ) to 1.0 ( upper ) . also plotted in figure 8 are model curves for explosion energies @xmath81 , 0.5 , and 1.0 ( in units of @xmath82 erg ) assuming a distance of 6 kpc ( solid curves ) and 2 kpc ( dotted curves ) under the assumptions that the remnant is in the sedov ( 1959 ) phase of evolution and the electrons and ions share the same temperature behind the shock front . we see that the derived upper limits are not inconsistent with the model curves for our prefered distance of 6 kpc . that is , for the range of parameter space shown , the lack of detectable thermal emission is expected for a sedov - phase snr ; as with sn 1006 , the thermal emission simply pales in comparison to the bright nonthermal emission . for such an interpretation , however , the density of the ambient medium must be very low , as we found above . the derived upper limits on emission measure appear to rule out a sedov solution for g347.3@xmath00.5 if the remnant is at a distance of 2 kpc . the only acceptable solution requires an unusually low explosion energy and a rather hot remnant , @xmath83 kev . however this conclusion is sensitive to the assumption of temperature equilibration between electrons and ions in the post - shock gas . the alternative scenario , wherein the electrons and ions acquire the same mean thermal velocity behind the shock front and then share energy slowly through coulomb collisions ( see spitzer 1978 ) , will result in the mean electron temperature in the remnant ( @xmath84 ) lagging considerably behind the post - shock temperature of the ions . in this situation the measured electron temperature does not reflect the actual shock velocity and , in fact , it has been shown ( cox & anderson 1982 ) that the electron temperature has an extremely weak dependence on the shock temperature . furthermore , because the equilibration timescale is inversely proportional to the electron density , the effects of non - equilibration are greatest for low density remnants , which is precisely the situation we have here for g347.3@xmath00.5 . although a detailed analysis of the effects of non - equilibration will be addressed in a future article , we make the following brief comments here . the curves for the sedov solution shown in figure 8 are substantially modified if equilibration on coulomb collisional timescales is assumed . first of all , there are no high temperature solutions for reasonable values of the explosion energy ( @xmath85 ) anywhere in the range of emission measures plotted . second , it is not possible to rule out a sedov solution as a model for the remnant if it is at a distance of 2 kpc . for example , for this distance and a mean temperature of @xmath86 kev , emission measures in the allowed range 0.01@xmath87 predict ranges for the explosion energy of @xmath882.2 and the remnant s age of 21003400 yrs . lower values of the mean temperature are also acceptable and result in a broad range of possible dynamical states that , in general , encompass smaller values of the explosion energy and larger values of the age . for a distance of 6 kpc , a reasonable dynamical state for g347.3@xmath00.5 is possible for a mean temperature of @xmath89 kev and emission measures in the allowed range 0.003@xmath90 . within this range the explosion energy varies in the range @xmath911.7 and the remnant s age lies between 18,700 yrs and 40,600 yrs . so in conclusion to this discussion , the lack of detected thermal emission from g347.3@xmath00.5 can be reconciled with a broad range of reasonable dynamical evolutionary states , possible remnant ages , and distances as long as the mean ambient density around the remnant is low . this would appear incompatible with the environment in the vicinity of a molecular cloud unless the remnant has involved in a cavity produced by the stellar wind of the progenitor star . we discuss this further in 4 . the limits on thermal emission used for the preceeding analyses represent an average of the faint parts of the remnant . clearly the limits will be higher in the bright regions of the snr . in the upper panel of figure 8 , the dashed histogram represents the upper limit for thermal emission from the bright northwestern region of the remnant using the spectral extraction region illustrated in figure 2 . treating the emission region as a simple slab through the remnant , the upper limit at the low end of the temperature scale limits the preshock density of material at this position to @xmath92 where @xmath93 represents the fraction of the slab volume that is actually responsible for the emission . this limit suggests that the ambient density toward the northwest could be considerably greater the best spectral fit for the unidentified central x - ray source 1wgaj1713.4@xmath03949 based upon a joint fit to the gis , sis , and pspc is obtained for a blackbody spectrum with @xmath94 kev ( see table 2 ) . the normalization yields a relatively small emitting area with radius @xmath95 km , suggesting a compact object . the lack of an optical counterpart supports the interpretation that the source is a neutron star associated with g347.3@xmath00.5 . however , the blackbody temperature is rather high for a cooling neutron star . temperatures similar to that observed are often associated with emission from heated polar caps , but in the case of 1wgaj1713.4@xmath03949 the emitting area @xmath96 is considerably larger than that observed from other sources ( _ e.g. _ psr 1919 + 10 ; yancopoulos , hamilton , & helfand 1994 ) , implying that 1wgaj1713.4@xmath03949 may be a rapidly spinning object ( @xmath97 , where @xmath98 is the rotation period ) . there is no evidence of long - term flux variability between the rosat and asca observations ( although uncertainties in the spectral model do not allow us to rule out small variations ) . we extracted a total of 1100 counts in the gis data from the region around the source for use in timing analysis . a search for pulsations for periods longer than @xmath99 ms ( the highest time resolution for the entire data set ) yielded negative results implying a pulsed fraction upper limit of @xmath100 for the extracted events assuming a sinusoidal light curve ( vaughan et al . 1994 ) . based upon fits to the background - subtracted spectrum for the source , @xmath101 of the events in the extraction region were attributable to the source . the derived upper limit to the source pulsed fraction is thus of order 100% and offers no meaningful constraint . we note that adequate spectral fits for the central source emission can also be obtained for power law and raymond - smith models . for the power law model , the spectral index is extremely steep ( photon index @xmath102 ) and the column density is large ( @xmath103 ) . for the thermal plasma model , the best - fit column density is @xmath104 which is comparable with the total absorption through the galaxy in this direction . thus it is possible that the source is extragalactic , although the steep x - ray spectral slope and the lack of radio emission makes this interpretation somewhat implausible . based upon all observations to date , a complete picture of g347.3@xmath00.5 is only just beginning to unfold . initial x - ray observations ( pfeffermann & aschenbach 1996 ) indicated a nearby ( @xmath105 kpc ) remnant which was subsequently suggested as a counterpart to a guest star from ad393 ( wang , qu , & chen 1997 ) . the latter authors argue that the limited information available for such a scenario would seem to suggest a type ib / ic progenitor explosion . they further suggest that the morphology of the remnant may show some evidence of a central ring - like structure which could be associated with evolution in a wind - blown bubble . we see no evidence of such structure in either the original pspc maps of g347.3@xmath00.5 nor in our maps obtained with asca . further , as argued above , a distance as small as 1 kpc appears to be very unlikely . the asca spectra for g347.3@xmath00.5 clearly indicate the presence of very high energy electrons , providing evidence for direct acceleration of particles by the snr shock . reynolds ( 1998 ) provides a thorough discussion of the maximum energy constraints for diffusive shock acceleration in snrs and produces model results for the nonthermal radiation assuming a constant magnetic field and explosion characteristics typical of type ia progenitors . the maximum energy for the accelerated particles is determined by : ( 1 ) radiative losses by the particles in the ambient magnetic / photon field ; ( 2 ) the finite acceleration time associated with the snr age ; and ( 3 ) particle escape due to the weakening or absence of magnetohydrodynamic waves ( beyond some maximum wavelength ) from which the electrons scatter . reynolds ( 1998 ) shows that all young snrs should show some evidence of nonthermal x - rays due to shock acceleration , but that the maximum ratio of x - ray / radio nonthermal emission will occur when the snr age is roughly equal to the loss time for electrons in the postshock magnetic field a condition that apparently holds for sn 1006 but not for other historical snrs . the nonthermal emission persists to ages beyond @xmath106 yr . a similar scenario for the nonthermal emission in g347.3@xmath00.5 presents some difficulties . we have argued that the distance to this snr is of order 6 kpc and that it appears to be in the vicinity of a molecular cloud complex . the conditions are thus considerably different than the high - latitude , type ia environment for sn 1006 . however , if g347.3@xmath00.5 were the result of an explosion in a stellar wind bubble , there are at least qualitative similarities . the surrounding medium would have low density and a toroidal magnetic field which may yield conditions for efficient particle acceleration . tenario - tagle et al . ( 1991 ) have performed 2d numerical hydrodynamical simulations of young type ii snrs interacting with wind - blown bubbles and find that the freely moving ejecta undergo only partial thermalization while the remnant as a whole ages rapidly since early evolutionary phases are essentially accelerated due to the low density medium in the interior of the bubble . if , upon encountering the bubble wall , diffusive shock acceleration is enhanced , the under - thermalization of the ejecta along with the small swept - up mass may result in a large ratio of nonthermal / thermal x - ray emission . thus , if the remnant is indeed interacting with the observed molecular clouds , this could enhance the nonthermal component . simulations indicate that such interactions result in additional shocks in the cloud which can enhance the nonthermal electron population ( jun & jones 1999 ) although jones & kang ( 1993 ) argue that such enhancements occur primarily when the incident shock is already dominated by cosmic - ray pressure . an alternative interpretation for the featureless x - ray spectrum from g347.3@xmath00.5 could be thermal emission from material dominated by low-@xmath107 elements ( h , he , c , o ) which are fully ionized ( hamilton , sarazin , szymkowiak , & vartanian 1985 ) . hamilton , sarazin , & szymkowiak ( 1986 ) proposed a scenario for sn 1006 in which the ejecta from the type ia explosion were stratified in such a way as to yield an outer region dominated by such featureless emission . in addition they were able to obtain an approximately power - law spectral form by integrating over the power - law distribution of thermal components that came out of their dynamical model . however , koyama et al . ( 1995 ) argue convincingly that their asca observations of sn 1006 rule this model out . a similar scenario has been revisited by laming ( 1998 ) who concludes that the flux associated with the featureless components of sn 1006 and g347.3@xmath00.5 , if interpreted as thermal in nature , would imply an x - ray emitting mass of low-@xmath107 elements far in excess of that expected to be ejected by a type ia supernova explosion . thus , while this mechanism could potentially explain some small featureless components ( _ e.g. _ for ic443 or cas a ) it does not appear to be a viable explanation for the spectrum of g347.3@xmath00.5 . recent model calculations by baring et al . ( 1999 ) suggest that the soft x - ray flux of young remnants with parameters typical of type ia explosions should be dominated by bremsstrahlung emission . however , these authors do find models that produce spectra dominated by synchrotron emission up to and beyond 10 kev . one of these ( their model `` d '' ) uses @xmath108 , ambient density @xmath109 , ejecta mass @xmath110 , and an age of 40,000 yrs . with the exception of the explosion energy ( which is probably unrealistically large anyway ) , these parameters are moderately consistent with a picture in which g347.3@xmath00.5 is a moderate age snr which is in a low density environment . the characteristics of the unidentified x - ray source 1wgaj1713.4@xmath03949 are marginally consistent with those for a neutron star . the high x - ray to optical flux ratio virtually rules out a stellar counterpart but is consistent with the source being a neutron star . the lack of detected pulsations at current sensitivities does not provide a strong discriminator in the interpretation . the spectral characteristics of the blackbody model provide a curious scenario if accurate , however . the luminosity implied by the spectral fit is considerably higher than expected for hot polar cap models unless the neutron star is spinning with a period of @xmath111 ms . with the rather sparse spectrum , it is conceivable that an additional spectral component ( e.g. , softer blackbody emission from the entire surface ) is contributing to the somewhat high flux . alternatively , as mentioned earlier , the source could be an extragalactic background object unrelated to the snr . more sensitive x - ray measurements are needed to address this . if 1wgaj1713.4@xmath03949 is indeed a neutron star associated with g347.3@xmath00.5 , then g347.3@xmath00.5 differs from sn 1006 in a very fundamental way in that the latter is the remnant of a type ia explosion . association with the molecular cloud seen in co would lead to a similar conclusion ; the remnant presumably then originated from a high mass progenitor which , in its short lifetime , did not migrate far from its birthplace . models for cosmic ray production from shell - type remnants would thus need to accommodate the very different environments expected from the two types of progenitors . in this case , we expect the evolution of g347.3@xmath00.5 to have initially been dominated by the effects of a precursor wind cavity . if our preferred distance of 6 kpc , as suggested by the apparent association with the molecular cloud complex , is correct , then the remnant now has a radius of over 50 pc which is still well within the expected size of the stellar wind bubble for massive o - type stars ( chevalier & liang 1989 ) . this could provide a logical explanation for both the low density inferred from the lack of thermal x - ray emission as well as the presence of a molecular cloud interaction site . the strong nonthermal x - ray emission observed from g347.3@xmath00.5 provides tantalizing evidence for the second case of a direct signature of cosmic ray acceleration in snrs . like sn 1006 , this remnant is very faint in the radio . the peak radio surface brightness of sn 1006 is @xmath112 at the same frequency and resolution , which is comparable to arc 2 of g347.3@xmath00.5 . however , in distinct contrast , g347.3@xmath00.5 shows no evidence of the thermal x - ray emission typically associated with ism and ejecta material heated by the snr shock , nor is the morphology as distinctly shell - like as sn 1006 . we have shown that nonthermal x - ray emission pervades the entire image of g347.3@xmath00.5 . faint radio emission appears to delineate a nearly complete shell with brighter arc - like features along regions of bright x - ray emission . co observations suggest that these features coincide with a region in which the snr may be interacting with a molecular cloud found along the line of sight . several lines of evidence , including an association with this molecular cloud complex as well as an region , suggest that the most appropriate distance to the remnant is 6 kpc . the lack of thermal emission sets a strong limit on the mean density around the remnant : @xmath113 . although this is rather low for the near vicinity of a molecular cloud , we argue that it means the majority of the remnant is evolving in the low density interior of the large circumstellar cavity driven by the wind of its massive progenitor star . simple sedov - phase models for the remnant s evolutionary state can be found that are consistent with this low density environment for reasonable values of the age and explosion energy . however in the absence of an actual measurement of the thermal properties of the remnant , any conclusions about the evolutionary state of the remnant remain premature . in the x - ray bright region toward the northwest edge of g347.3@xmath00.5 , where an interaction with denser gas may be occurring , the limits on the ambient density are higher @xmath114 and broadly consistent with the densities estimated around other snrs that are associated with molecular clouds ( e.g. , n132d in the large magellanic cloud , see discussion in hughes , hayashi , and koyama 1998 ) . regardless of its precise evolutionary state , our observations reveal g347.3@xmath00.5 to be the most extreme example known of a nonthermal shell - type x - ray remnant and may indicate that shock acceleration of cosmic rays by supernova remnants can occur in a diversity of interstellar environments . additional measurements will help shed light on the nature of g347.3@xmath00.5 . in particular , a search for oh maser emission ( frail et al . 1996 ) along the nw limb could confirm a molecular cloud interaction there . optical observations of this region could also provide evidence of such an interaction , but the extinction is rather high making such observations difficult . deeper x - ray observations which might reveal and characterize the thermal component are needed , as are additional observations of the centrally - located source in order to assess the potential for association as the compact relic of the explosion which produced g347.3@xmath00.5 . while the x - ray spectrum of the snr clearly indicates the presence of very high energy electrons , and by inference the acceleration of cosmic rays , further study is required to provide a sufficiently constrained picture of g347.3@xmath00.5 to which acceleration models can be compared . higher - energy x - ray observations to search for any change in the spectral index which might be associated with a bremsstrahlung component will be important in this regard , as will higher - resolution spatial and spectral observations to provide additional information on the shock structure , and more sensitive observations to search for thermal emission . we gratefully acknowledge t. handa and t. hasegawa of the university of tokyo who kindly shared their co(21 ) observations in the vicinity of g347.3@xmath00.5 prior to publication . the most is operated by the university of sydney with support from the australian research council and the science foundation for physics within the university of sydney . bmg acknowledges the support of an australian postgraduate award . this work was supported in part by the national aeronautics and space administration through contract nas8 - 39073 and grant nag5 - 4803 . jph acknowledges support through nasa grants nag 5 - 4794 and nag 5 - 4871 .

random privacy amplification based on the universal@xmath3 condition@xcite has been studied by many authors@xcite . this technique is originally aimed for random number extraction@xcite . it can be applied to secret key distillation ( agreement ) with public communication@xcite and wire - tap channel@xcite , which treats the secure communication in the presence of an eavesdropper . ( for details of its application , see e.g. hayashi @xcite . ) when random privacy amplification is implemented by a universal@xmath3 hash function , it can yield protocols for the above tasks with a relatively small amount of calculation . similar to the study @xcite for random privacy amplification based on the universal@xmath3 condition , hayashi@xcite focused only on the mutual information with the eavesdropper . however , as the secrecy criterion , many papers in cryptography community @xcite dopt the half of the @xmath0 norm distance , which is also called the variation distance or eve s distinguishability . because this criterion is closely related to universally composable security @xcite , it is required to evaluate the leaked information based on the @xmath0 norm distance from cryptography community viewpoint . in this paper , we adopt the @xmath0 norm distance as the secrecy criterion , and evaluate the secrecy for random privacy amplification while hayashi@xcite adopts the mutual information criterion . in the independent and identical distributed case , when the rate of generating random numbers is smaller than the entropy of the original information source , it is possible to generate the random variable whose @xmath0 norm distance to the uniform random number approaches zero asymptotically . however , in the real setting , we can manipulate only a finite size of random variables . so , the speed of this convergence is very important . in the community of information theory , in order to discuss the speed , we often focus on the the exponential rate of decrease . this rate is called the exponent , and is widely discussed among several topics in information theory , e.g. , channel coding@xcite , source coding@xcite , and mutual information criterion in wire - tap channel@xcite . however , the exponent has not been discussed in the community of cryptography as an important criterion . the purpose of this paper is establishing a systematic evaluating method for exponent for the @xmath0 norm distance in secure protocols . in subsection [ s2a ] , for a given generating rate and a given source distribution , in the i.i.d . setting , we derive a lower bound of the exponent of the average of the @xmath0 norm distance between the generated random number and the uniform random number when universal@xmath3 hash function is applied . then , we introduce a stronger condition for hash functions , which is called strongly universal@xmath3 function , and show that our lower bound is tight under this condition . since our bound realizes the optimal exponent , it is thought to be powerful even for the finite length setting . we consider the @xmath4-independent and identical extension , and show that the exponential rate of decrease for this bound is tight under a stronger condition by using the type method@xcite . however , if our protocol generating the random number depends on the original distribution , there is a possibility to improve the exponent while it is known that asymptotic generation can not be improved@xcite . in subsection [ s3 ] , we derive the optimal exponent in this setting by using the cramer s theorem @xcite and the type method @xcite . comparing these two exponents , we can compare the performances between the protocol based on universal@xmath3 hash function and the protocol depending on the information source . in information theory , the part constructing the protocol achieving the rate is called the direct part . and the proof for non - existence surpassing the rate is called the converse part . concerning the exponent of the @xmath0 norm distance with universal@xmath3 hash function , the direct part is relatively simple while the converse part is more complicated . the converse part is regarded as an important topic in information theory , while it has been regarded as a minor topic in other research areas . in the direct part , we focus on bennett et al@xcite s evaluation for random privacy amplification , which employs the rnyi entropy of order 2 . this evaluation was also obtained by hastad et al @xcite and is often called leftover hash lemma . using the same discussion as renner @xcite , we derive an upper bound for the @xmath0 norm distance . renner @xcite applied @xmath5 approximation idea ( smoothing ) to min - entropy , which bounds the rnyi entropy of order 2 . in this paper , we apply the same idea to the rnyi entropy of order 2 . however , for the evaluation of the smoothing error , we use a method different from the method invented by holenstein - renner@xcite , but our method is standard in large deviation theory@xcite . in fact , as is shown in subsection [ s2b ] , our evaluation is much better than that by holenstein - renner@xcite . then , we obtain a bound based on the rnyi entropy of order @xmath1 as the main theorem ( theorem [ thm1 ] ) . note that , hayashi@xcite obtained a similar result using the rnyi entropy of order @xmath1 , but it evaluates the shannon entropy not the @xmath0 norm distance . contrastly , in the converse part , we employ type method , which was invented by csiszr and krner @xcite . the discussion of this part is quite technical , but conventional in information theory . concerning the exponent of the protocol depending on the original distribution , the direct part can be shown by the combination of a new invented lemma ( lemma [ lemdi ] ) and cramer s theorem in large deviation theory@xcite . the converse part can be shown by very technical application of the type method@xcite . in section [ s42 ] , we consider the case when an eavesdropper has a random variable correlated to the random variable of the authorized user . in this case , when the authorized user applies universal@xmath3 hash function to his random variable , he obtain a secure random variable . when we apply theorem [ thm1 ] to the security by @xmath0 norm distance in this setting , we obtain a tighter evaluation than existing evaluation than that directly obtained from hayashi@xcite . in section [ s2 ] , we focus on wire - tap channel model , whose capacity has been calculated by wyner @xcite and csiszr and krner @xcite . csiszr @xcite showed the strong security , and hayashi @xcite derived bounds for both exponential rates of decrease for the mutual information between alice and eve and the security criterion based on the @xmath0 norm distance , which is simply called the @xmath0 security criterion . he obtained a bound for the exponential rate of decrease concerning the @xmath0 security criterion . however , their approaches are based on completely random coding . using the method of information spectrum@xcite , hayashi @xcite showed strong security when universal@xmath3 hash function is applied in the privacy amplification process of wire - tap channel . but , the analysis on @xcite was based on the mutual information not the @xmath0 security criterion . in the paper@xcite , he invented a systematic method to construct a wire - tap channel code based on universal@xmath3 hash function , and evaluated the performance based on a privacy amplification theorem concerning the shannon entropy given in @xcite . so , we can expect that theorem [ thm1 ] plays the same role in evaluation of the @xmath0 security criterion in wire - tap channel model as a privacy amplification theorem concerning the shannon entropy given in @xcite . in this paper , we apply theorem [ thm1 ] to wire - tap channel model , and obtain the evaluation of the exponent of the @xmath0 security criterion . in fact , we can evaluate the exponent of the @xmath0 security criterion from the evaluation for exponent concerning the mutual information in @xcite by applying the pinsker inequality to the mutual information . it is also shown in section [ s6 - 1 ] , that the evaluation obtained in this paper is better than that by @xcite and that obtained by application of pinsker inequality . in section [ s6 ] , the same discussion as sections v and vi in @xcite can be applied to the @xmath0 security criterion with a suitable modification . in the additive noise case , the obtained bound can be attained by linear operation . further , we obtain the bound for the @xmath0 security criterion in one - way secret key distillation by the same discussion as sections vi in @xcite . in appendix [ as2 ] , we prove theorem [ thm2 ] mentioned in subsection [ s2a ] . in appendix [ al9 - 9 - 1 ] , we prove lemma [ l9 - 9 - 1 ] given in subsection [ s3 ] . firstly , we consider the uniform random number generation problem from a biased random number @xmath6 , which obeys a probability distribution @xmath7 when its cardinality @xmath8 is finite . there are two types protocols for this problem . one is a protocol specialized for the given distribution @xmath7 . the other is a universal protocol that does not depends on the given distribution @xmath7 . the aim of this section is evaluate the performance of the latter setting . in the latter setting , our protocol is given by a function @xmath9 from @xmath10 to @xmath11 . in order to evaluate the protocol @xmath9 , we use the @xmath0 distance ( the variational distance ) @xmath12 and the @xmath13 distance @xmath14 these definitions can be extended when the total measure is less than @xmath15 i.e. , @xmath16 . in the following , we call such @xmath7 a sub - distribution . this extension for sub distributions is crucial for the later discussion . then , the quality of the random number obeying the distribution @xmath7 is evaluated by @xmath17 where @xmath18 is the uniform distribution on @xmath10 . we also use the quantitty @xmath19 and the rnyi entropy order @xmath1 : @xmath20 the @xmath13 distance is written by using the rnyi entropy order @xmath21 as follows . @xmath22 now , we focus on an ensemble of the functions @xmath23 from @xmath10 to @xmath24 , where @xmath25 denotes a random variable describing the stochastic behavior of the function @xmath23 . an ensemble of the functions @xmath23 is called universal@xmath3 hash when it satisfies the following condition@xcite : @xmath26 , the collision probability that @xmath27 is at most @xmath28 . we sometimes require the following additional condition : for any @xmath25 , the cardinality of @xmath29 does not depend on @xmath30 . this condition will be used in section iii . indeed , when the cardinality @xmath8 is a power of a prime power @xmath31 and @xmath32 is another power of the same prime power @xmath31 , as is shown in appendix ii of hayashi @xcite , the ensemble @xmath33 can be given by the concatenation of toeplitz matrix and the identity @xmath34@xcite only with @xmath35 random variables taking values in the finite field @xmath36 . that is , the function can be obtained by the multiplication of the random matrix @xmath34 taking values in @xmath36 . in this case , condition [ c12 ] can be confirmed because the rank of @xmath34 is constant . for condition [ c1 ] , see appendix ii of hayashi @xcite . bennett et al@xcite essentially showed the following lemma . a universal@xmath3 function @xmath23 satisfies @xmath37 this was also shown by hastad et al @xcite and is often called leftover hash lemma . now , we follow the derivation of theorem 5.5.1 of renner @xcite when one classical random variable is given . the schwarz - inequality implies that @xmath38 the jensen inequality yields that @xmath39 substituting ( [ 5 - 14 - 3 ] ) and ( [ 5 - 14 - 4 ] ) into the above inequality , we obtain @xmath40 using ( [ 5 - 14 - 6 ] ) , we can show the following theorem as a generalization of ( [ 5 - 14 - 6 ] ) . a universal@xmath3 function @xmath23 satisfies @xmath41 substituting @xmath42 , we obtain @xmath43 since the difference between ( [ 5 - 14 - 6 ] ) and ( [ 5 - 14 - 7 ] ) is only the coefficient , theorem [ thm1 ] can be regarded as a kind of generalization of bennett et al@xcite s result ( [ 5 - 14 - 4 ] ) . for any @xmath44 , we choose subset @xmath45 , and define the subdistribution @xmath46 by @xmath47 since @xmath48 , @xmath49 the inequality ( [ 5 - 14 - 6 ] ) yields @xmath50 for @xmath51 , we can evaluate @xmath52 and @xmath53 as @xmath54 combining ( [ 5 - 14 - 10 ] ) , ( [ 5 - 14 - 8 ] ) , and ( [ 5 - 14 - 9 ] ) , for @xmath55 , we obtain @xmath56 where we substitute @xmath57 into @xmath58 . next , we consider the case when our distribution @xmath59 is given by the @xmath4-fold independent and identical distribution of @xmath60 , i.e , @xmath61 . when the random number generation rate @xmath62 is @xmath63 , we focus on the _ exponential rate of decrease _ of @xmath64 , and consider the supremum . when an ensemble @xmath65 is a universal@xmath3 hash functions from @xmath66 to @xmath67 , theorem [ thm1 ] yields that @xmath68 on the other hand , when we apply the pinsker inequality@xcite to the upper bound for the mutual information obtained in hayashi@xcite , we obtain another bound @xmath69 , which is smaller than ( [ 4 - 19 - 2 ] ) . the right hand side of ( [ 4 - 19 - 2 ] ) is equal to the optimal exponential rate in the fixed - length source coding . the following lemma is known @xcite : assume that @xmath60 is a probability distribution and the random number generation rate @xmath63 is greater than the critical rate @xmath70 , where @xmath71 . using kullback - leibler divergence @xmath72 , we have the following expression . @xmath73 here the last equation of ( [ 9 - 7 - 1 ] ) can be checked in the following way . @xmath74 is equal to the minimum exponent for correct decoding probability when the compression rate is @xmath63@xcite . this minimum exponent is equal to @xmath75@xcite . hence , we obtain the last equation . in oder to show the tightness of the exponential rate decrease ( [ 9 - 7 - 1 ] ) under the universal@xmath3 condition , we consider the following ensemble . for any @xmath76 , @xmath77 . the random variable @xmath78 is independent of @xmath79 for different arbitrary two elements @xmath80 . under the strongly universal@xmath3 ensemble , and any subset @xmath81 with @xmath82 satisfies @xmath83 its proof is given in appendix [ as2 ] . using theorem [ thm2 ] and lemma [ lem9 ] with @xmath84 and applying the type method for fixed length source coding @xcite , we can show the following proposition : let @xmath85 be the set of empirical distributions on @xmath10 with @xmath4 trials . assume the following conditions : @xmath86 . when @xmath87 , any sequence of strongly universal@xmath3 ensemble @xmath88 from @xmath66 to @xmath67 satisfies the equation @xmath89 when @xmath90 , applying the type method for fixed length source coding @xcite to the random variable @xmath91 , we obtain @xmath92 using theorem [ thm2 ] with @xmath93 , we obtain @xmath94 thus , @xmath95 in particular , when @xmath87 , the equations @xmath96 hold . since @xmath87 , lemma [ lem9 ] yields that @xmath89 therefore , we can conclude that our exponential rate of decrease is tight for the strongly universal@xmath3 condition when @xmath87 . in the above derivation , the key point is evaluating @xmath97 in the @xmath4-i.i.d . setting , i.e. , @xmath98 . in the community of cryptography , they often use holenstein - renner @xcite evaluation for @xmath53 in the @xmath4-i.i.d . setting . they proved the following theorem . when @xmath99 , @xmath100 further , when @xmath101 and @xmath102 , @xmath103 when @xmath104 , the inequality yields the following evaluation . when @xmath105 , @xmath106 for even @xmath4 . our evaluation ( [ 5 - 14 - 9 ] ) of @xmath98 contains the parameter @xmath107 . since this parameter is arbitrary , it is natural to compare the upper bound @xmath108 given by ( [ 5 - 14 - 9 ] ) with that by theorem [ holenstein ] . that is , using ( [ 5 - 14 - 9 ] ) , we obtain the exponential evaluation @xmath109 while theorem [ holenstein ] yields that @xmath110 in this case , the upper bound is @xmath111 for @xmath101 and @xmath112 for @xmath113 . in fact , the probability @xmath97 is the key quantity in the method of information spectrum , which is a unified method in information theory@xcite . when the method of information spectrum is applied to the i.i.d . source , the probability @xmath97 is evaluated by applying cramr theorem ( see @xcite ) to the random variable @xmath114 . then , we obtain @xmath115 for @xmath116 . since @xmath117 is concave , when @xmath118 , the maximization can be attained in @xmath119 $ ] , i.e. , @xmath120 which implies that our evaluation ( [ 5 - 14 - 9 ] ) gives the tight bound for exponential rate of decrease for the probability @xmath98 . in fact , the difference among these bounds is numerically given in fig . [ f2 ] . therefore , we can conclude that our evaluation ( [ 5 - 14 - 9 ] ) is much better than that by holenstein - renner @xcite . that is , combination of lemma 1 and ( [ 5 - 14 - 9 ] ) is essential for deriving the tight exponential bound . next , we consider a function @xmath9 from @xmath10 to @xmath121 specialized to a given probability distribution @xmath7 . this problem is called intrinsic randomness , which was studied with general source by vembu and verd @xcite . hayashi@xcite discussed the relation between the second order asymptotic rate and central limit theorem . in the following , for the comparison with the exponential rate of decrease for ( [ 9 - 7 - 1 ] ) , we derive the optimal exponential rate of decrease for a given rate generating uniform random number . the following lemma is useful for this purpose . any probability distribution @xmath7 and any function @xmath9 from @xmath10 to @xmath121 satisfy that @xmath122 when @xmath123 is not empty , @xmath124 thus , we obtain @xmath125 which implies ( [ 9 - 8 - 1 ] ) . concerning the opposite inequality , we consider the case when our distribution @xmath59 is given by the @xmath4-fold independent and identical distribution of @xmath60 , i.e , @xmath61 . in this setting , we have the following lemma . for any probability distribution @xmath7 , there exists a function @xmath126 from @xmath66 to @xmath127 such that @xmath128 where @xmath85 is the set of empirical distribution with @xmath4 trials , and @xmath129 is the set of data with the empirical distribution @xmath130 . in the first step , we define the function @xmath126 . in the second step , we show that the function satisfies ( [ 9 - 9 - 1 ] ) . using the integer @xmath131 , we divide @xmath85 into three parts : @xmath132 the condition that @xmath133 is equivalent with the condition that @xmath134 for @xmath135 . hence , @xmath136 so , @xmath137 since @xmath138 we have @xmath139 therefore , we can choose @xmath140 on @xmath141 satisfying the following conditions . ( 1 ) for @xmath142 , @xmath143 . ( 2 ) @xmath144 injective for @xmath145 . ( 3 ) @xmath146 for @xmath147 . further , we choose @xmath140 satisfying the additional condition . ( 4 ) any type @xmath147 satisfies that @xmath148 for @xmath149 . then , for @xmath147 , we obtain @xmath150 next , we define @xmath126 on the whole set by modifying @xmath140 as follows . ( 5 ) @xmath126 is the same as @xmath140 on @xmath151 . ( 6 ) @xmath152 . therefore , our remaining task is to evaluate the value @xmath153 , where @xmath154_+,\\ [ x]_+&:=\left\ { \begin{array}{ll } x & \hbox { if } x \ge 0 \\ 0 & \hbox { if } x < 0 . \end{array } \right.\end{aligned}\ ] ] then , ( [ 2 - 2 - 6 ] ) implies that @xmath155 for @xmath156 , ( [ 2 - 2 - 1 ] ) implies @xmath157 for @xmath158 , @xmath159 combining ( [ 2 - 2 - 2 ] ) , ( [ 2 - 2 - 3 ] ) , and ( [ 2 - 2 - 4 ] ) , we obtain @xmath160 which implies ( [ 9 - 9 - 1 ] ) . using the above two lemmas , we obtain the following theorem . assume that @xmath161 . @xmath162 where the above minimum is taken among functions @xmath126 from @xmath163 to @xmath164 . combining lemma [ lem9 ] and theorems [ thm2 ] and [ t9 - 9 - 1 ] , we can compare the performances between random universal protocol and specialized protocol . so , our exponential rate of decrease for the protocol based on universal@xmath3 hash functions is slightly smaller than the optimal exponential rate of decrease for specialized protocols . using cramer s theorem@xcite , we obtain @xmath165 further , @xmath166 so , the remaining task is show that @xmath167 which can be shown from lemma [ l9 - 9 - 1 ] . ( both cases in lemma [ l9 - 9 - 1 ] are required for this derivation . ) when @xmath161 , @xmath168 when @xmath169 , @xmath170 its proof is given in appendix [ al9 - 9 - 1 ] . next , we consider the secure key generation problem from a common random number @xmath6 which has been partially eavesdropped on by eve . for this problem , it is assumed that alice and bob share a common random number @xmath6 , and eve has another random number @xmath171 , which is correlated to the random number @xmath172 . the task is to extract a common random number @xmath173 from the random number @xmath6 , which is almost independent of eve s random number @xmath171 . here , alice and bob are only allowed to apply the same function @xmath9 to the common random number @xmath6 . then , when the initial random variables @xmath174 and @xmath175 obey the distribution @xmath176 , eve s distinguishability can be represented by the following value : @xmath177 where @xmath178 is the product distribution of both marginal distributions @xmath179 and @xmath180 , and @xmath181 is the uniform distribution on @xmath121 . while the half of this value directly gives the probaility that eve can distingushes the alice s information , we call @xmath182 eve s distinguishability in the following . this criterion was proposed by @xcite and was used by @xcite . since the half of this quantity @xmath182 is closely related to universally composable security , we adopt it as the secrecy criterion in this paper . as another criterion , we sometimes treat @xmath183 since @xmath184 , we have @xmath185 further , @xmath179 is the uniform distribution , the above both criteria coincide with each other . in order to evaluate the average performance , we define the quantity @xmath186 thus , using theorem [ thm1 ] and putting @xmath187 , we obtain the inequality : @xmath188 for @xmath189 and any universal@xmath3 hash function @xmath33 . note that when eve s random variable @xmath175 takes a continuous value in the set @xmath190 , the relation ( [ 4 - 25 - 1 ] ) holds by defining @xmath191 in the following way . @xmath192 next , we consider the case when our distribution @xmath193 is given by the @xmath4-fold independent and identical distribution of @xmath194 , i.e , @xmath195 . ahlswede and csiszr @xcite showed that the optimal generation rate @xmath196 equals the conditional entropy @xmath197 . that is , the generation rate @xmath198 is smaller than @xmath197 . the quantity @xmath199 goes to zero . in order to treat the speed of this convergence , we focus on the supremum of the _ exponential rate of decrease ( exponent ) _ for @xmath199 for a given @xmath63 @xmath200 since the relation @xmath201 holds , the inequality ( [ 4 - 25 - 1 ] ) implies that @xmath202 when we apply pinsker inequality to the lower bound given in ( 5 ) of @xcite , we obtain @xmath203 where we define the function @xmath204 for @xmath2 $ ] . since @xmath205 , the right hand sides of ( [ 4 - 16 - 4 ] ) and ( [ 2 - 3 - 1 ] ) are strictly greater than @xmath15 for @xmath206 . concerning the comparison of both bounds , the following lemma holds . @xmath207 for @xmath206 . for example , we consider the following case : @xmath208 equals @xmath190 , the set @xmath208 has the module structure , and the conditional distribution @xmath209 has the form @xmath210 . then , both bounds can be simplified to @xmath211 in particular , the both exponents are numerically plotted in fig . [ f3 ] when @xmath212 , and @xmath213 , @xmath214 . we choose @xmath215 . then , applying jensen inequality to the concave function @xmath216 , we have @xmath217 where @xmath218 . thus , obtain ( [ 2 - 4 - 2 ] ) . next , we consider the wire - tap channel model , in which the eavesdropper ( wire - tapper ) , eve and the authorized receiver bob receive information from the authorized sender alice . in this case , in order for eve to have less information , alice chooses a suitable encoding . this problem is formulated as follows . let @xmath219 and @xmath220 be the probability spaces of bob and eve , and @xmath221 be the set of alphabets sent by alice . then , the main channel from alice to bob is described by @xmath222 , and the wire - tapper channel from alice to eve is described by @xmath223 . in this setting , in order to send the secret message in @xmath121 subject to the uniform distribution , alice chooses @xmath32 distributions @xmath224 on @xmath221 , and she generates @xmath225 subject to @xmath226 when she wants to send the message @xmath227 . bob prepares @xmath32 disjoint subsets @xmath228 of @xmath219 and judges that a message is @xmath30 if @xmath229 belongs to @xmath230 . therefore , the triplet @xmath231 is called a code , and is described by @xmath232 . its performance is given by the following three quantities . the first is the size @xmath32 , which is denoted by @xmath233 . the second is the average error probability @xmath234 : @xmath235 and the third is eve s distinguishability @xmath236 : @xmath237 ) \\ w^e_{\phi}(e ) : = & \sum_i \frac{1}{m } w_{q_i}^e(e ) , \quad w^e[\phi](i , e ) : = \frac{1}{m } w_{q_i}^e(e).\end{aligned}\ ] ] the quantity @xmath236 gives an upper bound for the probability that eve can succeed in distinguishing whether alice s information belongs to a given subset . so , the value can be regarded as eve s distinguishability . in order to calculate these values , we introduce the following quantities . @xmath238 where @xmath239 . the following lemma gives the property of @xmath240 . when the random variable @xmath241 takes a continuous value in the set @xmath242 while @xmath243 takes discrete value , the above definition can be changed to @xmath244 these definitions do not depend on the choice of the measure on @xmath242 . that is , when @xmath245 for a positive function @xmath9 , @xmath246 the function @xmath247 is convex for @xmath248 $ ] , and is concave for @xmath249 $ ] . see lemma 1 of hayashi @xcite . then , we obtain the following theorem . there exists a code @xmath232 for any integers @xmath250 , and any probability distribution @xmath251 on @xmath221 such that @xmath252 and @xmath253 we can construct the code satisfying the above conditions by the same way as section iii of @xcite . this theorem can be shown by replacing the concavity of @xmath254 and ( 3 ) in @xcite by the concavity of @xmath255 and the inequality ( [ 4 - 25 - 1 ] ) , respectively if we take care of the following point . when @xmath251 is the uniform distribution on the set @xmath256 and the joint distribution @xmath257 is given by @xmath258 , the equations @xmath259 hold . in the discussion in section iii of @xcite , the main point is ( 12 ) of @xcite , which is essentially equivalent with ( [ 12 - 26 - 1-a ] ) . in the proof of theorem [ 3 - 6 ] , we need to modify ( 12 ) of @xcite by using ( [ 12 - 26 - 1 ] ) . in the @xmath4-fold discrete memoryless channels @xmath260 and @xmath261 of the channels @xmath262 and @xmath263 , the additive equation @xmath264 holds . thus , there exists a code @xmath265 for any integers @xmath266 , and any probability distribution @xmath251 on @xmath221 such that @xmath267 and @xmath268 since @xmath269 , the rate @xmath270 can be asymptotically attained . therefore , when the sacrifice information rate is @xmath63 , i.e. , @xmath271 , the exponential rate for decrease of eve s distinguishability is greater than @xmath272 . in the @xmath4-fold discrete memoryless channels @xmath260 and @xmath261 of the channels @xmath262 and @xmath263 , the additive equation @xmath264 holds . thus , there exists a code @xmath265 for any integers @xmath266 , and any probability distribution @xmath251 on @xmath221 such that @xmath267 and @xmath268 since @xmath269 , the rate @xmath270 can be asymptotically attained . therefore , when the sacrifice information rate is @xmath63 , i.e. , @xmath271 , the exponential rate of decrease for eve s distinguishability is greater than @xmath272 . now , we compare the lower obtained bound @xmath273 for the exponential rate of decrease for eve s distinguishability with existing lower bounds @xcite . as existing lower bounds , hayashi @xcite derived a lower bound of this exponential rate of decrease @xmath274 . hayashi @xcite also derived a lower bound @xmath275 of this exponential rate of decrease for the mutual information . application of pinsker inequality to this bound yields the bound @xmath276 , which is smaller than our lower bound @xmath277 because @xmath278 for @xmath51 . in the following , we compare the two bounds @xmath273 and @xmath277 for this purpose , we treat @xmath279 and @xmath280 for @xmath189 . reverse hlder inequality @xcite with the measurable space @xmath281 is given as @xmath282 for @xmath283 . using this inequality , we obtain @xmath284 w_p(y)^{-s } \\ \ge & \left ( \sum_y \left [ \sum_x p(x ) ( w_x(y))^{1+s } \right]^{\frac{1}{1+s } } \right)^{1+s } \cdot \left ( \sum_y w_p(y)^{-s\cdot -\frac{1}{s } } \right)^{-s } \\ = & \left ( \sum_y \left [ \sum_x p(x ) ( w_x(y))^{1+s } \right]^{\frac{1}{1+s } } \right)^{1+s}.\end{aligned}\ ] ] substituting @xmath285 , we obtain @xmath286 w_p(y)^{\frac{-t}{1-t } } \\ \ge & \left ( \sum_y \left [ \sum_x p(x ) ( w_x(y))^{\frac{1}{1-t } } \right]^{1-t } \right)^{\frac{1}{1-t}},\end{aligned}\ ] ] which implies @xmath287 w_p(y)^{\frac{-t}{1-t } } \right)^{1-t } \\ \ge & \sum_y \left [ \sum_x p(x ) ( w_x(y))^{\frac{1}{1-t } } \right]^{1-t } = e^{\phi(t|w^e , p)}.\end{aligned}\ ] ] thus , our bound @xmath273 for the exponential rate of decrease is better than the existing bound @xmath277 @xcite . assume that @xmath288 . we consider the following channel . @xmath289 when @xmath290 , @xmath291 then , the three bounds @xmath292 , @xmath293 , and @xmath294 with @xmath295 are numerically compared as in fig . [ f4 ] . next , we consider a more specific case . when @xmath296 and @xmath297 is a module and @xmath298 , the channel @xmath299 is called _ any additive channel @xmath263 satisfies @xmath300 for the uniform distribution @xmath301 on @xmath297 because @xmath302 we consider a more general case . eve is assumed to have two random variables @xmath303 and @xmath304 . the first random variable @xmath305 is the output of an additive channel depending on the second variable @xmath304 . that is , the channel @xmath306 can be written as @xmath307 , where @xmath308 is a joint distribution . hereinafter , this channel model is called a general additive channel . this channel is also called a regular channel@xcite . for this channel model , the equalities @xmath309 hold because @xmath310 furthermore , in a practical sense , we need to take into account the decoding time . for this purpose , we often restrict our codes to linear codes . in the following , we consider the case where the sender s space @xmath221 has the structure of a module . here , we should remark that hayashi @xcite considered the completely random ensemble of codes . he did not consider universal@xmath3 condition . so , we can not derive the following discussion from his result @xcite while his bound for exponential rate is the same as our bound when eve s channel is additive or general additive . such a required code can be constructed by the same discussion as section v of @xcite . when we generate codes @xmath232 by the same method as in @xcite , the average of @xmath236 is bounded by @xmath311 . this result can be applied to one - way secret key distillation by the same way as section vi of @xcite . in this problem , alice , bob , and eve are assumed to have initial random variables @xmath312 , @xmath313 , and @xmath314 , respectively . the task for alice and bob is to share a common random variable almost independent of eve s random variable @xmath315 by using a public communication . for this purpose , we assume that alice and bob can perform local data processing in the both sides and alice can send messages to bob via public channel . that is , only one - way communication is allowed . we call such a combination of these operations a code and denote it by @xmath232 . the quality is evaluated by three quantities : the size of the final common random variable , the probability that their final variables coincide , and eve s distinguishability @xmath236 of the final joint distribution between alice and eve . then , we can construct the same code @xmath232 as given in section vi of@xcite for any numbers @xmath32 and @xmath316 . it follows from the similar discussion that this code satisfies that @xmath317 and @xmath318 we have derived the tight evaluation for exponent for the average of the @xmath0 norm distance between the generated random number and the uniform random number when universal@xmath3 hash function is applied and the key generation rate is less than the critical rate @xmath319 . using this evaluation , we have obtained an upper bound for eve s distinguishability in secret key generation from a common random number without communication when a universal@xmath3 hash function is applied . since our bound is based on the rnyi entropy of order @xmath1 for @xmath2 $ ] , it can be regarded as an extension of bennett et al @xcite s result with the rnyi entropy of order 2 . applying this bound to the wire - tap channel , we obtain an upper bound for eve s distinguishability , which yields an exponential upper bound . this exponent improves on the existing exponent @xcite . further , when the error correction code is given by a linear code and when the channel is additive or general additive , the privacy amplification is given by a concatenation of toeplitz matrix and the identity matrix . this method can be applied to secret key distillation with public communication . the author is grateful to professors ryutaroh matsumoto and takeshi koshiba for a helpful comments . this research was partially supported by a mext grant - in - aid for young scientists ( a ) no . the centre for quantum technologies is funded by the singapore ministry of education and the national research foundation as part of the research centres of excellence programme . first , for a fixed element @xmath320 , we consider the condition @xmath321 we denote the probability that this condition holds and the expectation under this condition by @xmath322 and @xmath323 . then , @xmath324 and @xmath325 now , using the function @xmath342 , we make a code for the wire - tap channel based on the random coding method . for this purpose , we make a protocol to share a random number . first , we generate the random code @xmath343 with size @xmath344 , which is described by the @xmath344 independent and identical random variables @xmath345 subject to the distribution @xmath251 on @xmath221 . for integers @xmath346 and @xmath347 , let @xmath348 be the maximum likelihood decoder of the code @xmath343 . gallager @xcite showed that the ensemble expectation of the average error probability concerning decoding the input message @xmath174 is less than @xmath349 for @xmath350 . after sending the random variable @xmath174 taking values in the set with the cardinality @xmath351 , alice and bob apply the above universal@xmath3 function @xmath23 to the random variable @xmath174 and generate another piece of data of size @xmath32 . here , we assume that the ensemble @xmath352 satisfies condition [ c12 ] . then , alice and bob share random variable @xmath353 with size @xmath32 . this protocol is denoted by @xmath354 let @xmath175 be the random variable of the output of eve s channel @xmath263 . for simplicity , we simplify the uniform distribution @xmath355 on the message space @xmath356 and the distribution @xmath357 on @xmath221 by @xmath358 and @xmath359 . therefore , since the random variable @xmath174 obeys the uniform distribution @xmath358 , the joint distribution @xmath360 concerning @xmath174 and @xmath175 satisfies @xmath361 for a given code @xmath343 , we apply the inequality ( [ 4 - 25 - 1 ] ) to eve s distinguishability . then , @xmath362 for @xmath363 . the concavity of @xmath279 ( lemma [ l1 ] ) guarantees that @xmath364 for @xmath363 . now , we make a code for wire - tap channel by modifying the above protocol @xmath354 . first , we choose the distribution @xmath226 to be the uniform distribution on @xmath29 . when alice wants to send the secret message @xmath30 , before sending the random variable @xmath174 , alice generates the random number @xmath174 subject to the distribution @xmath226 . alice sends the random variable @xmath174 . bob recovers the random variable @xmath174 , and applies the function @xmath23 . then , bob decodes alice s message @xmath30 , and this code for wire - tap channel @xmath365 is denoted by @xmath366 . since the ensemble @xmath352 satisfies condition [ c12 ] and the secret message @xmath30 obeys the uniform distribution on @xmath367 , this protocol @xmath366 has the same performance as the above protocol @xmath354 . finally , we consider what code is derived from the above random coding discussion . using the markov inequality , we obtain @xmath368 therefore , the existence of a good code is guaranteed in the following way . that is , we give the concrete performance of a code whose existence is shown in the above random coding method . we choose @xmath326 such that @xmath327 , where @xmath328 . when @xmath130 satisfies @xmath329 , @xmath330 hence , @xmath331 the last equation follows from the concavity of @xmath332 concerning @xmath333 . assume that @xmath161 . then , @xmath334 . when @xmath335 , @xmath336 which implies ( [ 9 - 9 - 4 ] ) . assume that @xmath169 . when @xmath335 , @xmath337 further , when @xmath338 , @xmath339 which implies ( [ 9 - 9 - 5 ] ) . further , the concavity of @xmath340 and the condition @xmath169 imply that @xmath341 . thus , we obtain ( [ 2 - 2 - 8 ] ) . first , we regard a submodule @xmath369 as an encoding for the usual sent message , and focus on its decoding @xmath370 by the authorized receiver . we construct a code for a wire - tap channel @xmath371}\}_{[x]\in c_1/c_2 } , \{\cd_{[x]}\}_{[x]\in c_1/c_2})$ ] based on a submodule @xmath372 of @xmath373 as follows . the encoding @xmath374}$ ] is given as the uniform distribution on the coset @xmath375:=x+c_2 $ ] , and the decoding @xmath376}$ ] is given as the subset @xmath377 . next , we assume that a submodule @xmath378 of @xmath373 with cardinality @xmath379 is generated by a random variable @xmath25 satisfying the following condition . any element @xmath380 is included in @xmath378 with probability at most @xmath381 . then , the performance of the constructed code is evaluated by the following theorem . choose the subcode @xmath378 according to condition [ c3 ] . we construct the code @xmath382 by choosing the distribution @xmath374}$ ] to be the uniform distribution on @xmath375 $ ] for @xmath375\in c_1/c_2({{\bf x}})$ ] . then , we obtain @xmath383 where @xmath384 is the uniform distribution on the subset @xmath385 . this inequality can be shown by ( [ 4 - 25 - 1 ] ) as follows . now , we define the joint distribution @xmath386 . the choice of @xmath374}$ ] corresponds to a hashing operation satisfying condition 1 . then , ( [ 4 - 25 - 1 ] ) yields that @xmath387 is bounded by @xmath388 , which implies ( [ 4 - 27 - 1 ] ) . next , we consider a special class of channels . when the channel @xmath263 is additive , i.e. , @xmath389 , the equation @xmath390 holds for any @xmath391 . thus , the concavity of @xmath279 ( lemma [ l1 ] ) implies that @xmath392 thus , combining ( [ 4 - 27 - 1 ] ) , ( [ 2 - 13 - 1 ] ) , and ( [ 12 - 26 - 2 ] ) , we obtain @xmath393 for @xmath394 . similarly , when the channel @xmath263 is general additive , i.e. , @xmath395 , combining ( [ 4 - 27 - 1 ] ) , ( [ 2 - 13 - 1 ] ) , and ( [ 5 - 14 - 1 ] ) , we obtain @xmath396 for @xmath394 . in the following discussion , we assume that @xmath221 is an @xmath4-dimensional vector space @xmath397 over the finite field @xmath36 . in this assumption , as is mentioned in @xcite , the bound can be attained by combination of linear code and the concatenation of toeplitz matrix and the identity @xmath34 of the size @xmath398 . next , we explain the detail of construction of one - way secret key distillation protocol . first , alice generates another uniform random variable @xmath391 and sends the random variable @xmath399 . then , the distribution of the random variables @xmath400 ( @xmath401 ) accessible to bob ( eve ) can be regarded as the output distribution of the channel @xmath402 ( @xmath403 ) . the channels @xmath404 and @xmath405 are given as follows . @xmath406 where @xmath407 ( @xmath408 ) is the joint probability between alice s initial random variable @xmath172 and bob s ( eve s ) initial random variable @xmath409 ( @xmath315 ) . hence , the channel @xmath263 is general additive . in particular , when @xmath221 is an @xmath4-dimensional vector space @xmath397 over the finite field @xmath36 and the joint distribution between @xmath174 and @xmath413(@xmath175 ) is the @xmath4-fold independent and identical distribution ( i.i.d . ) of @xmath414 ( @xmath176 ) , respectively , the relations @xmath415 and @xmath416 hold . thus , there exists a code @xmath265 for any integers @xmath266 , and any probability distribution @xmath251 on @xmath221 such that @xmath267 and @xmath417 hence , the achievable rate of this protocol is equal to @xmath418 which was obtained by maurer@xcite and ahlswede - csiszr@xcite . here , since the channels @xmath262 and @xmath263 can be regarded as general additive , we can apply the discussion in section [ s3 ] . that is , the bound ( [ 2 - 13 - 2 ] ) can be attained with the combination of a linear code and random privacy amplification , which is given in section [ s3 ] . u. maurer and s. wolf , `` infromation - theoretic key agreement : from weak to strong secrecy for free , '' _ advances in cryptology eurocrypt 2000 _ , lecture notes in computer science , vol.1807 , pp.351 - 368 , springer - verlag ( 2000 ) . r. renner and s. wolf , `` simple and tight bounds for information reconciliation and privacy amplification , '' _ asiacrypt 2005 _ , lecture notes in computer science , springer - verlag , vol . 3788 , pp . 199 - 216 , 2005 . s. watanabe , t. saitou , r. matsumoto , t. uyematsu `` strongly secure privacy amplification can not be obtained by encoder of slepian - wolf code , '' _ proceedings of the 2009 ieee international symposium on information theory _ , volume 2 , seoul , korea , pp . 1298 - 1302 ( 2009 ) ( arxiv:0906.2582 ) a. winter , a. c. a. nascimento , and h. imai , `` commitment capacity of discrete memoryless channels , '' _ proc . 9th cirencester crypto and coding conf . _ , lncs 2989 , pp 35 - 51 , springer , berlin 2003 ; cs.cr/0304014 ( 2003 ) m. hayashi , `` general non - asymptotic and asymptotic formulas in channel resolvability and identification capacity and its application to wire - tap channel , '' _ ieee trans . inform . theory _ , vol . * 52*(4 ) , 15621575 , 2006 . kuptsov , `` holder inequality '' , in hazewinkel , michiel , encyclopaedia of mathematics , springer , ( 2001 ) . t. holenstein and r. renner , `` on the randomness of independent experiments , '' arxiv : cs/0608007 ( 2006 )

in the hot field of computer simulation of biological macromolecules , available potential energy functions are often not accurate enough to properly describe complex processes such as the folding of proteins @xcite . in order to improve the situation , it is convenient to extract ab initio information from quantum mechanical calculations with the hope of being able to devise less computationally demanding methods that can be used to tackle large systems . in this spirit , the effective potential for the nuclei calculated in the non - relativistic born - oppenheimer approximation is typically considered as a good reference to assess the accuracy of cheaper potentials @xcite . the study of molecules at this level of theoretical detail and the design of computationally efficient approximations for solving the demanding equations that appear constitute the major part of the field called _ quantum chemistry _ @xcite . in this work , we voluntarily circumscribe ourselves to the basic formalism needed for the ground - state quantum chemical calculations that are typically performed in this context . for more general expositions , we refer the reader to any of the thorough accounts in refs . @xcite . in sec . [ sec : qc_molecular_h ] , we introduce the molecular hamiltonian and a special set of units ( the atomic ones ) that are convenient to simplify the equations . in sec . [ sec : qc_born - oppenheimer ] , we present in an axiomatic way the concepts and expressions related to the separation of the electronic and nuclear problems in the born - oppenheimer scheme . in sec . [ sec : qc_variational ] , we introduce the variational method that underlies the derivation of the basic equations of the hartree and hartree - fock approximations , discussed in sec . [ sec : qc_hartree ] and [ sec : qc_hartree - fock ] respectively . the computational implementation of the hartree - fock approximation is tackled in sec . [ sec : qc_roothaan ] , where the celebrated roothaan - hall equations are derived . in sec . [ sec : qc_basis_sets ] , the main issues related to the construction and selection of gaussian basis sets are discussed , and , finally , in sec . [ sec : qc_modern_developments ] , the hottest areas of modern research are briefly reviewed and linked to the issues in the rest of the work , with a special emphasis in the development of linear - scaling methods . since 1960 , the international scientific community has agreed on an ` official ' set of basic units for measurements : _ le ystme nternational dunits _ , or si for short ( see ` http://www.bipm.org/en/si/ ` and ref . @xcite ) . the meter ( m ) , the kilogram ( kg ) , the second ( s ) , the ampere ( a ) , the kelvin ( k ) , the mole ( mol ) , the joule ( j ) and the pascal ( pa ) are examples of si units . sticking to the si scheme , the non - relativistic quantum mechanical hamiltonian operator of a molecule consisting of @xmath0 nuclei ( with atomic numbers @xmath1 and masses @xmath2 , @xmath3 ) and @xmath4 electrons ( i.e. , the _ molecular hamiltonian _ ) is expressed as note that the non - relativistic molecular hamiltonian does not depend on spin - like variables . ] : @xmath5 where @xmath6 stands for @xmath7 , being @xmath8 planck s constant , @xmath9 denotes the electron mass , @xmath10 the proton charge , @xmath11 the position of the @xmath12-th electron , @xmath13 that of the @xmath14-th nucleus , @xmath15 the vacuum permittivity and @xmath16 the laplacian operator with respect to the coordinates of the @xmath12-th particle . although using a common set of units presents obvious communicative advantages , when circumscribed to a particular field of science , it is common to appeal to non - si units in order to simplify the most frequently used equations by getting rid of some constant factors that always appear grouped in the same ways and , thus , make the numerical values in any calculation of the order of unity . in the field of quantum chemistry , _ atomic units _ ( see table [ tab : atomic_units ] ) , proposed in ref . @xcite and named in ref . @xcite , are typically used . in these units , ( [ eq : chqc_ham_tot_si ] ) is substantially simplified to @xmath17 [ tab : atomic_units ] since all the relevant expressions in quantum chemistry are derived in one way or another from the molecular hamiltonian , the simplification brought up by the use of atomic units propagates to the whole formalism . consequently , they shall be the choice all throughout this work . [ tab : unit_conversion ] apart from the atomic units and the si ones , there are some other miscellaneous units that are often used in the literature : the _ ngstrm _ , which is a unit of length defined as , and the units of energy _ _ cm__@xmath18 ( which reminds about the spectroscopic origins of quantum chemistry and , even , quantum mechanics ) , _ electronvolt _ ( ev ) , _ kilocalorie per mole _ ( kcal / mol ) and _ kilojoule per mole _ ( kj / mol ) . the last two are specially used in the field of macromolecular simulations and quantify the energy of a mole of entities ; for example , if one asserts that the torsional barrier height for @xmath19 is @xmath20 kcal / mol , one is really saying that , in order to make a mole of @xmath19 ( i.e. , @xmath21 molecules ) rotate @xmath22 around the o o bond , one must spend @xmath20 kcal . for the conversion factors between the different energy units , see table [ tab : unit_conversion ] . finally , to close this section , we rewrite eq . ( [ eq : chqc_ham_tot_au ] ) introducing some self - explanatory notation that will be used in the subsequent discussion : [ eq : chqc_ham_tot_op ] @xmath23 to think of a macromolecule as a set of quantum objects described by a _ wavefunction _ @xmath24 dependent on the spatial and [ page : foot : spin_convenient ] spin one convenient way of thinking about functions that depend on spin - like variables is as an @xmath25-tuple of ordinary @xmath26 functions , where @xmath25 is the finite number of possible values of the spin . in the case of a one - particular wavefunction describing an electron , for example , @xmath27 can take two values ( say , @xmath28 and @xmath29 ) in such a way that one may picture any general _ spin - orbital _ @xmath30 as a 2-tuple @xmath31 . of course , another valid way of imagining @xmath30 is simply as a function of four variables , three real and one discrete . ] degrees of freedom , , of the electrons and on those of the nuclei , @xmath32 , would be too much for the imagination of physicists and chemists . all the language of chemistry would have to be remade and simple sentences in textbooks , such as `` rotation about this single bond allows the molecule to avoid steric clashes between atoms '' or even `` a polymer is a long chain - like molecule composed of repeating monomer units '' , would have to be translated into long and counter - intuitive statements involving probability and ` quantum jargon ' . conscious or not , we think of molecules as classical objects . more precisely , we are ready to accept that electrons are quantum ( we know of the interference experiments , electrons are light , we are accustomed to draw atomic ` orbitals ' , etc . ) , however , we are reluctant to concede the same status to nuclei . nuclei are heavier than electrons ( at least @xmath33 times heavier , in the case of the single proton nucleus of hydrogen ) and we picture them in our imagination as ` classical things ' that move , bond to each other , rotate around bonds and are at precise points at precise times . we imagine nuclei ` slowly moving ' in the field of the electrons , which , for each position of the first , immediately ` adjust their quantum state ' . the formalization of these ideas is called _ orn - ppenheimer _ ( bo ) _ approximation _ @xcite and the confirmation of its being good for many relevant problems is a fact that supports our intuitions about the topic and that lies at the foundations of the vast majority of the images , the concepts and the research in quantum chemistry there are many phenomena , however , in which the born - oppenheimer approximation is broken . for example , in striking a flint to create a spark , mechanical motion of the nuclei excites electrons into a plasma that then emits light @xcite . ] . like any approximation , the born - oppenheimer one may be either _ derived _ from the exact problem ( in this case , the entangled behaviour of electrons and nuclei as the same quantum object ) or simply _ proposed _ on the basis of physical intuition , and later confirmed to be good enough ( or not ) by comparison with the exact theory or with the experiment . of course , if it is possible , the first way should be preferred , since it allows to develop a deeper insight about the terms we are neglecting and the specific details that we will miss . however , although in virtually every quantum chemistry book @xcite hand - waving derivations up to different levels of detail are performed and the born - oppenheimer approximation is typically presented as unproblematic , it seems that the fine mathematical details on which these ` standard ' approaches are based are far from clear @xcite . this state of affairs does not imply that the final equations that will need to be solved are ill - defined or that the numerical methods based on the theory are unstable ; in fact , it is just the contrary ( see the discussion below ) , because the problems are related only to the precise relation between the concepts in the whole theory and those in its simplified version . nevertheless , the many subtleties involved in a _ derivation _ of the born - oppenheimer approximation scheme from the exact equations suggest that the second way , that of _ proposing _ the approximation , be taken . hence , in the following paragraphs , an _ axiomatic _ presentation of the main expressions , aimed mostly to fix the notation and to introduce the language , will be performed . first of all , if we examine the hamiltonian operator in eq . ( [ eq : chqc_ham_tot_au ] ) , we see that the term @xmath34 prevents the problem from being separable in the nuclear and electronic coordinates , i.e. , if we define @xmath35 as the set of all electronic coordinates ( spatial and spin - like ) and do likewise with the nuclear coordinates @xmath36 , the term @xmath34 prevents any wavefunction @xmath37 solution of the _ time - independent schrdinger equation _ , @xmath38 from being written as a product , @xmath39 , of an electronic wavefunction and a nuclear one . if this were the case , the problem would still be difficult ( because of the coulomb terms @xmath40 and @xmath41 ) , but we would be able to focus on the electrons and on the nuclei separately . the starting point for the born - oppenheimer approximation consists in assuming that a less strict separability is achieved , in such a way that , for a pair of suitably chosen @xmath42 and @xmath43 , any wavefunction solution of eq . ( [ eq : chqc_schrodinger ] ) ( or at least those in which we are interested ; for example , the eigenstates corresponding to the lowest lying eigenvalues ) can be expressed as @xmath44 where we have used a ` ; ' to separate the two sets of variables in the electronic part of the wavefunction in order to indicate that , in what follows , it is convenient to use the image that ` from the point of view of the electrons , the nuclear degrees of freedom are fixed ' , so that the electronic wavefunction depends ` parametrically ' on them . in other words , that the @xmath36 are not quantum variables in eq . ( [ eq : chqc_clamped ] ) below . of course , it is just a ` semantic ' semicolon ; if anyone feels uncomfortable about it , she may drop it and write a normal comma . notably , in ref . @xcite , hunter showed that any solution of the schrdinger equation can in fact be written exactly in the form of eq . ( [ eq : chqc_bo_wf ] ) , and that the two functions , @xmath42 and @xmath43 , into which @xmath37 is split may be interpreted as marginal and conditional probability amplitudes respectively . however , despite the insight that is gained from this treatment , it is of no practical value , since the knowledge of the exact solution @xmath37 is required in order to compute @xmath42 and @xmath43 in hunter s approach . in the born - oppenheimer scheme , an additional assumption is made in order to avoid this drawback : _ the equations obeyed by the electronic and nuclear parts of the wavefunction are supposed to be known_. hence , @xmath43 is assumed to be a solution of the time - independent _ clamped nuclei schrdinger equation _ , @xmath45 where the _ electronic hamiltonian operator _ @xmath46 and the _ electronic energy _ @xmath47 ( both dependent on the nuclei positions ) have been defined , and , since the nuclear spins do not enter the expression , we have explicitly indicated that @xmath48 depends parametrically on @xmath49 and not on @xmath36 . the common interpretation of the clamped nuclei equation is , as we have advanced at the beginning of the section , that the nuclei are much ` slower ' than the electrons and , therefore , the latter can automatically adjust their quantum state to the instantaneous positions of the former . physically , eq . ( [ eq : chqc_clamped ] ) is just the time - independent schrdinger equation of @xmath4 particles ( the electrons ) of mass @xmath9 and charge @xmath50 in the _ external electric field _ of @xmath0 _ point charges _ ( the nuclei ) of size @xmath51 at locations @xmath13 . mathematically , it is an eigenvalue problem that has been thoroughly studied in the literature and whose properties are well - known @xcite . in particular , it can be shown that , in the case of neutral or positively charged molecules ( i.e. , with @xmath52 ) , the clamped nuclei equation has an infinite number of normalizable solutions in the discrete spectrum of @xmath46 ( _ bound - states _ ) for every value of @xmath49 @xcite . these solutions must be regarded as the different _ electronic energy levels _ , and a further approximation that is typically made consists in , not only accepting that the electrons immediately ` follow ' nuclear motion , but also that , for each value of the nuclear positions @xmath49 , they are in the electronic ground - state this is customarily assumed in the literature and it is supported by the general fact that electronic degrees of freedom are typically more difficult to excite than nuclear ones . hence , in the vast majority of the numerical implementations of the theory , only the electronic ground - state is sought . we will see this in the forecoming sections . ] , i.e. , the one with the lower @xmath47 . consequently , we define @xmath53 to be the _ effective electronic field _ in which the nuclei move , in such a way that , once we have solved the problem in eq . ( [ eq : chqc_clamped ] ) and know @xmath54 , the time - independent _ nuclear schrdinger equation _ obeyed by @xmath42 is : @xmath55 where the _ effective nuclear hamiltonian _ @xmath56 has been implicitly defined . now , to close the section , we put together the main expressions of the born - oppenheimer approximation for quick reference and we discuss them in some more detail : [ eq : chqc_bo_summary ] @xmath57 to start , note that the above equations are written in the logical order in which they are imagined and used in any numerical calculation . first , we assume the nuclei fixed at @xmath49 and we ( hopefully ) solve the clamped nuclei electronic schrdinger equation ( eq . ( [ eq : chqc_bo_summary_a ] ) ) , obtaining the electronic ground - state @xmath58 with its corresponding energy @xmath59 . next , we repeat this procedure for all possible values of course , this can not be done in practice . due to the finite character of available computational resources , what is customarily done is to define a ` grid ' in @xmath49-space and compute @xmath54 in a finite number of points . ] of @xmath49 and end up with an hyper - surface @xmath54 in @xmath49-space . finally , we add this function to the analytical and easily computable @xmath60 and find the _ effective potential _ that determines the nuclear motion : @xmath61 it is , precisely , this effective potential that is called _ otential nergy urface _ ( pes ) ( or , more generally , _ otential nergy yper - urface _ ( pehs ) ) in quantum chemistry and that is the central object through which scientists picture chemical reactions or conformational changes of macromolecules @xcite . in fact , the concept is so appealing and the classical image so strong that , after ` going quantum ' , we can ` go classical ' back again and think of nuclei as perfectly classical particles that move in the classical potential @xmath62 . in such a case , we would not have to solve eq . ( [ eq : chqc_bo_summary_c ] ) but , instead , integrate the newtonian equations of motion . this is the basic assumption of every typical force field used for classical ground - state _ molecular dynamics _ , such as the ones in the popular charmm @xcite , amber @xcite or opls @xcite packages . finally , we would like to remind the reader that , despite the hand - waving character of the arguments presented , up to this point , every computational step has a clear description and eqs . ( [ eq : chqc_bo_summary_a ] ) through ( [ eq : chqc_bo_summary_c ] ) could be considered as _ definitions _ involving a certain degree of notational abuse . to assume that the quantities obtained through this process are close to those that proceed from a rigorous solution of the time independent schrdinger equation ( eq . ( [ eq : chqc_schrodinger ] ) ) is where the approximation really lies . hence , the more accurate eqs . ( [ eq : chqc_bo_summary_d ] ) are , the better the born - oppenheimer guess is , and , like any other one , if one does not trust in the heuristic grounds on which the final equations stand , they may be taken as axiomatic and judged a posteriori according to their results in particular cases until now , two approximations have been done : the non - relativistic character of the objects studied and the born - oppenheimer approximation . in the forecoming , many more will be done . the a priori quantification of their goodness in large molecules is a formidable task and , despite the efforts in this direction , in the end , the comparison with experimental data is the only sound method for validation . ] . in quantum chemistry , the born - oppenheimer approximation is assumed in a great fraction of the studies and it allows the central concept of potential energy surface to be well - defined , apart from considerably simplifying the calculations . the same decision is taken in this work . there exists a mathematically appealing way of deriving the time independent schrdinger equation ( eq . ( [ eq : chqc_schrodinger ] ) ) from an extremal principle . to this end , we first define the functional ( see appendix a ) that corresponds to the expected value of the energy , @xmath63 : = \langle\psi|\,\hat{h}\,|\psi\rangle \ , \ ] ] where the traditional _ bra _ and _ ket _ notation is read as @xmath64 @xmath65 being any operator in the space of wavefunctions and @xmath66 a dummy variable representing all possible coordinates on which @xmath67 depends . the norm of @xmath67 , in this notation is expressed as @xmath68 , and we shall say that @xmath67 is _ normalized _ if @xmath69 . if we want to optimize the energy functional above restricting the search space to the normalized wavefunctions , the constrained - extremals problem that results can be solved via the _ lagrange multipliers method _ ( see appendix b ) by constructing the _ associated functional _ @xmath70 $ ] , where we introduce a _ lagrange multiplier _ @xmath71 to force normalization : @xmath72 : = \mathcal{f}[\psi ] + \lambda \,\big ( \langle \psi|\psi\rangle - 1 \big ) \ .\ ] ] if we now ask that the functional derivative of @xmath70 $ ] with respect to the complex conjugate @xmath73 of the wave [ page : foot : deriving_conjugate ] function a function of a complex variable @xmath74 ( or , analogously , a functional on a space of complex functions ) may be regarded as depending on two different sets of independent variables : either @xmath75 and @xmath76 or @xmath74 and @xmath77 . the choice frequently depending on technical issues . ] be zero , i.e. , we look for the stationary points of @xmath78 $ ] conditioned by , we obtain the eigenvalues equation for @xmath79 , i.e. , the time - independent schrdinger equation . additionally , it can be shown , first , that , due to the self - adjointedness of @xmath79 , the equation obtained from the stationarity condition with respect to @xmath67 ( not @xmath73 ) is just the complex conjugate and adds no new information . moreover , one can see that the reverse implication is also true @xcite , so that , if a given normalized wavefunction @xmath67 is a solution of the eigenvalue problem and belongs to the discrete spectrum of @xmath79 , then the functional in eq . ( [ eq : chqc_functional_ftilde ] ) is stationary with respect to @xmath73 : @xmath80}{\delta\psi^ { * } } = 0 \quad \longleftrightarrow \quad \hat{h}\,\psi=-\lambda\,\psi:=e\,\psi \quad \mathrm{and } \quad \langle\psi|\psi\rangle = 1 \ .\ ] ] this result , despite its conceptual interest , is of little practical use , because it does not indicate an operative way to solve the schrdinger equation different from the ones that we already knew . the equivalence above simply illustrates that mathematical variational principles are over - arching theoretical statements from which the differential equations that actually contain the details of physical systems can be extracted . nevertheless , using similar ideas , we will derive another simple theorem which is indeed powerfully practical : the _ variational theorem_. let @xmath81 be a basis of eigenstates of the hamiltonian operator @xmath79 and @xmath82 their corresponding eigenvalues . since @xmath79 is self - adjoint , the eigenstates @xmath83 can be chosen to be orthonormal ( i.e. , @xmath84 ) and any normalized wavefunction @xmath67 in the hilbert space can be written as a linear combination of them we assume here , for the sake of simplicity and in order to highlight the relevant concepts , that @xmath79 has only discrete spectrum . the ideas involved in a general derivation are the same , but the technical details and the notation are more complicated @xcite . ] : @xmath85 if we now denote by @xmath86 the lowest @xmath87 ( i.e. , the energy of the ground - state ) its existence is not guaranteed : it depends on the particular potential in @xmath79 . however , for the physically relevant cases , there is indeed a minimum energy in the set @xmath82 . ] and calculate the expected value of the energy on an arbitrary state @xmath67 such as the one in eq . ( [ eq : chqc_decomposition ] ) , we obtain @xmath88 this simple relation is the _ variational theorem _ and it states that any wavefunction of the hilbert space has an energy larger than the one of the ground - state ( the equality can only be achieved if @xmath89 ) . however trivial this fact may appear , it allows a very fruitful ` everything - goes ' strategy when trying to approximate the ground - state in a difficult problem . if one has a procedure for finding a promising guess wavefunction ( called _ variational ansatz _ ) , no matter how heuristic , semi - empirical or intuitive it may be , one may expect that the lower the corresponding energy , the closer to the ground - state it is of course , this not necessarily so ( and , in any case , it depends on the definition of ` closer ' ) , since it could happen that the @xmath90 landscape in the constrained subset of the hilbert space in which the search is performed be ` rugged ' . in such a case , we may have very different wavefunctions ( say , in the sense of the @xmath91-norm ) with similar energies @xmath90 . the only ` direction ' in which one can be sure that the situation improves when using the variational procedure is the ( very important ) energetic one . that one is also moving towards better values of any other observable is , in general , no more than a _ bona fide _ assumption . ] this provides a systematic strategy for improving the test wavefunction which may take a number of particular forms . one example of the application of the variational theorem is to propose a family of normalized wavefunctions @xmath92 parametrically dependent on a number @xmath93 and calculate the @xmath93-dependent expected value of the energy note that , if the functions @xmath92 were not normalized , then we should deal with the constrained problem as in ( [ eq : chqc_functional_ftilde ] ) , or , equivalently , we could include a dividing _ overlap term _ @xmath94 in ( [ eq : chqc_e_theta ] ) . ] : @xmath95 then , one may use the typical tools of one - variable calculus to find the minimum of @xmath96 and thus make the best guess of the energy @xmath86 constrained to the family @xmath92 . if the ansatz is cleverly chosen , this estimate could be rather accurate , however , for large systems that lack symmetry , it is very difficult to write a good enough form for @xmath92 . when dealing with a large number of particles , there exists another protocol based on the variational theorem that will permit us to derive the hartree and hartree - fock equations for the electronic wavefunction @xmath48 ( see secs . [ sec : qc_hartree ] and [ sec : qc_hartree - fock ] , respectively ) . the first step is to devise a restricted way ( a function @xmath97 with no free parameters ) to express @xmath48 in terms of _ one - electron wavefunctions _ , also called _ orbitals _ and denoted in principle , there could be more orbitals than electrons , however , in both the hartree and hartree - fock applications of this formalism , the index @xmath98 runs , just like @xmath12 , from 1 to @xmath4 . ] by @xmath99 , thus reducing the search space to a ( typically small ) subset of the whole hilbert space : @xmath100 the second step consists in establishing a ( possibly infinite ) number of constraints on the one - electron functions actually , both restrictions ( the one at the level of the total wavefunction in eq . ( [ eq : chqc_constraint_total ] ) and the one involving the one - particle ones in eq . ( [ eq : chqc_constraint_mono ] ) ) are simply constraints ( see appendix b ) . the distinction is not fundamental but operative , and it also helps us to devise variational ansatzs separating the two conceptual playgrounds . ] , @xmath101 with these two ingredients , we can now write the lagrange functional that describes the constrained problem in terms of the orbitals @xmath102 ( see eq . ( [ eq : chqc_functional_ftilde ] ) ) : @xmath103 = \big\langle f\big(\{\psi_{a}(x_{i})\}\big ) \,\big|\ , \hat{h}\ , \big|\ , f\big(\{\psi_{a}(x_{i})\}\big ) \big\rangle + \sum_{k } \lambda_{k } l_{k}\big(\{\psi_{a}(x_{i})\}\big ) \ .\ ] ] finally , we take the derivatives of @xmath104 $ ] with respect to every @xmath105 ( normally , with respect to the complex conjugate @xmath106 , see footnote [ foot : deriving_conjugate ] in page ) and we ask each one to be zero ( see appendix a ) . this produces the final equations that must be solved in order to find the stationary one - electron orbitals . of course , these final equations may have multiple solutions . in the cases discussed in this work , there exist procedures to check that a particular solution ( found computationally ) is , not only stationary , but also minimal @xcite . however , to assure that it is , not only locally minimal , but also globally ( i.e. , that is _ optimal _ ) , could be , in general , as difficult as for any other multi - dimensional optimization problem @xcite . in the hartree and hartree - fock cases , discussed in secs . [ sec : qc_hartree ] and [ sec : qc_hartree - fock ] respectively , the aufbau principle and a clever choice of the starting guess constitute particular techniques intended to alleviate this problem . assuming the born - oppenheimer approximation ( see sec . [ sec : qc_born - oppenheimer ] and eqs . ( [ eq : chqc_bo_summary ] ) ) , the central problem that one must solve in quantum chemistry is _ to find the ground - state of the electronic hamiltonian for a fixed position _ @xmath49 _ _ of the nuclei _ _ since , from now on , we will only be dealing with the ` electronic problem ' , the notation has been made simpler by dropping superfluous subindices @xmath10 where there is no possible ambiguity . as a consequence , for example , the electronic hamiltonian is now denoted by @xmath79 , the electronic kinetic energy by @xmath107 and the electronic wavefunction by @xmath108 ( dropping the parametric dependence on @xmath49 in the same spirit ) . ] : @xmath109 as already remarked in sec . [ sec : qc_born - oppenheimer ] , this problem is well posed for neutral and positively charged molecules , and , in the same way in which the term @xmath34 prevented the total wavefunction to be a product of an electronic and a nuclear part , the term @xmath41 in the expression above breaks the separability in the one - electron variables @xmath110 of the electronic time - independent schrdinger equation associated to @xmath79 . hence , a general solution @xmath108 can not be a product of orbitals and the search must be a priori performed in the whole hilbert space . however , this is a much too big place to look for @xmath108 , since the computational requirements to solve the schrdinger equation grow exponentially on the number of electrons . partially recognizing this situation , in the first days of quantum mechanics , dirac wrote that , _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the underlying physical laws necessary for the mathematical theory of a large part of physics and the whole of chemistry are thus completely known , and the difficulty is only that the exact application of these equations leads to equations much too complicated to be soluble . it therefore becomes desirable that approximate practical methods of applying quantum mechanics should be developed , which can lead to an explanation of the main features of complex atomic systems without too much computation @xcite . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the description of the most popular approximate methods , which the great physicist envisaged to be necessary , will be the objective of the following sections . two basic points responsible of the relative success of such an enterprise are the severe reduction of the space in where the ground - state is sought ( which , of course , leads to only an approximation of it ) and the availability of computers unimaginably faster than anything that could be foreseen in times of dirac . one of the first and most simple approximations aimed to solve the problem posed in the previous section is due to hartree in 1927 @xcite ( although the way in which the hartree equations will be derived here , using the variational theorem , is due to slater @xcite ) . in this approximation , the total wavefunction is constrained to be a product ( typically referred to as _ hartree product _ ) of @xmath4 one - electron orbitals ( see eq . ( [ eq : chqc_constraint_total ] ) ) , where the spin of the electrons and the antisymmetry ( i.e. , the pauli exclusion principle ) are not taken into account we shall denote with capital greek letters the wavefunctions depending on all the electronic variables , and with lowercase greek letters the one - electron orbitals . in addition , by @xmath67 ( or @xmath111 ) , we shall indicate wavefunctions containing spin part ( called _ spin - orbitals _ ) and , by @xmath112 ( or @xmath113 ) , those that depend only on spatial variables . ] : @xmath114 where the @xmath98 index in the orbitals has been substituted by @xmath12 due to the fact that each function is paired to a specific set of electron coordinates , consequently being the same number of both of them . also , the additional requirement that the one - particle wavefunctions be normalized is imposed ( see eq . ( [ eq : chqc_constraint_mono ] ) ) : @xmath115 with these two ingredients , we can construct the auxiliary functional whose zero - derivative condition produces the solution of the constrained stationary points problem ( see eq . ( [ eq : chqc_constrained_functional ] ) ) . to this effect , we introduce @xmath4 lagrange multipliers @xmath116 that force the normalization constraints note that the normalization of the total wavefunction is a consequence of the normalization of the one - electron ones and needs not to be explicitly asked . ] : @xmath117 = \left \langle \prod_{i=1}^{n}\phi_{i}({\bm r}_{i } ) \right | \hat{h } \left | \prod_{i=1}^{n}\phi_{i}({\bm r}_{i } ) \right \rangle - \sum_{i=1}^{n } \varepsilon_{i } \big ( \langle\phi_{i}|\phi_{i}\rangle-1 \big ) \ , \ ] ] where the minus sign in the lagrange multipliers term is chosen in order to get to the most common form of the final equations without having to define new quantities this functional may be considered to depend on @xmath118 independent functions : the @xmath4 one - electron @xmath119 and their @xmath4 complex conjugates ( see footnote [ foot : deriving_conjugate ] in page ) . the _ hartree equations _ are then obtained by imposing that the functional derivative of @xmath120 with respect to @xmath121 be zero for @xmath122 . in order to obtain them and as an appetizer for the slightly more complicated process in the more used hartree - fock approximation , the functional derivative will be here computed in detail following the steps indicated in appendix a. first , we write out the limits in sums and products are dropped if there is no possible ambiguity . ] the first term in the right - hand side of eq . ( [ eq : chqc_constrained_functional_h ] ) : @xmath123 where @xmath124 denotes the euclidean @xmath125 volume element @xmath126 . now , we realize that the products outside the integrals can be dropped using the constraints in eq . ( [ eq : chqc_constraint_mono_h ] ) ( see the last paragraphs of appendix b for a justification that this can be done before taking the derivative ) . then , using the previous expression and conveniently rearranging the order of the integrals and sums , we write out the first term in the numerator of the left - hand side of eq . ( [ eq : def_funct_der ] ) that corresponds to an infinitesimal variation of the function @xmath121 : @xmath127:= } \nonumber \\ \lefteqn { \widetilde{\mathcal{f}}\,\big[\phi_{1},\phi_{1}^{*},\ldots , \phi_{k},\phi_{k}^{*}+ \epsilon \delta \phi_{k}^{*},\ldots , \phi_{n},\phi_{n}^{*}\big ] = } \nonumber \\ & & \mbox { } -\frac{1}{2}\sum_{i } \int\phi_{i}^{*}({\bm r})\nabla^{2}\phi_{i}({\bm r})\,\mathrm{d}{\bm r } - \sum_{i } \int\big|\phi_{i}({\bm r})\big|^{2 } \left ( \sum_{a=1}^{n_{n}}\frac{z_{a}}{|{\bm r}-{\bm r}_{a}| } \right ) \mathrm{d}{\bm r } \nonumber \\ & & \mbox { } + \frac{1}{2}\sum_{i}\int\big|\phi_{i}({\bm r})\big|^{2 } \left ( \int\frac{\sum_{j \ne i}\big|\phi_{j}({\bm r}\ , ' ) \big|^{2 } } { |{\bm r}-{\bm r}\ , ' |}\,\mathrm{d}{\bm r}\ , ' \right ) \mathrm{d}{\bm r } - \sum_{i}\varepsilon_{i } \left ( \int\big|\phi_{i}({\bm r})\big|^{2 } \mathrm{d}{\bm r } - 1 \right ) \nonumber \\ & & \mbox { } -\frac{1}{2}\epsilon\int\delta\phi_{k}^{*}({\bm r})\nabla^{2 } \phi_{k}({\bm r})\,\mathrm{d}{\bm r } - \epsilon\int\delta\phi_{k}^{*}({\bm r})\,\phi_{k}({\bm r } ) \left ( \sum_{a=1}^{n_{n}}\frac{z_{a}}{|{\bm r}-{\bm r}_{a}| } \right ) \mathrm{d}{\bm r } \nonumber \\ & & \mbox { } + \epsilon\int\delta\phi_{k}^{*}({\bm r})\,\phi_{k}({\bm r } ) \left ( \int\frac{\sum_{i \ne k}\big|\phi_{i}({\bm r}\ , ' ) \big|^{2 } } { |{\bm r}-{\bm r}\ , ' |}\mathrm{d}{\bm r}\ , ' \right ) \mathrm{d}{\bm r } -\epsilon\,\varepsilon_{k}\int\delta\phi_{k}^{*}({\bm r})\,\phi_{k}({\bm r } ) \,\mathrm{d}{\bm r } \ .\end{aligned}\ ] ] we subtract from this expression the quantity @xmath128 $ ] , so that the first four terms cancel , and we can write @xmath129 - \widetilde{\mathcal{f}}\,\big[\phi_{k}^{*}\big]}{\epsilon } = } \nonumber \\ & & \int \bigg { [ } -\frac{1}{2 } \nabla^{2 } \phi_{k}({\bm r } ) - \left ( \sum_{a=1}^{n_{n}}\frac{z_{a } } { |{\bm r}-{\bm r}_{a}| } \right ) \phi_{k}({\bm r } ) \nonumber \\ & & \mbox { } + \left ( \int\frac{\sum_{i \ne k}|\phi_{i}({\bm r}\ , ' ) |^{2 } } { |{\bm r}-{\bm r}\ , ' |}\,\mathrm{d}{\bm r}\ , ' \right ) \phi_{k}({\bm r } ) - \varepsilon_{k}\phi_{k}({\bm r } ) \bigg { ] } \,\delta\phi_{k}^{*}({\bm r } ) \,\mathrm{d}{\bm r } \ .\end{aligned}\ ] ] now , by simple inspection of the right - hand side , we see that the functional derivative ( see eq . ( [ eq : def_funct_der ] ) ) is the part enclosed by square brackets : @xmath130 } { \delta \phi_{k}^{*}}= \left ( - \frac{1}{2}\nabla^{2 } + \hat{v}_{e}({\bm r } ) + \hat{v}_{e}^{k}({\bm r } ) - \varepsilon_{k } \right ) \phi_{k}({\bm r } ) \ , \ ] ] where the _ nuclear potential energy _ and the _ electronic potential energy _ have been respectively defined as compare the notation with the one in eqs . ( [ eq : chqc_ham_tot_op ] ) , here a subindex @xmath10 has been dropped to distinguish the new objects defined . ] [ eq : chqc_potentials_h ] @xmath131 finally , if we ask the functional derivative to be zero for @xmath122 , we arrive to the equations that the stationary points must satisfy , the _ hartree equations _ : @xmath132 \ , \phi_{k}({\bm r } ) : = \left ( - \frac{1}{2}\nabla^{2 } + \hat{v}_{n}({\bm r } ) + \hat{v}_{e}^{k}({\bm r } ) \right ) \phi_{k}({\bm r } ) = \varepsilon_{k}\,\phi_{k}({\bm r})\ , \ ] ] for all @xmath122 . let us note that , despite the fact that the object @xmath133 $ ] defined above is not a operator strictly speaking , since , as the notation emphasizes , it depends on the orbitals @xmath134 , we will stick to the name _ hartree operator _ for it , in order to be consistent with most of the literature . now , some remarks related to the hartree equations are worth making . first , it can be shown that , if the variational ansatz in eq . ( [ eq : chqc_constraint_total_h ] ) included the spin degrees of freedom of the electrons , all the expressions above would be kept , simply changing the orbitals @xmath135 by the spin - orbitals @xmath136 . secondly , and moving into more conceptual playgrounds , we note that the special structure of @xmath137 in eq . ( [ eq : chqc_potentials_h_b ] ) makes it mandatory to interpret the hartree scheme as one in which each electron ` feels ' only the average effect of the rest . in fact , if the _ quantum charge density _ @xmath138 is regarded for a moment as a classical continuum distribution , then the potential produced by all the electrons but the @xmath139-th is precisely the one in eq . ( [ eq : chqc_potentials_h_b ] ) . supporting this image , note also the fact that , if we write the joint probability density of electron 1 being at the point @xmath140 , electron @xmath141 being at the point @xmath142 and so on ( simply squaring eq . ( [ eq : chqc_constraint_total_h ] ) ) , @xmath143 we see that , _ in a probabilistic sense _ , the electrons are independent ( they could not be independent in a physical , complete sense , since we have already said that they ` see ' each other in an average way ) . anyway , despite these appealing images and also despite the fact that , disguised under the misleading ( albeit common ) notation , these equations seem ` one - particle ' , they are rather complicated from a mathematical point of view . on the one hand , it is true that , whereas the original electronic schrdinger equation in ( [ eq : chqc_bo_summary_a ] ) depended on @xmath144 spatial variables , the expressions above only depend on 3 . this is what we have gained from drastically reducing the search space to the set of hartree products in eq . ( [ eq : chqc_constraint_total_h ] ) and what renders the approximation tractable . on the other hand , however , we have paid the price of greatly increasing the mathematical complexity of the expressions , so that , while the electronic schrdinger equation was one linear differential equation , the hartree ones in ( [ eq : chqc_hartree_eqs ] ) are @xmath4 coupled non - linear integro - differential equations @xcite . this complexity precludes any analytical approach to the problem and forces us to look for the solutions using less reliable iterative methods . typically , in computational studies , one proposes a _ starting guess _ for the set of @xmath4 orbitals @xmath145 ; with them , the hartree operator @xmath146 $ ] in the left - hand side of eq . ( [ eq : chqc_hartree_eqs ] ) is constructed for every @xmath139 and the @xmath4 equations are solved as simple eigenvalue problems . for each @xmath139 , the @xmath147 that corresponds to the lowest @xmath148 is selected and a new hartree operator @xmath149 $ ] is constructed with the @xmath150 . the process is iterated until ( hopefully ) the @xmath151-th set of solutions @xmath152 differ from the one @xmath153 less than a reasonably small amount . many technical issues exist that raise doubts about the possible success of such an approach . the most important ones being related to the fact that a proper definition of the _ hartree problem _ should be : _ find the global minimum of the energy functional @xmath154 under the constraint that the wavefunction @xmath112 be a hartree product _ and not : _ solve the hartree equations _ ( [ eq : chqc_hartree_eqs ] ) , whose solutions indeed include the global minimum sought but also all the rest of stationary points . while the possibility that a found solution be a maximum or a saddle point can be typically ruled out @xcite , as we remarked in sec . [ sec : qc_variational ] and due to the fact that there are an infinite number of solutions to the hartree equations @xcite , to be sure that any found minimum is the global one is impossible in a general case . there exists , however , one way , related to a theorem by simon and lieb @xcite , of hopefully biasing a particular found solution of the hartree equations to be the global minimum that we are looking for . they showed that , first , for neutral or positively charged molecules ( @xmath155 ) , the hartree global minimization problem has a solution ( its uniqueness is not established yet @xcite ) and , second , that the minimizing orbitals @xmath156 correspond to the lowest eigenvalues of the @xmath133 $ ] operators self - consistently constructed with them in quantum chemistry , where the number of electrons considered is typically small , the version of the hartree equations that is used is the one derived here , with the hartree operators depending on the index @xmath139 in a non - trivial way . however , if the number of electrons is large enough ( such as in condensed matter applications ) , is customary to add to the effective electronic repulsion in eq . ( [ eq : chqc_potentials_h_b ] ) the self - interaction of electron @xmath139 with himself . in such a case , the hartree operator is independent of @xmath139 so that , after having achieved self - consistency , the orbitals @xmath157 turn out to be eigenstates corresponding to different eigenvalues of the same hermitian operator , @xmath158 $ ] , and , therefore , mutually orthogonal . ] . now , although _ the reverse of the second part of the theorem is not true in general _ ( i.e. , from the fact that a particular set of orbitals are the eigenstates corresponding to the lowest eigenvalues of the associated hartree operators , does not necessarily follow that they are the ones that minimize the energy ) @xcite , in practice , the insight provided by lieb and simon s result is invoked to build each successive state in the iterative procedure described above , choosing the lowest lying eigenstates each time . in this way , although one can not be sure that the global minimum has been reached , the fact that the found one has a property that the former also presents is regarded as a strong hint that it must be so ( see also the discussion for the hartree - fock case in the next section ) . this drawback and all the problems arising from the fact that an iterative procedure such as the one described above could converge to a fixed point , oscillate eternally or even diverge , are circumvented in practice by a clever choice of the starting guess orbitals @xmath145 . if they are extracted , for example , from a slightly less accurate theory , one may expect that they could be ` in the basin of attraction ' of the true hartree minimum ( so that the stationary point found will be the correct one ) and close to it ( so that the iterative procedure will converge ) . this kind of wishful thinking combined with large amounts of heuristic protocols born from many decades of trial - and - error - derived knowledge pervade and make possible the whole quantum chemistry discipline . the hartree theory discussed in the previous section is not much used in quantum chemistry and many textbooks on the subject do not even mention it . although it contains the seed of almost every concept underlying the _ hartree - fock approximation _ discussed in this section , it lacks an ingredient that turns out to be essential to correctly describe the behaviour of molecular species : _ the indistinguishability of the electrons_. this was noticed independently by fock @xcite and slater @xcite in 1930 , and it was corrected by proposing a variational ansatz for the total wavefunction that takes the form of a so - called _ slater determinant _ ( see eq . ( [ eq : chqc_constraint_total_sd_ghf ] ) below ) . the most important mathematical consequence of the indistinguishability among a set of @xmath4 quantum objects of the same type is the requirement that the total @xmath4-particle wavefunction must either remain unchanged ( _ symmetric _ ) or change sign ( _ antisymmetric _ ) when any pair of coordinates , @xmath110 and @xmath159 , are swapped . in the first case , the particles are called _ bosons _ and must have integer spin , while in the second case , they are called _ fermions _ and have semi - integer spin . electrons are fermions , so the total wavefunction must be antisymmetric under the exchange of any pair of one - electron coordinates . this is a property that is certainly not met by the single hartree product in eq . ( [ eq : chqc_constraint_total_h ] ) but that can be easily implemented by forming linear combinations of many of them . the trick is to add all the possible hartree products that are obtained from eq . ( [ eq : chqc_constraint_total_h ] ) changing the order of the orbitals labels while keeping the order of the coordinates ones it is immaterial whether the orbitals labels are kept and the coordinates ones changed or vice versa . ] , and assigning to each term the sign of the permutation @xmath160 needed to go from the natural order @xmath161 to the corresponding one @xmath162 . the sign of a permutation @xmath160 is 1 if @xmath160 can be written as a composition of an even number of two - element transpositions , and it is @xmath163 if the number of transpositions needed is odd . therefore , we define @xmath164 as the minimum number it can be shown that the parity of all decompositions of @xmath160 into products of elementary transpositions is the same . we have chosen the minimum only for @xmath164 to be well defined . ] of transpositions needed to perform the permutation @xmath160 , and we write the sign of @xmath160 as @xmath165 . using this , an antisymmetric wavefunction constructed from hartree products of @xmath4 different orbitals may be written as @xmath166 where the factor @xmath167 enforces normalization of the total wavefunction @xmath67 ( if we use the constraints in eq . ( [ eq : chqc_constraint_mono_ghf ] ) ) and @xmath168 denotes the _ symmetric group _ of order @xmath4 , i.e. , the set of all permutations of @xmath4 elements ( with a certain multiplication rule ) . the above expression is more convenient to perform the calculations that lead to the hartree - fock equations , however , there is also a compact way of rewriting eq . ( [ eq : chqc_constraint_total_ghf ] ) which is commonly found in the literature and that is useful to illustrate some particular properties of the problem . it is the _ slater determinant _ : @xmath169 now , having established the constraints on the form of the total wavefunction , we ask the hartree - fock one - electron orbitals to be , not only normalized , like we did in the hartree case , but also mutually orthogonal : @xmath170 additionally note that , contrarily to what we did in the previous section , we have now used one - electron wavefunctions @xmath171 dependent also on the spin @xmath27 ( i.e. , spin - orbitals ) to construct the variational ansatz . general _ spin - orbital note that , if we had not included the spin degrees of freedom , the search space would have been half as large , since , where we now have @xmath118 functions of @xmath172 ( i.e. , @xmath173 and @xmath174 , with @xmath175 ) , we would have had just @xmath4 ( the @xmath176 ) . ] may be written as ( see also footnote [ foot : spin_convenient ] in page ) @xmath177 where the functions @xmath14 and @xmath178 correspond to the _ spin - up _ and _ spin - down _ eigenstates of the operator associated to the @xmath74-component of the one - electron spin . they are defined as @xmath179 the formalism obtained when these general spin - orbitals are used is accordingly called _ eneral artree - ock _ ( ghf ) . the first part of the mathematical treatment in the following paragraphs shall be performed assuming this situation . the advantage of such a choice is that , later on , by imposing additional constraints to the spin part of the one - electron orbitals , we will be able to derive the basic equations for some other flavours of the hartree - fock theory , such as uhf , rhf and rohf , in a very direct way . now , to calculate the expected value of the energy in a state such as the one in eqs . ( [ eq : chqc_constraint_total_ghf ] ) and ( [ eq : chqc_constraint_total_sd_ghf ] ) , let us denote the one - particle part ( that operates on the @xmath12-th coordinates ) of the total electronic hamiltonian @xmath79 in eq . ( [ eq : chqc_el_ham_1 ] ) by @xmath180 in such a way that , @xmath181 where @xmath182 . [ page : one_e_calculation ] we shall compute separately each one of the sums in the expression above . let us start now with the first one : for a given @xmath12 in the sum , and due to the structure of the electronic wavefunction in ( [ eq : chqc_constraint_total_ghf ] ) , the expected value @xmath183 is a sum of @xmath184 terms of the form @xmath185 but , since @xmath186 operates only on @xmath187 and due to the orthogonality of the spin - orbitals with different indices , we have that the only non - zero terms are those with @xmath188 . taking this into account , all permutations @xmath160 appear still as terms of the sum , and we see that every orbital @xmath189 occurs depending on every coordinate @xmath187 . given a particular pair @xmath12 and @xmath190 , this happens in the terms for which @xmath191 and one of such terms may be expressed as @xmath192 where we have used that @xmath193 , and we have dropped the index @xmath12 from @xmath186 noticing that the integration variables in @xmath194 are actually dummy . next , we use again the one - electron wavefunctions constraints in eq . ( [ eq : chqc_constraint_mono_ghf ] ) to remove the product of norms in brackets , and we realize that , for each @xmath190 , there are as many terms like the one in the expression above as permutations of the remaining @xmath195 orbital indices ( i.e. , @xmath196 ) . in addition , we recall that every @xmath190 must appear and perform the first sum in eq . ( [ eq : chqc_expected_h_ghf ] ) , yielding @xmath197 where all factorial terms have canceled out . the next step is to calculate the second sum in eq . ( [ eq : chqc_expected_h_ghf ] ) . again , we have that , for each pair @xmath198 , @xmath199 is a sum of @xmath184 terms like @xmath200 [ page : two_e_calculation ] for this expected value , contrarily to the case of @xmath186 and due to the two - body nature of the operator @xmath201 , not only do the terms with @xmath188 survive , but also those in which @xmath160 and @xmath202 differ over only a pair of values @xmath12 and @xmath190 , i.e. , those for which @xmath203 , @xmath204 and @xmath205 . the reason for this is that , even if @xmath206 and @xmath207 , the integral @xmath208 does not vanish . now , using that @xmath201 operates only on @xmath187 and @xmath209 , the orthonormality conditions in eq . ( [ eq : chqc_constraint_mono_ghf ] ) and the fact that @xmath193 , we have that , when @xmath210 depends on @xmath187 and @xmath211 depends on @xmath209 , the @xmath188 part of the corresponding terms in eq . ( [ eq : chqc_calculation_rij_1 ] ) reads @xmath212 where we have defined @xmath213 next , we see that , for each pair @xmath214 , and keeping @xmath215 , we can make @xmath216 permutations among the @xmath217 indices of the orbitals on which @xmath201 does not operate and still find the same expression ( [ eq : chqc_calculation_rij_2 ] ) . therefore , for each pair @xmath214 , we have a sum of @xmath216 identical terms . in addition , if we perform the sum on @xmath12 and @xmath190 in eq . ( [ eq : chqc_expected_h_ghf ] ) and remark that the term in eq . ( [ eq : chqc_calculation_rij_2 ] ) does not depend on the pair @xmath198 ( which is obvious from the suggestive notation above ) , we have that the @xmath188 part of the second sum in eq . ( [ eq : chqc_expected_h_ghf ] ) , which is typically called _ coulomb energy _ , reads @xmath218 where we have used that the sum @xmath219 is performed on @xmath220 identical terms which do not depend neither on @xmath12 nor on @xmath190 . on the other hand , in the case in which @xmath160 and @xmath202 only differ in that the indices of the orbitals that depend on @xmath110 and @xmath159 are swapped , all the derivation above applies except for the facts that , first , @xmath221 and , second , the indices @xmath139 and @xmath222 must be _ exchanged _ in eq . ( [ eq : chqc_calculation_rij_3 ] ) ( it is immaterial if they are exchanged in the bra or in the ket , since the indices are summed over and are dummy ) . henceforth , the remaining part of the second sum in eq . ( [ eq : chqc_expected_h_ghf ] ) , typically termed _ exchange energy _ , may be written as @xmath223 finally , the expected value of the energy in the ghf variational state @xmath67 turns out to be @xmath224 where the _ one - electron integrals _ @xmath225 have been defined together with the _ two - electron integrals _ , @xmath226 and @xmath227 , and the fact that @xmath228 has been used to include the diagonal terms in the second sum . now , the energy functional above is the quantity that we want to minimize under the orthonormality constraints in eq . ( [ eq : chqc_constraint_mono_ghf ] ) . so we are prepared to write the auxiliary functional @xmath120 , introducing @xmath229 lagrange multipliers @xmath230 ( see eq . ( [ eq : chqc_constrained_functional ] ) and compare with the hartree example in the previous section ) : @xmath231 = \sum_i h_i + \frac{1}{2 } \sum_{i , j } ( j_{ij}-k_{ij } ) - \sum_{i , j } \lambda_{ij}\ , \big ( \langle\psi_{i}|\psi_{j}\rangle-\delta_{ij } \big ) \ .\ ] ] in order to get to the _ hartree - fock equations _ that the stationary orbitals @xmath210 must satisfy , we impose that the functional derivative of @xmath232 $ ] with respect to @xmath233 be zero . to calculate @xmath234 , we follow the procedure described in appendix a , using the same notation as in eq . ( [ eq : chqc_h_step_2 ] ) . the variation with respect to each @xmath233 shall yield the hartree - fock equations for the unconjugated @xmath171 . the equations for the @xmath235 are obtained either by differentiating @xmath232 $ ] with respect to each @xmath210 or , if the @xmath230 matrix is hermitian ( which will turn out to be the case ) , by simply taking the complex conjugate of both sides of the final equations in ( [ eq : chqc_hf_eqs_v1_ghf ] ) . now , @xmath236 - \widetilde{\mathcal{f}}\,\big[\psi_{k}^{*}\big]}{\epsilon } = } \nonumber \\ & & \langle\delta\psi_{k}|\,\hat{h}\,|\psi_{k}\rangle + \sum_j\bigg(\langle\delta\psi_k\psi_j|\,\frac{1}{r}\,|\psi_k\psi_j\rangle - \langle\delta\psi_k\psi_j|\,\frac{1}{r}\,|\psi_j\psi_k\rangle\bigg ) - \sum_j \lambda_{kj } \langle\delta\psi_{k}|\psi_{j}\rangle = \nonumber \\ & & \int \bigg { [ } \hat{h}\,\psi_{k}(x ) + \sum_j \left ( \psi_k(x ) \!\ ! \int \frac{|\psi_{j}(x^{\prime})|^{2 } } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } - \psi_j(x ) \!\ ! \int \frac{\psi^{*}_{j}(x^{\prime})\,\psi_{k}(x^{\prime } ) } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \right ) \nonumber \\ & & \mbox { } - \sum_j \lambda_{kj}\,\psi_{j}(x ) \bigg { ] } \,\delta\psi_{k}^{*}(x ) \,\mathrm{d}x \ , \end{aligned}\ ] ] where we have used the more compact notation @xmath237 instead of @xmath238 . then , like in the previous section , by simple inspection of the right - hand side , we see that the functional derivative is the part enclosed by square brackets ( see eq . ( [ eq : def_funct_der ] ) ) : @xmath239 } { \delta \psi_{k}^{*}}= \left[\hat{h } + \sum_j \big(\hat{j}_{j}[\psi]- \hat{k}_{j}[\psi]\big)\right]\psi_{k}(x)- \sum_j\lambda_{kj}\psi_{j}(x ) \ , \ ] ] where the _ coulomb _ and _ exchange operators _ are respectively defined by their action on an arbitrary function @xmath240 as follows like in the hartree case in the previous section , the word _ operator _ is a common notational abuse if they act upon the very @xmath171 on which they depend . this is again made explicit in the notation . ] : [ eq : chqc_operators_ghf ] @xmath241\,\varphi(x):=\left(\int\frac{|\psi_{j}(x^{\prime})|^{2 } } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \right ) \varphi(x)\ , \label{eq : chqc_operators_ghf_a } \\ & \hat{k}_{j}[\psi]\,\varphi(x):= \left ( \int\frac{\psi^{*}_{j}(x^{\prime})\,\varphi(x^{\prime } ) } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \right ) \psi_{j}(x)\ . \label{eq : chqc_operators_ghf_b } \end{aligned}\ ] ] therefore , if we define the ghf _ fock operator _ as @xmath242:= \hat{h}+\sum_j \big(\hat{j}_{j}[\psi]-\hat{k}_{j}[\psi]\big ) \ , \ ] ] we arrive to a first version of the _ hartree - fock equations _ by asking that the functional derivative in eq . ( [ eq : chqc_funct_der_ghf ] ) be zero : @xmath243\,\psi_{i}(x)=\sum_j\lambda_{ij}\psi_{j}(x)\ , \qquad i=1,\ldots , n\ .\ ] ] now , in order to obtain a simpler version of them , we shall take profit from the fact that the whole problem is invariant under a unitary transformation among the one - electron orbitals . if we repeat the calculation in eq . ( [ eq : chqc_limit_ghf ] ) but varying @xmath210 this time , instead of @xmath244 , and use the following relation : @xmath245\psi_j(x)\mathrm{d}x \ , \ ] ] we arrive to the ghf equations for the conjugated spin - orbitals : @xmath246\,\psi^*_{i}(x)=\sum_j\lambda_{ji}\psi^*_{j}(x)\ . \qquad i=1,\ldots , n\ .\ ] ] then , we may subtract the complex conjugate of eq . ( [ eq : chqc_hf_eqs_v1_ghf_conj ] ) from eq . ( [ eq : chqc_hf_eqs_v1_ghf ] ) yielding @xmath247 therefore , since the set of the @xmath189 is orthogonal and hence linearly independent , we have that the @xmath248 matrix @xmath249 of lagrange multipliers is hermitian : @xmath250 this actually means that we have a set of three equations that the stationary spin - orbitals satisfy , but only two of them are independent . these equations are the ghf equations for the @xmath171 and @xmath235 , in ( [ eq : chqc_hf_eqs_v1_ghf ] ) and ( [ eq : chqc_hf_eqs_v1_ghf_conj ] ) , respectively , and ( [ eq : chqc_lambdakjjk_ghf_herm ] ) . any pair of them could be in principle be chosen as the basic equations , however , in common practice the first and the last one of them are typically picked . in any case , due to ( [ eq : chqc_lambdakjjk_ghf_herm ] ) , a unitary matrix @xmath251 exists that _ diagonalizes _ @xmath252 ; in the sense that @xmath253 is a diagonal matrix , i.e. , @xmath254 . using this unitary matrix @xmath251 , we can transform the set of orbitals @xmath255 into a new one @xmath256 : @xmath257 this transformation is physically legitimate since it only changes the @xmath4-electron wavefunction @xmath67 in an unmeasurable phase @xmath258 . to see this , let us denote by @xmath259 the @xmath260-element of the matrix inside the slater determinant in eq . ( [ eq : chqc_constraint_total_sd_ghf ] ) , i.e. , @xmath261 . then , after using the expression above , the @xmath260-element of the new matrix @xmath262 can be related to the old ones via @xmath263 , in such a way that @xmath264 and the desired result follows : @xmath265 now , we insert eq . ( [ eq : chqc_transformation_phi_ghf ] ) into the first version of the hartree - fock equations in ( [ eq : chqc_hf_eqs_v1_ghf ] ) : @xmath266\,\left(\sum_j u_{ij}\psi^{\prime}_{j}(x)\right)= \sum_{j , k}\lambda_{ij}u_{jk}\psi^{\prime}_{k}(x)\ , \qquad i=1,\ldots , n\ .\ ] ] next , we multiply by @xmath267 each one of the @xmath4 expressions and sum in @xmath12 : @xmath268\,\left(\sum_{i , j } \underbrace{u^{-1}_{li}u_{ij}}_{\displaystyle \delta_{lj } } \psi^{\prime}_{j}(x)\right)= \sum_{i , j , k}\underbrace{u^{-1}_{li}\lambda_{ij}u_{jk } } _ { \displaystyle \varepsilon_{lk}=\delta_{lk}\varepsilon_{l } } \psi^{\prime}_{k}(x ) } \nonumber \\[6pt ] & & \longrightarrow\quad \hat{f}^\mathrm{ghf}[u\psi^{\prime}]\,\psi^{\prime}_{l}(x)= \varepsilon_{l}\psi^{\prime}_{l}(x)\ , \qquad l=1,\ldots , n\ .\end{aligned}\ ] ] although this new version of the hartree - fock equations can be readily seen as a _ pseudo - eigenvalue problem _ and solved by the customary iterative methods , we can go a step further and show that , like the @xmath4-particle wavefunction @xmath67 ( see eq . ( [ eq : chqc_phi_invariant_ghf ] ) ) , the fock operator @xmath269 $ ] , as a function of the one - electron orbitals , is invariant under a unitary transformation such as the one in eq . ( [ eq : chqc_transformation_phi_ghf ] ) . in fact , this is true for each one of the sums of coulomb and exchange operators in eq . ( [ eq : chqc_fock_op_ghf ] ) separately : @xmath270\,\varphi(x)= \sum_j \bigg(\int\frac { \left |\sum_k u_{jk } \,\psi^{\prime}_{k}(x^{\prime } ) \right|^{2 } } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \bigg)\,\varphi(x ) = } \nonumber \\ & & \sum_j \bigg(\int\frac { \sum_{k , l } u^{*}_{jk}u_{jl } \,\psi^{\prime * } _ { k}(x^{\prime } ) \,\psi^{\prime}_{l}(x^{\prime } ) } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \bigg ) \,\varphi(x)= \nonumber \\ & & \bigg(\int\frac { \sum_{j , k , l } u^{-1}_{kj}u_{jl } \,\psi^{\prime * } _ { k}(x^{\prime } ) \,\psi^{\prime}_{l}(x^{\prime } ) } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \bigg)\,\varphi(x)= \nonumber \\ & & \sum_k \bigg(\int\frac { |\psi^{\prime}_{k}(x^{\prime})|^{2 } } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \bigg)\,\varphi(x)= \sum_j \hat{j}_{j}[\psi^{\prime}]\,\varphi(x ) \ , \qquad \forall \varphi(x ) \ , \end{aligned}\ ] ] where , in the step before the last , we have summed on @xmath190 and @xmath222 , using that . performing very similar calculations , one can show that @xmath271\,\varphi(x ) = \sum_j \hat{k}_{j}[\psi^{\prime}]\,\varphi(x)\ , \qquad \forall \varphi(x ) \ , \ ] ] and therefore , that @xmath272=f^\mathrm{ghf}[\psi^{\prime}]$ ] . in such a way that any unitary transformation on a set of orbitals that constitute a solution of the hartree - fock equations in ( [ eq : chqc_hf_eqs_v1_ghf ] ) yields a different set that is also a solution of _ the same _ equations . for computational and conceptual reasons ( see , for example , koopmans theorem below ) , it turns out to be convenient to use this freedom and choose the matrix @xmath251 in such a way that the lagrange multipliers matrix is diagonalized ( see eq . ( [ eq : chqc_lambdakjjk_ghf ] ) and the paragraph below it ) . the particular set of one - electron orbitals @xmath256 obtained with this @xmath251 are called _ canonical orbitals _ and their use is so prevalent that we will circumscribe the forecoming discussion to them and drop the prime from the notation . using the canonical orbitals , the _ hartree - fock equations _ can be written as @xmath273\,\psi_{i}(x)=\varepsilon_{i}\psi_{i}(x)\ , \qquad i=1,\ldots , n\ .\ ] ] many of the remarks related to these equations are similar to those made about the hartree ones in ( [ eq : chqc_hartree_eqs ] ) , although there exist important differences due to the inclusion of the indistinguishability of the electrons in the variational ansatz . this is clearly illustrated if we calculate the joint probability density , associated to a wavefunction like the one in eq . ( [ eq : chqc_constraint_total_ghf ] ) , of the coordinates with label 1 taking the value @xmath274 , the coordinates with label @xmath141 taking the value @xmath275 , and so on : @xmath276 if we compare this expression with eq . ( [ eq : chqc_hartree_rho ] ) , we see that the antisymmetry of @xmath67 has completely spoiled the statistical independence among the one - electron coordinates . however , there is a weaker quasi - independence that may be recovered : if , using the same reasoning about permutations that took us to the one - electron part @xmath277 of the energy functional in page , we calculate the marginal probability density of the @xmath12-th coordinates taking the value @xmath187 , we find @xmath278 now , since the coordinates indices are just immaterial labels , the actual probability density of finding _ any _ electron with coordinates @xmath66 is given by @xmath279 which can be interpreted as a _ charge density _ ( except for the sign ) , as , in atomic units , the charge of the electron is @xmath280 . the picture being consistent with the fact that @xmath281 is normalized to the number of electrons @xmath4 : @xmath282 additionally , if we perform the same type of calculations that allowed to calculate the two - electron part of the energy functional in page , we have that the two - body probability density of the @xmath12-th coordinates taking the value @xmath187 and of the @xmath190-th coordinates taking the value @xmath209 reads @xmath283 and , if we reason in the same way as in the case of @xmath284 , in order to get to the probability density of finding _ any _ electron with coordinates @xmath66 at the same time that _ any other _ electron has coordinates @xmath285 , we must multiply the function above by @xmath286 , which is the number of immaterial @xmath198-labelings , taking into account that the distinction between @xmath66 and @xmath285 is also irrelevant : @xmath287 finally , taking eq . ( [ eq : chqc_qrhoi_ghf ] ) to this one , we have @xmath288 where the first term corresponds to independent electrons and the second one , called the _ interference term _ , could be interpreted as an exchange correction . although , in general , this is the furthest one may go , when additional constraints are imposed on the spin part of the one - electron wavefunctions ( see the discussion about restricted hartree - fock in the following pages , for example ) , the exchange correction in eq . ( [ eq : chqc_qrhoij_qrhoi_ghf ] ) above vanishes for electrons of opposite spin , i.e. , electrons of opposite spin turn out to be pairwise independent . however , whereas it is true that more correlation could be added to the hartree - fock results by going to higher levels of the theory and , in this sense , hartree - fock could be considered the first step in the ` correlation ladder ' , one should not regard it as an ` uncorrelated ' approximation , since , even in the simplest case of rhf ( see below ) , hartree - fock electrons ( of the same spin ) are statistically correlated . all of this has its roots in the _ pauli principle _ , which states that no pair of electrons can share all the quantum numbers . let us now point out that , like in the hartree case , the left - hand side of the hartree - fock equations in ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) is a complicated , non - linear function of the orbitals @xmath255 and the notation chosen is intended only to emphasize the nature of the iterative protocol that is typically used to solve the problem . however , note that , while the hartree operator @xmath133 $ ] depended on the index of the orbital @xmath157 on which it acted , the fock operator in eq . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) is the same for all the spin - orbitals @xmath171 . this is due to the inclusion of the @xmath289 terms in the sum of the coulomb and exchange two - electron integrals in eq . ( [ eq : chqc_expected_h_2_ghf ] ) and it allows to perform the iterative procedure solving only one eigenvalue problem at each step , instead of @xmath4 of them like in the hartree case ( see however the uhf and rohf versions of the hartree - fock problem in what follows ) . the one - particle appearance of eqs . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) is again strong and , whereas the ` eigenvalues ' @xmath116 are not the energies of the individual electrons , they are called _ orbital energies _ due to the physical meaning they receive via the well - known _ koopmans theorem _ @xcite . to get to this result , let us multiply eq . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) from the left by @xmath290 , for a given @xmath12 , and then integrate over @xmath66 . using the definition of the fock operator in eq . ( [ eq : chqc_fock_op_ghf ] ) together with the coulomb and exchange ones in eqs . ( [ eq : chqc_operators_ghf ] ) , we obtain @xmath291 where we have used the same notation as in eq . ( [ eq : chqc_expected_h_2_ghf ] ) and the fact that the one - electron orbitals are normalized . if we next sum on @xmath12 and compare the result with the expression in eq . ( [ eq : chqc_expected_h_2_ghf ] ) , we found that the relation of the eigenvalues @xmath116 with the actual hartree - fock energy is given by @xmath292 [ page : koopman ] finally , if we assume that upon ` removal of an electron from the @xmath139-th orbital ' the rest of the orbitals will remain unmodified , we can calculate the _ ionization energy _ using the expression in ( [ eq : chqc_expected_h_2_ghf ] ) together with the equations above : @xmath293 and this is koopmans theorem , namely , that _ the @xmath139-th ionization energy in the frozen - orbitals approximation is @xmath294_. moving now to the issue about the solution of the hartree - fock equations in ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) , we must remark that the necessity of using the relatively unreliable iterative approach to tackle them stems again from their complicated mathematical form . like in the hartree case , we have managed to largely reduce the dimension of the space on which the basic equations are defined : from @xmath144 in the electronic schrdinger equation in ( [ eq : chqc_bo_summary_a ] ) to 3 in the hartree - fock ones . however , to have this , we have payed the price of dramatically increasing their complexity @xcite , since , while the electronic schrdinger equation was one linear differential equation , the hartree - fock ones in ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) are @xmath4 coupled non - linear integro - differential equations , thus precluding any analytical approach to their solution . a typical iterative procedure the process described in this paragraph must be taken only as an outline of the one that is performed in practice . it is impossible to deal in a computer with a general function as it is ( a non - countable infinite set of numbers ) , and the problem must be discretized in some way . the truncation of the one - electron hilbert space using a finite basis set , described in secs . [ sec : qc_roothaan ] and [ sec : qc_basis_sets ] , is the most common way of doing this . moreover , the iterative procedure is normally performed using not the spin - orbitals but the spatial ones . in this sense , the restricted versions of the hartree - fock problem , discussed below , are closer to the actual implementation of the theory in computer applications . ] begins by proposing a _ starting guess _ for the set of @xmath4 spin - orbitals @xmath295 . with them , the fock operator @xmath296 $ ] in the left - hand side of eq . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) is constructed and the set of @xmath4 equations is solved as one simple eigenvalue problem . then , the @xmath297 that correspond to the @xmath4 lowest eigenvalues @xmath298 are selected ( see the discussion of the aufbau principle below ) and a new fock operator @xmath299 $ ] is constructed with them . the process is iterated until ( hopefully ) the @xmath151-th set of solutions @xmath300 differs from the @xmath301-th one @xmath302 less than a reasonably small amount ( defining the distance among solutions in some suitable way typically combined with a convergence criterium related to the associated energy change ) . when this occurs , the procedure is said to have converged and the solution orbitals are called _ self - consistent _ ; also , a calculation of this kind is commonly termed _ elf - onsistent ield _ ( scf ) . again , like in the hartree case , many issues exist that raise doubts about the possible success of such an approach . the most important ones are related to the fact that a proper definition of the _ hartree - fock problem _ should be : _ find the global minimum of the energy functional @xmath303 under the constraint that the wavefunction @xmath67 be a slater determinant of one - electron spin - orbitals _ , and not : _ solve the hartree - fock equations _ ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) . the solutions of the latter are all the stationary points of the constrained energy functional , while we are interested only in the particular one that is the global minimum . even ruling out the possibility that a found solution may be a maximum or a saddle point ( which can be done @xcite ) , one can never be sure that it is the global minimum and not a local one . there exists , however , one way , related to the hartree - fock version of the theorem by simon and lieb @xcite mentioned in the previous section , of hopefully biasing a particular found solution of eqs . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) to be the global minimum that we are looking for . they showed , first , that for neutral or positively charged molecules ( @xmath304 ) , the hartree - fock global minimization problem has a solution ( its uniqueness is not established yet @xcite ) and , second , that the minimizing orbitals @xmath255 correspond to the @xmath4 lowest eigenvalues of the fock operator @xmath269 $ ] that is self - consistently constructed with them . therefore , although _ the reverse is not true in general _ @xcite ( i.e. , from the fact that a particular set of orbitals are the eigenstates corresponding to the lowest eigenvalues of the associated fock operator , does not necessarily follow that they are the optimal ones ) , the information contained in simon and lieb s result is typically invoked to build each successive state in the iterative procedure described above by keeping only the @xmath4 orbitals that correspond to the @xmath4 lowest eigenvalues @xmath116 . indeed , by doing that , one is effectively constraining the solutions to have a property that the true solution does have , so that , in the worst case , the space in which one is searching is of the same size as the original one , and , in the best case ( even playing with the possibility that the reverse of simon and lieb s theorem be true , though not proved ) , the space of solutions is reduced to the correct global minimum alone . this wishful - thinking way of proceeding is termed the _ aufbau principle _ @xcite , and , together with a clever choice of the starting - guess set of orbitals @xcite ( typically extracted from a slightly less accurate theory , so that one may expect that it could be ` in the basin of attraction ' of the true hartree - fock minimum ) , constitute one of the many heuristic strategies that make possible that the aforementioned drawbacks ( and also those related to the convergence of iterative procedures ) be circumvented in real cases , so that , in practice , most of scf calculations performed in the field of quantum chemistry do converge to the true solution of eqs . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) in spite of the theoretical notes of caution . now , to close this ghf part , let us discuss some points regarding the imposition of constraints as a justification for subsequently introducing three commonly used forms of the hartree - fock theory that involve additional restrictions on the variational ansatz ( apart from those in eqs . ( [ eq : chqc_constraint_total_ghf ] ) and ( [ eq : chqc_constraint_mono_ghf ] ) ) . in principle , the target systems in which we are interested in our group and to which the theory developed in this work is meant to be applied are rather complex ( short peptides , small ligands , etc . ) . they have many degrees of freedom and the different interactions that drive their behaviour typically compete with one another , thus producing complicated , ` frustrated ' energy landscapes ( see refs . @xcite , but note , however , that we do not need to think about macromolecules ; a small molecule like co@xmath305 already has 22 electrons ) . this state of affairs renders the a priori assessment of the accuracy of any approximation to the exact equations an impossible task . as researchers calculate more and more properties of molecular species using quantum chemistry and the results are compared to higher - level theories or to experimental data , much empirical knowledge about ` how good is theory a for calculating property x ' is being gathered . however , if the characterization of a completely new molecule that is not closely related to any one that has been previously studied is tackled with , say , the hartree - fock approximation , it would be very unwise not to ` ask for a second opinion ' . all of this also applies , word by word , to the choice of the constraints on the wavefunction in variational approaches like the one discussed in this section : for example , it is impossible to know a priori what will be the loss of accuracy due to the requirement that the @xmath4-particle wavefunction @xmath67 be a slater determinant as in eq . ( [ eq : chqc_constraint_total_ghf ] ) . however , in the context of the hartree - fock approximation , there exists a way of proceeding , again , partly based on wishful thinking and partly confirmed by actual calculations in particular cases , that is almost unanimously used to choose additional constraints which are expected to yield more _ efficient _ theories . it consists of imposing constraints to the variational wavefunction that are _ properties that the exact solution to the problem does have_. in such a way that the obvious loss of accuracy due to the reduction of the search space is expected to be minimized , while the decrease in computational cost could be considerable . this way of thinking is clearly illustrated by the question of whether or not one should allow that the one - electron spin - orbitals @xmath171 ( and therefore the total wavefunction @xmath67 ) be complex valued . indeed , due to the fact that the electronic hamiltonian in eq ( [ eq : chqc_el_ham_1 ] ) is self - adjoint , the real and imaginary parts of any complex eigenfunction solution of the time independent schrdinger equation in ( [ eq : chqc_bo_summary_a ] ) are also solutions of it @xcite . therefore , the ground - tate , which is the exact solution of the problem that we are trying to solve , may be chosen to be real valued . nevertheless , the exact minimum will not be achieved , in general , in the smaller space defined by the hartree - fock constraints in eqs . ( [ eq : chqc_constraint_total_ghf ] ) and ( [ eq : chqc_constraint_mono_ghf ] ) , so that there is no a priori reason to believe that allowing the hartree - fock wavefunction to take complex values would not improve the results by finding a lower minimum . in fact , in some cases , this happens @xcite . nevertheless , if one constrains the search to real orbitals , the computational cost is reduced by a factor two , and , after all , the whole formalism discussed in this work profits from the imposition of constraints ( starting by the consideration of only one slater determinant ) , all of which save some computational effort at the expense of a reduction in the accuracy . the search for the most efficient of these approximations constitutes the main part of the quantum chemistry field . apart from these ` complex vs. real ' considerations , there exist three further restrictions that are commonly found in the literature and that affect the spin part of the one - electron orbitals @xmath171 . the @xmath4-electron wavefunction @xmath67 of the ghf approximation ( which is the one discussed up to now ) is not an eigenstate of the total - spin operator , @xmath306 , nor of the @xmath74-component of it , @xmath307 @xcite . however , since both of them commute with the electronic hamiltonian in eq . ( [ eq : chqc_el_ham_1 ] ) , the true ground - state of the exact problem can be chosen to be an eigenstate of both operators simultaneously . therefore three additional types of constraints on the spin part of the ghf wavefunction in ( [ eq : chqc_constraint_total_ghf ] ) are typically made that force the variational ansatz to satisfy these ground - state properties and that should be seen in the light of the above discussion , i.e. , as reducing the search space , thus yielding an intrinsically less accurate theory , but also as being good candidates to hope that the computational savings will pay for this . the first approximation to ghf ( in a logical sense ) is called _ nrestricted artree - ock _ ( uhf ) and it consists of asking the orbitals @xmath171 to be a product of a part @xmath308 depending on the positions @xmath172 times a spin eigenstate of the one - electron @xmath309 operator , i.e. , either @xmath310 or @xmath311 ( see eq . ( [ eq : chqc_alpha_beta ] ) ) . this is denoted by @xmath312 , where @xmath313 is either the @xmath14 or the @xmath178 function . now , if we call @xmath314 and @xmath315 the number of spin - orbitals of each type , we have that , differently from the ghf one , the uhf wavefunction @xmath67 is an eigenstate of the @xmath307 operator with eigenvalue @xmath316 ( in atomic units , see sec . [ sec : qc_molecular_h ] ) . however , it is not an eigenstate of @xmath317 and , when this deviation results into a poor description of the observables in which we are interested , we talk about _ spin contamination _ although the uhf wavefunction can be projected into pure @xmath317-states , the result is multideterminantal @xcite and will not be considered here since it spoils many of the properties that render hartree - fock methods a low - cost choice . regarding the computational cost of the uhf approximation , it is certainly lower than that of ghf , since the search space is half as large : in the latter case , we had to consider @xmath118 ( complex or real ) functions of @xmath125 ( the @xmath318 and the @xmath319 , see eq . ( [ eq : chqc_general_so ] ) ) , while in uhf we only have to deal with @xmath4 of them : the @xmath320 . now , if we introduce into the expression for the ghf constrained functional in ( [ eq : chqc_constrained_functional_ghf ] ) the following relations that hold for the uhf spin- and spatial orbitals of course , the average values at both sides of the expressions are taken over different variables : over @xmath66 and @xmath321 on the left - hand side , and over @xmath172 and @xmath322 on the right - hand side . also , let us remark that , although placing _ functions _ as arguments of the kronecker s delta @xmath323 is a bit unorthodox mathematically , it constitutes an intuitive ( and common ) notation . ] [ eq : chqc_help_relations_uhf ] @xmath324 we may perform a derivation analogous to the one performed for the ghf case , and get to a first version of the _ uhf equations _ [ eq : chqc_hf_eqs_vfinal_uhf ] @xmath325\,\phi_{i}({\bm r}):= \left(\hat{h}+\sum^n_j \hat{j}_j[\phi ] - \sum^n_{j}\delta_{\gamma_i \gamma_j } \hat{k}_j[\phi ] \right)\,\phi_{i}({\bm r})= \sum_j \lambda_{ij}\phi_{j}({\bm r})\ , \label{eq : chqc_hf_eqs_vfinal_uhf_a}\\ & \hat{f}^{\mathrm{uhf}}_{i}[\phi]\,\phi^*_{i}({\bm r})= \sum_j \lambda_{ji}\phi^*_{j}({\bm r})\ , \label{eq : chqc_hf_eqs_vfinal_uhf_b}\end{aligned}\ ] ] for all @xmath175 and where the coulomb and exchange operators dependent on the spatial orbitals @xmath326 are defined by their action on an arbitrary function @xmath327 as follows : [ eq : chqc_operators_uhf ] @xmath328\,\varphi({\bm r}):= \left ( \int\frac{|\phi_{j}({\bm r}^{\prime})|^{2 } } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}{\bm r}^{\prime } \right ) \varphi({\bm r})\ , \label{eq : chqc_operators_uhf_a } \\ & \hat{k}_{j}[\phi]\,\varphi({\bm r}):= \left ( \int\frac{\phi^{*}_{j}({\bm r}^{\prime})\,\varphi({\bm r}^{\prime } ) } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}{\bm r}^{\prime } \right ) \phi_{j}({\bm r})\ . \label{eq : chqc_operators_uhf_b } \end{aligned}\ ] ] now , one must note that , differently from the ghf case , due to the fact that the exchange interaction only takes place between orbitals ` of the same spin ' , the _ uhf fock operator _ $ ] depends on the index @xmath12 . this precludes the solution of uhf as a _ pseudoeigenvalue problem ( c.f . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) ) and makes necessary some further considerations in order to arrive to a more directly applicable form of the expressions : first , although the equations for @xmath326 and @xmath330 in ( [ eq : chqc_hf_eqs_vfinal_uhf ] ) can be combined in the same way as in the ghf case to yield the hermiticity conditions @xmath331 , there are fewer lagrange multipliers in the uhf scheme than in the previous derivation . to see this , one must notice that the orthogonality constraints must be imposed _ on the spin - orbitals , not on the spatial orbitals _ ( see eq . ( [ eq : chqc_constraint_mono_ghf ] ) ) . therefore , since two uhf spin - orbitals @xmath332 and @xmath333 are orthogonal no matter the value of @xmath334 due to the different spin parts , the corresponding lagrange multiplier @xmath230 needs not to be included in the constrained functional from which the uhf equations come . this may be incorporated into the formalism by simply using that the matrix @xmath249 in eqs . ( [ eq : chqc_hf_eqs_vfinal_uhf ] ) presents the following block - diagonal form : @xmath335 0 & \lambda^\beta \end{array } \right ) \ , \ ] ] where we have assumed ( without loss of generality ) that the uhf spin - orbitals are ordered in such a way that the @xmath14 ones occur first , and @xmath336 indicates a block of zeros of the appropriate size . ( of course , redundant constraints may be imposed on the orbitals by , for example in this case , including matrix terms in @xmath252 that connect the @xmath14 and @xmath178 spaces . however , in order to know the exact freedom we have in the choice of the constraints , it is convenient to use the minimal number of lagrange multipliers . if this approach were not followed , for example , the discussion below about the diagonalization of @xmath337 and @xmath338 would become much less direct . ) the next step consists in noticing that , despite the dependence of @xmath339 on the orbital index @xmath12 in eqs . ( [ eq : chqc_hf_eqs_vfinal_uhf ] ) , there are actually only _ two _ different fock operators : one for the @xmath14 orbitals and one for the @xmath178 ones . defining the sets of indices @xmath340 and @xmath341 , we can write these two @xmath14 and @xmath178 operators : [ eq : chqc_alpha_beta_f_uhf ] @xmath342=\hat{h}+\sum^{n}_{j=1 } \hat{j}_j[\phi ] - \sum_{j \in a } \hat{k}_j[\phi ] \ , \label{eq : chqc_alpha_beta_f_uhf_a } \\ & \hat{f}^{\mathrm{uhf}}_\beta[\phi]=\hat{h}+\sum^{n}_{j=1 } \hat{j}_j[\phi ] - \sum_{j \in b } \hat{k}_j[\phi ] \ . \label{eq : chqc_alpha_beta_f_uhf_b } \end{aligned}\ ] ] with them , and using the particular structure of the matrix @xmath343 in ( [ eq : chqc_lambda_uhf ] ) , the original uhf equations in ( [ eq : chqc_hf_eqs_vfinal_uhf ] ) are split into two disjoint sets of expressions that are only coupled through the fock operators on the left hand sides , plus the hermiticity condition : [ eq : chqc_hf_eqs_vfinal_1_uhf ] @xmath342\phi_i({\bm r } ) = \sum_{j \in a } \lambda_{ij}^\alpha \phi_j({\bm r } ) \ , \quad \mathrm{if } \ i \in a \ , \label{eq : chqc_hf_eqs_vfinal_1_uhf_a } \\ & \hat{f}^{\mathrm{uhf}}_\beta[\phi]\phi_i({\bm r } ) = \sum_{j \in b } \lambda_{ij}^\beta \phi_j({\bm r } ) \ , \quad \mathrm{if } \ i \in b \ , \label{eq : chqc_hf_eqs_vfinal_1_uhf_b } \\ & \lambda_{ij } = \lambda_{ji}^ * \ . \label{eq : chqc_hf_eqs_vfinal_1_uhf_c}\end{aligned}\ ] ] the last step needed to arrive to the final form of the uhf equations ( found by pople and nesbet @xcite and named after them ) is the diagonalization of both the @xmath344 and @xmath338 hermitian matrices above . in order to achieve this , the orbitals @xmath326 must be transformed similarly to the ghf case . however , this is trickier than it was then , since not only the @xmath4-electron wavefunction @xmath67 and the fock operators must remain invariant under the sought transformation , but also the uhf constraints must be kept . as we saw before , any unitary transformation @xmath251 in the set of spin - orbitals @xmath171 like the one in eq . ( [ eq : chqc_transformation_phi_ghf ] ) is physically legitimate , since it changes the slater determinant by only an unmeasurable phase and leave the fock operators invariant . then , if we write each spin - orbital as in ( [ eq : chqc_general_so ] ) : @xmath345 we can make use of ( [ eq : chqc_transformation_phi_ghf ] ) , to obtain @xmath346 \ .\ ] ] by setting @xmath347 and @xmath348 in this expression , we see that any transformation @xmath251 in the spin - orbitals @xmath171 induces exactly the same transformation in their spatial components @xmath349 and @xmath350 , @xmath351 now , if we order the sets of spin and spatial orbitals : @xmath352 and @xmath353 , with @xmath354 , we can express the _ uhf constraints _ by saying that the @xmath355 must have the form [ eq : chqc_constraints_uhf ] @xmath356 then , since the fact that the @xmath14 orbitals appear first constitutes no loss of generality , we must ask the transformed @xmath357 , with @xmath354 , to have also the structure in ( [ eq : chqc_constraints_uhf ] ) if we want to remain inside the uhf scheme . as a consequence , and due to the linear independence among the orbitals , not every unitary matrix @xmath251 is allowed , but only those of the form @xmath358 0 & u^\beta \end{array } \right ) \ , \ ] ] using the same notation as in eq . ( [ eq : chqc_lambda_uhf ] ) . [ eq : qc_rohf_argument ] this can be easily proved by focusing on a particular value for @xmath12 and @xmath359 in eq . ( [ eq : chqc_transformation_phi_i ] ) , say , @xmath360 and @xmath361 . due to the uhf constraints in ( [ eq : chqc_constraints_uhf ] ) , we know that the left - hand side of such an expression is zero and that only spatial orbitals with @xmath362 appear in the sum on the right - hand side , yielding the relation @xmath363 . but the @xmath364 , with @xmath365 , form an orthonormal , and therefore linearly independent set , so that the only possibility that such a relation can hold is that all coefficients @xmath366 be zero . by repeating this for all @xmath360 and , then , for @xmath367 , the result follows . fortunately , this limited freedom in the choice of @xmath251 is still enough to independently diagonalize @xmath344 and @xmath338 in eq . ( [ eq : chqc_hf_eqs_vfinal_1_uhf ] ) ( which are both hermitian ) by suitably choosing the unitary submatrices @xmath368 and @xmath369 respectively . this takes us to the final , diagonal form of the uhf equations , the _ pople - nesbet equations _ @xcite : [ eq : chqc_hf_eqs_vfinal_2_uhf ] @xmath342\phi_i({\bm r } ) = \varepsilon_{i}^\alpha \phi_i({\bm r } ) \ , \quad \mathrm{if } \ i \in a \ , \label{eq : chqc_hf_eqs_vfinal_2_uhf_a } \\ & \hat{f}^{\mathrm{uhf}}_\beta[\phi]\phi_i({\bm r } ) = \varepsilon_{i}^\beta \phi_i({\bm r } ) \ , \quad \mathrm{if } \ i \in b \ . \label{eq : chqc_hf_eqs_vfinal_2_uhf_b}\end{aligned}\ ] ] although these equations are coupled through the coulomb term in the fock operators on the left - hand side , at each step of the iterative scf procedure , they can be solved as two independent eigenvalue problems . this has allowed to implement them in most quantum chemical packages and it has made uhf calculations now routine . to close the uhf discussion , we shall now study the statistical properties of the probability densities appearing in this model . if we introduce the special form of the uhf orbitals in ( [ eq : chqc_constraints_uhf ] ) into the general expression in ( [ eq : chqc_qrhoij_ghf ] ) , we can calculate the two - body probability density of finding _ any _ electron with coordinates @xmath66 at the same time that _ any other _ electron has coordinates @xmath285 : @xmath370 if we compute this probability density for ` electrons of the same spin ' , i.e. , for @xmath371 , we obtain placing _ a function and a coordinate _ as arguments of the kronecker s delta @xmath372 is even more unorthodox mathematically than placing two functions ( in fact , @xmath372 is exactly the same as @xmath373 ) , however , the intuitive character of the notation compensates again for this . ] @xmath374 & & \frac{1}{2 } \sum_{k , l \in i_\sigma } \left ( \big|\phi_k({\bm r})\big|^{2}\big|\phi_l({\bm r}^{\prime})\big|^{2 } - \phi^{*}_k({\bm r})\,\phi^{*}_l({\bm r}^{\prime})\,\phi_l({\bm r})\ , \phi_k({\bm r}^{\prime } ) \right ) = \nonumber \\ & & \frac{1}{2}\left(\rho^{\mathrm{uhf}}({\bm r},\sigma)\,\rho^{\mathrm{uhf}}({\bm r}^{\prime},\sigma ) - \sum_{k , l } \delta_{\gamma_k \sigma}\delta_{\gamma_l \sigma } \phi^{*}_k({\bm r})\ , \phi^{*}_l({\bm r}^{\prime})\,\phi_l({\bm r})\ , \phi_k({\bm r}^{\prime})\right ) \ , \end{aligned}\ ] ] where the following expression for the one - electron charge density has been used : @xmath375 at this point , note that eq . ( [ eq : chqc_qrhoij_samespins_uhf ] ) contains , like in the ghf case , the exchange correction to the first ( independent electrons ) term . nevertheless , as we advanced , if we calculate the two - body @xmath376 for ` electrons of opposite spin ' , i.e. , for @xmath377 , we have that @xmath378 i.e. , that _ uhf electrons of opposite spin are statistically pairwise independent_. there is another common approximation to ghf that is more restrictive than uhf and is accordingly called _ estricted artree - ock _ ( rhf ) . apart from asking the orbitals @xmath171 to be a product of a spatial part times a spin eigenstate of the one - electron @xmath309 operator ( like in the uhf case ) , in rhf , the number of ` spin - up ' and ` spin - down ' orbitals is the same , @xmath379 ( note that this means that rhf may only be used with molecules containing an even number of electrons ) , and each spatial wavefunction occurs twice : once multiplied by @xmath310 and the other time by @xmath311 . this is typically referred to as a _ closed - shell _ situation and we shall denote it by writing @xmath380 if @xmath381 , and @xmath382 if @xmath383 ; in such a way that there are @xmath384 different spatial orbitals denoted by @xmath385 , with , @xmath386 . using the same notation as in ( [ eq : chqc_constraints_uhf ] ) , the _ rhf constraints _ on the spin - orbitals are [ eq : chqc_constraints_rhf ] @xmath387 due to these additional restrictions , we have that , differently from the ghf and uhf ones , the rhf wavefunction @xmath67 is an eigenstate of both the @xmath306 and the @xmath307 operators , with zero eigenvalue in both cases @xcite , just like the ground - state of the exact problem . i.e. , _ there is no spin - contamination in rhf_. regarding the computational cost of the rhf approximation , it is even lower than that of uhf , since the size of the search space has been reduced to one quarter that of ghf : in the latter case , we had to consider @xmath118 ( complex or real ) functions of @xmath125 ( the @xmath318 and the @xmath319 , see eq . ( [ eq : chqc_general_so ] ) ) , while in rhf we only have to deal with @xmath384 of them : the @xmath388 ( see fig . [ fig : hf_computational_cost ] ) . now , using again the relations in ( [ eq : chqc_help_relations_uhf ] ) , we can derive a first version of the _ rhf equations _ : @xmath389\,\phi_{i}({\bm r}):= \left [ \hat{h}+\sum^{n/2}_j \bigg ( 2\hat{j}_j[\phi ] - \hat{k}_j[\phi ] \bigg ) \right]\phi_{i}({\bm r})= \sum_{j}^{n/2}\lambda_{ij}\phi_{j}({\bm r})\ , \ ] ] for all @xmath386 , where we have used that the _ rhf fock operator _ @xmath390 $ ] , differently from the uhf case , does not depend on the index @xmath391 of the orbital on which it operates , and following the same steps as before , the minimal lagrange multipliers matrix needed to enforce the orthogonality constraints among the _ _ spin__orbitals in the rhf case is @xmath392 0 & \lambda^{(n/2)}/2 \end{array } \right ) \ , \ ] ] where , this time , @xmath393 is an arbitrary @xmath394 hermitian matrix , and the @xmath29 has been included in order to get to the classical rhf equations in ( [ eq : chqc_hf_eqs_vfinal_1_rhf ] ) without irrelevant numerical factors . using the same arguments as for uhf , it is clear that , in order to ` remain inside rhf ' upon an unitary transformation of the spin - orbitals @xmath171 , not every unitary matrix @xmath251 is allowed , but only those of the form @xmath395 0 & u^{(n/2 ) } \end{array } \right ) \ .\ ] ] finally , by suitable choosing the @xmath394 unitary block @xmath396 , the hermitian matrix @xmath397 can be diagonalized and the final , diagonal form of the _ rhf equations _ can be written : @xmath398\,\phi_{i}({\bm r}):= \left [ \hat{h}+\sum^{n/2}_j \bigg ( 2\hat{j}_j[\phi ] - \hat{k}_j[\phi ] \bigg ) \right]\phi_{i}({\bm r})= \varepsilon_{i}\phi_{i}({\bm r})\ , \ ] ] with @xmath386 . as we can see , this version of the hartree - fock theory can be numerically solved as a _ single _ pseudoeigenvalue problem . this , together with the aforementioned small size of the rhf space , has made the rhf approximation ( in its real - valued version ) the first one cast into a computationally manageable form @xcite and the most used one in recent literature @xcite ( for molecules with an even number of electrons ) . now , in order to investigate the statistical features of rhf , if we introduce the special form of the orbitals in ( [ eq : chqc_constraints_rhf ] ) into the general expression in ( [ eq : chqc_qrhoij_ghf ] ) , we can calculate the rhf two - body probability density of finding _ any _ electron with coordinates @xmath66 at the same time that _ any other _ electron has coordinates @xmath285 : @xmath399 where the following expression for the rhf one - electron charge density has been used : @xmath400 we notice that the situation is the same as in the uhf case : for rhf electrons with equal spin , there exists an exchange term in @xmath401 that corrects the ` independent ' part , whereas _ rhf electrons of opposite spin are statistically pairwise independent_. finally , if we follow the same steps as for ghf , in page , we can relate the _ rhf energy _ to the eigenvalues @xmath402 and the two - electron spatial integrals : @xmath403 to close this section , we shall discuss a fourth flavour of hartree - fock which is called _ estricted pen - shell artree - ock _ ( rohf ) . compared to the rest of variants , rohf is the most difficult to derive theoretically and it shall be described here only in an introductory manner . the rohf wavefunction is somewhat between the rhf and the uhf ones , and both of them can be obtained as particular cases of the rohf scheme ( the rhf one provided that the molecule has an even number of electrons ) . in the ( monodeterminantal ) rohf case , the one - electron spin orbitals @xmath171 are constrained to be of two different types : @xmath404 of them are _ doubly occupied _ , like in the rhf case , in such a way that they are formed by only @xmath405 spatial orbitals @xmath406 , each one of them appearing once multiplied by @xmath310 and once by @xmath311 . the associated @xmath404 spin - orbitals are said to belong to the _ closed shell _ part of the wavefunction . the remaining @xmath407 ones are _ singly occupied _ , like in the uhf case , and are said to belong to the _ open shell_. among them , @xmath314 are multiplied by an @xmath310 spin part , and @xmath315 by @xmath311 . if we number the whole set of rohf spatial orbitals @xmath406 , with @xmath408 , in such a way that the doubly occupied ones occur first , with @xmath409 , then the alpha ones , with @xmath410 , and finally the beta ones , with @xmath411 , we can express the _ rohf constraints _ on the spin - orbitals using the same notation as in ( [ eq : chqc_constraints_uhf ] ) and ( [ eq : chqc_constraints_rhf ] ) : [ eq : chqc_constraints_rohf ] @xmath412 where the particular ordering has been chosen in order to facilitate the forecoming calculations . the general monodeterminantal rohf wavefunction considered here and constructed using the above constraints has the same spin properties as the uhf one , i.e. , it is an eigenstate of the the @xmath307 operator with eigenvalue @xmath316 , but it is not an eigenstate of @xmath317 . however , in the particular ( and common ) case in which all the @xmath413 open shell orbitals are constrained to present parallel spin parts ( either all @xmath14 or all @xmath178 ) , the rohf wavefunction becomes an eigenstate of the @xmath317 operator too , with eigenvalue @xmath414 , thus avoiding the problem of spin contamination @xcite . in order to construct wavefunctions with the same @xmath415 but lower @xmath416 , several rohf slater determinants must be linearly combined . the subtleties arising from such a procedure are beyond the scope of this review ; the interested reader may want to check references @xcite , which discuss this topic . regarding the size of the variational space in rohf , it is somewhere between uhf and rhf , depending on the @xmath417 ratio . now , if the rohf constraints in ( [ eq : chqc_constraints_rohf ] ) are imposed on the ghf energy in eq . ( [ eq : chqc_expected_h_2_ghf ] ) , the rohf analogue in terms of the spatial orbitals @xmath406 can be calculated : @xmath418= \sum_{a}f_a\langle\phi_{a}|\hat{h}|\phi_{a}\rangle } \nonumber \\ & & \mbox { } + \frac{1}{2 } \sum_{a , b } \bigg(f_a f_b \langle\phi_a\phi_b|\frac{1}{r}|\phi_a\phi_b\rangle - g_{ab}\langle\phi_a\phi_b|\frac{1}{r}|\phi_b\phi_a\rangle \bigg ) \ , \end{aligned}\ ] ] where the @xmath419 are a sort of ` occupation numbers ' that take the value @xmath420 when @xmath421 ( i.e. , when it corresponds to a closed shell orbital ) and @xmath422 otherwise . the matrix @xmath423 is defined as @xmath424 \hline\\[-8pt ] 1 & \begin{array}{c@{\hspace{5pt}}c } 1^\alpha & 0 \\[5pt ] 0 & 1^\beta \end{array } \end{array } \right ) \ , \ ] ] being @xmath425 a @xmath426 box of 2 s , @xmath427 and @xmath428 , @xmath429 and @xmath430 boxes of 1 s respectively . the off - diagonal blocks contain in all elements the number indicated and are of the appropriate size . then , as we did for ghf , in order to derive the rohf equations , we construct the functional for the conditioned stationary values problem , adding to the energy in ( [ eq : chqc_energy_rohf ] ) the lagrange multipliers terms needed to enforce the orthonormality constraints : @xmath431 = e^{\mathrm{rohf}}\big[\{\phi_a\}\big ] - \sum_{a , b } \lambda_{ab}\ , \big ( \langle\phi_a|\phi_b\rangle-\delta_{ab } \big ) \ .\ ] ] now , like in all restricted hf cases , special attention must be payed to the structure of the matrix @xmath432 , since the requirement is that all _ spin - orbitals _ be orthogonal , not the spatial orbitals . in the rohf case , this leads to explicitly impose the orthonormality conditions ( of course ) inside the three sets of @xmath406 : the doubly occupied , the alpha and the beta ones ; but also between the doubly occupied and the alpha ones , and between the doubly occupied and the beta ones . the orthormality between the alpha and beta sets , however , needs not to be enforced , since the associated spin - orbitals are already orthogonal due to the different spin parts . these considerations lead to the following form for the minimal lagrange multipliers matrix : @xmath433 ( \lambda^{d\alpha})^+ & \lambda^\alpha & 0 \\[5pt ] ( \lambda^{d\beta})^+ & 0 & \lambda^\beta \end{array } \right ) \ , \ ] ] where the notation used for the different blocks connecting the doubly occupied , alpha and beta shells is self - explanatory , and the fact that @xmath434 is hermitian after reaching stationarity has been advanced . next , we impose the condition of zero functional derivative on the functional in ( [ eq : chqc_constrained_functional_rohf ] ) , and obtain a first version of the _ rohf equations _ : [ eq : chqc_hf_eqs_v1_rohf ] @xmath435\phi_{a}({\bm r})= \sum_{b}\lambda_{ab}\phi_{b}({\bm r})\ , \qquad a = 1,\ldots , n_d+n_\alpha+n_\beta \ , \label{eq : chqc_hf_eqs_v1_rohf_a } \\ & \hat{f}_a^{\mathrm{rohf}}[\phi]\,\phi^*_{a}({\bm r})= \sum_{b}\lambda_{ba}\phi^*_{b}({\bm r})\ , \qquad a = 1,\ldots , n_d+n_\alpha+n_\beta \ , \label{eq : chqc_hf_eqs_v1_rohf_b}\end{aligned}\ ] ] where the _ rohf operator _ @xmath436 $ ] is defined as @xmath437:= f_a \hat{h}+ \sum_b \bigg ( f_a f_b \hat{j}_b[\phi ] - g_{ab}\hat{k}_b[\phi ] \bigg ) \ , \ ] ] and the hermiticity property of the lagrange multipliers matrix @xmath438 follows from conjugation and subtraction in eqs . ( [ eq : chqc_hf_eqs_v1_rohf ] ) . again , although the fock operator depends on the index @xmath98 of the orbital upon which it operates , this dependence presents a very particular structure , yielding _ only three _ different types of operators : [ eq : chqc_rohf_ops ] @xmath439:= 2 \hat{h}+ \sum_b \bigg ( 2 f_b \hat{j}_b[\phi ] - f_{b}\hat{k}_b[\phi ] \bigg ) \ , \qquad a \in d \ , \label{eq : chqc_hf_rohf_ops_a } \\ & \hat{f}_\alpha^{\mathrm{rohf}}[\phi]:= \hat{h}+ \sum_{b \in ( d \cup a ) } \bigg(\hat{j}_b[\phi ] - \hat{k}_b[\phi ] \bigg ) \ , \qquad a \in a \ , \label{eq : chqc_hf_rohf_ops_b } \\ & \hat{f}_\beta^{\mathrm{rohf}}[\phi]:= \hat{h}+ \sum_{b \in ( d \cup b ) } \bigg ( \hat{j}_b[\phi ] - \hat{k}_b[\phi ] \bigg ) \ , \qquad a \in b \ . \label{eq : chqc_hf_rohf_ops_c}\end{aligned}\ ] ] note that , in the particular case that we had no open - shell orbitals , the operator for the doubly occupied ones does not reduce ( as it should ) to the rhf one in ( [ eq : chqc_hf_eqs_vfinal_1_rhf ] ) . this is because we hid a factor 2 in the definition of the lagrange multipliers in the rhf derivation . at this point , it would be desirable to continue with the same program that we followed in the uhf and rhf cases and diagonalize the matrix @xmath438 in order to arrive to a system of three pseudoeigenvalue equations , coupled only via the fock operators . nevertheless , this is not possible in rohf , as we shall show in the following lines , and it is the root of rohf being the most tricky flavour in the hartree - fock family . the obstruction to achieve this diagonalization comes from the fact , already used in the uhf case , that not every unitary transformation of the _ spin - orbitals _ is allowed if we want to remain inside the rohf scheme , i.e. , if we want that the transformed spin - orbitals satisfy the same rohf constraints in ( [ eq : chqc_constraints_rohf ] ) that the untransformed ones did . to begin with , if we recall that the doubly occupied spatial orbitals must be orthogonal to both the alpha and the beta ones , we can use the same reasoning as in page to show that the alpha - beta connecting parts of the allowed unitary matrix in this case must be zero . hence , we can write @xmath440 u^{\prime d\alpha } & u^\alpha \end{array } & 0 \\[5pt ] \hline\\[-8pt ] 0 & \begin{array}{c@{\hspace{5pt}}c } u^{\prime d } & u^{d\beta } \\[5pt ] u^{\prime d\beta } & u^\beta \end{array } \end{array } \right ) \ , \ ] ] where the size of each block may be easily found from ( [ eq : chqc_constraints_rohf ] ) and the notation is again self - explanatory . now , in order to obtain further restrictions to the form of @xmath441 , we write the transformation of the _ two _ spin - orbitals in the closed shell that correspond to _ the same _ spatial orbital @xmath406 , with @xmath442 . to this end , we use ( [ eq : chqc_u_rohf ] ) , ( [ eq : chqc_constraints_rohf ] ) , and the appropriate lines in the first and third lines of blocks in ( [ eq : chqc_u_rohf ] ) : @xmath443 using the orthogonality relations inside and among the three sets of spatial orbitals , we multiply the above expression by any @xmath444 , with @xmath445 and integrate on @xmath172 . from this , the equality of @xmath446 and @xmath447 follows , and the corresponding sums in ( [ eq : chqc_rohf_proof1 ] ) subtract to zero , yielding @xmath448 which equates one vector in the linear space spanned by the alpha orbitals to another in the linear space spanned by the beta ones . however , only the zero vector can belong to both spaces if we want the rohf assumptions regarding the spin - orbitals to hold . to see this , we can use a _ reductio ad absurdum _ type of argument : assume that both sides of ( [ eq : chqc_rohf_proof2 ] ) are different from zero . then , we may perform a unitary transformation changing only the alpha and beta spaces , and with no elements connecting the two . this is allowed , since it does not change neither the @xmath4-electron wavefunction , nor the fock operators , nor the rohf constraints . now , if we select the partial unitary transformations in the alpha and beta sets so that they turn the vectors at both sides of ( [ eq : chqc_rohf_proof2 ] ) into single elements in the bases of their respective spaces , we have that a single alpha orbital _ equals _ a beta one . although this can happen in particular cases , we can not ask it or we would be changing the fundamental assumptions made in ( [ eq : chqc_constraints_rohf ] ) . therefore , both sides of ( [ eq : chqc_rohf_proof2 ] ) must be zero , and , since the alpha and beta sets are linearly independent , all coefficients must be zero too . the proof that @xmath449 and @xmath450 are also zero is performed using similar arguments , and the final form of @xmath451 satisfying all the restrictions reads @xmath452 0 & u^\alpha \end{array } & 0 \\[5pt ] \hline\\[-8pt ] 0 & \begin{array}{c@{\hspace{5pt}}c } u^d & 0 \\[5pt ] 0 & u^\beta \end{array } \end{array } \right ) \ .\ ] ] finally , if we write the associated matrix using the @xmath98 indices , i.e. , operating on the set of spatial orbitals with the doubly occupied ones _ unrepeated _ : @xmath453 & u^\alpha & \\[5pt ] 0 & & u^\beta \end{array } \right ) \ , \ ] ] then , the transformed @xmath454 is related to the original one in ( [ eq : chqc_lambda_rohf ] ) through simple matrix multiplication : @xmath455 ( dropping the rohf superindices ) . it is clear that such a restricted @xmath456 does not operate on the off - diagonal blocks of @xmath438 in ( [ eq : chqc_lambda_rohf ] ) and , therefore , _ the sought diagonalization is not possible_. using the fact that , however , the @xmath457 , @xmath344 and @xmath338 do allow to be diagonalized , we can write the final , simplest possible form of the _ rohf equations _ ( forgetting their complex conjugate counterparts ) : [ eq : chqc_hf_eqs_vfinal_rohf ] @xmath439\phi_{a}({\bm r})= \varepsilon^d_a \phi_{a}({\bm r } ) + \sum_{b \in a}\lambda^{d\alpha}_{ab}\phi_{b}({\bm r } ) + \sum_{b \in b}\lambda^{d\beta}_{ab}\phi_{b}({\bm r})\ , \quad a \in d \ , \label{eq : chqc_hf_eqs_vfinal_rohf_a } \\ & \hat{f}_\alpha^{\mathrm{rohf}}[\phi]\phi_{a}({\bm r})= \varepsilon^\alpha_a \phi_{a}({\bm r } ) + \sum_{b \in d}(\lambda^{d\alpha}_{ba})^*\phi_{b}({\bm r})\ , \quad a \in a \ , \label{eq : chqc_hf_eqs_vfinal_rohf_b } \\ & \hat{f}_\beta^{\mathrm{rohf}}[\phi]\phi_{a}({\bm r})= \varepsilon^\beta_a \phi_{a}({\bm r } ) + \sum_{b \in d}(\lambda^{d\beta}_{ba})^*\phi_{b}({\bm r})\ , \quad a \in b \ , \label{eq : chqc_hf_eqs_vfinal_rohf_c}\end{aligned}\ ] ] where the superindices in the matrix elements are only written for visual convenience when comparing with ( [ eq : chqc_lambda_rohf ] ) . in this final form , it is evident that the off - diagonal elements of the lagrange multipliers matrix , i.e. , those related to the orthogonality constraints between the closed and open shells , introduce a coupling in the right - hand side of the rohf equations that completely spoils the possibility of casting them into pseudoeigenvalue ones . in the previous uhf and rhf versions , _ all _ orthogonality constraints were handled by simply choosing a special basis in the space spanned by the spatial orbitals in which the lagrange multipliers matrix was diagonal . however , in the rohf scheme , there is no such basis and we must deal with the problem in a different way , resulting into higher computational and theoretical difficulty . since the pioneering work by roothaan @xcite , the solution of the rohf problem has been attempted by distinct means , ranging from directly tackling the rohf equations in ( [ eq : chqc_hf_eqs_vfinal_rohf ] ) explicitly forcing the orthogonality constraints @xcite , to the construction of a so - called _ unified coupling operator _ @xcite , which allows to turn the rohf scheme into a single pseudoeigenvalue problem at the price of introducing certain ambiguities in the one - electron orbital energies @xcite . the details and subtleties involved in these issues are beyond the scope of a review of the fundamental topics such as this one . the interested reader may want to check the more specialized accounts in @xcite . the hartree - fock equations in the rhf form in expression ( [ eq : chqc_hf_eqs_vfinal_rhf ] ) are a set of @xmath384 coupled integro - differential equations . as such , they can be tackled by finite - differences methods and solved on a discrete grid ; this is known as _ numerical hartree - fock _ @xcite , and , given the present power of computers , it is only applicable to very small molecules . in order to deal with larger systems , such as biological macromolecules , independently proposed by roothaan @xcite and hall @xcite in 1951 , a different kind of discretization must be performed , not in @xmath458 but in the hilbert space @xmath459 of the one - electron orbitals . hence , although the actual dimension of @xmath459 is infinite , we shall approximate any function in it by a finite linear combination of @xmath460 different functions @xmath461 in all this section and the next one , the indices belonging to the first letters of the alphabet , @xmath462 , etc . , run from 1 to @xmath460 ( the number of functions in the finite basis set ) ; whereas those named with capital letters from @xmath391 towards the end of the alphabet , @xmath463 , etc . , run from 1 to @xmath384 ( the number of spatial wavefunctions @xmath464 , also termed the number of _ occupied orbitals _ ) . ] . in particular , the one - electron orbitals that make up the rhf wavefunction , shall be approximated by @xmath465 in both cases , numerical hartree - fock and discretization of the function space , the correct result can be only be reached asymptotically ; when the grid is very fine , for the former , and when @xmath466 , for the latter . this exact result , which , in the case of small systems , can be calculated up to several significant digits , is known as the _ hartree - fock limit _ @xcite . in practical cases , however , @xmath460 is finite ( often , only about an order of magnitude larger than @xmath384 ) and the set @xmath467 in the expression above is called the _ basis set_. we shall devote the next section to discuss its special characteristics , but , for now , it suffices to say that , in typical applications , the functions @xmath461 are _ atom - centred _ , i.e. , each one of them has non - negligible value only in the vicinity of a particular nucleus . therefore , like all the electronic wavefunctions we have dealt with in the last sections , they parametrically depend on the positions @xmath49 of the nuclei ( see sec . [ sec : qc_born - oppenheimer ] ) . this is why , sometimes , the functions @xmath461 are called _ _ tomic rbitals _ _ some authors @xcite suggest that , being strict , the term _ atomic orbitals _ should be reserved for the one - electron wavefunctions @xmath464 that are the solution of the hartree - fock problem ( or even to the exact schrdinger equation of the isolated atom ) , and that the elements @xmath461 in the basis set should be termed simply _ localized functions_. however , it is very common in the literature not to follow this recommendation and choose the designation that appear in the text @xcite . we shall do the same for simplicity . ] ( ao ) ( since they are localized at individual atoms ) , the @xmath464 are referred to as _ olecular rbitals _ ( mo ) ( since they typically have non - negligible value in the whole space occupied by the molecule ) , and the approximation in eq . ( [ eq : chqc_lcao ] ) is called _ inear ombination of tomic rbitals _ ( lcao ) . in addition , since we voluntarily circumscribe to real - rhf , we assume that both the coefficients @xmath468 and the functions @xmath461 in the above expression are real . now , if we introduce the linear combination in eq . ( [ eq : chqc_lcao ] ) into the hartree - fock equations in ( [ eq : chqc_hf_eqs_vfinal_rhf ] ) , multiply the result from the left by @xmath469 ( for a general value of @xmath470 ) and integrate on @xmath471 , we obtain @xmath472 where we denote by @xmath473 the @xmath474-element of the _ _ fock matrix _ _ note that the rhf superindex has been dropped from @xmath475 . ] , and by @xmath476 the one of the _ overlap matrix _ , defined as @xmath477 \ , | \chi_a \rangle \qquad \mathrm{and } \qquad s_{ba}:=\langle \chi_b | \chi_a \rangle \ , \ ] ] respectively . note that we do not ask the @xmath461 in the basis set to be mutually orthogonal , so that the overlap matrix is not diagonal in general . next , if we define the @xmath478 matrices @xmath479:=(f_{ab})$ ] and @xmath480 , together with the ( column ) @xmath460-vector @xmath481 , we can write eq . ( [ eq : chqc_rh_1 ] ) in matricial form : @xmath482 c_i=\varepsilon_i\,sc_{i}\ .\ ] ] hence , using the lcao approximation , we have traded the @xmath384 coupled integro - differential hartree - fock equations in ( [ eq : chqc_hf_eqs_vfinal_rhf ] ) for this system of @xmath384 algebraic equations for the @xmath384 orbital energies @xmath402 and the @xmath483 coefficients @xmath468 , which are called _ roothaan - hall equations _ @xcite and which are manageable in a computer . now , if we forget for a moment that the fock matrix depends on the coefficients @xmath468 ( as stressed by the notation @xmath479 $ ] ) and also that we are only looking for @xmath384 vectors @xmath484 while the matrices @xmath475 and @xmath485 are @xmath486 , we may regard the above expression as a @xmath460-dimensional _ generalized eigenvalue problem_. many properties are shared between this kind of problem and a classical eigenvalue problem ( i.e. , one in which @xmath487 ) @xcite , being the most important one that , due to the hermiticity of @xmath479 $ ] , one can find an orthonormal set of @xmath460 vectors @xmath488 corresponding to real eigenvalues @xmath489 ( where , of course , some eigenvalue could be repeated ) . in fact , it is using this formalism how most of actual hartree - fock computations are performed , although the reader must also note that other approaches , in which the orthonormality constraints are automatically satisfied due to the choice of variables also exist in the literature @xcite . the general outline of the iterative procedure is essentially the same as the one discussed in sec . [ sec : qc_hartree - fock ] : choose a _ starting - guess _ for the coefficients @xmath468 ( let us denote it by @xmath490 ) , construct the corresponding fock matrix @xmath491 $ ] note ( in eq . ( [ eq : chqc_matrices_2 ] ) , for example ) that the fock matrix only depends on the vectors @xmath488 with @xmath492 . ] and solve the generalized eigenvalue problem in eq . ( [ eq : chqc_rh_2 ] ) . by virtue of the aufbau principle discussed in the previous section , from the @xmath460 eigenvectors @xmath488 , keep the @xmath384 ones @xmath493 that correspond to the @xmath384 lowest eigenvalues @xmath494 , construct the new fock matrix @xmath495 $ ] and iterate ( by convention , the eigenvalues @xmath496 , for all @xmath151 , are ordered from the lowest to the largest as @xmath98 runs from 1 to @xmath460 ) . this procedure ends when the @xmath151-th solution is close enough ( in a suitable defined way ) to the @xmath301-th one . also , note that , after convergence has been achieved , we end up with @xmath460 orthogonal vectors @xmath497 . only the @xmath384 ones that correspond to the lowest eigenvalues represent real one - electron solutions and they are called _ occupied orbitals _ ; the @xmath498 remaining ones do not enter in the @xmath4-electron wavefunction ( although they are relevant for calculating corrections to the hartree - fock results ) and they are called _ virtual orbitals_. regarding the mathematical foundations of this procedure , let us stress , however , that , whereas in the finite - dimensional ghf and uhf cases it has been proved that the analogous of lieb and simon s theorem ( see the previous section ) is satisfied , i.e. , that the global minimum of the original optimization problem corresponds to the lowest eigenvalues of the self - consistent fock operator , in the rhf and rohf cases , contrarily , no proof seems to exist @xcite . of course , in practical applications , the positive result is assumed to hold . finally , if we expand @xmath499 in eq . ( [ eq : chqc_matrices_1 ] ) , using the shorthand @xmath500 for @xmath501 , we have @xmath502 } \underbrace{\left ( 2 \langle ac|\,\frac{1}{r}\,|bd\rangle - \langle ac|\,\frac{1}{r}\,|db\rangle\right ) } _ { \displaystyle g^{cd}_{ab } } \ , \ ] ] where we have introduced the _ density matrix _ @xmath503 $ ] , and also the matrix @xmath504 , made up by the _ two - electron four - centre integrals _ @xmath505 ( also called _ lectron epulsion ntegrals _ ( eris ) ) defined by @xmath506 it is also convenient to introduce the _ coulomb _ ( @xmath507 $ ] ) and _ exchange _ ( @xmath508 $ ] ) _ matrices _ [ eq : chqc_coulomb_exchange ] @xmath509 : = \sum_{c , d}d_{cd}[c]\langle ac|\,\frac{1}{r}\,|bd\rangle \ , \label{eq : chqc_coulomb_exchange_a } \\ & k_{ab}[c ] : = \sum_{c , d}d_{cd}[c]\langle ac|\,\frac{1}{r}\,|db\rangle \ , \label{eq : chqc_coulomb_exchange_b } \end{aligned}\ ] ] in terms of which , the fock operator in eq . ( [ eq : chqc_matrices_2 ] ) may be expressed as @xmath510 - k_{ab}[c ] \ .\ ] ] after scf convergence has been achieved , the _ rhf energy _ in the finite - dimensional case can be computed using the discretized version of eq . ( [ eq : chqc_energy_rhf ] ) : @xmath511d_{bd}[c]\langle ab|\,\frac{1}{r}\,|cd\rangle\ , \end{aligned}\ ] ] where a convenient rearrangement of the indices in the two sums has been performed from the first to the second line . in principle , arbitrary functions may be chosen as the @xmath461 to solve the roothaan - hall equations in the previous section , however , in eq . ( [ eq : chqc_matrices_2 ] ) , we see that one of the main numerical bottlenecks in scf calculations arises from the necessity of calculating the @xmath512 four - centre integrals if the symmetry properties of the integrals are used , the precise number of eris is found to be @xmath513 @xcite . ] @xmath514 , since the solution of the generalized eigenvalue problem in eq . ( [ eq : chqc_rh_2 ] ) typically scales only like @xmath515 , and there are two - centre @xmath516 integrals ( see however the next section ) . either if these integrals are calculated at each iterative step and directly taken from ram memory ( _ direct scf _ ) or if they are calculated at the first step , written to disk , and then read from there when needed ( _ conventional scf _ ) , an appropriate choice of the functions @xmath461 in the finite basis set is essential if accurate results are sought , @xmath460 is intended to be kept as small as possible and the integrals are wanted to be computed rapidly . when one moves into higher - level theoretical descriptions and the numerical complexity scales with @xmath460 even more unpleasantly , the importance of this choice greatly increases . in this section , in order to support that study , we shall introduce some of the concepts involved in the interesting field of basis - set design . for further details not covered here , the reader may want to check refs . @xcite . the only analytically solvable molecular problem in non - relativistic quantum mechanics is the _ hydrogen - like atom _ , i.e. , the system formed by a nucleus of charge @xmath517 and only one electron ( h , he@xmath518 , li@xmath519 , etc . ) . therefore , it is not strange that all the thinking about atomic - centred basis sets in quantum chemistry is much influenced by the particular solution to this problem . the spatial eigenfunctions of the hamiltonian operator of an hydrogen - like atom , in atomic units and spherical coordinates , read for consistency with the rest of the text , the born - oppenheimer approximation has been also assumed here . so that the _ reduced mass _ @xmath520 that should enter the expression is considered to be the mass of the electron @xmath521 ( recall that , in atomic units , @xmath522 and @xmath523 ) . ] @xmath524 ^ 3 } } \left ( \frac{2z}{n } r \right)^l l^{2l+1}_{n - l-1}\left ( \frac{2z}{n } r \right ) e^{-zr / n } y_{lm}(\theta,\varphi ) \ , \ ] ] where @xmath151 , @xmath222 and @xmath25 are the energy , total angular momentum and @xmath74-angular momentum quantum numbers , respectively . their ranges of variation are coupled : all being integers , @xmath151 runs from 1 to @xmath525 , @xmath222 from 0 to @xmath526 and @xmath25 from @xmath527 to @xmath222 . the function @xmath528 is a _ generalized laguerre polynomial _ @xcite , for which it suffices to say here that it is of order @xmath529 ( thus having , in general , @xmath529 zeros ) , and the function @xmath530 is a _ spherical harmonic _ , which is a simultaneous eigenfunction of the total angular momentum operator @xmath531 ( with eigenvalue @xmath532 ) and of its @xmath74-component @xmath533 ( with eigenvalue @xmath25 ) . the hope that the one - electron orbitals that are the solutions of the hartree - fock problem in many - electron atoms could not be very different from the @xmath534 [ page : foot : orbitals_only_in_hf ] above note that the @xmath4-electron wavefunction of the exact ground - state of a non - hydrogen - like atom depends on 3@xmath4 spatial variables in a way that can not be written , in general , as a slater determinant of one - electron functions . the image of single electrons occupying definite orbitals , together with the possibility of comparing them with the one - particle eigenfunctions of the hamiltonian of hydrogen - like atoms , vanishes completely outside the hartree - fock formalism . ] , together with the powerful chemical intuition that states that ` atoms - in - molecules are not very different from atoms - alone ' , is what mainly drives the choice of the functions @xmath461 in the basis set , and , in the end , the variational procedure that will be followed is expected to fix the largest failures coming from these too - simplistic assumptions . hence , it is customary to choose functions that are centred at atomic nuclei and that partially resemble the exact solutions for hydrogen - like atoms . in this spirit , the first type of aos to be tried @xcite were the _ later - ype rbitals _ ( stos ) , proposed by slater @xcite and zener @xcite in 1930 : @xmath535 where @xmath536 is a normalization constant and @xmath537 is an adjustable parameter . the index @xmath538 is that of the nucleus at which the function is centred , and , of course , in the majority of cases , there will be several @xmath539 corresponding to different values of @xmath98 centred at the same nucleus . the integers @xmath540 and @xmath541 can be considered quantum numbers , since , due to the fact that the only angular dependence is in @xmath542 ( see below for a definition ) , the sto defined above is still a simultaneous eigenstate of the one - electron angular momentum operators @xmath531 and @xmath533 ( with the origin placed at @xmath543 ) . the parameter @xmath544 , however , should be regarded as a ` principal ( or energy ) quantum number ' only by analogy , since , on the one hand , it does not exist a ` one - atom hamiltonian ' whose exact eigenfunctions it could label and , on the other hand , only the leading term of the laguerre polynomial in eq . ( [ eq : chqc_hydrogen - like ] ) has been kept in the sto if we notice that , within the set of all possible stos ( as defined in eq . ( [ eq : chqc_sto ] ) ) , every hydrogen - like energy eigenfunction ( see eq . ( [ eq : chqc_hydrogen - like ] ) ) can be formed as a linear combination , we easily see that the stos constitute a complete basis set . this is important to ensure that the hartree - fock limit could be actually approached by increasing @xmath460 . ] . additionally , in the above notation , the fact that @xmath539 parametrically depends on the position of a certain @xmath538-th nucleus has been stressed , and the functions @xmath542 , which are called _ real spherical harmonics _ @xcite ( remember that we want to do real rhf ) , are defined in terms of the classical _ spherical harmonics _ @xmath545 by [ eq : chqc_real_spherical_harmonics ] @xmath546 where @xmath547 stands for _ cosine _ , @xmath548 for _ sine _ , the functions @xmath549 are the _ associated legendre polynomials _ @xcite , and the spherical coordinates @xmath550 and @xmath551 also carry the @xmath538-label to remind that the origin of coordinates in terms of which they are defined is located at @xmath543 . also note that , using that @xmath552 , there is the same number of real spherical harmonics as of classical ones . these @xmath539 have some good physical properties . among them , we shall mention that , for @xmath553 , they present a _ cusp _ ( a discontinuity in the radial derivative ) , as required by kato s theorem @xcite ; and also that they decay at an exponential rate when @xmath554 , which is consistent with the image that , an electron that is taken apart from the vicinity of the nucleus must ` see ' , at large distances , an unstructured point - like charge ( see , for example , the sto in fig . [ fig : contraction ] ) . finally , the fact that they do not present radial nodes ( due to the aforementioned absence of the non - leading terms of the laguerre polynomial in eq . ( [ eq : chqc_hydrogen - like ] ) ) can be solved by making linear combinations of functions with different values [ page : foot : r_exponent_idontcare ] of @xmath537 this way of proceeding renders the choice of the exponent carried by the @xmath555 part ( @xmath556 in the case of the sto in eq . ( [ eq : chqc_sto ] ) ) a rather arbitrary one . as a consequence , different definitions may be found in the literature and the particular exponent chosen in actual calculations turns out to be mostly a matter of computational convenience . ] . now , despite their being good theoretical candidates to expand the mo @xmath464 that make up the @xmath4-particle solution of the hartree - fock problem , these stos have serious computational drawbacks : whereas the two - centre integrals ( such as @xmath516 in eq . ( [ eq : chqc_matrices_2 ] ) ) can be calculated analytically , the four - centre eris @xmath505 can not @xcite if functions like the ones in eq . ( [ eq : chqc_sto ] ) are used . this fact , which was known as `` the nightmare of the integrals '' in the first days of computational quantum chemistry @xcite , precludes the use of stos in practical ab initio calculations of large molecules . a major step to overcome these difficulties that has revolutioned the whole field of quantum chemistry @xcite was the introduction of _ artesian aussian - ype rbitals _ ( cgto ) : @xmath557 where the @xmath558 and the @xmath559 , with @xmath560 , are the euclidean coordinates of the electron and the @xmath538-th nucleus respectively , and the integers @xmath561 , @xmath562 and @xmath563 , which take values from 0 to @xmath525 , are called _ _ orbital quantum numbers _ _ since the harmonic - oscillator energy eigenfunctions can be constructed as linear combinations of cartesian gtos , we have that the latter constitute a complete basis set and , like in the case of the stos , we may expect that the hartree - fock limit is approached as @xmath460 is increased . ] . although these gtos do not have the good physical properties of the stos ( compare , for example , the sto and the gto in fig . [ fig : contraction ] ) , in 1950 , boys @xcite showed that all the integrals appearing in scf theory could be calculated analytically if the @xmath461 had the form in eq . ( [ eq : chqc_gto ] ) . the enormous computational advantage that this entails makes possible to use a much larger number of functions to expand the one - electron orbitals @xmath326 if gtos are used , partially overcoming their bad short- and long - range behaviour and making the gaussian - type orbitals the universally preferred choice in scf calculations @xcite . to remedy the fact that the angular behaviour of the cartesian gtos in eq . ( [ eq : chqc_gto ] ) is somewhat hidden , they may be linearly combined to form _ pherical aussian - ype rbitals _ ( sgto ) : @xmath564 which are proportional to the real spherical harmonic @xmath565 , and to which the same remarks made in footnote [ foot : r_exponent_idontcare ] in page for the stos , regarding the exponent in the @xmath555 part , may be applied . the fine mathematical details about the linear combination that relates the cartesian gtos to the spherical ones are beyond the scope of this introduction . we refer the reader to refs . @xcite and @xcite for further information and remark here some points that will have interest in the subsequent discussion . first , the cgtos that are combined to make up a sgto must have all the same value of @xmath566 and , consequently , this sum of the three orbital quantum numbers @xmath561 , @xmath562 and @xmath563 in a particular cartesian gto is typically ( albeit dangerously ) referred to as the _ angular momentum _ of the function . in addition , apart from the numerical value of @xmath540 , the spectroscopic notation is commonly used in the literature , so that cgtos with @xmath567 are called _ s _ , _ p _ , _ d _ , _ f _ , _ g _ , _ h _ , , respectively . where the first four come from the archaic words _ harp _ , _ rincipal _ , _ iffuse _ and _ undamental _ , while the subsequent ones proceed in alphabetical order . second , for a given @xmath568 , there are more cartesian gtos ( @xmath569 ) than spherical ones ( @xmath570 ) , in such a way that , from the @xmath569 functionally independent linear combinations that can be formed using the cgtos of angular momentum @xmath540 , only the angular part of @xmath570 of them turns out to be proportional to a real spherical harmonic @xmath571 ; the rest of them are proportional to real spherical harmonic functions with a different value of the angular momentum quantum number . for example , from the six different d - cartesian gtos , whose polynomial parts are @xmath572 , @xmath573 , @xmath574 , @xmath575 , @xmath576 and @xmath577 ( using an evident , compact notation ) , only five different spherical gtos can be constructed : the ones with polynomial parts proportional to @xmath578 , @xmath576 , @xmath577 , @xmath579 and @xmath575 @xcite . among these new sgtos , which , in turn , are proportional ( neglecting also powers of @xmath580 , see footnote [ foot : r_exponent_idontcare ] in page ) to the real spherical harmonics @xmath581 , @xmath582 , @xmath583 , @xmath584 and @xmath585 , the linear combination @xmath586 is missing , since it presents the angular behaviour of an s - orbital ( proportional to @xmath587 ) . finally , let us remark that , whereas cartesian gtos in eq . ( [ eq : chqc_gto ] ) are easier to be coded in computer applications than sgtos@xcite , it is commonly accepted that these spurious spherical orbitals of lower angular momentum that appear when cgtos are used do not constitute efficient choices to be included in a basis set @xcite ( after all , if we wanted an additional s - function , why include it in such an indirect and clumsy way instead of just designing a specific one that suits our particular needs ? ) . consequently , the most common practice in the field is to use cartesian gtos removing from the basis sets the linear combinations such as the @xmath586 above . now , even if the integrals involving cgtos can be computed analytically , there are still @xmath512 of them in a scf calculation . for example , in the model dipeptide hco - l - ala - nh@xmath305 , which is commonly used to mimic an alanine residue in a protein @xcite , there are 62 electrons and henceforth 31 rhf spatial orbitals @xmath464 . if a basis set with only 31 functions is used ( this is a lower bound that will be rarely reached in practical calculations due to symmetry issues , see below ) , near a million of four - centre @xmath588 integrals must be computed . this is why , one must use the freedom that remains once the decision of sticking to cgtos has been taken ( namely , the choice of the _ exponents _ @xmath537 and the _ angular momentum _ @xmath540 ) to design basis sets that account for the relevant behaviour of the systems studied while keeping @xmath460 below the ` pain threshold ' . the work by nobel prize john pople s group has been a major reference in this discipline , and their sto-@xmath151 g family @xcite , together with the split - valence gaussian basis sets , 3 - 21 g , 4 - 31 g , 6 - 31 g , etc . @xcite , shall be used here to exemplify some relevant issues . however , note that most of the concepts introduced are also applicable to more modern basis sets , such as those by dunning @xcite . to begin with , let us recall that the short- and long - range behaviour of the slater - type orbitals in eq . ( [ eq : chqc_sto ] ) is better than that of the more computationally efficient gtos . in order to improve the physical properties of the latter , it is customary to linearly combine @xmath589 cartesian gtos , denoted now by @xmath590 ( @xmath591 ) , and termed _ rimitive aussian - ype rbitals _ ( pgto ) , having the same atomic centre @xmath543 , the same set of orbital quantum numbers , @xmath561 , @xmath562 and @xmath563 , but different exponents @xmath592 , to make up a _ ontracted aussian - ype rbitals _ ( cgto ) , defined by @xmath593 where the normalization constants @xmath594 have been kept inside the sum because they typically depend on @xmath595 . also , we denote now by @xmath596 the number of contracted gtos and , by @xmath597 , the number of primitive ones . in the sto-@xmath151 g family of basis sets , for example , @xmath151 primitive gtos are used for each contracted one , fitting the coefficients @xmath598 and the exponents @xmath592 to resemble the radial behavior of slater - type orbitals @xcite . in fig . [ fig : contraction ] , the 1s - contracted gto ( see the discussion below ) of the hydrogen atom in the sto-3 g basis set is depicted , together with the three primitive gtos that form it and the sto that is meant to [ page : foot : basisform ] be approximated basis sets were obtained from the extensible computational chemistry environment basis set database at ` http://www.emsl.pnl.gov/forms/basisform.html ` , version 02/25/04 , as developed and distributed by the molecular science computing facility , environmental and molecular sciences laboratory which is part of the pacific northwest laboratory , p.o . box 999 , richland , washington 99352 , usa , and funded by the u.s . department of energy . the pacific northwest laboratory is a multi - program laboratory operated by battelle memorial institute for the u.s . department of energy under contract de - ac06 - 76rlo 1830 . contact karen schuchardt for further information . ] . we can see that the contracted gto has a very similar behavior to the sto in a wide range of distances , while the single gto that is also shown in the figure ( with the same norm and exponent as the sto ) has not . typically , the fitting procedure that leads to contracted gtos is performed on isolated atoms and , then , the already mentioned chemical intuition that states that ` atoms - in - molecules are not very different from atoms - alone ' is invoked to keep the linear combinations fixed from there on . obviously , better results would be obtained if the contraction coefficients were allowed to vary . moreover , the number of four - centre integrals that need to be calculated depends on the number of _ primitive _ gtos ( like @xmath599 ) , so that we have not gained anything on this point by contracting . however , the size of the variational space is @xmath596 ( i.e. , the number of _ contracted _ gtos ) , in such a way that , once the integrals @xmath588 are calculated ( for non - direct scf ) , all subsequent steps in the iterative self - consistent procedure scale as powers of @xmath596 . also , the disk storage ( again , for non - direct schemes ) depends on the number of _ contracted _ gtos and , frequently , it is the disk storage and not the cpu time the limiting factor of a calculation . an additional chemical concept that is usually defined in this context and that is needed to continue with the discussion is that of _ shell _ : atomic shells in quantum chemistry are defined analogously to those of the hydrogen atom , so that each electron is regarded as ` filling ' the multi - electron atom ` orbitals ' according to hund s rules @xcite . hence , the _ occupied shells _ of carbon , for example , are defined to be 1s , 2s and 2p , whereas those of , say , silicon , would be 1s , 2s , 2p , 3s an 3p . each shell may contain @xmath600 electrons if complete , where @xmath601 accounts for the orbital angular momentum multiplicity and the 2 factor for that of electron spin . [ tab : contraction ] thus , using these definitions , all the basis sets in the aforementioned sto-@xmath151 g family are _ minimal _ ; in the sense that they are made up of only @xmath601 contracted gtos for each completely or partially occupied shell , so that the sto-@xmath151 g basis sets for carbon , for example , contain two s - type contracted gtos ( one for the 1s- and the other for the 2s - shell ) and three p - type ones ( belonging to the 2p - shell ) . moreover , due to rotational - symmetry arguments in the isolated atoms , all the @xmath601 functions in a given shell are chosen to have the same exponents and the same contraction coefficients , differing only on the polynomial that multiplies the gaussian part . such @xmath601 cgtos shall be said to constitute a _ aussian hell _ ( gs ) , and we shall also distinguish between the _ rimitive _ ( pgs ) and _ ontracted _ ( cgs ) versions . in table [ tab : contraction ] , the exponents @xmath595 and the contraction coefficients @xmath598 of the primitive gtos that make up the three different shells in the sto-3 g basis set for carbon are presented ( see ref . @xcite and footnote [ foot : basisform ] in page ) . the fact that the exponents @xmath595 in the 2s- and 2p - shells are constrained to be the same is a particularity of some basis sets ( like this one ) which saves some computational effort and deserves no further attention . next , let us introduce a common notation that is used to describe the contraction scheme : it reads ( _ primitive shells _ ) / [ _ contracted shells _ ] , or alternatively ( _ primitive shells _ ) @xmath602 [ _ contracted shells _ ] . according to it , the sto-3 g basis set for carbon , for example , is denoted as ( 6s,3p ) @xmath602 [ 2s,1p ] , or ( 6,3 ) @xmath602 [ 2,1 ] . moreover , since for organic molecules it is frequent to have only hydrogens and the 1st - row atoms c , n and o ( whose occupied shells are identical ) in proteins , one may also have sulphur in cysteine and methionine residues . ] , the notation is typically extended and the two groups of shells are separated by a slash ; as in ( 6s,3p/3s ) @xmath602 [ 2s,1p/1s ] for sto-3 g . [ tab : contraction2 ] the first improvement that can be implemented on a minimal basis set such as the ones in the sto-@xmath151 g is the _ splitting _ , which consists in including more than one gaussian shell for each occupied one . if the splitting is evenly performed , i.e. , each shell has the same number of gss , then the basis set is called _ ouble eta _ ( dz ) , _ riple eta _ ( tz ) , _ uadruple eta _ ( qz ) , _ uintuple eta _ ( 5z ) , _ extuple eta _ ( 6z ) , and so on ; where the word _ zeta _ comes from the greek letter @xmath603 used for the exponents . a hypothetical tz basis set in which each cgto is made by , say , four primitive gtos , would read ( 24s,12p/12s ) @xmath602 [ 6s,3p/3s ] in the aforementioned notation . at this point , the already familiar intuition that says that ` atoms - in - molecules are not very different from atoms - alone ' must be refined with another bit of chemical experience and qualified by noticing that ` core electrons are less affected by the molecular environment and the formation of bonds than valence electrons ' recall that , for the very concept of ` core ' or ` valence electrons ' ( actually for any label applied to a single electron ) to have any sense , we must be in the hartree - fock formalism ( see footnote [ foot : orbitals_only_in_hf ] in page ) . ] . in this spirit , the above evenness among different shells is typically broken , and distinct basis elements are used for the energetically lowest lying ( _ core _ ) shells than for the highest lying ( _ valence _ ) ones . on one side , the contraction scheme may be different . in which case , the notation used up to now becomes ambiguous , since , for example , the designation ( 6s,3p/3s ) @xmath602 [ 2s,1p/1s ] , that was said to correspond to sto-3 g , would be identical for a _ different _ basis set in which the 1s - gaussian shell of heavy atoms be formed by 4 pgss and the 2s - gaussian shell by 2 pgss ( in 1st - row atoms , the 2s- and 2p - shells are defined as valence and the 1s - one as core , while in hydrogen atoms , the 1s - shell is a valence one ) . this problem can be solved by explicitly indicating how many primitive gss form each contracted one , so that , for example , the sto-3 g basis set is denoted by ( 33,3/3 ) @xmath602 [ 2,1/1 ] , while the other one mentioned would be ( 42,3/3 ) @xmath602 [ 2,1/1 ] ( we have chosen to omit the angular momentum labels this time ) . the other point at which the core and valence gaussian shells may differ is in their respective ` zeta quality ' , i.e. , the basis set may contain a different number of contracted gaussian shells in each case . for example , it is very common to use a single cgs for the core shells and a multiple splitting for the valence ones . these type of basis sets are called _ split - valence _ and the way of naming their quality is the same as before , except for the fact that a capital v , standing for _ alence _ , is added either at the beginning or at the end of the acronyms dz , tz , qz , etc . , thus becoming vdz , vtz , vqz , etc . or dzv , tzv , qzv , etc . pople s 3 - 21 g @xcite , 4 - 31 g @xcite , 6 - 31 g @xcite and 6 - 311 g @xcite are well - known examples of split - valence basis sets that are commonly used for scf calculations in organic molecules and that present the two characteristics discussed above . their names indicate the contraction scheme , in such a way that the number before the dash represents how many primitive gss form the single contracted gss that is used for core shells , and the numbers after the dash how the valence shells are contracted , in much the same way as the notation in the previous paragraphs . for example , the 6 - 31 g basis set ( see table [ tab : contraction2 ] ) , contains one cgs made up of six primitive gss in the 1s - core shell of heavy atoms ( the 6 before the dash ) and two cgs , formed by three and one pgss respectively , in the 2s- and 2p - valence shells of heavy atoms and in the 1s - shell of hydrogens ( the 31 after the dash ) . the 6 - 311 g basis set , in turn , is just the same but with an additional single - primitive gaussian shell of functions in the valence region . finally , to fix the concepts discussed , let us mention that , using the notation introduced above , these two basis sets may be written as ( 631,31/31 ) @xmath602 [ 3,2/2 ] and ( 6311,311/311 ) @xmath602 [ 4,3/3 ] , respectively . two further improvements that are typically used and that may also be incorporated to pople s split - valence basis sets are the addition of _ polarization _ @xcite or _ diffuse functions _ @xcite . we shall discuss them both to close both this section and the work . up to now , neither the contraction nor the splitting involved gtos of larger angular momentum than the largest one among the occupied shells . however , the molecular environment is highly anisotropic and , for most practical applications , it turns out to be convenient to add these _ polarization _ ( large angular momentum ) gaussian shells to the basis set , since they present lower symmetry than the gss discussed in the preceding paragraphs . typically , the polarization shells are single - primitive gss and they are denoted by adding a capital p to the end of the previous acronyms , resulting into , for example , dzp , tzp , vqzp , etc . , or , say , dz2p , tz3p , vqz4p if more than one polarization shell is added . in the case of pople s basis sets @xcite , these improvements are denoted by specifying , in brackets and after the letter g , the number and type of the polarization shells separating heavy atoms and hydrogens by a comma there also exists an old notation for the addition of a single polarization shell per atom that reads 6 - 31 g * * and that is equivalent to 6 - 31g(d , p ) . ] . for example , the basis set 6 - 31g(2df , p ) contains the same gaussian shells as the original 6 - 31 g plus two d - type shells and one f - type shell centred at the heavy atoms , as well as one p - type shell centred at the hydrogens . finally , for calculations in charged species ( specially anions ) , where the charge density extends further in space and the tails of the distribution are more important to account for the relevant behaviour of the system , it is common to augment the basis sets with _ diffuse functions _ , i.e. , single - primitive gaussian shells of the same angular momentum as some preexisting one but with a smaller exponent @xmath603 than the smallest one in the shell . in general , this improvement is commonly denoted by adding the prefix _ aug_- to the name of the basis set . in the case of pople s basis sets , on the other hand , the insertion of a plus sign ` + ' between the contraction scheme and the letter g denotes that the set contains one diffuse function in the 2s- and 2p - valence shells of heavy atoms . a second + indicates that there is another one in the 1s - shell of hydrogens . for example , one may have the doubly augmented ( and doubly polarized ) 6 - 31++g(2d,2p ) basis set . the ground - breaking advances reviewed in the previous sections allow to routinely calculate , using hartree - fock scf methods , physical properties of molecules of tens of atoms in present day computers . however , the simplest algorithms that can be devised to perform the limiting steps in such calculations are far from optimal . this large room for improvement , which may be enough to accommodate the exciting possibility of linearly scaling approaches that could open the door to thousands atoms computations , has been steering many innovative lines of research in the last years to close this review , we shall briefly outline here some of the hottest areas of modern development related to the topics discussed , specially those aimed to the reduction of the in principle @xmath512 complexity associated to the calculation of the @xmath588 eris in eq . ( [ eq : chqc_twoe_integrals ] ) , as well as the @xmath515 cost of the diagonalization of the fock operator in eq . ( [ eq : chqc_rh_2 ] ) in principle , careful distinction must be made between the number of primitive gtos @xmath604 and the number of contracted ones @xmath596 ( see the previous section ) . however , since in this section we will be dealing only with approximate scalings without worrying much about the prefactor , the ` neutral ' notation @xmath460 has been chosen to denote a quantity which is linear on both @xmath604 and @xmath596 . ] . for wider reviews on these topics , we recommend to the interested reader the accounts in refs . @xcite . the first class of attempts to reduce the cost associated to the calculation of the @xmath512 eris @xmath505 are those aimed to the improvement of the algorithms for analytically calculating them _ without approximations_. there are basically two issues that render the construction of these algorithms a non trivial task : first , the fact that the only four - center eris that can be straightforwardly computed are the ones corresponding to a product of four s - type gtos , while higher angular momentum eris may obtained from them in a non unique way . after using the _ gaussian product rule _ @xcite ( see the previous section for the notation used ) @xmath605 which allows the four - centre integral to be turned into a two - center one , and whose absence for the case of stos is the essential reason for their being non practical , we can turn the 6-dimensional eri into a simple one - dimensional integral that can be readily calculated by different means @xcite : @xmath606 with [ eq : chqc_fund_expl ] @xmath607 such an integral is called a _ fundamental _ eri and , as we said , eris involving gtos with higher angular momentum than @xmath608 are obtained from this fundamental one through iterative differentiations of eq . ( [ eq : chqc_fund ] ) with respect to the nuclear positions , leading to recurrence relations expressing eris of a given angular momentum as a function of the lower angular momentum ones . the particular flavour of these recurrence relations that is used is one of the matters in which the algorithms for calculating eris differ . the second issue that renders the construction of algorithms for computing eris non - trivial is related to how the contraction of primitive gtos is handled . in the geminal paper by boys @xcite , the most naive procedure was suggested , namely , the conversion of each eri of contracted gtos into a quadruple sum of eris of primitive gtos . however , this does not take profit , for example , from the fact , mentioned in the previous section , that all cgtos in the same contracted gaussian shell are formed by pgtos with the same set of exponents @xmath609 . much profit can be taken from this and other constraints and , in fact , the cost - scaling profile of each algorithm with the contraction degree @xmath589 of the cgtos is strongly correlated to the moment at which the transformation between cgtos and pgtos is performed @xcite . among the most used of these analytical algorithms , we can mention the pople - hehre ( ph ) one @xcite , which additionally exploits the fact that for each four - center eri of low angular momentum there is a privileged cartesian axis system in which many primitive integrals vanish by symmetry ; the _ cmurchie - avidson _ ( md ) approach @xcite , which avoids the rotation in ph thus being more efficient for high angular momentum eris ; the _ bara - aika - chlegel _ ( oss ) algorithm @xcite ; and a better defined and improved version of it : the _ ead - ordon - ople _ ( hgp ) algorithm @xcite . finally , if the moment at which the contraction is handled is chosen dynamically depending on the type of gto appearing in the eri , we have the prism modifications of md and hgp : the md - prism @xcite and hgp - prism @xcite algorithms , as well as a generalization of all the previous methods , called cold prism @xcite . now , even if we implement any of these efficient methods for calculating the eris , there is still @xmath512 of them . this scaling is simply too harsh for a too large class of applications . therefore , the next natural step is to try to devise approximate methods that minimize the necessary decrease in accuracy at the same time that maximize the savings in computer time . thanks to the particular characteristics of the eris , the gaussian basis functions and the concrete physical problem intended to solve the locality of many - electron quantum mechanics @xcite , which is related to the _ nearsightedness _ concept introduced by kohn @xcite , is the main property that allows to finally achieve linearity . ] , this endeavour has been successfully pursued by many researchers and the ` holy grail ' @xcite of linear scaling with @xmath460 is asymptotically approaching current calculations in large systems . here , we shall discuss the basic issues that make this possible . for more in depth reviews , we point the readers to the accounts in references @xcite . the first point to consider in order to reduce the @xmath512 scaling attains the so - called _ radial overlap_. if we take a look at eq . ( [ eq : chqc_contracted_gto ] ) , we can see that , irrespective of the polynomial prefactor which contains all the angular dependence of the gto , every function @xmath610 contains a radial exponential part ( actually , a sum of exponentials ) . therefore , if we consider any product @xmath611 of two gtos , we will always find a multiplying sum of terms such as those depicted in the expression for the gaussian product rule in ( [ eq : chqc_gpr ] ) . the exponential decay of all quantities @xmath612 in those terms with the distance @xmath613 between the nuclei on which the gtos are centred indicates that , among the @xmath614 possible pair products @xmath615 , only @xmath616 of them will be non - negligible . to see this , note that if we fix ( say ) @xmath98 , the only values of @xmath470 that will yield a non - negligible product @xmath611 are those for which @xmath613 is ` small ' . since atoms do not interpenetrate in most of the conformations that shall be studied , this can only happen for a number of different @xmath470 s which is not @xmath616 but a constant independent of the size of the molecule . there are @xmath460 different possible values of @xmath98 for which the above reasoning can be repeated , and the result follows . as a consequence , if only @xmath616 pairs @xmath617 are non - negligible , then only @xmath618 eris @xmath588 may in principle contribute in a significant way to the fock operator in eq . ( [ eq : chqc_matrices_2 ] ) and not @xmath512 . ( for similar estimations based on slightly different hypotheses , see @xcite . ) one must also note that the discussion is complicated by the fact that the eris do not appear alone , but contracted with the density matrix elements @xmath619 . this allows for further improvements of the scaling beyond @xmath618 which are discussed later in this section . now , knowing that most of the eris are too small to be relevant , _ we do not know which ones _ and , if we calculated the @xmath512 of them in order to spot the little ones , then we would have not gained anything . this simple argument shows the necessity of finding a set of _ estimators _ that allow us to selectively drop eris _ without calculating them_. of course , the number of estimators must also scale at worst like @xmath618 in order for the scheme to be useful . one of the first and simplest such estimators , was introduced by almlf et al . at the same time that they proposed the direct scf method @xcite . in their scheme , each eri @xmath588 was approximated by the corresponding radial overlap factor @xmath620 ( note that there are only @xmath618 numbers @xmath612 ) . although this estimator was relatively successful , it presented the important drawback of not being an upper - bound for the eris , thus rendering the control of errors an a priori impossible task . in order to overcome this problem , hser and ahlrichs @xcite later proposed a different estimator which can be assured to be always greater than the associated eri . following them , after the proof of positive definiteness by roothaan @xcite , the electrostatic interaction energy between two continuous charge distributions , @xmath621 can be easily shown to satisfy the properties of an inner product . hence , if we choose @xmath622 , and @xmath623 , we can use the well - known schwarz inequality to show that @xmath624 in such a way that , by calculating only the @xmath614 different two - index eris in the last term , we can safely bound from above the whole set containing @xmath512 of them . after these seminal works , more sophisticated and tighter bounds have been developed through the years @xcite , their application being nowadays routine in quantum chemistry packages . in a second generation of methods , the @xmath618 formal scaling achieved in practical calculations @xcite by using the above ideas has been recently attacked . to this end , more specifically physical properties of the problem are exploited ( see footnote [ locality ] in page ) , and different strategies are used to deal with the coulomb and exchange parts of the fock operator in ( [ eq : chqc_matrices_2 ] ) . the main difference between the behaviours of these two contributions lies in the way in which the eris are contracted with the density matrix @xmath619 : in the ` classical ' coulomb part , the relative sizes of the eris @xmath588 are largely correlated with the relative sizes of the associated elements @xmath619 @xcite , so that no decrease in the @xmath625 scaling is expected from density matrix considerations . differently , in the exchange terms , the fact that the elements @xmath619 couple ` exchanged ' indices in the @xmath626 eris produces cancellations which make this type of ` quantum ' contributions rather short - range ( for non - metallic species ) @xcite . hence , as for every short - range interaction , a number of terms scaling _ linearly _ with the system size is expected . apart from the different treatment that this difference suggests , note that the separation of the coulomb and exchange tasks allows for improved parallelization the computer codes @xcite . for the exchange part , the aforementioned short - range behaviour allows to devise @xmath616 algorithms just by intelligently ordering the loops in which the eris are calculated . the fine details of these methods are rather technical and they are beyond the scope of this review . we point the reader to @xcite and references therein for further information . in the coulomb case , on the other hand , @xmath618 terms still enter the sum in ( [ eq : chqc_matrices_2 ] ) , and more physically - based approximations must be used . the _ ontinuous ast ultipole ethod _ ( cfmm ) by white et al . @xcite , for example , is probably one of the most celebrated algorithms for calculating the coulomb part of the fock operator . it is a generalization for continuous charge distributions of the _ ast ultipole ethod _ ( fmm ) , introduced by greengard and rokhlin @xcite in a ground - breaking paper and aimed for point - like charges . in both fmm and cfmm , a clever hierarchical tree - like division of the space into cells is performed , and the far away regions are approximated via truncated multipole expansions . these two ingredients , which allow to calculate the coulomb contribution in @xmath616 steps for large systems , are common to most of the fast coulomb algorithms . despite this similarity , the room for improvement seems still large enough to accommodate a vigorous field with many publications appearing each year . let us mention here , for example , the _ eneralized ell ultipole ethod _ ( gcmm ) by kutteh et al . , which can use moments higher than monopole @xcite ; the _ uantum hemical ree ode _ ( qctc ) by challacombe et al . @xcite , which independently thresholds ` bra ' and ` ket ' distributions ; and the _ aussian ery ast ultipole ethod _ ( gvfmm ) by strain et al . @xcite , which benefits from the idea , introduced in @xcite for point - like charges , of using a dynamical maximum angular momentum to further speed up the calculations note however , that the near - field contributions in these methods are still calculated without approximations and represent a great portion of the computer time . in this line , some modern algorithms are appearing to alleviate this part of the work , such as , for example , the _ j matrix engine _ by white and head - gordon @xcite , which uses and improves the ideas discussed in the first part of this section about analytically calculating the eris ; or the method by izmaylov et al . @xcite , which implements a hierarchy of screening levels to eliminate negligible integrals . according to recent reports @xcite , the combination of cfmm , with the j matrix engine technique and with the _ ourier ransform oulomb _ ( ftc ) method by fsti - molnr and pulay @xcite is nowadays probably the fastest way for assembling the coulomb matrix . now , once the construction process of the whole fock matrix @xmath499 ( via its coulomb and exchange parts ) has been cast into the form of an @xmath616 algorithm , the importance of the second rate - limiting step in scf procedures , the diagonalization of @xmath499 , comes into focus . although the prefactor of the ( in principle ) @xmath515 diagonalization step is very small and , for systems of less than a few thousands of atoms , the absolute time spent on it is smaller than the one needed for the formation of the fock matrix @xcite , it is clear that , in the long run , it will dominate the calculations in larger systems and will become the relevant bottleneck @xcite . we shall point out the essentials regarding the methods aimed to reduce the scaling the diagonalization step . for more in depth reviews on the topic , we suggest to the reader the accounts in @xcite . the @xmath515 complexity of classical diagonalization methods ( such as the givens - householder one @xcite ) can be easily understood if we think that the core of the algorithm is just the multiplication of @xmath486 matrices . nevertheless , if the matrices multiplied are _ sparse _ , i.e. , they have a number of non - negligible elements that scale not as @xmath618 but as @xmath616 , then the product can be obtained in @xmath616 steps . as a result of the already mentioned locality properties of many - body quantum mechanics ( see footnote [ locality ] in page ) , some matrices appearing in the scf methods discussed in this review , namely , the density matrix @xmath627 and the fock operator @xmath499 , are indeed sparse if the system is non - metallic , i.e. , if it has a non - vanishing homo - lumo gap @xcite . the idea behind most of the modern algorithms for achieving ( or avoiding ) diagonalization with @xmath616 effort consists essentially in performing all operations using only these ( local ) sparse matrices , and avoiding ( non - local ) dense ones , such as the mo coefficients matrix @xmath628 . the attempts to improve the scaling of the diagonalization steps fall basically in two groups @xcite . in the first one , profit is taken from the use of mos which , instead of being extended over the whole molecule , such as the canonical orbitals used in the previous sections , are _ localized _ in a small region of space . in these class of methods , diagonalization @xcite or pseudo - diagonalization ( annihilation of the occupied - virtual blocks of the fock matrix ) @xcite is still performed , and the @xmath616 scaling is achieved because the representation of all operators in the basis of localized mos is sparse . the second group of algorithms do not use the mos as variables but the density matrix itself . among them , two subfamilies of methods may be found : in the first one , the search for the optimal density matrix is simply treated as a standard optimization problem , being the score function the hf energy , and the variables the density matrix elements or a set of parameters of some suitable truncated expansion of it @xcite . see , for example , the approaches by ochsenfeld and head - gordon @xcite , by salek et al . @xcite , by millam and scuseria @xcite , or by ordejn et al . the other subfamily of density matrix - based methods use iterative procedures , in such a way that , at each step , the fock operator is considered fixed and the equations are solved for the density matrix ( much in the spirit of the mos - based algorithms discussed in the previous sections ) . in this group , we can find , for example , the approach in @xcite using the lanczos algorithm @xcite , or the method by helgaker et al . @xcite . these algorithms , combined with new strategies that also avoid diagonalization and improve scf convergence properties , such as the one described in @xcite , represent the final step towards linear hartree - fock methods in quantum chemistry . to close this section , although we have been concerned , up to now , with the calculation of the electronic ground - state given a fixed position of the nuclei , let us stress that it is also very common to use quantum chemical methods for finding the local energy minima of molecules . to this end , geometry optimizations must be performed and not only must we be able to compute the energy of the molecule , but also their derivatives with respect to the nuclear coordinates . additionally , these derivatives are also needed to do ab initio molecular dynamics in the ground - state born - oppenheimer pes @xcite . the most naive approach , namely , the computation of the gradient of the energy @xmath629 ( using a simpler notation for it than @xmath630 in ( [ eq : chqc_v_eff ] ) ) by finite differences , is very inefficient for anything but the smallest molecules . to see this , one just need to notice that the gradient has as many components as the system degrees of freedom @xmath151 hence , in order to obtain it , say , at a point @xmath631 , we would have to compute @xmath632 times a single point energy , in order to know @xmath633 and @xmath634 , being @xmath635 , with @xmath636 , a small displacement in each of the nuclear degrees of freedom . this drawback was overcome in the late 60s by pulay and others ( see @xcite and references therein ) with the introduction of the so - called _ analytical derivatives _ , in which the gradient ( and higher - order derivatives ) are expressed , like the energy itself , just as a function of eris involving the wavefunction at the point @xmath631 . this marked an inflexion point in the development of optimization and molecular dynamics algorithms that continues nowadays , as analytical derivatives are routinely introduced together with almost any new method for calculating the energy . in relation to the improvements reviewed in this work , for example , let us note that , in @xcite , the extension of the j matrix engine method to calculate the derivatives of the coulomb part with respect of the nuclei coordinates is introduced ; in @xcite , linear scaling exchange gradients are developed ; analytic derivatives for the gvfmm are provided in @xcite ; the hgp algorithm for calculating the eris is extended to the computation of derivatives of the eris as well in @xcite ; and we may find similar developments for the ftc method @xcite or for algorithms that achieve diagonalization with linear effort @xcite . this work has been supported by the research projects e24/3 and pm048 ( aragn government ) , mec ( spain ) and mcyt ( spain ) . p. echenique has been supported by a bifi research contract and by a mec ( spain ) postdoctoral grant . a _ functional _ @xmath78 $ ] is a mapping that takes functions to numbers ( in this work , only functionals in the real numbers are considered ) : @xmath637 \end{array}\ ] ] for example , if the function space @xmath638 is the hilbert space of square - integrable functions @xmath91 ( the space of states of quantum mechanics ) , the objects in the domain of @xmath639 ( i.e. , the functions in @xmath91 ) can be described by infinite - tuples @xmath640 of complex numbers and @xmath639 may be pictured as a function of infinite variables . when dealing with function spaces @xmath638 that meet certain requirements we will not discuss the issue further but let it suffice to say that @xmath91 does satisfy these requirements . ] , the limit on the left - hand side of the following equation can be written as the integral on the right - hand side : @xmath641 - \mathcal{f}[\psi_{0}]}{\varepsilon } : = \int \frac{\delta \mathcal{f}[\psi_{0}]}{\delta\psi}(x ) \delta\psi(x ) \mathrm{d}x\ , \ ] ] where @xmath66 denotes a point in the domain of the functions in @xmath638 , and the the object @xmath642 / \delta\psi)(x)$ ] ( which is a function of @xmath66 not necessarily belonging to @xmath638 ) is called the _ functional derivative _ of @xmath78 $ ] in the in the _ point _ @xmath643 . one common use of this functional derivative is to find stationary points of functionals . a function @xmath643 is said to be an _ stationary point _ of @xmath78 $ ] if : @xmath644}{\delta\psi}(x ) = 0 \ .\ ] ] in order to render this definition operative , one must have a method for computing @xmath645 / \delta\psi)(x)$ ] . interestingly , it is possible , in many useful cases ( and in all the applications of the formalism in this work ) , to calculate the sought derivative directly from the left - hand side of eq . ( [ eq : def_funct_der ] ) . the procedure , in such a situation , begins by writing out @xmath646 $ ] and clearly separating the different orders in @xmath647 . secondly , one drops the terms of zero order ( by virtue of the subtraction of the quantity @xmath648 $ ] ) and those of second order or higher ( because they vanish when divided by @xmath647 and the limit @xmath649 is taken ) . the remaining terms , all of order one , are divided by @xmath647 and , finally , @xmath645 / \delta\psi)(x)$ ] is identified out of the resulting expression ( which must written in the form of the right - hand side of eq . ( [ eq : def_funct_der ] ) . for a practical example of this process , see secs . [ sec : qc_hartree ] and [ sec : qc_hartree - fock ] . very often , when looking for the stationary points of a function ( or a functional ) , the search space is not the whole one , in which the derivatives are taken , but a certain subset of it defined by a number of constraints . an elegant and useful method for solving the constrained problem is that of the _ lagrange multipliers_. although it can be formally generalized to infinite dimensions ( i.e. , to functionals , see appendix a ) , here we will introduce the method in @xmath650 in order to gain some geometrical insight and intuition . the general framework may be described as follows : we have a differentiable function @xmath651 that takes points in @xmath650 to real numbers and we want to find the stationary points of @xmath97 restricted to a certain subspace @xmath652 of @xmath650 , which is defined by @xmath653 _ constraints _ if the constraints are functionally independent , one must also ask that @xmath654 . if not , @xmath652 will be either a point ( if @xmath655 ) or empty ( if @xmath656 ) . ] : the points that are the solution of the constrained problem are those @xmath658 belonging to @xmath652 where the first order variation of @xmath97 would be zero if the derivatives were taken ` along ' @xmath652 . in other words , the points @xmath658 where the gradient @xmath659 has only components ( if any ) in directions that ` leave ' @xmath652 ( see below for a rigorous formalization of these intuitive ideas ) . thus , when comparing the solutions of the unconstrained problem to the ones of the constrained problem , three distinct situations arise ( see fig . [ fig : lagrange ] ) : 1 . a point @xmath658 is a solution of the unconstrained problem ( i.e. it satisfies @xmath660 ) but it does not belong to @xmath652 . hence , it is not a solution of the constrained problem . this type of point is depicted as a white - filled circle in fig . [ fig : lagrange ] . a point @xmath658 is a solution of the unconstrained problem ( i.e. it satisfies @xmath660 ) and it belongs to @xmath652 . hence , it is also a solution of the constrained problem , since , in particular , the components of the gradient in directions that do not leave @xmath652 are zero . this type of point is depicted as a gray - filled circle inside a white - filled circle in fig . [ fig : lagrange ] . 3 . a point @xmath658 is not a solution of the unconstrained problem ( i.e. , one has ) but it belongs to @xmath652 and the only non - zero components of the gradient are in directions that leave @xmath652 . hence , it is a solution of the constrained problem . this type of point is depicted as a gray - filled circle in fig . [ fig : lagrange ] . from this discussion , it can be seen that , in principle , no conclusions about the number ( or existence ) of solutions of the constrained problem may be drawn only from the number of solutions of the unconstrained one . this must be investigated for each particular situation . in fig . [ fig : lagrange ] , an schematic example in @xmath125 is depicted . the constrained search space @xmath652 is a 2-dimensional surface and the direction note that , only if @xmath661 , i.e. , if the dimension of @xmath652 is @xmath195 , there will be a vector perpendicular to the constrained space . for @xmath662 , the dimensionality of the vector space of the directions in which one ` leaves ' @xmath652 will be also larger than 1 . ] in which one leaves @xmath652 is shown at several points as a perpendicular vector @xmath663 . in such a case , the criterium that @xmath659 has only components in the direction of leaving @xmath652 may be rephrased by asking @xmath659 to be parallel to @xmath663 , i.e. , by requiring that there exists a number @xmath71 such that @xmath664 . the case @xmath665 is also admitted and the explanation of the minus sign will be given in the following . it is clear that , requiring the gradient of @xmath667 to be zero , one recovers the condition @xmath669 , which is satisfied by the points solution of the constrained stationary points problem . if one also asks that the derivative of @xmath667 with respect to @xmath71 be zero , the constraint @xmath670 that defines @xmath652 is obtained as well . this process illustrates the _ lagrange multipliers _ method in this particular example . in the general case , described by eq . ( [ eq : constraints ] ) and the paragraph above it , it can be proved that the points @xmath658 which are stationary subject to the constraints imposed satisfy also , it is worth remarking here that any two pair of functions , @xmath674 and @xmath675 , of @xmath650 whose restrictions to @xmath652 are equal ( i.e. , that satisfy @xmath676 ) obviously represent the same constrained problem and they may be used indistinctly to construct the auxiliary function @xmath667 . this fact allows us , after having constructed @xmath667 from a particular @xmath97 , to use the equations of the constraints to change @xmath97 by another simpler function which is equal to @xmath97 when restricted to @xmath652 . this freedom is used to derive the hartree and hartree - fock equations , in secs . [ sec : qc_hartree ] and [ sec : qc_hartree - fock ] , respectively . the formal generalization of these ideas to functionals ( see appendix a ) is straightforward if the space @xmath650 is substituted by a functions space @xmath639 , the points @xmath658 by functions , the functions @xmath97 , @xmath677 and @xmath667 by functionals and the requirement that the gradient of a function be zero by the requirement that the functional derivative of the analogous functional be zero . finally , let us stress something that is rarely mentioned in the literature : _ there is another ( older ) method , apart from the lagrange multipliers one , for solving a constrained optimization problem : simple substitution . i.e. , if we can find a set of @xmath678 independent _ adapted coordinates _ that parameterize @xmath652 and we can write the score function @xmath97 in terms of them , we would be automatically satisfying the constraints . actually , in practical cases , the method chosen is a suitable combination of the two ; in such a way that , if substituting the constraints in @xmath97 is difficult , the necessary lagrange multipliers are introduced to force them , and vice versa . as a good example of this , the reader may want to check the derivation of the hartree equations in sec . [ sec : qc_hartree ] ( or the hartree - fock ones in sec . [ sec : qc_hartree - fock ] ) . there , we start by proposing a particular form for the total wavefunction @xmath112 in terms of the one - electron orbitals @xmath326 ( see eq . ( [ eq : chqc_constraint_total_h ] ) ) and we write the functional @xmath639 ( which is the expected value of the energy ) in terms of that special @xmath112 ( see eq . ( [ eq : chqc_constrained_functional_h ] ) ) . in a second step , we impose the constraints that the one - particle orbitals be normalized ( @xmath679 ) and force them by means of @xmath4 lagrange multipliers @xmath680 . despite the different treatments , both conditions are constraints standing on the same footing . the only difference is not conceptual , but operative : for the first condition , it would be difficult to write it as a constraint ; while , for the second one , it would be difficult to define adapted coordinates in the subspace of normalized orbitals . so , in both cases , the easiest way for dealing with them is chosen . mackerell jr . , a. d. , brooks , b. , brooks iii , c. l. , nilsson , l. , roux , b. , won , y. , and karplus , m. in _ the encyclopedia of computational chemistry _ , p. v. r. schleyer , p. r. schreiner , n. l. allinger , t. clark , j. gasteiger , p. kollman and h. f. schaefer iii ( eds ) , pp . 217277 , john wiley & sons , chichester ( 1998 ) . cancs , e. , defranceschi , m. , kutzelnigg , w. , le bris , c. , and maday , y. in _ handbook of numerical analysis . volume x : special volume : computational chemistry _ , p. ciarlet and c. le bris ( eds ) , pp . 3270 , elsevier ( 2003 ) . schlegel , h. b. and mcdouall , j. j. w. in _ computational advances in organic chemistry : molecular structure and reactivity _ , c. gretir and i. g. csizmadia ( eds ) , pp . 167185 , kluwer academic , the netherlands ( 1991 ) . rodrguez , a. m. , baldoni , h. a. , suvire , f. , nieto vzquez , r. , zamarbide , g. , enriz , r. d. , farkas , . , perczel , a. , mcallister , m. a. , torday , l. l. , papp , j. g. , and csizmadia , i. g. , * 455 * , 275301 ( 1998 ) . garca de la vega , j. m. and miguel , b. in _ introduction to advanced topics of computational chemistry _ , l. a. montero , l. a. daz and r. bader ( eds ) , chapt . 3 , pp . 4180 , editorial de la universidad de la habana ( 2003 ) . shao , y. , molnar l. f. , jung , y. , kussmann , j. , ochsenfeld , c. , brown , s. t. , gilbert , a. t. b. , slipchenko , l. v. , levchenko , s. v. , oneill , d. p. , distasio , r. a. , lochan , r. c. , wang , t. , beran , g. j. o. , besley , n. a. , herbert , j. m. , lin , c. y. , van voorhis , t. , chien , s. h. , sodt , a , steele , r. p. , rassolov , v. a. , maslen , p. e. , korambath , p. p. , adamson , r. d. , austin , b. , baker , j. , byrd , e. f. c. , dachsel , h. , doerksen , r. j. , dreuw , a , dunietz , b. d. , dutoi , a. d. , furlani , t. r. , gwaltney , s. r. , heyden , a. , hirata , s. , hsu , c. p. , kedziora , g. , khalliulin , r. z. , klunzinger , p. , lee , a. m. , lee , m. s. , liang , w. , lotan , i. , nair , n. , peters , b. , proynov , e. i. , pieniazek , p. a. , rhee , y. m. , ritchie , j. , rosta , e. , sherrill , c. d. , simmonett , a. c. , subotnik , j. e. , woodcock , h. l. , zhang , w. , bell , a. t. , chakraborty , a. k. , chipman , d. m. , keil , f. j. , warshel , a. , hehre , w. j. , schaefer , h. f. , kong , j. , krylov , a. i. , gill , p. m. w. , head - gordon , m. , * 8 * , 31723191 ( 2006 ) .

in recent years , more and more effort has been devoted to finding and studying field theories which are candidates for effective descriptions of multiple m - branes . inspired by the proposal of basu and harvey@xcite of an m - theory version of the nahm equation , bagger and lambert@xcite and independently gustavsson@xcite ( blg ) found a three - dimensional chern - simons matter theory with @xmath0 supersymmetry that is invariant under parity transformations . this theory was based on a matter multiplet taking values in a 3-lie algebra@xcite and a topological gauge field with a natural action on the matter fields . it was soon realized that the number of 3-lie algebras with desirable properties , such as admitting a positive definite metric , is very limited and generalizations of the blg model were sought . the most promising such generalization is the aharony - bergman - jafferis - maldacena ( abjm ) model@xcite , which is again a chern - simons matter theory . in this model the matter fields live in the bifundamental representation of the product gauge group @xmath1 , which is endowed with a metric of split signature . one can , however , rewrite the abjm model in 3-algebra language if one uses a generalization of 3-lie algebras known as hermitian 3-algebras@xcite . it seems that supersymmetry in the abjm model is reduced to @xmath2 , which makes it less appealing than the blg model . this could be remedied by the inclusion of monopole operators , see @xcite and references therein . furthermore , the abjm model passed a non - trivial consistency check when the scaling @xmath3 of the degrees of freedom with the number @xmath4 of m2-branes was reproduced@xcite . for a comprehensive review of m2-brane models , see @xcite . given that there is now a reasonable candidate for the description of multiple m2-branes , it is clearly tempting to look for a similar candidate for the effective description of multiple m5-branes . there are various arguments in the literature suggesting that such a theory has to be intrinsically quantum and that it should not possess a lagrangian description@xcite . however , recent results suggest that there might be loopholes to these arguments , as e.g. the inclusion of non - local components in the model@xcite . as mentioned above , the development of the m2-brane models used as a guideline the basu - harvey equation that captures the dynamics of self - dual strings , i.e. m2-branes ending on m5-branes . it is therefore reasonable to expect that the description of self - dual strings from the perspective of the m5-branes might shed light on potential m5-brane models . moreover , self - dual strings are interesting by themselves , as they should lead to new integrable models with a wealth of non - trivial mathematical structures behind them . in the following , we will review a description of self - dual strings using loop space@xcite . loop space appears very naturally within m - theory , as one can see by the following nave argument : recall that the interactions of d - branes are governed by fundamental strings , whose endpoints induce an effective theory on the set of points in the d - branes worldvolume . the interactions of m5-branes , however , are governed by m2-branes , whose boundaries form loops called self - dual strings in the m5-branes worldvolume . it is therefore natural to expect an effective description on the loop space of the worldvolume . the use of loop space in m - theory has been advocated before@xcite . moreover , loop space equations for self - dual strings ( which differ from the ones presented here ) have been proposed@xcite . we will give more arguments in favor of a loop space description of m5-branes in the last section . a configuration of ( flat ) m2-branes ending on m5-branes corresponds in string theory to d2-branes ending on d4-branes or d1-branes ending on d3-branes . the latter configuration has been identified@xcite as the string theory interpretation of the atiyah - drinfeld - hitchin - manin - nahm ( adhmn ) construction of monopoles@xcite . the loop space description of self - dual strings that we present in the following is a rather literal lift of the original adhmn construction to m - theory . there are quite a few interesting structures related to the adhmn construction , such as twistor descriptions , lax pairs , spectral curves , etc . our m - theory lift of the adhmn construction should therefore allow us to find the m - theory analogues of these structures . we will list some more recent results related to this point in the last section . as a further motivation for loop space , we also review the reformulation@xcite of a set of equations of motion that was proposed@xcite as an effective description of m5-branes on loop space . this reformulation leads to natural interpretations for some of the fields appearing in these equations as well as for some of the equations themselves . the bps equations in this model essentially agree with the self - dual string equation on loop space . this review is structured as follows . in section 2 , we recall the definition of self - dual strings as well as their description from the perspective of the m2-branes in terms of the basu - harvey equation . just as monopoles can be seen as characteristic classes of principal fiber bundles , self - dual strings yield characteristic classes of abelian gerbes , which we review in section 3 . section 4 discusses the lift of an abelian gerbe to a line bundle over loop space , which yields a loop space version of the self - dual string equation . in section 5 , we demonstrate how to lift all the ingredients in the adhmn construction of monopoles to m - theory . we also perform a few consistency checks and demonstrate that this lift reduces to the adhmn construction in a very natural way . section 6 gives a brief impression of how the construction of section 5 is extended to self - dual strings in the abjm model . in section 7 , we review the loop space interpretation of a proposal of m5-brane equations . we close in section 8 by summarizing the presented results and putting them into the context of some more recent findings related to m - theory . as stated in the introduction , it was the study of the bps equation for multiple m2-branes@xcite that ultimately led to the construction of the blg model@xcite . analogously , one might hope that studying the bps equations for multiple m5-branes might provide crucial hints on their effective description . this motivates the study of self - dual strings@xcite and their non - abelian extensions . in this section , we provide a concise review of the self - dual string soliton and its relation to monopoles . we start from a d - brane configuration of @xmath5 d1-branes ending on a d3-brane . the worldvolumes of the branes extend in ten - dimensional flat target space of type iib superstring theory as follows : @xmath6 we have coordinates @xmath7 , and , as usual , we mostly write @xmath8 instead of @xmath9 . the symbol @xmath10 indicates that the d1-branes end on the d3-brane located at @xmath11 . in the following , we will be interested in the vacua of this configuration . from the perspective of the d3-brane , the endpoints of the d1-branes appear like @xmath5 dirac monopoles . the configuration is thus described by the _ bogomolny monopole equation _ @xmath12 , where @xmath13 is the curvature of the electromagnetic potential @xmath14 and @xmath15 is the higgs field of the monopole configuration . from the perspective of the d1-brane , however , the configuration is described by the _ nahm equation_@xcite @xmath16~.\ ] ] the @xmath17 are scalar fields taking values in the adjoint representation of @xmath18 and describe the fluctuations of the @xmath5 d1-branes in the directions parallel to the worldvolume of the d3-branes . note that the configuration is invariant under @xmath19 rotations in the @xmath20-directions , and this is reflected in the nahm equation . both descriptions are linked by the nahm transform@xcite . one can see this link very explicitly when comparing solutions to the bogomolny monopole equation and the nahm equation . note that @xmath12 , together with the bianchi identity , implies that @xmath15 is a harmonic function on @xmath21 , where we allow for sources at the positions of the monopoles . for a monopole at @xmath22 , we therefore have a higgs field @xmath23 , where @xmath5 is the _ monopole charge _ and the number of d1-branes . on the other hand , we find a simple solution to the nahm equation by factorizing @xmath24 , where @xmath25 . plugging this ansatz into , we obtain @xmath26~.\ ] ] this solution is called the _ fuzzy funnel_@xcite : equations imply that the @xmath27 form a representation of @xmath28 , which suggests to interpret them as coordinates on a fuzzy sphere @xmath29 . the radius of this sphere is given by the function @xmath30 . thus , each point in the worldvolume of the d1-branes polarizes into a fuzzy sphere with a radius that diverges at @xmath31 . because the higgs field @xmath15 describes fluctuations of the d3-brane s worldvolume in the @xmath32-direction ( and therefore the coordinate @xmath8 should be identified with @xmath15 ) , the profile of the d1-branes given by @xmath30 agrees precisely with that given by the higgs field from the d3-brane s perspective . let us now lift this configuration to m - theory . for this , we t - dualize along @xmath33 and interpret @xmath34 as the m - theory direction . we arrive at the configuration @xmath35 here , the boundary of the m2-branes in the m5-brane s worldvolume looks like a string . together with the self - duality of the three - form curvature in the m5-brane s worldvolume , this led to the name _ self - dual string_. we will assume that the boundary of the m2-brane is stable in the @xmath33-direction , and therefore restrict our attention to the @xmath36-directions . from the perspective of the m5-brane , the vacua of the configuration are described by the self - dual string equation@xcite @xmath37 . the configuration is obviously invariant under @xmath38 rotations in the @xmath36-directions , and this should be reflected in the extension of the nahm equation . basu and harvey therefore suggested essentially the following equation@xcite @xmath39~.\ ] ] here , we assume that the bracket @xmath40 $ ] is totally antisymmetric and linear in each slot . before discussing this 3-algebra structure in any further detail , let us briefly compare the profiles of the scalar fields in the two descriptions . we make again a product ansatz @xmath41 and plug it into . the solution is @xmath42~.\ ] ] on the other hand , the self - dual string equation @xmath43 implies that @xmath15 is a harmonic function on @xmath44 with sources at the location of the self - dual string . we therefore have @xmath45 , which is again compatible with the profile obtained from the solution given in . the components @xmath46 of the solution @xmath47 take values in a _ 3-lie algebra_@xcite : they are elements in a vector space @xmath48 endowed with a totally antisymmetric trilinear bracket @xmath40:{\mathcal{a}}^{\wedge 3}\rightarrow { \mathcal{a}}$ ] . we also impose the _ fundamental identity _ , @xmath49=[[a , b , c],d , e]+[c,[a , b , d],e]+[c , d,[a , b , e]]~,\ ] ] which plays the role of a higher jacobi identity . it guarantees that the inner derivations @xmath50 spanned by the operators @xmath51 with @xmath52 $ ] , @xmath53 , form a lie algebra . this lie algebra can be used as a gauge algebra in the basu - harvey equation : analogously to the nahm equation , we can write in gauge covariant form after introducing a gauge potential @xmath54 taking values in @xmath50 : @xmath55~.\ ] ] the most prominent example of a 3-lie algebra is the 3-lie algebra @xmath56 whose generators @xmath57 , @xmath58 , satisfy the relation @xmath59={{\varepsilon}}_{\mu\nu\kappa\lambda}\tau_\lambda~.\ ] ] it underlies the solution , and its lie algebra of inner derivations @xmath60 is @xmath38 . the basu - harvey equation is a bps equation in the blg model@xcite , a chern - simons matter theory with @xmath0 supersymmetry that was proposed as a candidate for the effective description of stacks of multiple m2-branes . in this theory , the matter supermultiplet , which takes values in a 3-lie algebra , is coupled to a chern - simons action . to write down an action for this theory , it is necessary to equip the 3-lie algebra with a metric structure . unfortunately , @xmath61 and its direct sums are the only examples of finite - dimensional 3-lie algebras that can be endowed with a positive definite , @xmath50-invariant metric . generalizations of 3-lie algebras were therefore proposed that relax the total antisymmetry of the 3-bracket@xcite . _ real 3-algebras_@xcite are a generalization of 3-lie algebras obtained by allowing the 3-bracket to be antisymmetric only in the first two slots . they evidently contain 3-lie algebras . _ hermitian 3-algebras_@xcite are complex vector spaces endowed with a bracket @xmath62 $ ] , which is linear and antisymmetric in its first two slots and antilinear in the last one . it satisfies the following form of the fundamental identity : @xmath63,d;e]=[[a , d;e],b;c]+[a,[b , d;e];c]-[a , b;[c , e;d]]~.\ ] ] the 3-lie algebra @xmath61 is contained in this class . both for real and hermitian 3-algebras , we have again a lie algebra of inner derivations and we can construct chern - simons matter theories using them . in the case of hermitian 3-algebras , these theories contain the abjm model@xcite , that is now regarded as the most promising candidate for the description of multiple m2-branes . more details on 3-algebras can be found in @xcite . all our constructions in the following can be applied to models using real or hermitian 3-algebras . while monopoles are described in terms of certain connections on principal fiber bundles , self - dual strings are described by certain connective structures on ( abelian ) gerbes . to explain this statement , let us briefly recall the bundle description of magnetic monopoles . for simplicity , we restrict ourselves to the abelian case . consider a manifold @xmath64 covered by patches @xmath65 . let @xmath66 be a circle- or principal @xmath67-bundle over @xmath64 . the _ first chern class _ of @xmath66 can be identified with the curvature @xmath68 of a connection @xmath69 . here , @xmath68 is a globally defined closed two - form with values in @xmath70 . the poincar lemma allows us to derive gauge potentials @xmath71 on the patches @xmath72 with @xmath73 . another application of the poincar lemma yields transition functions @xmath74 on the intersections of patches @xmath75 taking values in the group @xmath67 . altogether we have @xmath76 an example of a principal @xmath67-bundle over @xmath77 is the hopf fibration @xmath78 . consider now a dirac monopole in @xmath21 . as we have seen before , the higgs field becomes singular at the position of the monopole . let us therefore restrict to the unit sphere @xmath77 with the monopole at its center . we cover @xmath77 with the standard patches @xmath79 and @xmath80 centered at the north and south poles , respectively . the overlap @xmath81 can be contracted to the equator @xmath82 . the transition function is therefore of the form @xmath83 with @xmath84 and @xmath85 the coordinate on the equator , this bundle is therefore the @xmath5-fold tensor product of the principal @xmath67-bundle given by the hopf fibration . for @xmath86 , we obtain tensor products of the dual of the hopf fibration . ] . the first chern number of this principal @xmath67-bundle is calculated as @xmath87 where we used stokes theorem . the integer @xmath5 is a topological invariant and called the _ monopole charge_. let us now come to the description of self - dual strings via abelian gerbes . here , we will restrict ourselves to abelian local or hitchin - chatterjee gerbes , see e.g. @xcite for more details . just as the first chern class @xmath68 of a principal @xmath67-bundle @xmath66 characterizes the bundle up to topological equivalence , the dixmier - douady class @xmath88 characterizes the abelian gerbe @xmath89 . here , @xmath88 is a globally defined , closed three - form with values in @xmath70 . repeated application of the poincar lemma yields the following objects : @xmath90 to describe a self - dual string , we restrict ourselves to a sphere @xmath91 with the self - dual string at its center . we cover @xmath91 by two patches @xmath79 and @xmath80 , and the overlap @xmath81 can be contracted to the equator , which is a two - sphere . instead of a transition function on the overlap , we have now a one - form potential of a connection and therefore a principal @xmath67-bundle . as we saw above , however , the principal @xmath67-bundles over @xmath77 are characterized by their monopole charge . thus , abelian gerbes over @xmath91 are also characterized by an integer , which we call the _ self - dual string charge_. cohomologically , this is reflected in @xmath92 . the self - dual string charge can be computed explicitly analogously to by integrating @xmath88 over @xmath91 . as mentioned in the introduction , it is rather natural to expect a description of self - dual strings in terms of loop space . to perform the transition from abelian gerbes to objects on loop space , we use the so - called _ transgression map_. the free loop space of a manifold @xmath64 is the space of all maps @xmath93 . we restrict ourselves to maps that are immersions and factor out reparametrization invariance . the latter is done by working only with formulas that are invariant under reparametrizations . the resulting space of loops @xmath94 is also called the space of singular oriented knots@xcite . acting on functions of @xmath94 are the derivations @xmath95 where @xmath96 are some local coordinates on @xmath64 . these derivations span the tangent space @xmath97 . we write dual elements and the exterior derivative on @xmath94 as @xmath98 respectively . for more details and the precise definition of the above objects , see @xcite . we have the following natural double fibration : @xmath99{$m$ } } \put(64.0,0.0){\makebox(0,0)[c]{${\mathcal{l}}m$ } } \put(32.0,33.0){\makebox(0,0)[c]{${\mathcal{l}}m\times s^1 $ } } \put(7.0,18.0){\makebox(0,0)[c]{$\mathsf{ev}$ } } \put(55.0,18.0){\makebox(0,0)[c]{$\mathsf{pr}$ } } \put(25.0,25.0){\vector(-1,-1){18 } } \put(37.0,25.0){\vector(1,-1){18 } } \end{picture } \end{aligned}\ ] ] here , @xmath100 is the obvious projection from a point @xmath101 to @xmath102 , and @xmath103 is the evaluation map @xmath104 . given a @xmath105-form @xmath106 on @xmath64 , we can pull it back to @xmath107 . subsequent integration along the fibers of @xmath100 yields a @xmath108-form on @xmath94 . locally , we have the following expressions : @xmath109 the transgression map @xmath110 is clearly reparametrization invariant . moreover , it is a chain map and maps closed forms to closed forms and exact forms to exact forms . it therefore descends to a map @xmath111 on de rham cohomology . this implies that an abelian gerbe on @xmath64 characterized by a dixmier - douady class @xmath88 is mapped via transgression to a first chern class @xmath68 of a principal @xmath67-bundle over the loop space @xmath94 . note , however , that @xmath112 is _ not _ surjective , and therefore there are in general more circle bundles over the loop space @xmath94 than there are gerbes on @xmath64 . forms on loop space in the image of @xmath112 are usually called _ ultra - local_. working with vector bundles instead of gerbes is a priori easier . for example , extending gauge structures to the non - abelian case is straightforward . it should be stressed , however , that in performing a transgression , we trade the difficulties of working with gerbes for the difficulties of working over infinite - dimensional spaces . let us now transgress the self - dual string equation @xmath113 on @xmath44 to the loop space @xmath114 . applying @xmath112 to both sides of the equation , we obtain in cartesian coordinates @xmath96 , @xmath58 , on @xmath44 the following expression : @xmath115 we would like to extend the above equation in various ways : first , @xmath116 should be a function on @xmath94 and not , as it stands , a function on @xmath64 or @xmath107 . moreover , we would like to extend the curvatures @xmath68 beyond abelian and ultra - local expressions . we will see that the appropriate generalization of reads as@xcite @xmath117}~ , \end{aligned}\ ] ] where @xmath118 and @xmath119 $ ] indicates antisymmetrization . we will refer to as the loop space self - dual string equation . note that this equation is reparametrization invariant . in the following , we will demonstrate how to construct solutions to this equation from a generalized adhmn construction . the atiyah - drinfeld - hitchin - manin - nahm ( adhmn ) construction of monopoles is a nahm transform , which is in turn a special fourier - mukai transform@xcite . the nahm transform takes an instanton bundle over a four - torus @xmath120 , pulls it back along the projection @xmath121 , where @xmath122 is the four - torus dual to @xmath120 , twists it with the poincar line bundle over @xmath123 and projects it down to an instanton bundle over @xmath122 . roughly speaking , we consider in the adhmn case a degenerate four - torus with radii @xmath124 yielding @xmath125 . the `` dual four - torus '' has then radii @xmath126 , giving @xmath21 . the instanton equation reduces to the nahm equation on @xmath125 and the bogomolny monopole equation @xmath12 on @xmath21 . in both cases , certain boundary conditions have to be imposed . the spaces @xmath125 and @xmath21 are to be seen as the spatial parts of the d1- and d3-branes in the configuration . let us now go through the construction in more detail , and present all the ingredients that we will have to lift to m - theory . we start from a @xmath18-valued solution @xmath17 to the nahm equation on an open subset @xmath127 of @xmath125 . for a dirac monopole , we have @xmath128 , while for an @xmath129-monopole , we choose @xmath130 . from this solution , we construct the _ twisted dirac operator _ @xmath131 which acts on @xmath132-sections of a trivial complex rank @xmath5 vector bundle tensored with the spinor bundle @xmath133 on @xmath120 . the coordinates @xmath134 describing a point on @xmath21 arise from the twist of the vector bundle with the poincar line bundle . one can readily show that the @xmath17 satisfying the nahm equation is equivalent to the laplace operator @xmath135 being positive and @xmath136=0 $ ] . we now determine all normalized zero modes of @xmath137 . that is , we find all @xmath4 solutions @xmath138 , @xmath139 such that @xmath140 a solution to the monopole equation on @xmath21 is then given by @xmath141 as one can verify by direct computation after introducing a green s function for @xmath142 . let us briefly consider two examples for @xmath143 and @xmath144 . for @xmath143 , the nahm equation reduces to @xmath145 , and we can put @xmath146 . a zero mode , yielding well - defined fields on the two domains@xmath147 and corresponding to the dirac string going through the north and the south pole , respectively . ] is given by @xmath148 and using formulas , we arrive at @xmath149 in the case @xmath144 , the nahm equation is nontrivial , and we choose the solution we obtained before from factorization : @xmath150 following the adhmn construction , one finds the solution @xmath151 and a more complicated expression for @xmath152 . recall that the higgs field is expected to be proportional to the monopole charge @xmath5 , which is indeed the case in the above examples . to lift the construction up to m - theory , let us follow the t - duality along @xmath33 and the lift to m - theory along @xmath34 and construct dirac operators at each step : @xmath153 iia & 0 & 1 & 2 & 3 & 4 & 5 & 6 \\ d2 & $ \times$ & & & & & $ \times$ & $ \times$ & $ { \setbox0\hbox{$\nabla$ } \rlap{\hbox to \wd0{\hss/\hss}}\box0}_{s , x}^{\rm iia}=-\gamma_5\dfrac{{\mathrm{d}}}{{\mathrm{d}}s}+\gamma^4\gamma^i ( x^i-{\mathrm{i}}x^i)$ \\ d4 & $ \times$ & $ \times$ & $ \times$ & $ \times$ & & $ \times$ & \\[0.5 cm ] m & 0 & 1 & 2 & 3 & 4 & 5 & 6 \\ m2 & $ \times$ & & & & & $ \times$ & $ \times$ & $ { \setbox0\hbox{$\nabla$ } \rlap{\hbox to \wd0{\hss/\hss}}\box0}^{\rm m}_{s , x}=-\gamma_5\dfrac{{\mathrm{d}}}{{\mathrm{d}}s}+\tfrac{1}{2}\gamma^{\mu\nu}(d(x^\mu , x^\nu)-{\mathrm{i}}x^{\mu\nu})$ \\ m5 & $ \times$ & $ \times$ & $ \times$ & $ \times$ & $ \times$ & $ \times$ & \end{tabular}\ ] ] the dirac operator @xmath154 is clear and has been introduced before@xcite . the @xmath38-invariant form of @xmath155 is equally evident , if we assume that the zero modes will take values in a 3-lie algebra @xmath48 . we can then contract @xmath156 $ ] with @xmath157 . it is merely the twisting element @xmath158 that is less clear . here , however , the transition to loop space suggests to use @xmath159}(\tau)~,\ ] ] where @xmath160 is an element of @xmath114 , i.e. an equivalence class of maps @xmath161 . altogether we obtain the twisted dirac operator @xmath162 this dirac operator acts on @xmath48-valued functions on @xmath127 tensored by the trivial bundle @xmath163 , potentially originating from the chiral spinors over @xmath164 . let us now try to build a generalized adhmn construction from this guess . first , one readily shows that for a solution @xmath47 to the basu - harvey equation , the dirac operator is positive and commutes with the generators @xmath157 of @xmath165 . the inverse statement is not true , as the map @xmath166 has a non - trivial kernel . solutions for the loop space self - dual string equation are now obtained as follows : we start from the normalized zero modes @xmath138 , @xmath167 , satisfying @xmath168 where @xmath169 denotes the inner product on @xmath170 . note that we have assumed here that @xmath48 is equipped with a compatible inner product @xmath169 . because of the block - diagonal structure of the dirac operator in a reasonable basis of the clifford algebra , we can sort the zero modes according to their chirality . we arrive at @xmath4 zero modes @xmath138 , @xmath139 , with @xmath171 and @xmath4 zero modes @xmath138 , @xmath172 , with @xmath173 . the solutions to are given by formulas very similar to : @xmath174 one readily verifies that these fields satisfy the loop space self - dual string equation by explicit calculation . note that they are anti - hermitian and , because of our sorting of zero modes , the fields take values in the gauge algebra @xmath175 . to streamline the discussion , we ignored the role of boundary conditions in the adhmn construction . there , one has to demand that the @xmath17 have simple poles @xmath176 at finite boundaries of @xmath127 and that they form irreducible representations of @xmath28 at these points . there are corresponding boundary conditions in the m - brane case : the @xmath47 should behave like @xmath177 at finite boundaries of @xmath127 and form irreducible representations of @xmath165 . to explore our construction in more detail , let us first construct the obvious solutions for @xmath143 and @xmath144 . in the case @xmath143 , the basu - harvey equation becomes again @xmath178 and we therefore put @xmath179 . after introducing the shorthand @xmath180 , the zero modes are given by @xmath181 \psi^-_{s , x(\tau)}&\sim{\mathrm{e}}^{-r^2_+ s}\left(\begin{array}{c } 0\\ 0 \\ { \mathrm{i}}\left(r_+^2+x^{12 } + x^{34}\right ) \\ x^{13}-x^{24}+{\mathrm{i}}(x^{23}+x^{14 } ) \end{array } \right)~. \end{aligned}\ ] ] formulas yield the following higgs field and gauge potential : @xmath182 where @xmath183 and @xmath184 is obtained from @xmath185 by substituting @xmath186 . for @xmath144 , we start from the solution @xmath47 to the basu - harvey equation with @xmath187 based on factorization : @xmath188 the computation of the zero modes and the corresponding fields is a little more complicated in this case@xcite , but eventually one finds @xmath189 as expected , this is twice the higgs field of . as mentioned above , the only finite - dimensional example of a 3-lie algebra with all physically desirable properties is @xmath61 . this 3-lie algebra is conjectured to yield the effective description of two m2-branes . it is therefore necessary to extend our construction , in particular to the case of hermitian 3-algebras . the extension to real 3-algebras is almost trivial : we merely assume that the solution @xmath47 to the basu - harvey equation lives in a real 3-algebra , and the map @xmath166 is again the map from two elements of the real 3-algebra to its associated lie algebra of inner derivations . above we have demonstrated that the description of the bps subsector of the m5-brane theory that is given by self - dual strings in terms of loop space works rather well . let us now try to apply loop space to a description of the full model . various models for an effective description of m5-branes have been proposed over the last years . particularly interesting for our purposes is a proposal@xcite that assumes that all fields in the tensor multiplet assume values in a 3-lie algebra @xmath48 . adding a gauge potential @xmath217 of a connection @xmath69 as well as a 3-algebra valued vector @xmath218 , the field equations @xmath219+[x^j , c^\nu,[x^j , c_\nu , x^i]]&=0~,\\ \gamma^\mu\nabla_\mu\psi-[x^i , c^\nu,\gamma_\nu\gamma^i\psi]&=0~,\\ \nabla_{[\mu}h_{\nu\kappa\lambda]}+\tfrac{1}{4}{{\varepsilon}}_{\mu\nu\kappa\lambda\sigma\tau}[x^i,\nabla^\tau x^i , c^\sigma]+\tfrac{{\mathrm{i}}}{8}{{\varepsilon}}_{\mu\nu\kappa\lambda\sigma\tau}[\bar{\psi},\gamma^\tau\psi , c^\sigma]&=0~,\\ f_{\mu\nu}-d(c^\lambda , h_{\mu\nu\lambda})&=0~,\\ \nabla_\mu c^\nu = d(c^\mu , c^\nu)&=0~,\\ d(c^\rho,\nabla_\rho x^i)=d(c^\rho,\nabla_\rho\psi)=d(c^\rho,\nabla_\rho h_{\mu\nu\lambda})&=0~. \end{aligned}\ ] ] are invariant under the supersymmetry transformations @xmath220{{\varepsilon}}~,\\ \delta h_{\mu\nu\lambda}&=3{\mathrm{i}}{{\bar{\varepsilon}}}\gamma_{[\mu\nu}\nabla_{\lambda]}\psi+{\mathrm{i}}{{\bar{\varepsilon}}}\gamma^i\gamma_{\mu\nu\lambda\kappa}[x^i,\psi , c^\kappa]~,\\ \delta a_\mu&={\mathrm{i}}{{\bar{\varepsilon}}}\gamma_{\mu\lambda } d(c^\lambda,\psi)~,\\ \delta c^\mu&=0~. \end{aligned}\ ] ] the field equation @xmath221 implies a factorization of the vector @xmath218 into @xmath222 , where @xmath223 is an element of the 3-lie algebra @xmath48 and @xmath224 is a vector in @xmath225 . the appearance of this additional vector @xmath224 on spacetime suggests to identify it with the m - theory direction . going to loop space , we correspondingly identify @xmath224 with the tangent @xmath226 to the loop @xmath160 . in this picture , we obtain a very natural interpretation for the equation @xmath227 in : @xmath228 which is a non - abelian form of the transgression map . here , we added a @xmath229 to indicate that the field @xmath230 lives on loop space . introducing moreover the fields @xmath231 the field equations @xmath232 + [ { { \mathring{x}}}^j,[{{\mathring{x}}}^j,{{\mathring{x}}}^i]]&=0~,\\ \tfrac{1}{r } \gamma^{\mu\nu}{\dot{x}}_\nu\nabla_\mu{{\mathring{\psi}}}-\gamma^i [ { { \mathring{x}}}^i,{{\mathring{\psi}}}]&=0~,\\ \nabla_\mu { { \mathring{f}}}^{\mu\nu}+[{{\mathring{x}}}^i,\nabla^\nu { { \mathring{x}}}^i]+{\mathrm{i}}\left ( [ \bar{{{\mathring{\psi}}}},\gamma^\nu{{\mathring{\psi } } } ] - \tfrac{1}{r^2}{\dot{x}}^\sigma { \dot{x}}^\nu[\bar { { \mathring{\psi } } } , \gamma_\sigma { { \mathring{\psi}}}]\right)&=0 \end{aligned}\ ] ] are invariant under the supersymmetry transformations @xmath233{{\varepsilon}}~. \end{aligned}\ ] ] note that all the above equations are local on the loop @xmath160 . there are , however , a few open questions concerning this rewriting : the additional conditions @xmath234 should merely state that the fields are invariant under reparametrizations . the covariant derivatives @xmath235 , however , acts only on the zero - modes of the fields : @xmath236 and this should be extended to a full loop space derivative . at the moment , it is not clear how to do this while preserving supersymmetry . to compare this model with our generalization of the adhmn construction , we can derive the bps equations from the supersymmetry transformations on loop space . this yields indeed the ultra - local part of the loop space self - dual string equation . the adhmn - like construction we reviewed above arose very naturally from the assumption that self - dual strings should be described in terms of loop space : the necessary lift of the dirac operator of the original construction as well as the non - abelian generalization of the equation were readily derived . also , the basu - harvey equation features in the construction exactly in the expected way and takes over the role the nahm equation played in the original adhmn construction . explicit examples of solutions to the loop space self - dual string equation can be constructed . moreover , the reduction to the adhmn construction of monopoles is rather obvious . our construction extends to the more general real and hermitian 3-algebras , and it is therefore compatible with the abjm model . also , we obtained naturally a gauge group structure @xmath237 , which fits well with a recent proposals for m5-brane equations@xcite . besides the nave argument for the appearance of loop space given in the introduction , there is further evidence that this approach is appropriate . first of all , the notion of a monopole bag together with the associated nahm construction have been lifted to m - theory@xcite . recall that a monopole bag is a double - scaling limit of infinitely many monopoles forming a surface diffeomorphic to a sphere , where the ratio of the monopole number and the yang - mills coupling constant is kept fixed . in this double - scaling limit , all equations become abelian , and the description of self - dual string bags is therefore known . it was checked that the infinite - charge nahm construction has a loop space version , which produces solutions to an abelian form of the loop space self - dual string equation@xcite . further motivation for the use of loop space also comes from a different perspective . one might wonder about the interpretation of the fuzzy funnel in the case of m2-branes ending on m5-branes . recall that in the @xmath19-invariant situation of d1-branes ending on d3-branes , the points in the worldvolume of the d1-branes polarize into fuzzy two - spheres . analogously one would expect that points on m2-branes polarize into fuzzy three - spheres . a satisfyingly precise notion of fuzzy 3-spheres is so far unknown . roughly speaking , the prequantum line bundle over @xmath77 should be replaced by a prequantum gerbe over @xmath91 . the hilbert space of the quantization , which in the case of @xmath77 is given by global holomorphic sections of the prequantum line bundle , should here be obtained from global holomorphic sections of the prequantum gerbe . while the latter notion seems rather opaque , one can transgress the problem to loop space : the loop space @xmath238 of @xmath91 carries the structure of a khler manifold@xcite , and the prequantum gerbe over @xmath91 is transgressed to a prequantum line bundle over @xmath238 . one can then quantized ( a subset of the ) functions on the space @xmath238 by identifying them with linear operators on the space of global holomorphic sections of the transgressed abelian gerbe over @xmath91 . such an approach to quantization has been suggested previously@xcite , where it was used to quantize the loop space of @xmath21 . it could be shown that the quantization of the loop space reduces as expected to the moyal quantization of @xmath239 upon imposing the usual restrictions reducing m - theory to string theory . the fact that an adhmn - like construction for self - dual strings exists at all is by now no longer surprising . underlying both the original adhm and adhmn constructions are corresponding twistor geometric interpretations of the instanton and monopole equations . here , a penrose - ward transform establishes a bijection between certain holomorphic bundles over a twistor space and solutions to the relevant equations . quite recently , a twistor space was found@xcite , that yields a bijection between certain holomorphic gerbes on this twistor space and ( abelian ) solutions to the self - dual string equation . in principle it is possible to transgress this picture to loop space and to obtain the twistor picture underlying our loop space self - dual string equation . moreover , the twistor construction using abelian gerbes@xcite has recently been extended to the non - abelian case@xcite . we therefore know that a full adhmn - like construction of self - dual strings based on ordinary space and non - abelian gerbes has to exist . the existence of our construction leads to another question . recall that the adhmn construction is a special case of a nahm transform mapping instanton solutions on a four - torus to instanton solutions on the dual four - torus . it is therefore a map between solutions to the same theory on different spaces and in the special case of the adhmn construction , one has a map between solutions to dimensional reductions of the instanton equation . therefore one would also expect the m - theory lift to be a map between solutions to closely related theories . the basu - harvey equation and the self - dual string equation , however , appear to be very different . a first step towards resolving this issue is the observation that all 3-lie algebras ( as well as the generalized real and hermitian 3-algebras ) are special cases of differential crossed modules or strict lie 2-algebras@xcite . these strict lie 2-algebras , in turn , are structure groups of categorified principal bundles , which contain in particular non - abelian gerbes as special cases . assuming that the gauge structure of the m5-brane model yet to be found will be defined by a non - abelian gerbe , we can conclude that at least the gauge structure of both m2- and m5-brane models do agree . moreover , in the m2-brane model the gauge two - form curvature carries no degrees of freedom , as it is fixed via the equations of motion by the matter fields . something analogous happens when working with non - abelian gerbes : here , a gauge two - form curvature is fixed by the two - form potential via the so - called fake curvature condition and just as in chern - simons matter theories , the gauge field does not carry any additional degrees of freedom . altogether , it seems that m2- and m5-brane models might not be as different as it appeared initially . this , together with the fact that more and more m5-brane models are being proposed@xcite , gives reason for careful optimism about the existence of an effective description of m5-branes . i would like to thank sam palmer and costis papageorgakis for very pleasant collaborations on the papers@xcite that i reviewed here . moreover , i am grateful to sergey cherkis for the invitation to berkeley as well as for many inspiring and entertaining discussions . i would also like to thank martin wolf for comments on a draft of this review . this work was supported by a career acceleration fellowship from the uk engineering and physical sciences research council . a. basu and j. a. harvey , _ nucl . b _ * 713 * , 136 ( 2005 ) . j. bagger and n. d. lambert , _ phys . d _ * 77 * , 065008 ( 2008 ) . a. gustavsson , _ nucl . b _ * 811 * , 66 ( 2009 ) . v. t. filippov , _ sib . * 26 * , 126 ( 1985 ) . o. aharony , o. bergman , d. l. jafferis and j. m. maldacena , _ jhep _ * 0810 * , 091 ( 2008 ) . j. bagger and n. lambert , _ phys . d _ * 79 * , 025002 ( 2009 ) . j. bagger , n. lambert , s. mukhi and c. papageorgakis , preprint , arxiv:1203.3546 , 2012 . n. drukker , m. marino and p. putrov , _ commun . phys . _ * 306 * , 511 ( 2011 ) . e. witten , preprint , arxiv:0712.0157 , 2007 . chu and s .- l . ko , _ jhep _ * 1205 * , 028 ( 2012 ) . c. saemann , _ commun . phys . _ * 305 * , 513 ( 2011 ) . s. palmer and c. saemann , _ jhep _ * 1110 * , 008 ( 2011 ) . u. schreiber , preprint , arxiv : hep - th/0509163 , 2005 . a. gustavsson , _ jhep _ * 0601 * , 165 ( 2006 ) . a. gustavsson , _ jhep _ * 0612 * , 066 ( 2006 ) . a. gustavsson , _ jhep _ * 0804 * , 083 ( 2008 ) . diaconescu , _ nucl . phys . _ * b503 * , 220 ( 1997 ) . d. tsimpis , _ phys . b _ * 433 * , 287 ( 1998 ) . w. nahm , talk presented at int . summer inst . on theoretical physics , freiburg , west germany , aug 31 - sep 11 , 1981 . n. j. hitchin , _ commun . phys . _ * 89 * , 145 . c. papageorgakis and c. saemann , _ jhep _ * 1105 * , 099 ( 2011 ) . n. lambert and c. papageorgakis , _ jhep _ * 1008 * , 083 ( 2010 ) . p. s. howe , n. d. lambert and p. c. west , _ nucl . b _ * 515 * , 203 ( 1998 ) . r. c. myers , _ jhep _ * 12 * , 022 ( 1999 ) . s. cherkis and c. saemann , _ phys . d _ * 78 * , 066019 ( 2008 ) . j. a. de azcarraga and j. m. izquierdo , _ j. phys . _ * a43 * , 293001 ( 2010 ) . m. k. murray ( 2007 ) , in : _ the many facets of geometry : a tribute to nigel hitchin , _ eds . o. garcia - prada _ ( oxford university press , 2010 ) . h. schenk , _ commun . phys . _ * 116 * , 177 . v. l. campos , g. ferretti and p. salomonson , _ jhep _ * 0012 * , 011 ( 2000 ) . s. mukhi and c. papageorgakis , _ jhep _ * 0805 * , 085 ( 2008 ) . j. gomis , d. rodriguez - gomez , m. van raamsdonk and h. verlinde , _ jhep _ * 0809 * , 113 ( 2008 ) . s. terashima , _ jhep _ * 0808 * , 080 ( 2008 ) . k. hanaki and h. lin , _ jhep _ * 0809 * , 067 ( 2008 ) . chu , preprint , arxiv:1108.5131 , 2011 . d. harland , s. palmer and c. saemann , preprint , arxiv:1204.6685 , 2012 . c. saemann and r. j. szabo , _ pos _ * cncfg2010 * , 005 ( 2010 ) . c. saemann and r. j. szabo , proc . of the 4-th annual meeting of the european non - commutative geometry network , bucharest , 25.4.-30.4 . 2011 . c. saemann and r. j. szabo , preprint , arxiv:1203.5921 , 2012 . c. saemann and m. wolf , preprint , arxiv:1111.2539 , 2011 . l. mason , r. reid - edwards and a. taghavi - chabert , preprint , arxiv:1111.2585 , 2011 . c. saemann and m. wolf ( 2012 ) , preprint , arxiv:1205.3108 , 2012 . s. palmer and c. saemann ( 2012 ) , preprint , arxiv:1203.5757 , 2012 . h. samtleben , e. sezgin and r. wimmer , _ jhep _ * 1112 * , 062 ( 2011 ) .

high - redshift radio galaxies ( hzrgs ) are rare objects in the cosmos , residing at the very brightest end of the radio luminosity function . wide - area surveys have identified powerful radio galaxies out to very early cosmic epochs . however , since flux - limited surveys tend to probe increasingly luminous objects with increasing redshift , our knowledge of lower - luminosity hzrgs at high redshift is sparse ( tnj0924 - 2201 has a 1.4ghz flux density @xmath4mjy , despite being at @xmath5 ; @xcite ) . we have sought to remedy this situation with deep , multifrequency radio maps of the botes field ( @xcite ; s. croft et al . ( 2008 , in preparation ) ; [ sec : radio ] ) . this field has substantial multiwavelength supporting data . many hzrgs show diffuse rest - frame optical morphologies and other characteristics of `` protogalaxies '' ( e.g. * ? ? ? * ) ; they appear to form when lyman - break - galaxy - sized clumps merge , and eventually evolve into large elliptical galaxies @xcite . identifying lower - luminosity radio galaxies at early cosmic epochs can help our understanding of the nature of active galactic nucleus ( agn ) hosts as a function of radio luminosity , and can help constrain cosmic evolution of the radio galaxy population . by applying a cut in radio spectral index , @xmath6 ( @xmath7 ) , we can preferentially select higher - redshift objects , since a combination of intrinsic evolutionary effects and an observational `` @xmath8-correction '' tend to cause @xmath6 to become steeper ( i.e. , more negative ) with increasing redshift ( for details see , e.g. * ? ? ? indeed , 35% of sources with @xmath9 and @xmath10mjy observed by @xcite were found to have @xmath2 . this paper presents @xmath0-band images of radio sources with steeper than normal radio spectra and no detections in deep optical data to @xmath1mag ( vega ) from the noao deep wide - field survey ( ndwfs ) botes field , in order to study the morphologies and properties of candidate hzrgs that are fainter than `` typical '' hzrgs . @xcite performed a complementary study of optically - invisible radio sources in the botes field . they used the very large array ( vla ) at 1.4ghz to map a 0.5 square degree area to a limiting sensitivity ( at field center ) of @xmath11@xmath12jybeam@xmath13 . thirty six of their 377 radio sources ( 10% ) were not visible in the ndwfs data , although 90% of these 36 had flux densities less than 1mjy ; in contrast ( [ sec : ndwfs ] ) we consider sources brighter than 1mjy , and with relatively steep spectral indices , from our radio maps ( [ sec : radio ] ) covering an order of magnitude larger area . like us , however , @xcite conclude that most of their optically - invisible objects are agns at relatively high redshifts . the region mapped by @xcite has very little spatial overlap with our 325mhz data , and none of their optically - invisible radio sources is present in our radio catalog . throughout this paper , we use vega magnitudes , j2000 coordinates , and assume an @xmath14 cosmology @xcite . observations of a 4.9 square degree region of the botes field were made with the vla a - array at 325mhz during four runs from 2003 june august . the clean beam had a fwhm of 5 and the limiting sensitivity was @xmath15@xmath12jy . the data analysis and reduction will be discussed in detail in an upcoming paper ( s. croft et al . ( 2008 , in preparation ) ) . a catalog of radio sources ( as opposed to components , i.e. , a double - lobed radio galaxy counts as one source ) was generated using the method described by @xcite . we matched the 325mhz catalog ( s. croft et al . ( 2008 , in preparation ) ) to the 1.4ghz westerbork synthesis radio telescope ( wsrt ) dataset of @xcite , which has @xmath16@xmath12jy , an ellipsoidal clean beam of 13@xmath17 , and an area of 6.68 square degrees . the typical positional accuracy at 1.4ghz for sources in the high signal - to - noise ratio ( s / n ) regime ( which is the case for our sources since their fluxes are @xmath18mjy ) is @xmath19 ( @xmath20 ) . six hundred and fifty three sources ( in the 4.9 square degree where the two radio datasets overlap ) were found to be in common and hence had measured radio spectral indices , @xmath6 . the properties of this matched catalog will be discussed by s. croft et al . ( 2008 , in preparation ) . to select hzrg candidates , we combined deep radio data ( [ sec : radio ] ) with optical data from the ndwfs @xcite , a deep ( @xmath21 , @xmath22 , @xmath23 , @xmath24mag ; @xmath25 , 5 diameter @xcite ) optical / near - ir survey in botes . the magnitude limits for point sources are deeper , but hzrgs are likely to be resolved in the @xmath26 seeing of ndwfs , so we used 4 diameter optical apertures to determine whether or not a source is identified . we used the @xcite @xmath27-band selected catalog , which has improved photometry for faint objects compared to the public ndwfs dr3 source extractor ( sextractor ) catalog . in computing photometric redshifts , we measured photometry from the ndwfs optical data in 4diameter apertures at each of the radio positions ( [ sec : photoz ] ) . the ndwfs botes field is also covered by deep infrared , x - ray , and other observations ( [ sec : archival ] ) , although these data were not available to us at the time of our near infrared camera ( nirc ) observations and were not used as part of the selection . as noted by @xcite , @xcite , and others , radio sources with spectral indices that are steeper than normal ( average @xmath28 ) have a higher probability of being unidentified in optical surveys , a result that holds true for the combination of our radio data with the ndwfs @xmath29-band ( fig . [ fig : spixid ] ; see s. croft et al . ( 2008 , in preparation ) for more details ) . we also see a similar drop in identification fraction for sources with inverted ( @xmath30 ) spectra . at @xmath31mjy , the radio source counts are dominated by agns @xcite . pure starbursts with flux densities @xmath32mjy must either be at relatively low redshift , or have an extremely high star - formation rate , and either ought to be easily seen in our deep optical data . one way to preferentially target hzrgs , therefore , is to look for radio sources brighter than 1mjy that are optically faint or invisible , although of course even in these objects there may be some contribution to the radio luminosity from a starburst component . deep @xmath33-band imaging can provide information on the morphologies of the hzrgs , and the well - known @xmath34 relation ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) can be used to estimate their redshifts . fraction of radio sources with @xmath31mjy that are identified in the ndwfs @xmath29-band data , as a function of radio spectral index , @xmath6 . radio sources with inverted ( @xmath35 ) and steep spectral indices ( @xmath36 ) have lower identification fractions than sources with `` normal '' spectral indices , suggesting that they are more distant and / or less luminous . this result will be discussed in more detail by s. croft et al . ( 2008 , in preparation ) . ] four hundred and thirty of the 434 radio sources with @xmath31mjy in the matched 325mhz / 1.4ghz sample are within the ndwfs imaging coverage . seventy two of these 430 have no identifications in the ndwfs optical / ir data ( which we define as no detections within 4 of the 1.4ghz radio position in @xmath37 to 3@xmath38 in a 4 diameter aperture ) . some of these 72 objects have confusing or extended radio contours , are near bright stars , or appear to be associated with extended objects ( e.g. nearby galaxies larger than 8 in diameter where the radio position is offset from the optical centroid ) . we inspected finding charts of the ndwfs + 325mhz radio data , and selected a subsample of 14 sources that were unambiguously unidentified . these 14 sources have radio morphologies that appear unresolved at 325mhz , with no nearby extended radio emission , and are not near bright objects . only one of the 14 has a steep spectrum , with @xmath39 ; four have @xmath40 , seven have @xmath41 , and two have @xmath42 . the five radio sources with the steepest spectra ( fig . [ fig : postage ] ) were followed up with infrared imaging ( [ sec : irim ] ) . here we refer to these sources as a e ; see table [ tab : obs ] for positions and names . llllllllll source a & j142631 + 341557 & @xmath43 & @xmath44 & @xmath45 & @xmath46 & @xmath43 & @xmath47 & 3840 & 0 + source b & j142647 + 345851 & @xmath48 & @xmath49 & @xmath50 & @xmath51 & @xmath52 & @xmath53 & 1920 & 0 + source c & j142741 + 342324 & @xmath54 & @xmath55 & @xmath56 & @xmath57 & @xmath58 & @xmath59 & 3840 & 3600 + source d & j143036 + 334026 & @xmath60 & @xmath61 & & 1920 & 0 + source e & j143258 + 342055 & @xmath62 & @xmath63 & @xmath64 & @xmath65 & @xmath62 & @xmath66 & 1920 & 4680 + despite the efficiency of the method used for combining multiple radio source components in order to eliminate radio lobes from the catalog ( [ sec : radio ] ) , inevitably a few lobes may slip through . one such object turns out to be source d ; it was undetected in our nirc observations , and later turned out to be undetected in any of the irac bands . when we overplotted our vla data , as well as data from the nrao vla sky survey ( nvss ) , on a wide - field digitized sky survey ( dss ) image , we found that this source is in fact a lobe of a large ( 45 diameter ) fr - ii that has a faint @xmath29-band identification in the dss . additionally , after the observations were taken , we replotted the radio data for source b on the ndwfs images , including a lower contour than in our original plots , and noticed that the radio emission appeared extended ( fig . [ fig : postage ] ) , and so this object ought not to have been included in our original sample of 14 objects with unresolved radio emission and no ndwfs matches . nevertheless , the radio emission appears to be associated with an ir - bright source which meets our criteria for optical faintness . source d ( the large fr - ii ) will not be discussed further here ; the remaining four fields resulted in the detection of candidate hzrgs , as discussed in [ sec : discuss ] . [ sec : nirc]imaging observations were made in @xmath0-band , using the nirc @xcite , a @xmath67 insb array with 015 pix@xmath13 on keck i during two half nights on ut 2004 april 29 and 30 . a 16-position ( @xmath68 ) dither , with dither spacing 3 and 60s per exposure ( @xmath69 ) was used . three fields were observed for four dithers ( 3840s ) and the remaining two fields observed for two dithers ( 1920s ) , as summarized in table [ tab : obs ] . seeing was 05 06 and the sky was clear . the data were reduced in the standard manner using dimsum in iraf . astrometry was bootstrapped from the ndwfs data ; the small offsets between the @xmath0 and 4.5 positions ( table [ tab : obs ] ) suggest that our nirc astrometry is accurate to @xmath26 . photometry was performed in a 35 diameter aperture ; the @xmath0 magnitudes are presented in table [ tab : targets ] , where we also present the measured ndwfs and iss photometry ( [ sec : photoz ] ) . llllllllllllll source a & -1.48 & @xmath70 & @xmath71 & @xmath72 & @xmath73 & @xmath74 & @xmath75 & @xmath76 & @xmath77 & @xmath78 & @xmath79 & @xmath80 + source b & -0.89 & @xmath81 & @xmath82 & @xmath83 & @xmath84 & & & @xmath85 & @xmath86 & @xmath87 & @xmath88 & @xmath89 + source c & -0.98 & @xmath90 & @xmath91 & @xmath92 & @xmath93 & @xmath94 & & @xmath95 & @xmath96 & @xmath97 & @xmath98 & @xmath99 + source e & -0.87 & @xmath100 & @xmath101 & @xmath102 & @xmath103 & @xmath104 & @xmath105 & @xmath106 & @xmath107 & @xmath108 & @xmath109 & @xmath110 + two of the four hzrg candidates were observed spectroscopically . the slitmasks were filled with secondary targets selected by their ir colors from the agn `` wedge '' of @xcite , along with a few hzrg candidates ; these will be discussed in future papers . these observations used the low resolution imaging spectrometer ( lris ; * ? ? ? * ) on keck i , with the d560 dichroic and a slit width of 13 . on the blue side , a 400 line mm@xmath13 grism , blazed at 3400 was employed , giving 1.09 / pixel and spectral resolution 8.1 . on the red side , a 400 line mm@xmath13 grating , blazed at 8500 was used , giving 1.86 / pixel and spectral resolution 7.3 . this setup gives spectral coverage of @xmath111 , with some variation from slitlet to slitlet . source c was observed on ut 2005 june 4 for 3600s ( @xmath112s ) under clear skies , in 14 seeing . source e was observed on ut 2005 june 6 for 4680s ( @xmath113s ) in 08 seeing and with clear skies . the observations are summarized in table [ tab : obs ] . the data were reduced in the standard manner using bogus in iraf , and spectra extracted in a 15 wide aperture for many of the secondary targets . the primary targets were undetected in both the red and blue side two - dimensional spectra , and for these no extractions were performed . the botes field is covered by the _ chandra _ x - ray observations of @xcite . the resulting catalogs , published by @xcite and @xcite , reach a limiting full - band ( @xmath114kev ) flux of @xmath115erg s@xmath13 @xmath116 for on - axis sources with four counts . we checked for counterparts to our targets , and found that source e was detected , @xmath117 ) @xmath118 ( five counts in the full band , corresponding to @xmath119erg s@xmath13 @xmath116 ) . this source is very likely to be a powerful high redshift and / or optically obscured quasar simply on the basis of its large x - ray - to - optical flux ratio @xcite . intriguingly , an object in the field of source b is also detected in x - rays with 5 counts in the full band ( @xmath120erg s@xmath13 @xmath116 ) . the x - ray position , @xmath121 ) @xmath122 is close to the peak of the 325mhz radio contours , but offset ( by 70 ) from the flux - weighted 1.4ghz position . however , it appears that the x - ray and radio sources are two distinct objects ; if they were the same ( or if the radio source were associated with the object nearest the peak of the radio contours ) , the radio source would be extremely core dominated , which is not supported by the comparatively steep spectral index . rather , it appears that the radio source is an extended , double - lobed source associated with the @xmath27-band dropout at the position given in table [ tab : obs ] , and that the x - ray source is associated with the optical object , @xmath123 ) ] nearest ( 12 ) the x - ray position , and clearly visible in the ndwfs @xmath27-band ( fig . [ fig : postage ] ) and nirc @xmath0-band ( fig . [ fig : nirccon ] ) data . both of the x - ray sources have too few counts to reliably determine a hardness ratio ( three counts in the soft band and two in the hard band in both cases ) . the other hzrg candidates are not detected in xbotes , although they lie within the coverage area . since the catalogs of @xcite and @xcite contain only sources with four or more counts , it is possible that they may be x - ray sources , but be insufficiently bright for inclusion in the full catalog . none of our sources is within the coverage area of the more sensitive ( @xmath124erg s@xmath13 @xmath116 ) , single - pointing _ chandra _ x - ray observations of @xcite . the zbotes survey @xcite provides @xmath125-band data for part of the botes field . we checked for counterparts to our four hzrg candidates , and found that three were undetected at the zbotes flux limits ( table [ tab : targets ] ) , while source b was outside the zbotes coverage area . [ sec : flamex]flamex @xcite provides @xmath126- and @xmath0-band catalogs for part of the botes field . the @xmath0-band catalogs are deeper than the @xmath33-band catalogs used for our selection , but our measured nirc @xmath0-band fluxes ( for the three sources in regions covered by flamex ) are below the flamex @xmath0-band limit ( @xmath127mag ) , so these sources are too faint to appear in the @xmath0-band flamex catalog . the @xmath126-band flamex limit ( @xmath128mag ) is insufficient to put anything but the weakest constraints on @xmath129 colors , and in any case , two of our sources are outside the flamex coverage area . in future , refining our selection criteria by requiring non - detections in flamex will further increase our efficiency at finding hzrgs , since @xmath130 corresponds to @xmath131 for sources on the @xmath34 relation ( fig . [ fig : kz ] ; [ sec : discuss ] ) . @xmath34 diagram of @xcite ( circles represent radio galaxies with spectroscopic redshifts from the 3crr , 6ce , 6c@xmath132 , and 7crs surveys ) . our hzrg candidates are plotted as points with error bars , at the redshifts obtained from photometric redshift fitting . vertical error bars represent the measured uncertainties in the @xmath0 band magnitudes ( where we assume @xmath133 is negligible ) , and horizontal error bars represent the 68% confidence interval in redshift from hyperz as tabulated in table [ tab : pz ] . ] [ sec : iss]we checked the irac @xcite catalogs of the irac shallow survey @xcite for infrared counterparts to our hzrg candidates . all were detected at greater than @xmath134 significance at 3.6 and 4.5 , and one also at 5.8 . we report the catalog 5 aperture magnitudes in table [ tab : targets ] , and use these measurements to assist with spectral energy distribution ( sed ) fitting ( [ sec : discuss ] ) . table [ tab : targets ] shows measured values , even in cases where the photometric uncertainties are large , since real measurements from the data reflect the relative weights given to the data points in the @xmath135 photometric redshift fitting . all objects are pointlike at irac resolution ( @xmath136 ) . [ sec : mips]we downloaded the post - bcd 24 mosaics from the _ spitzer _ archival observations of soifer et al . for the fields surrounding the four hzrg candidates , and performed quick - look photometry using the published zero points . three of our objects were undetected ( to a flux limit of @xmath137mjy ) but source a ( the ultra - steep spectrum source ) was detected with @xmath138mjy . we have @xmath0 and irac detections and optical upper limits for our hzrg candidates ( table [ tab : targets ] ) , which we use to make somewhat crude redshift estimates . @xmath0 @xmath139 [ 3.6 ] , which straddles the balmer and / or 4000 breaks for the two @xmath140 sources from table [ tab : pz ] , provides quite tight redshift constraints . the position of our sources on a diagram of [ 3.6 ] @xmath139 [ 4.5 ] versus @xmath0 @xmath139 [ 3.6 ] color ( fig . [ fig : cc ] ) suggests that three are at @xmath141 . llllllll source a & 4.97 & 4.92 - 4.98 & 3.94 - 5.69 & arp220 & 0.00 & -24.67 & 6.94 + source b & 3.76 & 3.68 - 3.81 & 1.77 - 4.08 & qso2 & 0.30 & -24.32 & 10.5 + source c & 1.21 & 0.00 - 2.76 & 0.00 - 6.99 & mrk231 & 0.00 & -19.05 & 0.298 + source e & 4.65 & 4.21 - 4.99 & 0.00 - 6.99 & arp220 & 0.00 & -24.07 & 30.2 + [ 3.6 ] @xmath139 [ 4.5 ] versus @xmath0 @xmath139 [ 3.6 ] diagram for our four hzrg candidates ( magenta crosses with photometric error bars ) , compared to the colors of field sources in botes with at least a @xmath142 detection at ndwfs @xmath33-band ( where we assume that @xmath133 is negligible ) , 3.6 and 4.5 ( black points ) . also plotted are the colors of non - evolving galaxy templates from @xcite , redshifted from @xmath143 to @xmath144 in steps of @xmath145 ( with redshifts marked in steps of @xmath146 ) . the red line is a 2gyr old elliptical template , the blue line is arp220 , and the green line is a qso2 . ] [ sec : photoz]we also obtained photometric redshifts using the publicly available software hyperz ( version 1.2 , * ? ? ? * ) . in order to obtain more stringent limits on the fluxes of our objects than the 5 aperture limits used in the selection criteria , we remeasured photometry from the ndwfs @xmath147 images , in a 4 aperture at the position of the radio centroid . empirical photometric errors were derived using the method of @xcite , a monte carlo technique which places apertures on random regions of sky . this provides more accurate estimates of uncertainties ( verified by the addition of simulated galaxies to the real data ) than methods which assume uncorrelated poisson background noise and no source confusion . although some of the resulting measurements have large uncertainties ( table [ tab : targets ] ) , they help to constrain the resulting photometric redshifts better than using the published survey limiting magnitudes as upper limits . in cases where they are large , the quoted uncertainties provide a real measure of the variations in the background values due to neighboring objects . since the uncertainties are used essentially as weights in the @xmath135 photometric fitting , large uncertainties downweight the effect of bands where the object flux is poorly constrained , as opposed to assigning these bands too high a weight by setting the fluxes to zero and obtaining the uncertainty from the survey flux limit . however , even when we re - run the photometric redshift fitting using the survey limits for those sources with uncertainties larger than 1mag , we find that there is essentially no change in the best - fit redshift ( the largest change is for source e , whose best - fitting template changes from arp220 to a 2gyr old elliptical , and whose redshift changes by just 0.04 ) , although the confidence intervals become marginally less well constrained . for the @xmath125- and @xmath126-bands , where we only have access to catalogs and not to the original images ( in order to remeasure photometry at the radio positions ) , we set the fluxes to zero and derive the associated uncertainties from the published survey limiting magnitudes . we find that whether or not the @xmath125- and @xmath126-band data points are included has essentially no effect on the best - fitting redshifts and confidence intervals ( due to the large uncertainties in these bands ) . the @xmath27-band point for source a seems rather high ( fig . [ fig : sed ] ) ; inspection of the ndwfs @xmath27-band postage stamp in fig . [ fig : postage ] suggests that this is probably due to the unrelated saturated star to the southwest . in fact , even though source a fulfils our initial selection criterion ( no source in the @xcite catalog closer than 4 ) , the aperture magnitude measured at the radio position ( table [ tab : targets ] ) is brighter than the @xmath25 survey flux limits quoted in [ sec : ndwfs ] , strongly suggesting that light from a nearby object is contributing to the measured aperture magnitude ( the ndwfs @xmath27-band point - spread function ( psf ) has wings which can cause problems for aperture photometry at surprisingly large radii ) . if the photometric redshift is correct , there could also be some contribution to the flux in this band from ly@xmath6 at @xmath148 , although an @xmath149 source ought to be visible in the postage stamps ( fig . [ fig : postage ] ) . excluding the @xmath27-band point from the sed fitting for this object has no effect on the best - fitting redshift , and little effect on the confidence intervals . instead of using the templates provided with hyperz , we used a selection of templates ( the qso2 , arp220 , mrk231 , ell2 , ell5 and ell13 templates ; the latter three are 2 , 5 , and 13gyr old ellipticals ) from @xcite , which do a better job of reproducing ir fluxes . the effects of dust extinction were simulated using the prescriptions of @xcite , with extinction @xmath150 allowed to vary between @xmath151 and @xmath152 mag ( although some of the templates already incorporate the effects of some intrinsic dust extinction ) . @xmath153-band absolute magnitude , @xmath154 was constrained to be @xmath155 . because of the faintness of these objects in the optical , and the corresponding large uncertainties for the optical data points compared to the ir , the range of plausible redshifts is rather large ( table [ tab : pz ] ) . however , we can still estimate the best - fitting redshifts , @xmath156 . the template seds for the best - fitting models are shown in fig . [ fig : sed ] . three of the four hzrg candidates have @xmath157 , and the fourth is rather poorly constrained ( due to low s / n ) to @xmath3 ( table [ tab : pz ] ) . we report precise values for @xmath156 in order that the reader may know what values were used for the calculation of the rest - frame radio luminosities and other such properties . however , as can be seen from the confidence limits in redshift reported in table [ tab : pz ] , we can not constrain redshifts for these objects very accurately using sed fitting , given the large number of bands with non - detections or large photometric errors . source b was not observed spectroscopically , so we can not say whether or not the emission lines seen in the template ( fig . [ fig : sed ] ) are really present in its spectrum . in the event of galaxy - wide dust extinction these spectral lines would be less prominent , but this ought not to affect our determination of @xmath156 too much . we can also use the 4.5 flux densities alone as an alternative to the @xmath34 relation . @xcite studied a sample of 69 hzrgs with spectroscopic redshift @xmath158 and a wide range of radio luminosities . they showed that the rest - frame @xmath159-band light for these objects is well described by models of elliptical galaxies with formation redshift @xmath160 and masses between @xmath161 and @xmath162@xmath163 . in fig . [ fig:4p5z ] , we plot the measured values of the 4.5 flux density from @xcite against redshift , along with tracks representing @xcite models with @xmath160 and solar metallicity , normalized to @xmath161 and @xmath162@xmath163 . plotting our hzrg candidates in this figure at their best - fit photometric redshifts , these redshifts seem reasonable , although the lower bounds on the redshift ( assuming our objects have masses @xmath164@xmath163 ) are still quite weak ( @xmath131 ) . source c , with @xmath165 , falls below the @xmath161@xmath163 line , suggesting that it is either unusually low mass , or that the sed fitting underestimates the true redshift ( certainly possible given the wide confidence interval and low s / n for this source ) . source c is the only outlier on the @xmath34 plot ( fig . [ fig : kz ] ) , again suggesting it is either very underluminous , or is in fact at higher redshift than suggested by the sed fits . 4.5 flux density versus redshift for hzrgs with spectroscopic redshifts from @xcite ( circles ) , and our hzrg candidates ( crosses , plotted at the best - fit photometric redshift with 68% confidence intervals shown ) . the dashed and dotted lines are evolutionary tracks for elliptical galaxies with formation redshift @xmath160 and masses @xmath161 and @xmath162@xmath163 , respectively . ] the nirc images ( fig . [ fig : nirccon ] ) provide much higher resolution than irac , and we are able to see that two of the detected sources have somewhat extended , `` fuzzy '' morphologies ( fig . [ fig : nirczoom ] ) . source e is somewhat more compact , but appears to have a faint trail of material extending to the northeast . source b seems to be quite compact and relatively bright . the extended sources have @xmath0-band morphologies suggestive of the optical morphologies seen in hzrgs such as b20902 + 34 @xcite , which hints that these objects may be galaxies still in the process of formation ( again arguing for a higher redshift for source c than that obtained from sed fitting ) . the compactness of the observed radio morphologies ( unresolved in our maps , except for source b ) suggests that we are probably seeing these radio galaxies relatively soon after the agn turn on . the deduced @xmath8-corrected rest - frame @xmath153-band magnitudes ( table [ tab : pz ] ) imply optical luminosities a factor of a few times fainter than the most powerful @xmath2 radio galaxies as studied by @xcite and others , and the extended emission is a factor of a few smaller ( roughly @xmath166kpc as compared to @xmath167kpc for the more powerful sources ) . the extended radio emission of source b ( @xmath168kpc if its @xmath156 is correct ) is typical of a classical double - lobed radio galaxy , which lends weight to the ( independent ) selection of a qso2 template in the sed fitting . radio galaxies are among the most massive objects at all redshifts @xcite , but there is some evidence for a dependence of host optical luminosity on radio power ( e.g. , * ? ? ? * ; * ? ? ? it is plausible that our objects are less powerful in the radio and hence might be expected to be fainter than the @xmath34 relation ( fig . [ fig : kz ] ) ; they might , in fact , be still forming the bulk of their stars ( as suggested by their morphologies ) but the agn has already switched on . source b , which is more compact in the near - ir , and has more extended radio emission ( suggesting that the agn activity is less recent than in the other three sources ) is slightly brighter than the @xmath34 relation , again as might be expected in this scenario . from the measured radio fluxes and spectral indices , and the best - fit photometric redshifts , we calculate rest - frame 151mhz luminosities ( for the @xmath157 sources ) of @xmath169w hz@xmath13 sr@xmath13 , which implies @xcite ly@xmath6 luminosities @xmath170ergs@xmath13 . the corresponding ly@xmath6 fluxes are @xmath171erg s@xmath13 @xmath116 , essentially independent of redshift ( due to similar effects of distance on the observed ly@xmath6 and radio fluxes , and hence the determination of rest - frame luminosities ) . a typical hzrg ly@xmath6 line ( rest - frame fwhm @xmath172 ) would be marginally unresolved in our spectra at @xmath143 , but beyond @xmath173 , observed flux densities would start to decrease due to broadening of the ly@xmath6 line in the observed frame . at @xmath148 , the expected ly@xmath6 flux density is @xmath174erg s@xmath13 @xmath116 @xmath13 . since the noise in our spectra has @xmath175erg s@xmath13 @xmath116 @xmath13 , the predicted s / n is rather low ( @xmath172 ) , so it is perhaps unsurprising that we see no emission lines in our spectra , especially given the uncertainties ( @xmath176dex ) in the emission line radio correlation . if the line happens to fall in a region of the spectrum where noise is higher ( at the position of a subtracted telluric emission line , for example ) , the s / n would be lower still . dust , if present in the source , would also reduce the predicted fluxes , and hence the s / n . source a has the steepest radio spectrum of our candidates , so is most similar to classical hzrgs , and remains a good candidate for future spectroscopy . as discussed above , source b may have prominent emission lines , and spectroscopic observations would also be worthwhile . the photometric redshifts and `` typical '' hzrg morphologies support the interpretation of these objects as hzrgs ( and their deduced radio luminosities and @xmath153-band absolute magnitudes are agn - like , as expected ) . at the fainter radio flux levels we probe , we seem to be seeing the more `` normal '' members of the hzrg population that perhaps make up the bulk of the luminosity function at these redshifts . indeed , comparing to the results of @xcite , we find that our three @xmath2 galaxies ( with rest - frame @xmath29-band magnitudes @xmath177mag , as computed from the photometric redshifts in table [ tab : pz ] ) ought to lie on the fr - i / fr - ii break ( @xmath178whz@xmath13 at these optical absolute magnitudes ) . as a sanity check , we obtain a rough estimate of the space density of these objects by calculating the comoving volume covered by the 4.9 square degree combined radio dataset , in the redshift range 1.21 4.97 ( the range of photometric redshifts in table [ tab : pz ] ) . in our chosen cosmology , this volume is @xmath179mpc@xmath180 . since we do not claim to have exhaustively studied the radio - loud optically - faint galaxy population in this field within this redshift range , the resulting space density for our four hzrg candidates of @xmath181mpc@xmath182 should be taken as a lower limit . this compares favorably with the lower limit on radio galaxies of similar radio power ( @xmath183whz@xmath13 ) but steeper spectral index ( @xmath184 ) obtained by @xcite , @xmath185mpc@xmath182 at @xmath186 . since only one out of four of our hzrg candidates has a spectral index as steep as this ( and since both estimates are lower limits ) , the space density is consistent with that measured by @xcite deeper optical and / or ir observations ( particularly @xmath126 , @xmath159 and / or mid - ir ) would provide better constraints on our photometric redshifts . as discussed above , our objects may be only marginally too faint to be detected in our lris observations , so it is possible that deeper ( red - sensitive ) 10 m spectroscopy could in future provide spectroscopic redshifts . mid - ir observations would also be of interest in constraining star - formation rates , and the presence and temperature of any dust in these objects . we thank s. kulkarni and collaborators for making the lris observations of ut 2005 june 6 ( in exchange for images of grb 0505096 on ut 2005 june @xmath187 ) . we thank the mips botes team for mips data on our sources . we thank m. polletta and collaborators for making the swire template library publicly available . data presented herein were obtained at the w. m. keck observatory , which is operated as a scientific partnership among the california institute of technology , the university of california , and the national aeronautics and space administration . the observatory was made possible by the generous financial support of the w. m. keck foundation . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . the work of sc , wvb , wdv and sas was performed under the auspices of the u. s. department of energy , national nuclear security administration by the university of california , lawrence livermore national laboratory under contract no . w-7405-eng-48 . sc and wvb acknowledge support for radio galaxy studies at uc merced , including the work reported here , with the hubble space telescope and the _ spitzer space telescope _ via nasa grants hst # 10127 , sst # 1264353 , sst # 1265551 , sst # 1279182 . the work of ds and pe was carried out at jet propulsion laboratory , california institute of technology , under a contract with nasa . this work is based in part on observations made with the spitzer space telescope , which is operated by the jet propulsion laboratory , california institute of technology under a contract with nasa . support for this work was provided by nasa through an award issued by jpl / caltech . this work made use of images and data products provided by the noao deep wide - field survey , which is supported by the national optical astronomy observatory ( noao ) . noao is operated by aura , inc . , under a cooperative agreement with the national science foundation . finally , we thank the referee , r. windhorst , for his comments . blumenthal , g. , & miley , g. 1979 , , 80 , 13 bolzonella , m. , miralles , j .- m . , & pell , r. 2000 , , 363 , 476 brand , k. , et al . 2006 , , 641 , 140 brodwin , m. , et al . 2006 , , 651 , 791 brown , m. j. i. , dey , a. , jannuzi , b. t. , brand , k. , benson , a. j. , brodwin , m. , croton , d. j. , & eisenhardt , p. r. 2007 , , 654 , 858 bruzual , a. g. , & charlot , s. 1993 , , 405 , 538 calzetti , d. , armus , l. , bohlin , r. c. , kinney , a. l. , koornneef , j. , & storchi - bergmann , t. 2000 , , 533 , 682 cool , r. j. 2007 , , 169 , 21 de breuck , c. , klamer , i. , johnston , h. , hunstead , r. w. , bryant , j. , rocca - volmerange , b. , & sadler , e. m. 2006 , , 366 , 58 de breuck , c. , van breugel , w. , rttgering , h. j. a. , & miley , g. 2000 , , 143 , 303 de breuck , c. , van breugel , w. , stanford , s. a. , rttgering , h. , miley , g. , & stern , d. 2002 , , 123 , 637 de vries , w. h. , becker , r. h. , & white , r. l. 2006 , , 131 , 666 de vries , w. h. , morganti , r. , rttgering , h. j. a. , vermeulen , r. , van breugel , w. , rengelink , r. , & jarvis , m. j. 2002 , , 123 , 1784 eisenhardt , p. r. , et al . 2004 , , 154 , 48 elston , r. j. , et al . 2006 , , 639 , 816 fazio , g. g. , et al . 2004 , , 154 , 10 higdon , j. l. , et al . 2005 , , 626 , 58 hopkins , a. , windhorst , r. , cram , l. , & ekers , r. 2000 , exp . astron . , 10 , 419 jannuzi , b. t. , & dey , a. 1999 , asp conf . ser . 191 : photometric redshifts and the detection of high redshift galaxies , ( san francisco , ca : asp ) , 191 , 111 kenter , a. , et al . 2005 , , 161 , 9 krolik , j. h. , & chen , w. 1991 , , 102 , 1659 ledlow , m. j. , & owen , f. n. 1996 , , 112 , 9 lilly , s. j. , & longair , m. s. 1984 , , 211 , 833 matthews , k. & soifer , b. t. 1994 , assl vol . 190 : astronomy with arrays , the next generation , 239 mccarthy , p. j. 1993 , , 31 , 639 mclure , r. j. , willott , c. j. , jarvis , m. j. , rawlings , s. , hill , g. j. , mitchell , e. , dunlop , j. s. , & wold , m. 2004 , , 351 , 347 murray , s. s. , et al . 2005 , , 161 , 1 oke , j. b. , et al . 1995 , , 107 , 375 pentericci , l. , mccarthy , p. j. , rttgering , h. j. a. , miley , g. k. , van breugel , w. j. m. , & fosbury , r. 2001 , , 135 , 63 polletta , m. , et al . 2007 , arxiv : astro - ph/0703255 seymour , n. , et al . 2007 , , 171 , 353 spergel , d. n. , et al . 2007 , , 170 , 377 stern , d. , et al . 2005 , , 631 , 163 tielens , a. g. g. m. , miley , g. k. , & willis , a. g. 1979 , , 35 , 153 van breugel , w. , de breuck , c. , stanford , s. a. , stern , d. , rttgering , h. , & miley , g. 1999 , , 518 , l61 van breugel , w. j. m. , stanford , s. a. , spinrad , h. , stern , d. , & graham , j. r. 1998 , , 502 , 614 wang , j. x. , et al . 2004 , , 127 , 213 willott , c. j. , rawlings , s. , blundell , k. m. , & lacy , m. 1999 , , 309 , 1017 willott , c. j. , rawlings , s. , jarvis , m. j. , & blundell , k. m. 2003 , , 339 , 173

substantial progress in fabricating high-@xmath0 mechanical resonators with high frequencies , as well as recent success in cooling them close to the motional ground state @xcite , inaugurate a new research field of manifold fundamental interest @xcite . the regime of very low temperature , where a quantum mechanical description predicts only few quanta of mechanical motion , promises potential insight into some fundamental questions of decoherence , as well as various technical applications that make use the of the expected quantum behavior @xcite . fundamental questions , concerning the border between the classical ( macroscopic ) and the quantum ( microscopic ) worlds @xcite , trigger a natural interest in preparing quantum states of `` as large as possible '' objects and demonstrating their distinct quantum behavior by appropriate measurements . see @xcite for a recent review . regarding this major goal , it is important to stress that the dynamics of a purely harmonic quantum system is analogous to its classical dynamics , in the sense that expectation values of canonical observables follow the classical equations of motion @xcite . therefore , it is a common approach to introduce nonlinearities in a quantum system in order to detect quantum behavior . while there may be the possibility to achieve the strong optomechanical coupling regime @xcite and make use of the nonlinear nature of the standard optomechanical coupling , or to couple to a nonlinear ancilla system @xcite , we propose here a different approach : the use of an optoelectromechanical system featuring a tunable _ mechanical _ nonlinearity per phonon . the latter originates from the intrinsic geometric nonlinearity of elastic systems @xcite and its amount per motional quanta is enhanced with the help of electrostatic fields . this has the advantage that the linear optomechanical coupling is preserved as a control channel providing techniques as , for example , the sideband driving technique used in @xcite . the regime of large mechanical nonlinearity then enables new means to control the mechanical motion at the quantum level , if combined with the coupling to a high - q optical cavity mode , as well as the application of suitable gradient fields . the intrinsic anharmonicity in the mechanical motion of micro- and nanomechanical resonators is usually small and therefore only relevant in the regime of large oscillation amplitudes . in order to render the anharmonicity relevant for displacements at the scale of the quantum mechanical zero point motion , we propose to use electrostatic gradient forces to enhance the latter @xcite . these forces result from the dielectric properties of the resonator material when an inhomogeneous external electric field is applied . they can be used to effectively reduce the resonator s stiffness and therefore its resonance frequencies . in turn , this has the effect that the zero point deflection is enhanced up to an extent that the nonlinear contribution becomes important . using this technique , the nonlinearity per phonon can be made large enough , that distinct transitions in the mechanical spectrum can be resonantly addressed while interacting with other quantum systems . examples are the selective sideband driving of transitions in the mechanical spectrum @xcite , or the resonant exchange of excitations within an array of nanoresonators via a common cavity mode @xcite . in this paper , we explicitly derive the fundamental mode properties of a nonlinear mechanical resonator , subject to the aforementioned gradient forces , to obtain a suitable model for the mechanical degree of freedom . we then derive the specifics of the optomechanical coupling to a high finesse cavity and analyze possible source of damping and decoherence in detail . we also summarize different control schemes , associated with suitable laser drives for the cavity and gradient fields from the tip electrodes . two applications of these control mechanisms have been proposed previously @xcite , considering state of the art experimental components . we finally discuss a readout scheme for the nonlinear mechanical resonators considered . the remainder of the paper is organized as follows . in section [ sec : elasticity ] we describe the nonlinear dynamics of thin rods starting from elasticity theory and derive the resulting fundamental mode hamiltonian . in section [ sec : softening ] we describe our approach to enhance the intrinsic mechanical nonlinearity by mode softening with gradient forces . then we introduce the optomechanical model with several laser driven cavity modes in section [ sec : optomech ] . after introducing a possible setup for an implementation in section [ sec : setup ] , we quantitatively discuss its central physical properties , namely the optomechanical coupling mechanism in section [ sec : coupling ] , as well as potential setup specific losses in section [ sec : losses ] . in section [ sec : controlmech ] we review some mechanisms that can be used to control the mechanical motion at the level of single phonons and which have been applied in previous works @xcite . finally , in section [ sec : measurement ] we introduce methods to obtain information on the mechanical state from the output spectrum of a probe laser . the harmonic description of the transverse motion of thin rods is based on only considering the bending energy for small deflections @xcite . we consider here thin rods , which means that the cross - sectional dimensions , such as the radius for circular cross sections or width and depth for rectangular cross sections , are much smaller than the length @xmath1 . we also consider the rod to be homogeneous along the longitudinal axis , here parametrized by @xmath2 , with constant mass line density @xmath3 and use thin rod elasticity theory . the planar deflection in the transverse direction is described by a field @xmath4 and we consider a bridge geometry where the end points at @xmath5 and @xmath6 are fixed , i.e. @xmath7 and @xmath8 . the lagrangian within this approximation reads @xmath9\ , , \label{lagrangian}\ ] ] with a kinetic part as well as the bending energy @xmath10=\frac{1}{2}\int\limits \mathcal{f}\tilde{\kappa}^2(y'')^2{\rm d}x\ , . \label{bending_energy}\ ] ] here , @xmath11 is the linear modulus of the rod given by the young modulus @xmath12 of the material times the cross - section area @xmath13 , and @xmath14 is the ratio between the bending and compressional rigidities and depends on the cross - sectional geometry , where @xmath15 is the in - plane coordinate within the cross - section that is directed along the deflection with origin at the neutral line , see figure [ fig : bending_fig1 ] . for a rectangular cross - section of thickness @xmath16 , @xmath17 , whereas for a circular cross - section with radius @xmath18 , @xmath19 . for a cylindrical shell like a nanotube one finds @xmath20 . the energy ( [ bending_energy ] ) results from the fact that for small curvature @xmath21 , the local strain inside the rod is linear with respect to the distance @xmath15 to the neutral surface ( cf . 1 ) while the free energy density is quadratic with respect to the strain . depends on the transverse coordinate @xmath15 and on the local curvature @xmath21 and determines the energy density @xmath22 . ] the lagrangian ( [ lagrangian ] ) leads to the equation of motion @xmath23 as this equation is linear in @xmath24 and its derivatives , it leads to harmonic dynamics characterized by the following eigenmodes @xmath25}\,,\nonumber\end{aligned}\ ] ] with frequencies @xmath26 , where @xmath27 is the phase velocity of compressional phonons along the rod . the @xmath28 are the roots of the transcendental equation @xmath29 , with @xmath30 the smallest one . the @xmath31 are normalization constants chosen such that @xmath32 . we choose this normalization so that the coefficients in a mode expansion represent the maximum amplitudes of the deflection associated to each mode . introducing now the canonical momentum @xmath33 , as well as the expansion of the field into the modes @xmath34 yields the hamilton function of a harmonic oscillator for each mode @xmath35 with the deflection @xmath36 and mode momentum @xmath37 for the @xmath38-th mode , as well as the effective mode masses @xmath39 . corrections to this harmonic description that lead to nonlinearities originate from a stretching effect that occurs due to the deflection if the end points of the rod are fixed @xcite . the resulting strain leads to an additional energy @xmath40 where the stretched length is @xmath41 with @xmath1 being the zero deflection length . including this streching energy leads to a nonlinear extension of the hamiltonian , which after inserting the modes given in eq . leads to @xmath42 where @xmath43 . we quantize this model by introducing bosonic mode operators @xmath44 and @xmath45 , given by @xmath46 where we introduced the zero point motion amplitudes @xmath47 for each mode . this leads to the description @xmath48 with nonlinearity @xmath49 as in a rigorous elasticity treatment this description arises from an adiabatic elimination of the stretching modes , the indices in eq . should run up to an @xmath50 corresponding to an `` ultraviolet '' cutoff @xmath51 . the terms involving higher order modes induce small frequency shifts and nonlinear mode coupling . however , the later is found to be negligible for the parameters considered ( see appendix [ appendix b ] ) and the shift of the fundamental mode can be ignored given the additional tunable electrostatic contribution ( see section [ sec : softening ] ) . therefore we can restrict our description to the fundamental mode with @xmath52 and drop this label to get the usual hamiltonian for the duffing oscillator @xmath53 where we have introduced the fundamental frequency @xmath54 and the effective mass of the fundamental mode @xmath55 . the anharmonicity is given by @xmath56 in terms of phonon creation and annihilation operators @xmath57 and @xmath58 this hamiltonian reads , @xmath59 with the nonlinearity parameter @xmath60 . here , the frequencies @xmath61 and @xmath62 refer to fundamental mode properties that result only from the intrinsic elastic forces in the absence of any externally applied forces on the rod . as we will describe in the next section , external forces can be used to tune the resonance frequency , @xmath63 , which will in turn change the zero point motion amplitude , @xmath64 , and hence the nonlinearity , @xmath65 . in this section we describe how electric fields generated by tip electrodes that are placed near the center of the doubly clamped nanobeam , see figure [ fig : bending_fig2 ] , can be employed to control its dynamical properties . in particular , inhomogeneous gradient fields can be used to enhance the nonlinearity per phonon . the later scales as @xmath66 and can thus be enhanced by lowering the harmonic oscillation frequency @xmath61 . one way to change the mode frequency that has been discussed previously , is to add an additional external force along the rod axis , that causes compressive or tensile strain @xcite . an alternative approach , that promises better control but yields the same potential for the fundamental mode , is to use a static electric field that is strongly inhomogeneous in the direction of deflection . if the rod material shows suitable dielectric properties , this produces an additional , inverted square potential with respect to the deflection , see figure [ fig : bending_fig2 ] . here we consider tip electrodes with suitable applied voltages at each side of the nanoresonator that generate an electrostatic field . in our calculation we model the electrodes by point charges @xmath67 and @xmath68 , which is valid given that the relevant tip radii are much smaller than the gap between the electrodes ( cf . section [ sec : setup ] ) . the electrostatic energy associated to the dielectric per unit length along the rod can be described by @xmath69\,,\ ] ] where @xmath70 are the co - ordinates along the resonator axis and the direction of its deflection . @xmath71 are external field components parallel and perpendicular to the beam axis and @xmath72 the respective screened polarizabilities . we can expand @xmath73 to second order in the displacement @xmath24 and get an additional contribution to the hamiltonian of the nanobeam that reads @xmath74{\rm d}x\,,\nonumber\end{aligned}\ ] ] where we dropped the displacement independent constant @xmath75 which is irrelevant for the dynamics . inserting the modes defined in equation we get @xmath76 with @xmath77 the contributions @xmath78 will cause weak interactions between the modes , as @xmath78 is not a diagonal matrix . by diagonalizing the hamiltonian of the rod in the presence of electrostatic fields , one can find new normal modes . for the parameters considered in this work , the induced corrections to the mode shapes are however found to be negligibly small . we thus focus on the fundamental mode contributions @xmath79 and @xmath80 , where @xmath81 . . the field profile ( red ) leads to an inverted parabola for the dielectric potential ( blue ) around the equilibrium position . ] one may also consider time dependent electric fields and it is convenient to separate between static and time dependent contributions , @xmath82 whereas the time dependent contributions @xmath83 and @xmath84 generate drives applied to the nanobeam and will be considered in section [ sec : controlmech ] , we now focus on the constant contributions @xmath85 and @xmath86 , which can be employed to tune the nanobeam . the electrostatic force @xmath85 causes a static deflection of the nanobeam and shifts its equilibrium position . however if the rod interacts with the photon fields of a nearby cavity , those fields will also cause a deflecting force . for convenience we choose @xmath85 such that these two forces compensate and the equilibrium position remains unshifted ( see section [ sec : optomech ] ) . the electrostatic potential associated to @xmath86 in turn is an inverted harmonic potential that lowers the harmonic oscillation frequency of the nanobeam . therefore , we consider @xmath87 as the `` tuned '' mechanical hamiltonian with a reduced frequency @xmath88 . in a phononic description this hamiltonian reads @xmath89 where the nonlinearity per phonon @xmath90 is now increased by a factor @xmath91 , cf . as an example , for maximum applied fields at the tube @xmath92vm@xmath93 and @xmath94vm@xmath93 and a gap between the electrodes of size @xmath95 nm ( cf . 3 ) , typical parameters discussed in section [ sec : setup ] , yield @xmath96 allowing to boost @xmath97 by at least an order of magnitude . for further calculations , it is convenient to express all observables in the energy eigenbasis of the hamiltonian @xmath98 , so that @xmath99 where the energy eigenstates @xmath100 and energy levels @xmath101 , as well as the displacement matrix elements @xmath102 have to be determined numerically . for small nonlinearites @xmath103 analytical expression can be obtained as the hamiltonian is approximately diagonal in fock basis since one may apply a rotating wave approximation in the nonlinear part @xmath104 where @xmath105 , @xmath106 and the eigen - energies are given by @xmath107 . the time dependent contributions @xmath108 and @xmath109 in turn can be used to drive or temporarily detune the resonator . this has been used in @xcite to perform local gate operations on nanobeams acting as qubits . in the following section we will now discuss an optomechanical interaction between the nano resonator described by the hamiltonian in equation ( [ hm1 ] ) and the resonance modes of a high finesse cavity . we consider a typical optomechanical setup with a micro toroid cavity coupled to several nanomechanical resonators , where the displacement of the latter modifies the frequencies of the cavity modes . the cavity modes with frequencies @xmath110 that contribute to the dynamics are described by photon creation and annihilation operators @xmath111 and @xmath112 . they are each driven by a classical laser of input power @xmath113 and frequency @xmath114 . the coupling strength between cavity mode @xmath115 and nanobeam @xmath116 is given by @xmath117 , where @xmath118 is the optical frequency shift per deflection @xmath119 . thus , in a frame rotating with the laser field modes , the hamiltonian describing the coupled system reads ( @xmath120 ) @xmath121+\sum\limits_jh_{\mathrm{m},j}\nonumber\\ & + \sum\limits_{ij}g_{0,ij}{a_i}^{\dag}a_i{\ensuremath{\mathcal{x}}}_j\ , . \label{hamiltonian}\end{aligned}\ ] ] here we have introduced the laser detunings denoted by @xmath122 and drive amplitudes @xmath123 , with @xmath124 being the cavity decay rates into the associated outgoing electromagnetic modes . both the field inside the cavity and the mechanical motion are subject to damping , which in the regime of weak optomechanical coupling , small nonlinearity and low mechanical occupation is well described by a master equation with lindblad form damping terms . with the decay rates for the cavity modes @xmath125 and the mechanical damping rates @xmath126 , this master equation reads @xmath127&+\sum\limits_i\frac{\kappa_i}{2}{\ensuremath{\mathcal{d}}}(a_i)\rho\,+\\ \nonumber&+\sum\limits_j\frac{\gamma_j}{2}\left[(\overline n_j+1){\ensuremath{\mathcal{d}}}(b_j)\rho+\overline n_j { \ensuremath{\mathcal{d}}}({\ensuremath{b^{\dag}}}_j)\rho\right]\ , . \end{aligned}\ ] ] here we introduced the lindblad form dissipator @xmath128 , as well as the bose occupation number @xmath129^{-1}$ ] of the phonon bath mode with frequency @xmath130 at temperature @xmath131 . a more precise treatment of the mechanical damping accounts for the fact that , due to the mechanical nonlinearity , there is more than one bath mode resonantly coupling to the mechanical mode however , for small nonlinearity , equation ( [ mg ] ) proofs to be sufficiently accurate . as usual , we expand the cavity field operators around their steady state values and adopt a shifted representation @xmath132 , with @xmath133 , in which the master equation has the same form as in equation ( [ mg ] ) but with the shifted system hamiltonian @xmath134 with @xmath135 and where we have dropped the nonlinear terms @xmath136 in the coupling , which is valid for @xmath137 . we have also assumed that the static electric fields for each beam are chosen such that @xmath138 , so that their equilibrium positions are undeflected . we now turn to discuss a possible experimental setup that would allow to explore the physics described by the model presented in equation . d ) schematic cross - sectional view of the nems - chip and the cavity rim ( yellow ) . the nems - chip is mounted in the evanescent cavity field at a distance @xmath16 to the cavity rim surface . the nanoresonator is displaced from the closest point to allow for an optomechanical coupling that is linear in the resonator s deflection.,title="fig : " ] + for the experimental realization of the model in equation , we envisage a setup as shown in fig . [ fig : setup ] , comprising a nems chip containing the nanobeam resonators and a high finesse toroidal microcavity @xcite . each nanobeam resonator consists of a suspended single - walled carbon nanotube ( cnt ) with radius @xmath18 ( @xmath140ms@xmath93 , @xmath141kgm@xmath142 ) @xcite , and has electrodes in its vicinity that generate the electric fields for controlling and driving it . furthermore , the nanoresonator interacts with the evanescent field of the microcavity via optical gradient forces ( see section [ sec : coupling ] ) . given the state of the art , a cnt is a favorable system to implement the proposed optoelectromechanical scheme @xcite . in particular , the intrinsic nonlinearity per phonon scales as @xmath143 , which favors small transverse dimensions and masses . additionally , carbon nanotubes show ultra - low dissipation @xcite , that is expected to decrease further in the regime of small amplitudes @xcite . it is convenient to use ( @xmath144,@xmath145 ) cnts with radius @xmath146 , as these tubes show relatively large static polarizabilities with @xmath147 , @xmath148 @xcite , where @xmath149 ( @xmath150 ) is the polarizability parallel ( perpendicular ) to the tube axis . to obtain nonlinearities that are large enough , it is convenient to use cnt lengths below @xmath151 for @xmath152 we have @xmath153 and @xmath154 . the cavity is a silica ( @xmath155 ) microtoroid with resonant wavelengths @xmath156 , circumference @xmath157 , finesse @xmath158 @xcite and mode volume @xmath159 . these parameters correspond to @xmath160 , @xmath161 and a decay length of the evanescent field @xmath162 [ cf . eq . in section [ sec : coupling ] ] . the nems chip is placed at a distance @xmath163 from the cavity rim . the cnt - resonator is displaced from the closest point to allow for a linear coupling , as the resonator moves in the plane of the chip s surface ( cf . section [ sec : coupling ] ) , and the electrodes are aligned parallel to the rim of the cavity to minimize additional cavity losses they might induce , see section [ sec : losses ] . in the following two sections we analyze important practical aspects of the envisioned implementation with a carbon nanotube coupled to a toroidal microcavity in more detail . thus , readers who are only interested in the results of the discussed mechanism may directly turn to section [ sec : controlmech ] . in this section we derive an estimate for the coupling between the mechanical displacement and the cavity field . the coupling arises as the energy of the dielectric oscillator in the evanescent electric field depends on the field strength at the location of the oscillator . as the evanescent field decays with distance to the cavity rim , altering the oscillator - cavity distance by displacing the oscillator results in a change of energy . thus , the coupling part of the hamiltonian is given by @xmath164 where the polarization is @xmath165 with @xmath166 the screened polarizability tensor and @xmath167 the cavity field . the integration is taken over the oscillator volume @xmath168 and without loss of generality we consider a single cavity mode with resonant frequency @xmath169 . given the dimensions of the toroidal microcavity , its torus can be locally modeled as a cylindrical waveguide of radius @xmath170 note that this differs from the definition of @xmath170 used in ref . by a factor of @xmath171 [ cf . below eq . we introduce cylindrical coordinates @xmath172 with the @xmath173-direction along the waveguide axis ( cf . [ fig : setup]c ) and consider te@xmath174 modes , as a transverse electric field is advantageous to suppress loss mechanisms that are discussed in section [ sec : losses ] . the corresponding transverse fields are given by @xcite @xmath175 for the field inside the waveguide , @xmath176 , and @xmath177 outside the waveguide , @xmath178 . the axial field reads @xmath179 with the modified bessel function @xmath180 and the bessel function of the first kind @xmath181 . here @xmath182 , @xmath183 is the wavevector component parallel to the waveguide axis and @xmath184 and @xmath185 are the transverse wave vectors outside and inside the waveguide , respectively . henceforth , we assume a refractive index such that @xmath186 and a frequency @xmath169 well above cutoff , i.e. @xmath187 where @xmath188 and @xmath189 is the first zero of @xmath190 . these assumptions imply that @xmath191 , @xmath192 , @xmath193 , @xmath194 , and @xmath195 . within these approximations , the ratio of the axial magnetic field at @xmath196 to its value at the origin is given by @xmath197 , and the evanescent field can be written as @xmath198 this field will later be used to estimate losses induced by the electrodes . in order to determine the optomechanical coupling strength , we write the electric field in its quantized form @xmath199 with photon creation ( annihilation ) operators @xmath200 and where @xmath201 is the corresponding normalized eigenmode , satisfying @xmath202 we consider the following definition for the mode volume @xmath203 from eqs . and and properties of the bessel functions , we obtain that the maximum electric field inside the cavity is given by @xmath204 where @xmath205 is the first positive root of @xmath206 . thus , neglecting the small contributions of the evanescent part to the integrations in eqs . and which are higher order in @xmath207 , we obtain from eqs . - @xmath208 for @xmath178 . here , @xmath209 and @xmath210 denotes the ratio of the electric field at the waveguide s surface to the maximum field @xmath211 . the zero point motion @xmath64 of the oscillator is small compared to the decay length @xmath212 of the evanescent field and the same holds for the transverse dimensions of the oscillator . thus , we can linearize @xmath213 around the equilibrium position of the nanoresonator and assume that the electric field is constant everywhere inside the resonator volume @xmath168 . subsequently , by comparing the hamiltonians and , and using eqs . and , we find for the opto - mechanical coupling rate , @xmath214 here , we have neglected the contribution of the perpendicular polarizability since , for carbon nanotubes , the perpendicular polarizability is typically one order of magnitude smaller than the parallel one , and again used @xmath215 so that only the derivative of the exponential factor is relevant . the geometry of the setup we consider here is illustrated in figure [ fig : setup ] and the dependence on the alignment and positioning of the nanotube is accounted for in the correction factor @xmath216 . for a te@xmath174 mode of the cavity field , the electric field is directed along @xmath217 , i.e. tangential to the cavity rim , while the nanotube is aligned along @xmath218 . in addition , the deflection of the nanotube in the direction @xmath219 is not aligned with the interaction - energy gradient , which is approximately along @xmath220 . finally , if @xmath16 is the distance of the chip to the cavity rim , the actual distance of the oscillator to the rim is @xmath221 . taking into account these various issues , we find for the correction factor @xmath222 this is maximized for @xmath223 and @xmath224 , where we consider the leading order in the small parameter @xmath225 . these optimal angles yield @xmath226 , resulting in @xmath227 for the parameters introduced in section [ sec : setup ] ( i.e. @xmath155 and @xmath228 ) . finally , for those typical values we obtain @xmath229 . in addition to the well known loss mechanisms of photon losses from the cavity and intrinsic phonon losses of the mechanical resonator @xcite , there can be further sources of loss in our setup due to the presence of the tip electrodes . in this section we show that these additional loss mechanisms are negligible for the parameters we envision . exploiting the nonlinearity of the nanoresonators to control their dynamics demands a high-@xmath0 optical cavity as the linewidth @xmath230 needs to be at most comparable to the mechanical nonlinearity @xmath97 . using conventional metallic electrodes to generate the inhomogeneous control fields can potentially increase the cavity losses . to achieve the necessary low losses it is crucial to have deep subwavelength transverse dimensions for the electrodes . we now give an estimate of the photon losses that may be induced by such electrodes and show that these are negligible . to do so , we model the electrodes as metallic cylinders and assume that their radius @xmath231 is much smaller than the decay length of the evanescent cavity field , @xmath232 . we assume the electrodes to be parallel to the waveguide representing the cavity rim , with a small misalignment angle @xmath233 . for the relevant te@xmath234 modes , the losses result solely from the misalignment , as in the small radius regime considered they arise only from the field along the electrode axis which vanishes for @xmath235 . the resulting finesse @xmath236 for the cavity can be obtained from the finesses @xmath237 associated to different decay channels , which can be assumed to be independent such that @xmath238 for each loss channel , the finesse @xmath237 can be determined from the ratio of the power circulating in the cavity @xmath239 to the fraction of power that it lost through the respective channel @xmath113 , using that the time - averaged stored energy and free spectral range are given , respectively , by @xmath240 and @xmath241 . thus , we arrive at @xmath242 the loss channels that will be considered subsequently are ( 1 ) scattering by the `` bulk '' of the electrodes modeled as a single metallic cylinder , ( 2 ) scattering by the `` gap '' between the electrodes and ( 3 ) absorption . the independence between the contributions ( 1 ) and ( 2 ) assumed in eq . amounts to neglecting the interference between them which is permissible when estimating an upper bound . [ [ incident - field - and - circulating - power ] ] incident field and circulating power + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for our calculations , we introduce new cylindrical co - ordinates @xmath243 for the electrode with the @xmath244-direction along its axis . we first express the cavity field that is incident on the electrode in these primed coordinates . the result will later be used to determine the scattered and absorbed fractions of the incident power . to this end , we consider the projection of the electric field determined in section [ sec : coupling ] onto the electrode s axis , see fig . [ fig : coordinates_electrode_wg ] , @xmath245 which completely determines the losses in the small radius regime considered . the origin of the primed axis lies at @xmath246 and the relevant unit vectors are given by @xmath247 for points on the @xmath244-axis we have @xmath248 for our further calculations it is convenient to express the incident field via its fourier transform @xmath249 . using this and equations ( [ coordinate_rel1])-([coordinate_rel2 ] ) in equations and we arrive at @xmath250 where we have substituted @xmath251 and @xmath252 . one can find an approximation for the integral @xmath253 by applying the method of steepest descents , using @xmath254 and @xmath255 , which for @xmath256 yields @xmath257 and waveguide coordinates @xmath258 . the misalignment is determined by a small angle @xmath233.,scaledwidth=40.0% ] finally , we determine the circulating power to leading order in @xmath207 , which is given by @xmath259 where we have used eqs . and , that @xmath260 , and the time - averaged poynting vector @xmath261 with the vacuum permeability @xmath262 . [ [ scattering - losses ] ] scattering losses + + + + + + + + + + + + + + + + + here , we model the electrodes as a single metallic cylinder which for simplicity is assumed to be perfectly conducting since this maximizes the scattering and , thus , provides an estimate of an upper bound to the corresponding losses that is independent of material properties naturally for the transparent electrode scenario considered below , in section [ electrode losses].d , these losses would be substantially smaller than this upper bound . we expand the scattered field into solutions of the wave equation in cylindrical coordinates for the electrode . as the radius of the electrode @xmath231 is much smaller than the wavelength of the incident field @xmath263 , all contributions to the scattered power are suppressed at least like @xmath264 , except for s - wave scattering of tm modes , for which the suppression is only logarithmic . this can be understood in terms of the taylor expansions of the corresponding cylindrical harmonics and the incident field . in turn , to determine the tm s - wave scattering to leading order in @xmath265 , the incident field can be assumed to be constant for a given cross section and determined by the field at the electrode s center @xmath266 . we neglect multiple scattering between the waveguide and the electrode and ignore the dielectric substrate of the latter . thus , the scattered field @xmath267 is determined from the homogeneous boundary condition at the surface of the electrode @xmath268 , and an outgoing - wave boundary condition at infinity for @xmath267 here @xmath269 is the evanescent contribution . the transverse fields of an outgoing tm solution with @xmath244-dependence @xmath270 are given by @xmath271 with @xmath272 . for @xmath273 the solution is evanescent and does not contribute to the scattered power . for s - wave scattering @xmath274 , where @xmath275 , and to leading order in @xmath265 , the scattered field can be written as @xmath276 where @xmath277 is the fourier transform of the incident field @xmath266 and we have used the approximation @xmath278 for @xmath279 . we calculate the scattered power by integrating the energy flux across a cylinder coaxial with the electrode with radius @xmath280 . thus , from eqs . - we obtain for the scattered power @xmath281}}_{g } \label{scatpower}\end{aligned}\ ] ] where we have used @xmath282 and @xmath283 $ ] for @xmath284 . the integral @xmath285 can be estimated by performing the substitution @xmath286/\ln(2kr)$ ] and considering @xmath287 , which yields @xmath288 . hence , from eqs . , - , and , and using @xmath289 and @xmath290 for @xmath256 , and @xmath291 we arrive at @xmath292}}{n_{\mathrm{c}}\left(k{\ensuremath{a_{\mathrm{c}}}}\sqrt{n_{\mathrm{c}}^2 - 1}\right)^3|\ln2kr'|}\,.\ ] ] if we now use that for relevant parameters @xmath293 , @xmath294 and @xmath295 , we find a lower bound for the finesse associated to scattering losses , @xmath296 which is independent of the ratio @xmath297 . thus , for @xmath298 , @xmath299 , @xmath155 and @xmath300 , we find @xmath301 . hence losses due to scattering are clearly sufficiently suppressed for electrodes of subwavelength radius that are approximately aligned with the cavity rim . [ [ gap - contribution - to - scattering - losses ] ] gap contribution to scattering losses + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + so far we have considered scattering from one single continuous electrode . actually , our setup comprises two electrodes , see figure [ fig : setup ] , separated by a gap in which the nanomechanical oscillator is positioned . we now estimate an upper bound to the additional losses that this gap may induce . here we assume as before @xmath302 so that the gap is much smaller than the distance over which substantial currents are induced in the electrodes and one may consider @xmath235 for this estimate . denoting the direction joining the electrodes by @xmath303 , we focus on the relevant regime @xmath304 so that one can assume that only the incident field component along @xmath303 is screened by them . we model the gap of size @xmath305 as a perfectly conducting sphere with radius @xmath306 subject to an external field along @xmath303 determined by @xmath307 . this should provide an upper bound for the magnitude of the total induced dipole @xmath308 , which in turn determines the leading contribution in @xmath309 to the scattered power . thus we find @xmath310 @xcite for optimal placement , which yields for the scattered power @xmath311 ^ 2 { \ensuremath{\mathrm{e}^}}{-2\kappa_{\bot}d}\cos^2\theta'_*\,.\ ] ] here we have used eq . , @xmath312 and @xmath313 . along the same lines as before , using eqs . and and @xmath223 , the associated finesse reads @xmath314 which yields @xmath315 for the parameters discussed in section [ sec : setup ] and @xmath95 nm . thus , additional scattering losses due to the gap between the electrodes are also negligible . [ [ absorption - losses ] ] absorption losses + + + + + + + + + + + + + + + + + we assume here transparent electrodes afforded by cylindrical shells with 2d conductivity @xmath316 that absorb the power @xmath317 since @xmath318 induces a current @xmath319 in each electrode here , as in [ electrode losses].b . , we neglect the small gap . from eqs . and - we obtain for the absorbed power @xmath320 ^ 2 { \ensuremath{\mathrm{e}^}}{-2\kappa_{\bot}d}|b_{z}(0)|^2\,\times\\ & \times\underbrace{\int \limits_{-\infty}^{\infty}{\rm d}x\frac{{\ensuremath{\mathrm{e}^}}{-2\kappa_{\bot}(d+{\ensuremath{a_{\mathrm{c } } } } ) ( \sqrt{1+x^2}-1)}}{(1+x^2)^{3/2}}}_{j}{\nonumber}\end{aligned}\ ] ] where we have again substituted @xmath251 . using the method of steepest descents we estimate @xmath321 for @xmath322 . thus using @xmath323 and eqs . , and , yields for the finesse associated to absorption @xmath324 ^ 2\,.\ ] ] we consider now the specific case where the transparent electrodes are provided by a pair of nanotubes . the latter exhibit a maximum in the conductivity @xmath325 @xcite . by assuming an off - resonant @xmath326 with @xmath327 , we get @xmath328 with the fine structure constant @xmath329 . if we consider the same values as before except that now @xmath330 and @xmath331 and assuming @xmath332 we find @xmath333 . hence even though absorption losses clearly dominate over scattering losses , their effect can still be neglected for electrode radii @xmath334 and alignment angles @xmath330 . here we give estimates of decoherence rates for the nanoresonator induced by noise in the inhomogeneous electric fields . such noise might originate from voltage fluctuations @xmath335 due to the electrodes resistance ( johnson - nyquist noise ) or from moving charges on the chip surface ( @xmath336-noise ) . we calculate the respective single - phonon decoherence rates @xmath337 and @xmath338 from the corresponding noise spectra @xmath339 and @xmath340 using the relation @xmath341 with @xmath342 where @xmath343 is the force fluctuation acting on the resonator . the electrostatic gradient force acting on a resonator can be expressed by @xmath344 where we have estimated the field gradient at a distance @xmath345 from the electrode by @xmath346 and used the fact that the field mainly acts on the nanotube in a region of length @xmath347 . for a field with fluctuations associated to different independent sources @xmath348 , the force fluctuations are then given by @xmath349 thus , the resulting decoherence rates read @xmath350 where the @xmath351 are the noise spectra for the different electric field fluctuations . [ [ johnson - nyquist - noise ] ] johnson - nyquist noise + + + + + + + + + + + + + + + + + + + + + for johnson - nyquist noise @xcite , we have fluctuating voltages @xmath335 with @xmath352 for an ambient temperature @xmath131 and an internal resistance @xmath353 . for our setup we find @xmath354 at @xmath355 , which is well below the relevant mechanical decoherence rate @xmath356 for relevant resistances @xmath357 . [ [ f - noise ] ] @xmath336-noise + + + + + + + + + + + + + + + + + + + + + + + the origin of @xmath336-noise is usually associated with surface charge fluctuations in the device . an electric field noise density @xmath358 has been measured at @xmath359 and at a distance of @xmath360 between a charged resonator and a gold surface @xcite . for a scaling @xmath361 @xcite this corresponds to @xmath362 for our conditions with @xmath355 and @xmath363 . thus we expect for the associated decoherence rate @xmath364 , which is again well below the mechanical decoherence rate @xmath365 . these results are also corroborated by recent estimates that were obtained for a related setup @xcite . the hamiltonian of the full optomechanical system with tuned nanobeams potentially leads to complex dynamics for photons and phonons . here , we focus on scenarios where driven cavity modes and suitable electric gradient fields are used to control the dynamics of one or several nanobeams . we summarize the basic principles for three conceptually different schemes , namely ( i ) the selective addressing of transitions in the mechanical spectrum by cavity sideband driving , ( ii ) the coherent interaction between several nonlinear nanobeams mediated by a common driven cavity mode and ( iii ) the manipulation of a single resonator s state by time dependent gradient fields . the first scheme represents a suitable extension of the standard sideband cooling technique @xcite to nonlinear resonators . here , the detuning of a red ( blue ) detuned laser drive is only resonant with one specific transition @xmath366 ( @xmath367 ) in the nonlinear mechanical spectrum , see figure [ fig : schemes]a . therefore , if the mechanical nonlinearity is resolved by the cavity linewidth , @xmath368 , appropriate laser drives can lead to highly nonclassical steady states for the mechanical motion . for a single nanobeam for example , this allows for the preparation of stationary fock states with high fidelity @xcite . the sideband driving technique could potentially also be applied to more complicated level structures , for example the collective modes of several interacting nanobeams , and thus constitutes a versatile control mechanism . the second scheme uses the cavity to mediate a coherent coupling between several nanobeams that all couple to the same photon mode . here , the photon mode is driven with a large detuning to be off - resonant to any mechanical transition frequency . the coherent photon background field that builds up inside the cavity leads to an effective interaction @xmath369 between any pairs @xmath370 of nanobeams . in order to exchange excitations via this coupling , proper resonance conditions have to be met . by tuning each of the nanobeams using their respective electrodes , interactions between desired pairs of beams can be realized @xcite . furthermore , due to the nonlinear spectra , it is possible to restrict the dynamics of the nanobeams to the `` qubit '' subspace built up by the states @xmath371 , c.f . equation , for each resonator . finally , beside the static tuning capability , see equation , the gradient fields provided by the tip electrodes can be used to perform coherent operations on any nanobeam . this becomes most obvious if one considers the qubit subspace @xmath372 for one nanobeam . here , a drive @xmath373 , where @xmath374 is the qubit transition frequency , implements a @xmath375 rotation , see figure [ fig : schemes]b . a temporary shift of the qubit transition frequency @xmath376 can be achieved by a temporary @xmath84 contribution , which corresponds to a @xmath377 rotation , see figure [ fig : schemes]c . note that the drive @xmath378 associated with the tip electrodes is a coherent drive , while the cavity sideband driving technique constitutes a stochastic drive . together with the coherent coupling of several nanobeams , the time dependent gradient fields can for example be employed to build up a universal set of quantum gates for quantum information processing @xcite . + the steady state of a nanoresonator can be probed with an additional laser , weakly driving one cavity mode on resonance , i.e. with @xmath379 . then , information about the state of the mechanical resonator can be extracted from the sideband structures in the power spectrum , which correspond to photons that have been up- or down converted during the interaction with the mechanical motion . the intensity of the sideband peaks depends on the population of the mechanical energy eigenstates . thus , the power spectrum only provides information about the diagonal entries @xmath380 of the density matrix describing the mechanical resonator represented in the basis formed by eigenstates of the hamiltonian @xmath98 , see eq . ( [ eq : diaghm ] ) . for a probe laser resonantly driving a cavity mode at frequency @xmath381 , the spectrum shows a sideband structure @xmath382 with lorentzian sideband peaks determined by @xmath383 ^ 2+({\ensuremath{\gamma_{\mathrm{eff}}}}^{nm}/2)^2}\,,\ ] ] with @xmath384 as given in equation ( [ gammaeff ] ) , see appendix [ appendix b ] for details . here , @xmath385 is the optomechanical coupling strength associated with the probe laser and the @xmath386 denote the mechanical transition frequencies . the peaks appear in groups with @xmath387 ; see figure [ fig : spectrum ] . the occupation probabilities @xmath388 for the eigenstates @xmath389 can be extracted from the peak intensities within the main sidebands with @xmath390 @xcite . for the readout of mechanical qubits as discussed in @xcite , a shelving technique can be used to determine whether a qubit is in state @xmath391 or in state @xmath392 . here , a balanced cycling transition between @xmath393 using a cooling laser on @xmath394 and a coherent rf - drive with local gradient fields on those two levels causes a continuous stream of up converted photons only if the resonator is found in the state @xmath392 . this can be detected by measuring the corresponding sideband spectrum . here , a large enough number of photons has to be collected before external damping destroys the intermediate state , which requires @xmath395 . we have introduced a scheme to access a novel regime of optomechanics where the motion of the nanomechanical resonator becomes anharmonic and thus allows to explore genuine quantum dynamics . in our approach inhomogeneous electrostatic fields are applied to the nanomechanical resonator to enhance its anharmonicity per phonon until it becomes comparable to the linewidth of a high finesse optical cavity . for realistic experimental conditions , sufficiently large optomechanical couplings can be realized and losses induced by the tip electrodes can be suppressed to a negligible level . furthermore populations of the energy eigenstates of such nonlinear mechanical oscillators can be extracted from the output spectrum of a probe laser . the approach thus paves the way towards exploring nonclassical dynamics of nanomechanical oscillators at the single - phonon level . this work is part of the emmy noether project ha 5593/1 - 1 and was supported the crc 631 , both funded by the german science foundation ( dfg ) . iwr acknowledges support by the german science foundation ( dfg ) via the grant wi 3859/1 - 1 and nanosystems initiative munich ( nim ) . the authors thank j. kotthaus , t.j . kippenberg , and a. schliesser for enlightening discussions . in order to estimate the strength of the mode coupling , we rewrite the nonlinearity as @xmath396\ , , \label{bracket}\ ] ] where @xmath397 is the physical mass of the rod . the term in brackets solely depends on the mode shape for the doubly clamped boundary conditions and is independent of the parameters @xmath398 , @xmath3 , @xmath1 , of the oscillator . this can be seen from substituting @xmath399 , which yields @xmath400 table [ couple_vals ] shows some numerically obtained values for the bracket in equation that are relevant for the fundamental mode . in the case of an electrostatically tuned resonator , the modified nonlinear couplings read @xmath401 where @xmath402 is the factor by which the frequency of mode @xmath115 is reduced due to the presence of the gradient fields . while this factor is usually intended to be larger than unity for the fundamental mode @xmath403 , it remains close to unity for the higher modes . phonon transfer between modes is strongly suppressed because of resonance mismatches , as @xmath404 for processes where the phonon number in each mode is not preserved . one should note that the relevant frequency ratios scale as @xmath405 and the dominant processes of this type affecting the fundamental mode involve its coupling to the next higher mode with the same symmetry , which is the third mode . in addition , the fundamental mode experiences a modification of its rigidity due to the thermal and quantum fluctuations of higher order modes with @xmath406 . this effect however can be taken into account by a proper redefinition of the fundamental mode s rigidity . only pairs of modes @xmath370 with the same parity yield a finite coupling of this type . [ cols="^,^,^,^,^,^",options="header " , ] a steady state of the mechanical resonator can be probed via a resonant laser drive on an additional cavity mode . the quantum motion of the nanoresonator is described by the reduced master equation @xcite , @xmath408+\frac{1}{2}\sum\limits_{nm , j}a^{nm}_j{\ensuremath{\mathcal{d}}}({\ensuremath{|n\rangle}}{\ensuremath{\langlem|}})\mu+{\ensuremath{\mathcal{d}}}_{\gamma}\mu\ , . \label{eq : reduced_mg}\ ] ] here , @xmath409 includes the external mechanical damping via standard lindblad terms and the influence of the lasers is given by the rates @xmath410 here , for example , @xmath411 labels the probe laser with @xmath412 and the other lasers with @xmath413 are used for the steady state preparation , see section [ sec : controlmech ] . the rates @xmath414 have to be small , i.e. @xmath415 , to assure a weak measurement . the output power spectrum is given by @xmath416 where the output fields @xmath417 are related to the intracavity fields @xmath418 via the standard input - output relation @xcite , @xmath419 here , we only focus on the output for the probe field and thus drop the index @xmath116 . the dynamics of the intra cavity field can be described by a quantum langevin equation @xmath420 where @xmath421 and @xmath422 are the fluctuations of the input field in the laser mode and the other bath modes . defining @xmath423 to be a solution for @xmath424 , we can integrate formally and apply a dyson series type expansion to first order in the optomechanical coupling strength @xmath385 , to find for the motion of the cavity modes , cf . @xcite , @xmath425\\ & \approx a^{(0)}(t)-{\ensuremath{\mathrm{i}}}\frac{g_{\mathrm{m}}}{2}\sum\limits_{n < m}x_{nm}\left[\frac{{\ensuremath{|n\rangle } } { \ensuremath{\langlem|}}(t)}{{\ensuremath{\mathrm{i}}}\delta_{nm}-\kappa/2}+\text{h.c.}\right]\,.\label{dyson}\end{aligned}\ ] ] the contribution of the fluctuations is included in the free field solution @xmath426 and the input field operators have already been written in a shifted representation @xmath427 , which splits off the coherent part of the input . we substitute and into and concentrate on the contributions to the sidebands , which to lowest order in @xmath428 are given by the first order two - time correlations of the mechanical motion . the latter can be calculated from the reduced master equation using the quantum regression theorem . we find that @xmath429 where @xmath430 are the probabilities to find the resonator in the eigenstate @xmath389 , are the only nonvanishing contributions . thus the sideband spectrum around the probe laser frequency reads @xmath431 the resulting peak linewidths @xmath432 satisfy @xmath433 , where the nonlinearity @xmath97 is the typical peak distance within the fine structure of one sideband , see figure [ fig : spectrum ] . t. rocheleau , t. ndukum , c. macklin , j.b . hertzberg , a.a . clerk , and k.c . schwab , nature * 463 * , 72 ( 2010 ) . oconnell , m. hofheinz , m. ansmann , r.c . bialczak , m. lenander , e. lucero , m. neeley , d. sank , h. wang , m. weides , j. wenner , j.m . martinis and a.n . cleland , nature * 464 * , 697 ( 2010 ) . teufel , t. donner , dale li , j.h . harlow , m.s . allman , k. cicak , a.j . sirois , j.d . whittaker , k.w . lehnert , and r.w . simmonds , nature * 475 * , 359 ( 2011 ) . j. chan , t.p . mayer alegre , a.h . safavi - naeini , j.t . hill , a. krause , s. grblacher , m. aspelmeyer , and o. painter , nature * 478 * , 89 ( 2011 ) . e. verhagen , s. deleglise , s. weis , a. schliesser , and t. j. kippenberg , nature * 482 , * 63 ( 2012 ) . t. j. kippenberg and k. j. vahala , science * 321 * , 1172 ( 2008 ) a.d . armour , m.p . blencowe , and k.c . schwab , phys . lett . * 88 * , 148301 ( 2002 ) . f. marquardt and s.m . girvin , physics * 2 * , 40 ( 2009 ) . teufel , t. donner , m.a . castellanos - beltran , j.w . harlow , and k.w . lehnert , nature nanotech . * 4 * , 820 ( 2009 ) . a. schliesser , r. rivire , g. anetsberger , o. arcizet , t.j . kippenberg , nat . phys . * 5 * , 509 ( 2009 ) . g. anetsberger , e. gavartin , o. arcizet , q.p . unterreithmeier , e.m . weig , m.l . gorodetsky , j.p . kotthaus , t.j . kippenberg , phys . a * 82 * , 061804 ( 2010 ) . adler and a. bassi , science * 325 * , 275 ( 2009 ) . m. aspelmeyer , t. j. kippenberg , f. marquardt , arxiv:1303.0733 ( 2013 ) . ehrenfest , z. phys . a * 45 * , 455 ( 1927 ) . p. rabl , phys . lett . * 107 * , 063601 ( 2011 ) . a. nunnenkamp , k. borkje , and s. m. girvin , phys . lett . * 107 * , 063602 ( 2011 ) . j. c. sankey , c. yang , b. m. zwickl , a. m. jayich and j. g. e. harris , nat . * 6 * 707 ( 2010 ) k. jacobs , phys . lett . * 99 * , 117203 ( 2007 ) . s.m . carr , w.e . lawrence , and m.n . wybourne , phys . b * 64 * , 220101 ( 2001 ) . v. peano and m. thorwart , new j. phys . * 8 * , 21 ( 2006 ) . e. babourina - brooks , a. doherty and g.j . milburn , new j. phys . * 10 * , 105020 ( 2008 ) . i. katz , a. retzker , r. straub , and r. lifshitz , phys . * 99 * , 040404 ( 2007 ) . m.j . hartmann and m.b . plenio , phys . * 101 * , 200503 ( 2008 ) . dykman , m. marthaler , and v. peano , phys . rev . a * 83 * , 052115 ( 2011 ) . q.p . unterreithmeier , e.m . weig , and j.p . kotthaus , nature * 458 * , 1001 ( 2009 ) . landau and e.m . lifshitz , _ theory of elasticity _ , pergamon , oxford ( 1986 ) . p. werner and w. zwerger , europhys . lett.*65 * ( 2004 ) . g. anetsberger , o. arcizet , q.p . unterreithmeier , r. rivire , a. schliesser , e.m . weig , j.p . kotthaus , and t.j . kippenberg , nat . phys . * 5 * , 909 ( 2009 ) . m. pllinger , d. oshea , f. warken and a. rauschenbeutel , phys . lett . * 103 * , 053901 ( 2009 ) . k. h. lee , t. g. mcrae , g.i . harris , j. knittel , and w. p. bowen , phys . lett . * 104 * , 123604 ( 2010 ) . steele , a.k . httel , b. witkamp , m. poot , h.b . meerwaldt , l.p . kouwenhoven , and h.s.j . van der zant , science * 325 * , 1103 ( 2009 ) . b. lassagne , y. tarakanov , j. kinaret , d. garcia - sanchez , and a. bachtold , science * 325 * , 1107 ( 2009 ) . m. s. hofmann , j. t. glckert , j. noe , c. bourjau , r. dehmel , and a. hogele , nature nanotech . * 8 , * 502 ( 2013 ) . j. waissman , m. honig , s. pecker , a. benyamini , a. hamo , and s. ilani , nature nanotech . * 8 , * 569 ( 2013 ) . s. stapfner , l. ost , d. hunger , j. reichel , i. favero , and e. m. weig , appl . phys . lett . * 102 * , 151910 ( 2013 ) . httel , g.a . steele , b. witkamp , m. poot , l.p . kouwenhoven and h.s.j . van der zant , nano lett . * 9 * , 2547 ( 2009 ) . a. eichler , j. moser , j. chaste , m. zdrojek , i. wilson - rae , and a. bachtold , nature nanotech . * 6 * , 339 ( 2011 ) . b. kozinsky and n. marzari , phys . * 96 * , 166801 ( 2006 ) . t. j. kippenberg , s. m. spillane , and k. j. vahala , app . . lett . * 85 * , 6113 ( 2004 ) j. d. jackson , _ classical electrodynamics _ , wiley , new york ( 1998 ) i. wilson - rae , phys . rev . b * 77 * , 245418 ( 2008 ) . l. g. remus , m. p. blencowe , and y. tanaka , phys . rev . b * 80 * , 174103 ( 2009 ) . joh , j. kinder , l.h . herman , s .- y . ju , m.a . segal , j.n . johnson , g.k .- l . chan and j. park , nature nanotech . * 6 * , 51 ( 2011 ) . h. nyquist , phys . lett . * 32 * , 110 ( 1928 ) b. c. stipe , h. j. mamin , t. d. stowe , t. w. kenny , and d. rugar , phys . * 87 * , 096801 ( 2001 ) . p. rabl , s. j. kolkowitz , f. h. l. koppens , j. g. e. harris , p. zoller , and m. d. lukin , nat . phys . * 6 * , 602 ( 2010 ) . i. wilson - rae , c. galland , w. zwerger , and a. imamolu , new j. phys . * 14 * , 115003 ( 2012 ) i. wilson - rae , n. nooshi , w. zwerger , and t.j . kippenberg , phys . * 99 * , 093901 ( 2007 ) . f. marquardt , j.p . chen , a.a . clerk , and s.m . girvin , phys . lett . * 99 * , 093902 ( 2007 ) . c. w. gardiner and m. j. collett , phys . a * 31 * , 3761 ( 1985 )

one of the dynamic reconfiguration processes of the magnetosphere is called substorm and is a key process responsible for the energy transport and release in the terrestrial magnetosphere . it is believed that reconnection is responsible for the break of stability in the magnetospheric configuration and also for the transfer from magnetic to kinetic energy , e.g. @xcite . during this process the magnetic field forms a so - called x - line surrounded by the diffusion region , where acceleration of particles takes place . in the diffusion region the electric field accelerates high - energy ions and electrons away from the magnetic reconnection region and electrons at low - energy flow toward the reconnection site simultaneously @xcite . the energetic particles are released as bursty bulk flows and are widely studied by e.g. @xcite . the bursty bulk flows often contain structures with closed magnetic field lines plasmoids . such a large tailward - moving loop - like magnetic structure is generated as the result of x - type reconnection on the closed field lines in the near magnetotail @xcite . in two dimensions the typical observational signatures of a plasmoid have been defined by a bipolar magnetic deflection in the meridional direction . the examination of the three - dimensional topology and morphology of plasmoids in the earth s magnetotail showed rather complicated configurations compared to the original 2-dimensional hones scenario : helical flux ropes and travelling compression regions ( tcrs ) detected as moving magnetic imprints of a plasmoid in the lobes @xcite . a tailward moving plasmoid in the central plasma sheet temporarily compresses the magnetic flux in the tail lobes because the cross - sectional area of the lobe is reduced . in the area between the central plasma sheet and lobe a layer of particles moving parallel to the magnetic field arises from a rapid change of the flux tube volume outside the diffusion region @xcite . however , these field - aligned beams could also be of a different origin than reconnection , e.g. due to fermi acceleration @xcite . thus in the terrestrial magnetosphere two distinct groups of field - aligned beams can be derived : + ( 1 ) field - aligned ion beams accompanied with field - aligned electrons and associated with the x - line reconnection ; + ( 2 ) ion field - aligned beams accompanied by isotropic electrons accelerated by the quasi - steady dawn - dusk electric field . substorm - like ( or ) reconfiguration processes have been observed at jupiter by the particle and fields instruments onboard the galileo spacecraft @xcite . it is believed that the reconfiguration processes in the jovian magnetosphere are rather internally driven by ion mass loading from the moon io and fast planetary rotation , although we can not fully exclude the influence of the solar wind @xcite . this internal driving mechanism leads to a periodic release of plasma in the magnetotail with a repetition period of about 3 days . the thinning of the plasma sheet prior to the mass release reminds the growth phase of the terrestrial substorm @xcite . this plasma sheet reconfiguration due to internal energy loading leads to a formation of the x - line in the jovian magnetotail at approximately 80 @xmath0 @xcite . the scale size of downtail released plasmoids was estimated to be about 9 @xmath0 @xcite . reconnection processes at jupiter seem to play an important role in the reconfiguration process as reported by @xcite . reconnection conditions in the jovian magnetotail are satisfied just before the energetic particle release @xcite . statistical studies of bursty bulk flows and plasmoids were carried out by @xcite . also data from the new horizons spacecraft during the jupiter flyby showed evidence of periodic plasma injections possibly caused by reconnection @xcite . some field - aligned beams were discussed in the substorm - like context by @xcite . in this paper , we show for the first time observations of a tcr and plasmoids together with field - aligned beams and consider their features more detailed in the context of x - line reconnection using data from the galileo energetic particle detector ( epd ) @xcite and from the magnetometer ( mag ) @xcite . the paper is organized as follows : in section [ sec:1 ] we introduce shortly the instruments onboard galileo used to identify field - aligned beam features in the measurements . in section [ scene ] we present a typical jovian reconfiguration event with field - aligned beams and dynamic processes preceding and following the beam . in sections [ timing]-[event:4 ] we show examples of four field - aligned beams and their wide spectrum of characteristics . section [ discussion ] discusses results and provides a statistical analysis . section [ summary ] summarizes the observations . the epd was an instrument onboard the galileo spacecraft designed to measure the characteristics of the charged particle population such as energies , intensities , ion composition and angular distribution to determine , in particular , the configuration of the jovian magnetosphere @xcite . the advantages of the epd instrument compared to similar instruments on previous missions were the @xmath1 steradian angular coverage for jovian energetic particles and the extended coverage of particle energies . epd consisted of two double - headed detector telescopes : the low energy magnetospheric measurement system ( lemms ) and the composition measurement system ( cms ) . cms measured the ion fluxes with discriminating ion species using the time - of - flight technique . its energy ranges were 80 to 1250 kev ( protons ) , 27 kev / nuc to 1 mev / nuc ( helium ions ) , 12 to 562 kev / nuc ( oxygen ions ) , and 16 to 310 kev / nuc ( sulfur ions ) . on the other hand , lemms observed the electron flux in the energy range between 15 and 11000 kev and the species - integrated ion flux between 22 and 12400 kev . in this case electrons and ions were separated by a permanent magnet . although these telescopes have narrow field - of - view ( the full opening angles are 18@xmath2 and 15@xmath2 for cms and lemms , respectively ) , an almost full - sphere coverage is achieved by the motion of a turntable combined with the spacecraft spin ( the exception is a small solid angle ( @xmath3 0.1 sr ) along the spin axis which is blocked to avoid direct sun light in the detector ) . the unit sphere is divided into 16 ( see figure 6.2 , in @xcite ) or 6 sectors dependent on the energy channel . to study the flow direction we analyze 16-sector resolution data in this work . data are accumulated within sampling times of @xmath411.5 min ( for this study ) and allocated to the angular sectors . each sector is @xmath545@xmath2 wide . in the record time mode epd / lemms count rates are sampled every 1.25 seconds over up to 64 sectors . the description of lemms and cms energy channels used in our analysis one can find in table [ t2le ] . the first order anisotropies were calculated using technique described by @xcite , and references therein . the pitch angle distributions ( pad ) are presented in the detector frame detector pitch angles ( dpa ) . the flow is parallel to the magnetic field when dpa is close to 180@xmath6 and respectively anti - parallel when dpa is close to 0@xmath6 . particle pitch angles are determined with the magnetic field data obtained by the galileo magnetometer . we discuss pitch angles only when the magnetic field direction varies insignificantly during a sampling time , since otherwise the pitch angle determination is unreliable . the magnetic field observations have a time resolution of @xmath424 s , which is much higher than that of the particle measurements . this does not affect our results , however , as we compare the observed field - aligned beams in the particle data with features in the magnetic field and not the other way around . still , it is very likely that we miss many events with time scales less than 11 min in the epd data set . a typical substorm - like or reconfiguration event in the jovian magnetosphere is shown in figure [ 15 ] . the figure shows the time interval from day 1997 151 , 06:00 to day 154 , 02:00 . during this time galileo was located in the midnight sector of the jovian magnetotail ( 0116 local time ( lt ) ) at about a radial distance r = 100 @xmath7 the upper panel displays omnidirectional ion intensities in seven energy channels covering the energy range between 22 kev ( a1 channel ) and 3.2 mev ( a7 channel ) . the second panel from the top shows the radial ( red ) and azimuthal ( green ) component of the ion directional flow anisotropy as derived from measurements of 65 - 120 kev ions . the lower two panels present the magnitude and components of the measured magnetic field in the siii - system . in this system the radial magnetic field component is positive in the radially outward direction , the azimuthal component is positive in the direction of jupiter s rotation and the south - north ( meridional ) component is positive southward @xcite . first signatures of a mass - release process are observed at day 152 , 07:00 with an unusual ( for the lobe region ) increase of the ion intensities ( see vertical dashed line ( 1 ) in figure [ 15 ] ) . the corotational flow , typical for the jovian magnetosphere ( see the ion anisotropy denoted by the green line in the second panel from the top ) is disrupted by a strong tailward ( radially outward ) directed anisotropy spike or an field - aligned ion beam ( the red line ) in the lobe / psbl ( that the beam is field - aligned will be proven later ) . this is followed by the bursty flow ion anisotropy event or bursty bulk flow ( denoted by the vertical dashed line ( 2 ) ) which arrives with bipolar features in the south - north magnetic field component at day 152 , 10:00 when galileo approached the vicinity of the current sheet center . about 16 hours later another group of tailward directed bursts with similar characteristics arrives ( see vertical dashed lines ( 3 ) and ( 4 ) in figure [ 15 ] ) . then in the subsequent plasma sheet encounter radially inward directed bursts ( the negative red spikes ) are observed implying that the x - line moved tailward over the spacecraft . the observations of the flow reversals are very typical for the jovian magnetotail , see e.g. @xcite . it is interesting that just before the bursty flows the plasma sheet changed its displacement relative to the spacecraft from being more in the southern part of the plasma sheet to being in the northern part of the plasma sheet almost all of the presented time ( according to the radial magnetic field component behavior ) . such plasma sheet displacements were discussed by @xcite . they suggest that these displacements could be associated to a change in the solar wind conditions . more details on the reconfiguration or substorm - like events can be found in @xcite . this example is shown to present the global picture of the magnetotail dynamics before going into details . one example of a field - aligned beam during reconfiguration events is shown in figure [ 15 ] . it was recorded at 100 @xmath0 away from the planet in the magnetotail , on day 97 152 , 06:30 - 09:00 . the detailed particle and the magnetic field observations for this event are presented in figure [ 152 ] and the location relative to the plasma sheet configuration is sketched in figure [ scetch ] . according to the strength and direction of the magnetic field the spacecraft was located in the north lobe / or psbl . first the electrons arrive with energies from 15 to 304 kev ( see electron intensities in figure [ 152 ] , panel 1 ) at 0659 ut . these electrons were flowing parallel to the magnetic field in the lobe region ( see the increase in electron intensities close to 180@xmath6 in dpa distributions , panel 4 ) . after 0659 ut electrons moved bi - directionally parallel to the magnetic field ( see the increase in intensities close to 0@xmath6 and 180@xmath6 ) . between 0710 ut and 0750 ut the observed electrons changed their direction of motion from being anti - parallel 0710 - 0721 ut to mainly parallel 0721 - 0750 ut relatively to the magnetic field . at 0750 ut the electron distribution becomes isotropic . this field - aligned electron beam is associated with the reconfiguration event from the list published by @xcite . after the field - aligned beam was observed the plasma sheet center moved towards galileo ( see section [ scene ] , figure [ 15 ] ) . as it is expected from the reconnection scenario the bulk release of particles ( see vertical dashed line ( 2 ) in figure [ 15 ] ) at the alfvnic speed is observed at 1200 ut on day 152 , as studied by @xcite . from the comparison of the flow speed of the bursty bulk flows was compared with the alfvn speed . we can assume that during this time the reconnection process was ongoing and that the electron beam has possibly been mirrored at the next x - point as in a magnetic bottle bouncing between two mirror points on loop field lines ( i.e. multiple x - line formation is assumed , see also kasahara et al . , submitted to jgr ) or launched from another x - line . also the bi - directional streaming electrons are commonly observed in the terrestrial plasma sheet boundary layers @xcite . ions arrived at least 11 min later at 0710 ut . the third panel of figure [ 152 ] shows the first order directional anisotropy components of ions ( 65 - 120 kev ) in radial and azimuthal directions . during this event the ions moved clearly predominantly radially outward with an anti - corotational component . those ions moved also along the magnetic field direction as indicated in the pitch angle distributions ( panel 5 of figure [ 152 ] ) . ion fluxes exhibit an energy dispersion in the sense that the intensities of high energy ions increase first . assuming that the electrons have a single source ( e.g. released at the reconnection site during the onset phase ) , the distance from galileo to the source ( e.g. the reconnection line ) can be roughly estimated from the delay time between different species . we must point out that the time resolution of the epd data is 11 minutes and therefore the following calculations have large error bars . ions measured in the a1 channel ( 42 - 65 kev ) , mainly protons , with a geometric mean energy of 52 kev ( speed of @xmath4 3200 km s@xmath8 ) arrive @xmath4 22 min later than electrons . this gives a distance of 60 @xmath9 30 @xmath7 according to these calculations the distance from the source to galileo was between 30 @xmath0 and 90 @xmath0 . consequently a source would be located at @xmath4 40 @xmath9 30 @xmath0 from the planet . the source location is slightly closer to the planet than those derived by @xcite but still similar within the error bar . even if the delay calculations are rough , this event presents an energy dispersion of different particles : electrons arrive first , and then high energy ions followed later by lower energy ions . this implies a single source of these field - aligned beams . one can claim that the observed dispersion is related to gyroradius effect or gradient effects . we estimate the gyroperiod of the proton of @xmath49 s in this magnetic environment . we see that the accumulation time of 11 min is much longer than the proton gyroperiod . additionally , the spacecraft passes 0.002 @xmath0 during one accumulation time of the epd instrument ( the galileo speed is @xmath43.2@xmath1010@xmath11 @xmath12 ) . this distance is the same order of magnitude as the proton gyroradius of a field - aligned beam . therefore , the spacecraft could be considered basically stationary relative to the ion motion . the ion dispersion is also not related to @xmath13 drift , as they would lead to much longer time delays . another field - aligned beam was observed during the reconfiguration event on doy 153 at 0620 and is shown by vertical dashed line ( 3 ) in figure [ 15 ] . details are presented in figure [ 153 ] in the same format as for event 1 in figure [ 152 ] . during this beam significant increases by more than one order of magnitude of the ion intensities ( 42 kev to 1.7 mev ) and by one order of magnitude of the electron intensities ( 15 to 304 kev ) are observed . again , the first order flow anisotropy shows that the ions during this event were flowing radially outward in anti - corotational direction . location of this event relative to the plasma sheet configuration is sketched in figure [ scetch ] . the meridional magnetic field component @xmath14 becomes negative at 0625 ut ( northward directed ) suggesting a tailward flow in the plasma sheet @xcite in case of a neutral line formation . at the same time the field - aligned electrons followed by the field - aligned ions ( arrived later at 0636 ut ) are observed . the delays between electrons and ions are the same as in the example in section [ timing ] suggesting similar nature of the field - aligned beam . at @xmath40640 ut a signature of a flux rope or tcr is seen in the magnetic field , i.e. increase in the radial magnetic field component , bipolar signature in the meridional magnetic field component . additionally , the disappearance of the anti - phase signature between the azimuthal and radial magnetic field components as in usual quiet configuration ( @xmath15 reaching zero ) is observed . the flux rope - like magnetic field ( increased radial magnetic field component together with the bipolar signature in the meridional magnetic field component , see e.g. , @xcite ) is associated with ion field - aligned beam parallel to the magnetic field . the proton intensity increased by almost 3 orders of magnitude . after the flux - rope structure decaying oscillations in the meridional magnetic field component are observed . these oscillations could indicate small scale magnetic islands ( probably created by magnetic reconnection ) , or they are imprints of the post - plasmoid plasma sheet structure as shown in @xcite . we conclude that these field - aligned electron and ion beams are ( 1 ) associated with tcr mentioned as a reconnection event in the study of @xcite , ( 2 ) followed by the bursty bulk flow reversal ( see vertical line ( 4 ) in figure [ 15 ] ) , and ( 3 ) show the energy dispersion . all these features imply that the observed field - aligned electron and ion beams are associated with a reconnection process . another example of field - aligned particle beams is shown in figure [ 283 ] observed on day 1996 283 from 1100 to 1500 ut . during this time galileo was located in the pre - dawn sector of the magnetotail ( 03:19 lt ) at r = 113 @xmath7 location of this event relative to the plasma sheet configuration is sketched in figure [ scetch ] . these field - aligned beams preceded the bursty bulk flows in the reconfiguration event from the list by @xcite . the first order flow anisotropy shows that the ions during this event flow radially outward in anti - corotational direction . here again the dispersion is observed . at @xmath41151 ut , a bi - directional , field - aligned electron flow ( 29 to 42 kev ) arrival was observed ( seen in the dpa distribution ) . then at @xmath41214 ut the electron flux was mono - directional field - aligned and changed to bi - directional field - aligned again between @xmath41225 and @xmath41259 ut . simultaneously with this strong electron beam ( in our time resolution ) protons with energies from 220 to 540 kev arrived at @xmath41214 ut . at @xmath41225 ut the field - aligned ions with energies of 65 - 120 kev and protons ( 80 to 220 kev ) followed by oxygen ( 26 to 51 kev / nuc ) arrived at @xmath41248 ut and eventually sulfur ( 16 to 30 kev / nuc ) at @xmath41236 ut . assuming that the x - line as the source of all these flows , this scenario can be checked . as the arrival of the electrons is not clearly determined , we estimate the distance x to the x - line using the formula @xmath16 where @xmath17 is the sulfur speed and @xmath18 is the proton speed , @xmath19 is the time delay between the arrival of protons and sulfur @xmath423 min . we obtain the distance from galileo to the x - line equal to 52@xmath20 26 @xmath0 . this also agrees with the statistical x - line derived by @xcite . for the given location it is approximately at 70 @xmath0 . therefore , it is likely that particles were released from a single extended source and the time - of - flight dispersion theory works and the presented example fits the x - line reconnection scenario . it is not likely that gyroradius effects and gradient effects influence the results for this event , as the magnetic field during the field - aligned beam was stable and the spacecraft was approximately at the same position in the lobe region . we also note that this event was not considered as a reconnection event by @xcite . figure [ hr ] presents an event on doy 235 of 1997 showing the spacial structure of the electron beam . during this event galileo was located in the pre - dawn sector of the magnetotail ( 00:56 lt ) at r = 129.9 @xmath7 as the particle data were acquired with the full time resolution in this particular period , we see the covered pitch angle and the associated particle flux due to the instrument rotation . the magnetic field is plotted in siii coordinate system . galileo was located southward of the current sheet center during most of this period . anti - parallel field - aligned electron beam is observed in several channels ( 15 to 188 kev ) , from @xmath21 to @xmath22 ut ( see in dpa distributions as red dots close to 0@xmath6 ) . the beam almost disappeared during 1427 - 1429 ut , indicating that this is the intermittent and/or localized beam . this beam is associated with the bipolar signatures in the meridional magnetic field component at @xmath41420 ut . this event shows a mono - directional electron beam , consistent with an x - line planetward of the galileo spacecraft . the field - aligned ion beam is seen at energies 65 - 120 kev and 42 - 65 kev during 1443 - 1455 ut . during the neutral sheet approaching at around 1443 - 1455 ut , electron flux is enhanced by a factor of @xmath42 compare to the period at @xmath41420 - 1435 ut in the plasma sheet boundary layer , the particle distribution is isotropic . the fluxes in the plasma sheet are higher compared to those at the boundary layer . all these suggests a closed field line topology @xcite . an unambiguous evidence of the beam layer structure is seen . namely , the electron beam propagates at the outer edge of the thin current sheet and an isotropic distribution is observed around the current sheet center . this can not be clearly evidenced by low - resolution data due to the ambiguous pitch angle determination around the magnetic field reversal . the field - aligned beam is preceded by the plasmoids and associated with the magnetic field dipolarization . the meridional magnetic field component is already negative before ( unfortunately the particle data are not available in that resolution before @xmath41408 ut ) . therefore , there could be a formation of a large plasmoid or post plasmoid plasma sheet followed by smaller asymmetric plasmoid ( at @xmath41414 ut ) and dipolarization ( at @xmath41420 ut ) characterized by a strong and steep increase of the meridional magnetic field component , preceded by a much less negative dip , as in the terrestrial cases observed and simulated by e.g. @xcite . this bipolar change of the meridian magnetic field component followed by a dipolarization front can also be explained to be caused by the tailward movement of reconnection site , followed further by dipolarization of the magnetic field at planetward side of reconnection site . during the plasmoid passage , the neutral sheet crossing took place during between 1414 and 1416 ut and no electron flux enhancement was observed . this is different from the terrestrial case where the plasmoids centers are associated with increases in the electron flux @xcite . an approximate location of the spacecraft trajectory relative to the plasma sheet configuration , between 14:00 and 14:18 ut , is sketched in figure [ scetch ] . in the previous section we presented examples of the field - aligned electron and ion beams in energetic particle data measured onboard the galileo spacecraft in the jovian magnetotail . we investigated the galileo orbits g2 , e6 , g8 , c9 , c10 and e16 , searched for such field - aligned beams and found 30 events . for these events we required : + ( 1 ) observation during more than one accumulation time ( that excludes spikes due to instrumental effects ) ; + ( 2 ) the radial flow anisotropy dominates the corotational one and + ( 3 ) the ion pitch angle distributions show that the particle flow is predominantly parallel / anti - parallel to the magnetic field . the distribution of these events in the jovian magnetosphere along the galileo orbits is shown in figure [ map ] . it is possible that due to the low time resolution of the epd instrument many field - aligned beams were not detected . we listed the duration of the field - aligned beams in table [ events ] , which is on average 55.6 minutes . statistically most events have a duration between 20 and 40 minutes ( see the occurrence rate of field - aligned beams versus its duration in figure [ hist ] ) . the distribution of events versus duration time confirms that it is likely that many field - aligned beams have shorter duration than 20 minutes . all 30 field - aligned beams investigated in this study are associated with those 34 jovian reconfiguration events listed in @xcite . during 15 of them we observe the field - aligned beams . during 8 of these 15 events several field - aligned beams are seen . these field - aligned beams seem to be a constituent part of the jovian reconfiguration process . in figure [ map ] we show the location of the x - lines derived by @xcite and @xcite . the field - aligned beams are concentrated close to these lines . therefore , we believe that the field - aligned beams are signatures of reconnection , which occur during the reconfiguration in the jovian magnetotail . now we would like to estimate the mass and energy input of the field - aligned beams in the content of the jovian energy budget . the mass which carries a field - aligned beam can be estimated as follows : @xmath23 where @xmath24 is a typical number density in the jovian magnetotail @xcite . we assume that oxygen is the average ion contained in the jovian plasma sheet . the volume , @xmath25 of the field - aligned beam can be assessed as the length of the beam @xmath26 where @xmath27 is the field - aligned beam speed and @xmath28 is the duration , multiplied with the assumed vertical extent 1 @xmath0 and the azimuthal extent 25 @xmath0 . as average speed of the field - aligned beam we use the speed of the ions in the plasma sheet boundary layers , 800 km s@xmath8 derived by @xcite and as duration we take an average duration of the field - aligned beam derived above , @xmath460 min . therefore , the mass release during one field - aligned beam is in the order of 100 tons . this is lower compared to the mass of plasmoids derived by @xcite , 800 tons and 2500 tons , respectively . the kinetic energy of the field - aligned beam can be evaluated from the formula @xmath29 this will lead to @xmath30j . therefore , the power of the field - aligned beam will be in the order of 9 tw , for an averaged duration of the beam of @xmath460 min . this power will be enough to supply for instance jovian polar auroral emissions , see ( e.g. , * ? ? ? in this paper we investigated the events in the psbl or lobe regions . according to the x - type reconnection scenario by @xcite , we observe the field - aligned electron and ion beams in the psbl . in the paper of @xcite , two types of electron distribution are reported : ( 1 ) when the electrons are isotropic and ( 2 ) when electrons flow along the magnetic field lines . the first case is associated with the non - adiabatic acceleration by the quasi - steady dawn - dusk electric field and is not necessary associated with substorm - like activity . the beams of the second type are usually generated close to the x - line . we check the distribution of the particle flux in the three directions relative to the magnetic field using the dpa : anti - parallel , dpa@xmath3140@xmath32 parallel , dpa@xmath33140@xmath6 and perpendicular , 70@xmath34dpa@xmath31110@xmath6 . we would like to note that due to the plasma corotation in the jovian plasma sheet , many of the field - aligned flows will still have quite a substantial corotational component . this is reflected in the pitch angle distributions as an increased perpendicular component . 23 events with field - aligned ion beams are associated with anisotropic electrons , i.e. which flow along the magnetic field and 13 of them are associated with reconnection events studied by @xcite derived from the magnetometer data . the characteristics of the field - aligned ion beams are seen in table [ events ] . the propagation of the ions and electrons parallel or antiparallel to the magnetic field confirm the quadrupolar x - line configuration , see figure [ scetch ] . slightly more than half of the field - aligned beams are associated with reconnection events observed by @xcite . why not more ? the reasons could be ( 1 ) different arrival times of field - aligned beams and plasmoids ; ( 2 ) the field - aligned beams are mainly located in the lobe region where reconnection signatures might be smoothed . in 7 cases field - aligned ion beams are associated with the isotropic electron distribution , see table [ events ] . therefore , they might be accelerated not at the x - line . most of these events , namely 5 , has no association with the reconnection events from @xcite . this is expected . \(1 ) we presented two examples ( event 1 , event 3 ) of the counterstreaming electron beams in the plasma sheet boundary layers . according to @xcite the counterstreaming electron beams occur due to the bouncing between two mirror points on closed field lines in the plasma sheet . they could also occur due to the travelling of electrons between multiple x - lines . ( 2 ) most of the field - aligned ion events are associated with the field - aligned electron beams which according to @xcite are related to electron acceleration in the x - line vicinity . ( 3 ) the timing of the arrival of the field - aligned beams for the different species in many events ( e.g. event 1 , event 2 , event 3 ) is in accordance to the time - dispersion when the source is at one point / line @xcite . also @xcite observed the time dispersion of the energetic ions of different species using data from new horizons . ( 4 ) observations of all types of the plasma release from plasmoid or flux rope , its imprint in the psbl tcr and dipolarization are in accordance to the reconnection theory and all are associated with the electron and ion field - aligned beams ( event 2 , event 3 , event 4 ) . ( 5 ) we show evidence that the electron beam propagates in the psbl while isotropic distribution is observed around the current sheet center ( event 4 ) . ( 6 ) observations of the flow reversals in 53@xmath35 of the jovian reconfiguration events from @xcite imply x - line crossing , see an example in section [ scene ] , figure [ 15 ] . ( 7 ) all field - aligned beams are associated with the jovian reconfiguration events . based on all these arguments we believe that reconfiguration processes in the jovian magnetotail are associated with reconnection . the authors thank a. lagg for the provided epd software and useful discussions . 49 natexlab#1#1url # 1`#1`urlprefix , a. , taylor , m. g. g. t. , borg , a. l. , lavraud , b. , friedel , r. w. h. , escoubet , c. p. , laakso , h. , daly , p. , fazakerley , a. n. , may 2008 . multispacecraft observation of electron beam in reconnection region . j. geophys . 113 , 7 . , v. , kennel , c. f. , coroniti , f. v. , pellat , r. , kivelson , m. g. , walker , r. j. , russell , c. t. , baumjohann , w. , feldman , w. c. , gosling , j. t. , nov . statistical characteristics of bursty bulk flow events . j. geophys . 99 , 2125721280 . , v. , mcfadden , j. p. , larson , d. , carlson , c. w. , mende , s. b. , frey , h. , phan , t. , sibeck , d. g. , glassmeier , k .- h . , auster , u. , donovan , e. , mann , i. r. , rae , i. j. , russell , c. t. , runov , a. , zhou , x .- z . , kepko , l. , aug . tail reconnection triggering substorm onset . science 321 , 931935 . , f. , delamere , p. a. , may 2011 . flow of mass and energy in the magnetospheres of jupiter and saturn . journal of geophysical research ( space physics ) 116 , a05209 . , l .- j . , bhattacharjee , a. , puhl - quinn , p. a. , yang , h. , bessho , n. , imada , s. , mhlbachler , s. , daly , p. w. , lefebvre , b. , khotyaintsev , y. , vaivads , a. , fazakerley , a. , georgescu , e. , jan . observation of energetic electrons within magnetic islands . nature physics 4 , 1923 . , a. j. , 1983 . coordinate systems . in : dessler , a.j . ( editor ) , physics of the jovian magnetosphere , cambridge university press . , l. a. , paterson , w. r. , khurana , k. k. , jan . 2002 . observations of thermal plasmas in jupiter s magnetotail . j. geophys . 107 , a1 , 1003 . , y. s. , jian , l. k. , russell , c. t. , dec . growth phase of jovian substorms . 34 , l23106 . , y. s. , raeder , j. , angelopoulos , v. , gilson , m. l. , runov , a. , jan . 2011 . interaction of dipolarization fronts within multiple bursty bulk flows in global mhd simulations of a substorm on 27 february 2009 . j. geophys . 116 , a00i23 . , y. s. , russell , c. t. , khurana , k. k. , sep . 2010 . reconnection sites in jupiter s magnetotail and relation to jovian auroras . space sci . 58 , 14551469 . , e. e. , hoshino , m. , hirai , m. , mukai , t. , zelenyi , l. m. , mar . . `` geography '' of ion acceleration in the magnetotail : x - line versus current sheet effects . j. geophys . 114 , a03203 . , m. e. , haggerty , d. k. , mcnutt , r. l. , paranicas , c. p. , nov . energetic particle evidence for magnetic filaments in jupiter s magnetotail . 114 , 11201 . , e. w. , may 1979 . transient phenomena in the magnetotail and their relation to substorms . space science reviews 23 , 393410 . , a. , rme , h. , dandouras , i. , bosqued , j. m. , sergeev , v. , sauvaud , j. , jacquey , c. , lavraud , b. , louarn , p. , moreau , t. , vallat , c. , escoubet , c. p. , parks , g. k. , mccarthy , m. , mbius , e. , amata , e. , klecker , b. , korth , a. , lundin , r. , daly , p. , zong , q. , may 2004 . new properties of energy - dispersed ions in the plasma sheet boundary layer observed by cluster . j. geophys . 109 , a05215 . , m. g. , khurana , k. k. , means , j. d. , russell , c. t. , snare , r. c. , 1992 . the galileo magnetic field investigation . space science reviews 60 , 357383 . , e. a. , , woch , j. , krupp , n. , lagg , a. , daly , p. w. , korth , a. , 2008 . comparison of periodic substorms at jupiter and earth . j. geophys . 113 , a04212 . , e. a. , glassmeier , k .- h . , woch , j. , krupp , n. , lagg , a. , dougherty , m. k. , may 2007 . a possible intrinsic mechanism for the quasi - periodic dynamics of the jovian magnetosphere . j. geophys . 112 , a05203 . , e. a. , woch , j. , krupp , n. , lagg , a. , oct . mass release process in the jovian magnetosphere : statistics on particle burst parameters . j. geophys . 113 , a10202 . , e. a. , woch , j. , krupp , n. , lagg , a. , jun . 2009 . a summary of observational records on periodicities above the rotational period in the jovian magnetosphere . annales geophysicae 27 , 25652573 . , e. a. , woch , j. , krupp , n. , lagg , a. , khurana , k. k. , glassmeier , k .- h . , mar . mass release at jupiter : substorm - like processes in the jovian magnetotail . j. geophys . 110 , a03211 . , n. , lagg , a. , livi , s. , wilken , b. , woch , j. , roelof , e. c. , williams , d. j. , nov global flows of energetic ions in jupiter s equatorial plane : first - order approximation . j. geophys . 106 , 2601726032 . , n. , woch , j. , lagg , a. , wilken , b. , livi , s. , williams , d. j. , apr . 1998 . energetic particle bursts in the predawn jovian magnetotail . 25 , 12491252 . , a. , 1998 . energiereiche teilchen in der inneren jupitermagnetosphre : simulation und ergebnisse des epd - experimentes an bord der raumsonde galileo . thesis , university of innsbruck . , m. , ieroset , m. , phan , t. d. , lin , r. p. , fujimoto , m. , mar . field - aligned electrons at the lobe / plasma sheet boundary in the mid - to - distant magnetotail and their association with reconnection . 33 , l05101 . , d. j. , allegrini , f. , bagenal , f. , crary , f. , ebert , r. w. , elliott , h. , stern , a. , valek , p. , oct . diverse plasma populations and structures in jupiter s magnetotail . science 318 , 217220 . , r. l. , haggerty , d. k. , hill , m. e. , krimigis , s. m. , livi , s. , ho , g. c. , gurnee , r. s. , mauk , b. h. , mitchell , d. g. , roelof , e. c. , mccomas , d. j. , bagenal , f. , elliott , h. a. , brown , l. e. , kusterer , m. , vandegriff , j. , stern , s. a. , weaver , h. a. , spencer , j. r. , moore , j. m. , oct . 2007 . energetic particles in the jovian magnetotail . science 318 , 220222 . , t. , 2000 . plasma distribution functions in plasmoids : geotail observations . advances in space research 26 , 415424 . , t. , shinohara , i. , fujimoto , m. , hoshino , m. , saito , y. , machida , s. , mukai , t. , nov . 2001 . geotail observations of the hall current system : evidence of magnetic reconnection in the magnetotail . j. geophys . 106 , 2592925950 . , reconnection in the jovian magnetosphere . 10 , 451454 . , a. , grodent , d. , grard , j .- c . , bonfond , b. , jul . auroral signatures of flow bursts released during magnetotail reconnection at jupiter . journal of geophysical research ( space physics ) 115 , a07214 . , a. , angelopoulos , v. , sitnov , m. i. , sergeev , v. a. , bonnell , j. , mcfadden , j. p. , larson , d. , glassmeier , k. , auster , u. , jul . themis observations of an earthward - propagating dipolarization front . 36 , a014106 . , c. t. , khurana , k. k. , huddleston , d. e. , kivelson , m. g. , may 1998 . localized reconnection in the near jovian magnetotail . science 280 , 10611064 . , c. t. , khurana , k. k. , kivelson , m. g. , huddleston , d. e. , 2000 . substorms at jupiter : galileo observations of transient reconnection in the near tail . advances in space research 26 , 14991504 . , d. v. , sarris , e. t. , angelopoulos , v. , yamamoto , t. , kokubun , s. , oct . 1997 . spatial structure of the plasma sheet boundary layer at distances greater than 180 r e as derived from energetic particle measurements on geotail . annales geophysicae 15 , 12461256 . , k. , 2007 . physics of space plasma activity . cambridge . , m. i. , swisdak , m. , divin , a. v. , apr . 2009 . dipolarization fronts as a signature of transient reconnection in the magnetotail . j. geophys . 114 , a04202 . , j. a. , owen , c. j. , dunlop , m. w. , bor " alv , e. , moldwin , m. b. , sibeck , d. g. , tanskanen , e. , goldstein , m. l. , fazakerley , a. , balogh , a. , lucek , e. , richter , i. , reme , h. , bosqued , j. m. , dec . 2003 . cluster four spacecraft measurements of small traveling compression regions in the near - tail . , j. a. , owen , c. j. , kuznetsova , m. m. , hesse , m. , 1995 . isee 3 observations of plasmoids with flux rope magnetic topologies . 22 , 20612064 . , v. m. , frank , l. a. , ackerson , k. l. , paterson , w. r. , 1997 . geometry of the plasma sheet in the midnight - to - dawn sector of the jovian magnetosphere : plasma observations with the galileo spacecraft . 24 , 869872 . , i. i. , fritz , t. a. , zong , q. , sarris , e. t. , aug . 2006 . two distinct energetic electron populations of different origin in the earth s magnetotail : a cluster case study . annales geophysicae 24 , 19311948 . , m. f. , kivelson , m. g. , khurana , k. k. , joy , s. p. , walker , r. j. , jun . reconnection and flows in the jovian magnetotail as inferred from magnetometer observations . j. geophys . 115 , a06219 . , l. s. , fritz , t. a. , kivelson , m. g. , khurana , k. , krupp , n. , lagg , a. , may 2005 . jovian plasma sheet morphology : particle and field observations by the galileo spacecraft . space sci . 53 , 681692 . , a. p. , owen , c. j. , fazakerley , a. n. , forsyth , c. , dandouras , i. , mar . 2011 . average magnetotail electron and proton pitch angle distributions from cluster peace and cis observations . 380 , l06103 . , d. j. , mcentire , r. w. , jaskulek , s. , wilken , b. , 1992 . the galileo energetic particles detector . space science reviews 60 , 385412 . , j. , krupp , n. , khurana , k. k. , kivelson , m. g. , roux , a. , perraut , s. , louarn , p. , lagg , a. , williams , d. j. , livi , s. , wilken , b. , jul . 1999 . plasma sheet dynamics in the jovian magnetotail : signatures for substorm - like processes ? 26 , 21372140 . , j. , krupp , n. , lagg , a. , apr particle bursts in the jovian magnetosphere : evidence for a near - jupiter neutral line . 29 , 421 . , j. , krupp , n. , lagg , a. , wilken , b. , livi , s. , williams , d. j. , apr . 1998 . quasi - periodic modulations of the jovian magnetotail . 25 , 12531256 . , g. , may 1993 . observable implications of tearing - mode instability in jupiter s nightside magnetosphere . space sci . 41 , 357361 . , q .- g . , fritz , t. a. , pu , z. y. , fu , s. y. , baker , d. n. , zhang , h. , lui , a. t. , vogiatzis , i. , glassmeier , k .- h . , korth , a. , daly , p. w. , balogh , a. , reme , h. , sep . cluster observations of earthward flowing plasmoid in the tail . 31 , l18803 . lp1.5cmlp1.5cmp2cmp4.5 cm time interval , year , doy , ut&radial distance , @xmath0&lt&duration , min & association with the reconnection events from @xcite&characteristics + + + 1996 264 19151955&88&0158&40 & y & protons @xmath36@xmath37 , electrons assymetric + 1996 264 20552143&88&0158&48 & y & protons , electrons @xmath36 , not clear case + 1996 270 23218271 0020&103&0230&62 & n & protons , electrons @xmath38@xmath39 , not clear case + 1996 271 09451022&0230&103&37 & n&protons @xmath38 , electrons bi - directional + 1996 272 2335273 0011&106&0238&36 & y & protons , electrons @xmath38 + 1996 275 09451145&110&0250&120 & y & protons , electrons @xmath38 + 1996 275 19312045&110&0250&74 & y & protons , electrons @xmath38 + 1996 277 23182357&112&0300&39 & n & protons , electrons @xmath38 + 1996 278 01250220&112&0300&55 & n & protons , electrons @xmath36 + 1996 280 16021730&113&0309&73 & n & protons , electrons from @xmath36 to @xmath38 , current sheet crossing + 1996 283 12201315&112&0330&50 & n & protons @xmath36 , electrons bi - directional + 1996 283 20182117&112&0330&59 & y & protons , electrons @xmath36 + 1996 284 00290205&112&0330&91 & y & protons @xmath38 , electrons bi - directional + 1996 285 04150450&111&0328&35 & n & asymmetric @xmath38 protons and electrons + 1996 285 05250725&111&0328&120 & y & protons , electrons @xmath38 + 1996 285 14151510&111&0328&55 & y & protons , electrons @xmath36 + 1997 079 02070229&84&0255&22 & n & asymmetric @xmath38 protons and electrons , not clear case + + lp1.5cmlp1.5cmp2cmp4.5 cm time interval , year , doy , ut&radial distance , @xmath0&lt&duration , min & association with the reconnection events from @xcite&characteristics + + + 1997 152 07030740&100&0120&37 & n & protons @xmath36 , electrons bi - directional + 1997 153 06200700&100&0120&40 & y & protons , electrons @xmath36 + 1997 158 09000922&97.5&0143&22 & y & protons @xmath38 , electrons assymetric + 1997 161 12001300&94&0155&60 & y & protons @xmath38 , electrons assymetric + 1997 166 20252115&79.5&0230&50 & y&protons @xmath36 , electrons assymetric + 1997 166 21552227&79.5&0230&32 & n&asymmetric @xmath36 protons and electrons + + 1996 262 15251629&79&0140&64 & n & protons @xmath38 + 1996 271 02330345&103&0230&72 & y & protons @xmath36 + 1996 273 03500430&106&0238&35 & n & protons @xmath38 + 1996 275 17191841&110&0250&82 & y & protons @xmath38 + 1996 275 22252338&110&0250&73 & n & protons @xmath38 , not clear case + 1997 161 18321920&94&0155&48 & n & protons @xmath38 + 1997 168 07200742&75&0240&22 & n & protons @xmath36 + + ) ; ( second panel ) first order ion directional flow anisotropies in the radial ( positive is outward ) and corotational direction ; ( third panel ) the magnetic field components in siii coordinates and ( fourth panel ) its magnitude . the vertical dashed lines denote the field - aligned beams and the bursty bulk flows.,width=453 ]

among the extensive literature on long - range dependence , relatively few publications are devoted to cyclical long - memory processes or long - range dependent random fields . however , models with singularities at non - zero frequencies are of great importance in applications . for example , many time series show cyclical / seasonal evolutions . singularities at non - zero frequencies produces peaks in the spectral density whose locations define periods of the cycles . a survey of some recent asymptotic results for cyclical long - range dependent random processes and fields can be found in @xcite and @xcite . in image analysis popular isotropic spatial processes with singularities of the spectral density at non - zero frequencies are wave , @xmath0-bessel , and gegenbauer models . @xcite investigated probabilistic properties of spatial gegenbauer models . a realization of the gegenbauer random field on @xmath1 grid is shown in figure [ fig1 ] . [ fig1 ] this article studies minimum contrast estimators ( mces ) of parameters of the gegenbauer random fields . the mce methodology has been widely applied in different statistical frameworks ( see , for example , @xcite ; @xcite ) . one of the first works which used a mce methodology for the parameter estimation of spectral densities of stationary processes was the paper by @xcite . @xcite introduced a class of mces for random fields . @xcite derived consistency and asymptotic normality of a class of mces for stationary processes within the class of fractional riesz - bessel motion ( see @xcite ) . results based on the second and third - order cumulant spectra were given by @xcite . they also provided asymptotic properties of second and third - order sample spectral functionals . these properties are of independent interest , since they can be applied to study the limiting properties of nonparametric estimators of processes with short or long - range dependence . @xcite applied the minimum contrast parameter estimation to approximate the drift parameter of the ornstein - uhlenbeck process , when the corresponding stochastic differential equation is driven by the fractional brownian motion with a specific hurst index . a burgeoning literature on spatio - temporal estimation has emerged in recent decades ( see @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , among others ) . one of the most popular estimation tools applied was the maximum likelihood estimation method ( mle ) . @xcite addressed the problem of parameter estimation of fractionally integrated processes with seasonal components . in order to estimate the fractional parameters , they propose several log - periodogram regression estimators with different bandwidths selected around and/or between the seasonal frequencies . the same methodology was used by @xcite for fractionally differenced autoregressive - moving average processes in the stationary time series context . several contributions have also been made for mle of long memory spatial processes ( see , for example , @xcite ) . for two - dimensional spatial data the paper by @xcite introduced a spatial unilateral first - order autoregressive moving average ( arma ) model . to implement mle they provided a proper treatment to border cell values with a substantial effect in estimation of parameters . @xcite addressed the problem of the least - squares estimation of autoregressive fractionally integrated moving - average ( farima ) processes with long - memory . @xcite investigated asymptotic properties of least - squares estimators in regression models for two - dimensional random fields . maximization of the whittle likelihood has been also considered in the recent literature on the mce ( see for example , @xcite , @xcite , @xcite ) . @xcite gave a continuous version of the whittle contrast functional supplied with a specific weight function for the estimation of continuous - parameter stochastic processes , deriving the consistency and asymptotic normality of such estimators . @xcite demonstrated that the whittle maximum likelihood estimator is consistent and asymptotically normal for stationary seasonal autoregressive fractionally integrated moving - average ( sarfima ) processes . parameter estimation of stationary gegenbauer random processes was considered by numerous authors , see , for example , @xcite , @xcite , @xcite , @xcite , @xcite . @xcite used the generating function of the gegenbauer polynomials to develop long memory gegenbauer autoregressive moving - average ( garma ) models that generalize the farima process . garma models were estimated by applying the mle methodology . @xcite also applied this methodology with slight modifications based on the conditional sum of squares method . @xcite extended these results to the two - parameter context within the garma process class . @xcite introduced a @xmath2-factor extension of the garma model that allowed to associate the long - memory behavior with each one of the @xmath2 gegenbauer frequencies involved . in this article we restrict our consideration to the estimation of long - range dependence parameters . it is motivated in part by cyclic processes , for which pole locations are known . also in some applications the spectral density singularity location can be estimated in advance . various methods , including semiparametric , wavelet , and pseudo - maximum likelihood techniques , of the estimation of a singularity location were discussed by , for example , @xcite , @xcite , and @xcite . this paper introduces and studies the mce of parameters of spatial gegenbauer processes . specifically , analogous of continuous - space results by @xcite are formulated for random fields defined on integer grids . the consistency and asymptotic normality of the mce are obtained using a spatial discrete version of the ibragimov contrast function . the article develops a methodology to practically verify general theoretical assumptions for consistency and asymptotic normality of mces for specific models . the results provide a rigorous platform to conduct model selection and statistical inference . the outline of the article is the following . in section 2 , we start by introducing the main notations of the paper . some fundamental definitions and the main results of this article , theorem 1 and 2 , are given in 3 . section 4 consists of the proofs of the main results . section 5 presents simulation studies which support the theoretical findings . the appendix provides auxiliary materials that specify for our case the conditions that ensure the consistency and asymptotic normality of the mce based on the ibragimov contrast function formulated in @xcite . in what follows we use the symbol @xmath3 to denote constants which are not important for our discussion . moreover , the same symbol @xmath3 may be used for different constants appearing in the same proof . all calculations in the article were performed using the software _ r _ version 3.0.2 and _ maple 16 , _ maplesoft . this section introduces some of the main definitions of the gegenbauer random fields given in @xcite ( see also , @xcite , @xcite , and @xcite , for the temporal case ) . let @xmath4 @xmath5 be a random field defined on the grid lattice @xmath6 consider the fractional difference operator @xmath7 defined by @xmath8^{d},\end{aligned}\ ] ] where @xmath9 is the backward - shift operator , @xmath10 i.e. @xmath11 @xmath12 and @xmath13 assume that @xmath14 satisfies the following state equation @xmath15 where @xmath16 @xmath17 is given by equation ( [ fog ] ) , with @xmath18 @xmath17 denoting the backward - shift operator for each spatial coordinate , i.e. @xmath19 , and @xmath20 here , @xmath21 @xmath22 is a zero - mean white noise field with the common variance @xmath23=\sigma^{2}_{\varepsilon}.$ ] the random field @xmath14 is called a spatial gegenbauer white noise in @xcite . by equation ( [ eqgegenbauer ] ) the gegenbauer random field @xmath14 can be defined in terms of the inverse of the operator @xmath24 expanded in a gegenbauer polynomial series as follows @xmath25 where @xmath26 @xmath17 and @xmath27 is the gegenbauer polynomial given by @xmath28 } ( -1)^{k } \frac{(2u)^{n-2k}\gamma(d - k+n)}{k!(n-2k)!\gamma(d)}.\ ] ] the generating function for the gegenbauer polynomials is given by @xmath29 which explains the expansion of the inverse operator in ( [ gegenb ] ) . in general , a random field @xmath14 is called invertible if the white noise @xmath30 @xmath22 can be expressed as the convergent sum @xmath31 where @xmath32 the gegenbauer random field has the following property , see @xcite . [ proposition ] if @xmath33 and @xmath34 @xmath17 then @xmath14 is a stationary invertible long range dependent random field . the spectral density of a stationary gegenbauer random field is given by @xcite , and @xcite , for the one - parameter case , and @xcite , for the two - parameter case : @xmath35 where @xmath36 @xmath37 , and @xmath38 @xmath39 using the spectral density function ( [ spectral_density ] ) , one can compute the auto - covariance function of @xmath14 as follows : @xmath40^{\frac{1}{2}-2d_{1 } } \left [ p_{j_{i}-\frac{1}{2}}^{2d_{i}-\frac{1}{2 } } ( u_{i } ) + ( -1)^{j_{i } } p_{j_{i}-\frac{1}{2}}^{2d_{i}-\frac{1}{2 } } ( -u_{i } ) \right],\ ] ] where @xmath41 is the associated legendre function of the first kind , consult 8 in @xcite . from @xcite , @xcite and @xcite , the following asymptotic approximation of the autocovariance function can be obtained @xmath42 . \nonumber\end{aligned}\ ] ] the random field @xmath14 is long range dependent as its auto - covariance function satisfies the condition @xmath43 a detailed discussion on relations between local specifications of spectral functions and the tail behaviour of auto - covariance functions of long range dependent random fields can be found in @xcite . figure [ fig2 ] gives an example of the spectral density and the auto - covariance function of the gegenbauer random field for the values of the parameters @xmath44 @xmath45 @xmath46 and @xmath47 and @xmath48,width=226 ] and @xmath48,width=226 ] very detailed discussions about estimation of parameters of seasonal / cyclical long memory time series were given in @xcite,@xcite , @xcite , and references therein . it was shown that the case of spectral singularities outside the origin is much more difficult comparing to the situation of a spectral pole at 0 . in particular , estimators of parameters @xmath49 and @xmath50 have different rates of convergence and their joint distributions are still unknown under general conditions . @xcite pointed out that , in practice , parameter estimation of seasonal / cyclical long memory data is done in two steps . the first step consists in estimation of singularity locations , which does not represent a difficult issue . then , the obtained values of location parameters are used in estimators of the long memory parameters . it has been shown that this 2-steps method provides quasi - similar results to simultaneous procedures . similarly , the results of this paper may also be used in a hierarchical modeling framework . namely , locations of the singularities can be included in the first level of the hierarchical procedure . then , on the second step of the hierarchical procedure , the long - range dependence parameters can be estimated by the mce method , conditioning on the locations obtained on the first step . it also allows to construct a suitable weight function before applying the mce methodology . therefore , in this paper we only concentrate our attention on long - range dependence parameters . suppose that the conditions imposed in proposition [ proposition ] to ensure stationarity , invertibility and long - range dependence hold . assume the value of the parameter @xmath51 is known a priori or was estimated on step 1 . then @xmath52 is the vector of parameters to estimate of the gegenbauer random field defined by equation ( [ eqgegenbauer ] ) ( see also equation ( [ spectral_density ] ) for the corresponding spectral density ) . let @xmath4 @xmath53 be a part of a realization of the gegenbauer random field . let @xmath54 , @xmath55^{2}$ ] , be a nonnegative function . suppose the condition * a3 * in the appendix holds . we define @xmath56^{2 } } f(\boldsymbol{\lambda},\boldsymbol{\theta})w(\boldsymbol{\lambda } ) \,\ , d\boldsymbol{\lambda}\ ] ] and consider the factorization @xmath57 for all @xmath58 the function @xmath59 has the following property @xmath60^{2 } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) w(\boldsymbol{\lambda } ) \ , \ , d\boldsymbol{\lambda } = 1.\ ] ] let @xmath61 be a non - random real - valued function , usually referred as the _ contrast function _ , given by @xmath62^{2 } } f(\boldsymbol{\lambda},\boldsymbol{\theta_{0 } } ) w(\boldsymbol{\lambda } ) \log \frac{\psi(\boldsymbol{\lambda},\boldsymbol{\theta_{0 } } ) } { \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) } \,\ , d\boldsymbol{\lambda},\ ] ] and let the _ contrast field _ be @xmath63^{2 } } f(\boldsymbol{\lambda},\boldsymbol{\theta_{0 } } ) w(\boldsymbol{\lambda } ) \log \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \,\ , d\boldsymbol{\lambda},\ ] ] where @xmath64 is the true parameter value . in what follows , @xmath65 denotes the probability distribution with the density function @xmath66 the empirical version @xmath67 @xmath68 @xmath69 is defined by @xmath70^{2 } } i_{t}(\boldsymbol{\lambda } ) w(\boldsymbol{\lambda } ) \log\psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \,\ , d\boldsymbol{\lambda},\ ] ] where @xmath71 is the periodogram of the observations @xmath4 @xmath53 of the gegenbauer random field , that is , @xmath72 the mce is defined by the empirical contrast field @xmath73 and the contrast function @xmath61 being @xmath74 and having a unique minimum at @xmath75 in particular , we choose @xmath76^{2},\ ] ] where @xmath77 @xmath17 @xmath78 is a positive function with continuous second order derivatives on @xmath79^{2}.$ ] notice that by ( [ weightfunction ] ) the weight function @xmath54 is nonnegative and symmetric about @xmath80 due to the boundedness of @xmath54 the product @xmath81 is integrable . thus , the choice of @xmath54 fulfils condition * a3 * in the appendix . the developed methodology is readily adjustable to other classes of weight functions . theorems [ theorem_1 ] and [ theorem_2 ] give consistency and asymptotic normality results for the mce . [ theorem_1 ] let @xmath82 be a stationary gegenbauer random field which spectral density satisfies equation ( [ spectral_density ] ) . if @xmath73 is the empirical contrast field defined by equation ( [ equation_ut ] ) , then * @xmath83 satisfies the conditions * a1-a6 * in the appendix ; * the minimum contrast estimator @xmath84 is a consistent estimator of the parameter vector @xmath85 that is , there is a convergence in @xmath65 probability : @xmath86 * @xmath87 where the variance estimator @xmath88 is given by @xmath89 ^ 2 } i_{t}(\boldsymbol{\lambda})w(\boldsymbol{\lambda})\,\ , d\boldsymbol{\lambda}.\ ] ] notice that assumption * a4 * is not required to prove the last two statements in theorem [ theorem_1 ] , but it will be used in the proof of theorem [ theorem_2 ] . to formulate theorem [ theorem_2 ] we introduce the following notations . the unbiased estimator of the correlation function @xmath90 @xmath91 of the gegenbauer random field @xmath83 is @xmath92 note that all indices of the random field in the sum above are within the set @xmath93 where the observations are available . the unbiased periodogram is given by @xmath94 and the corresponding empirical _ contrast field _ is @xmath95 ^ 2 } i_{t}^{*}(\boldsymbol{\lambda } ) w(\boldsymbol{\lambda } ) \log \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \,\ , d\boldsymbol{\lambda}.\ ] ] we also define @xmath96 ^ 2 } i_{t}^{*}(\boldsymbol{\lambda})w(\boldsymbol{\lambda})\,\ , d\boldsymbol{\lambda}$ ] and the associated adjusted mce @xmath97 [ theorem_2 ] if @xmath82 is a stationary gegenbauer random field which spectral density satisfies equation ( [ spectral_density ] ) with @xmath98 then * @xmath83 satisfies the conditions * a1-a9 * in the appendix ; * the adjusted mce defined by ( [ equation_theta_estimado2 ] ) is asymptotically normal . that is , @xmath99 where the entries of the matrices @xmath100 and @xmath101 are @xmath102^{2 } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) w(\boldsymbol{\lambda } ) \frac{\partial^{2}}{\partial \theta_{i}\partial \theta_{j } } \log \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \,\ , d\boldsymbol{\lambda}=\sigma^{2}(\boldsymbol{\theta } ) \displaystyle\int_{[-\pi,\pi]^{2 } } w(\boldsymbol{\lambda } ) \nonumber \\ & & \times\left [ \frac{\partial^{2}}{\partial \theta_{i}\partial \theta_{j } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) - \frac{1}{\psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) } \frac{\partial}{\partial \theta_{i } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \frac{\partial}{\partial \theta_{j } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right ] \ , d\boldsymbol{\lambda } , \label{sij}\end{aligned}\ ] ] @xmath103^{2 } } f^{2}(\boldsymbol{\lambda},\boldsymbol{\theta } ) w^{2}(\boldsymbol{\lambda } ) \frac{\partial}{\partial \theta_{i } } \log \left(\psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right ) \frac{\partial}{\partial \theta_{j } } \log \left(\psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right ) \,\ , d\boldsymbol{\lambda } \nonumber \\ & = & 8\pi^{2 } \sigma^{4}(\boldsymbol{\theta } ) \displaystyle\int_{[-\pi,\pi]^{2 } } w^{2}(\boldsymbol{\lambda } ) \frac{\partial}{\partial \theta_{i } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \frac{\partial}{\partial \theta_{j } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \,\ , d\boldsymbol{\lambda}. \label{aij}\end{aligned}\ ] ] to avoid the edge effect @xcite employed the modified periodogram approach suggested by @xcite . we use their assumptions in theorem [ theorem_2 ] . note that some authors pointed few problems in using @xmath104 see @xcite , @xcite , and references therein . various other modifications , for example , typed variograms , smoothed variograms , kernel estimators , to reduce the edge effect have been proposed . it would be interesting to prove analogous of the results by @xcite and theorem [ theorem_2 ] for these modifications , too . however , it is beyond the scope of this paper . moreover , it remains as an open problem whether the edge - effect modification is essential for the asymptotic normality or not , see @xcite . to prove the theorems we will use the following differentiability lemma . [ * theorem 11.5)][lem1 ] let @xmath105 be a measurable space , @xmath106 and @xmath107 be an open set . suppose the function @xmath108 satisfies the following conditions : 1 . for all @xmath109 @xmath110 2 . for almost all @xmath111 the derivative @xmath112 exists for all @xmath113 3 . there is an integrable function @xmath114 such that @xmath115 for almost all @xmath116 then there exists @xmath117 [ lem2 ] the function @xmath118 is bounded and separated from zero on @xmath119 moreover , its first and second order derivatives are bounded on @xmath120 and can be computed by @xmath121^{2 } } w(\boldsymbol{\lambda } ) \frac{\partial}{\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \ , d\boldsymbol{\lambda}\nonumber\\ & = & -2\int_{[-\pi,\pi]^{2 } } \log\left|2\cos\lambda_{i}-2u_{i}\right|\,w(\boldsymbol{\lambda } ) f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \ , d\boldsymbol{\lambda},\end{aligned}\ ] ] @xmath122^{2 } } w(\boldsymbol{\lambda } ) \frac{\partial^{2}}{\partial \theta_{j}\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \ , d\boldsymbol{\lambda}= 4\int_{[-\pi,\pi]^{2 } } \log\left|2\cos\lambda_{i}-2u_{i}\right|\nonumber\\ & & \times \log\left|2\cos\lambda_{j}-2u_{j}\right|\,w(\boldsymbol{\lambda } ) f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \ , d\boldsymbol{\lambda},\label{d2theta } \end{aligned}\ ] ] where @xmath123 by the choice ( [ weightfunction ] ) of the weight function we obtain @xmath124^{2}\times \theta}f(\boldsymbol{\lambda } , \boldsymbol{\theta } ) w(\boldsymbol{\lambda } ) < + \infty\ ] ] and @xmath125^{2}}\sup_{\theta\in \theta}f(\boldsymbol{\lambda } , \boldsymbol{\theta } ) w(\boldsymbol{\lambda } ) d\boldsymbol{\lambda } \leq 4\pi^{2 } \sup_{[-\pi,\pi]^{2}\times \theta}f(\boldsymbol{\lambda } , \boldsymbol{\theta } ) w(\boldsymbol{\lambda } ) < + \infty.\ ] ] hence , @xmath118 is bounded . note that @xmath126^{2}\times \theta}\left\{\left|2\cos\lambda_{1}-2u_{1}\right|\right\}^{2d_{1 } } \ , \left\{\left|2\cos\lambda_{2}-2u_{2}\right|\right\}^{2d_{2}}<+\infty . $ ] also , by the choice of the weight function , there exists @xmath127 and a set @xmath128^{2}$ ] of non - zero lebesgue measure such that @xmath129 for all @xmath130 therefore , @xmath131^{2}}f(\boldsymbol{\lambda } , \boldsymbol{\theta } ) w(\boldsymbol{\lambda } ) d\boldsymbol{\lambda } \nonumber\\ & \geq \frac{\delta\ , \boldsymbol{\lambda } ( a_0)}{\sup_{[-\pi,\pi]^{2}\times \theta}\left\{\left|2\cos\lambda_{1}-2u_{1}\right|\right\}^{2d_{1 } } \ , \left\{\left|2\cos\lambda_{2}-2u_{2}\right|\right\}^{2d_{2}}}>0,\nonumber \end{aligned}\ ] ] which means that @xmath118 is separated from zero on @xmath119 now , to study @xmath132 we compute @xmath133 @xmath134 @xmath135^{-2d_{i } } \nonumber\\ & \quad\times\left[2(\cos\lambda_{j}-u_{j } ) \right]^{-2d_{j } } = -2\log\left|2\cos\lambda_{i}-2u_{i}\right| f(\boldsymbol{\lambda},\boldsymbol{\theta } ) , \label{deriv_f_d } \end{aligned}\ ] ] where @xmath136 and @xmath137 using ( [ weightfunction ] ) and ( [ deriv_f_d ] ) we conclude that @xmath138^{2}\times \theta}\left|w(\boldsymbol{\lambda } ) \frac{\partial}{\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right|<+\infty.\ ] ] thus , by ( [ eq_sigma_factorization ] ) and lemma [ lem1 ] there exists @xmath139^{2 } } w(\boldsymbol{\lambda } ) \frac{\partial}{\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \,\ , d\boldsymbol{\lambda}\ ] ] and @xmath140^{2}\times \theta}\left|w(\boldsymbol{\lambda } ) \frac{\partial}{\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right|<+\infty.\ ] ] it is not difficult to find @xmath141 by ( [ deriv_f_d ] ) , for @xmath142 @xmath143 the second derivatives of @xmath144 are given by @xmath145 @xmath146 it follows from ( [ eqsdds3 ] ) , ( [ eqsdds4 ] ) , and ( [ weightfunction ] ) that @xmath147^{2}\times \theta}\left|w(\boldsymbol{\lambda } ) \frac{\partial^{2}}{\partial \theta_{j}\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right|<+\infty,\quad i , j=1,2.\ ] ] finally , by ( [ dtheta ] ) and lemma [ lem1 ] there exists @xmath148^{2 } } w(\boldsymbol{\lambda } ) \frac{\partial^{2}}{\partial \theta_{j}\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \,\ , d\boldsymbol{\lambda}\ ] ] and @xmath149^{2}\times \theta}\left|w(\boldsymbol{\lambda } ) \frac{\partial^{2}}{\partial \theta_{j}\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right|<+\infty.\vspace{-3mm}\ ] ] @xmath150 _ of theorem _ [ theorem_1 ] . we will prove that the conditions * a1-a6 * in the appendix are satisfied . therefore , we will be able to apply theorem 3 by @xcite and obtain the statement of theorem [ theorem_1 ] . the condition * a1 * holds , since @xmath151 belongs to the parameter space @xmath152 which is an interior of the compact set @xmath153^{2}$ ] . it follows from representation ( [ spectral_density ] ) of the spectral density that @xmath154 for @xmath155 thus , the condition * a2 * is satisfied . the class of non - negative weight functions @xmath54 defined by ( [ weightfunction ] ) consists of symmetric functions . note that @xmath156 when @xmath157 thus , by ( [ weightfunction ] ) and representation ( [ spectral_density ] ) of the spectral density we get @xmath158^{2})$ ] for all @xmath85 to verify * a4 * , that is , to prove @xmath159^{2 } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) w(\boldsymbol{\lambda})\,\ , d\boldsymbol{\lambda } = \displaystyle\int_{[-\pi,\pi]^{2 } } \nabla_{\boldsymbol{\theta } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) w(\boldsymbol{\lambda})\,\ , d\boldsymbol{\lambda } = 0,\ ] ] we find @xmath160}_{s_{1}(\boldsymbol{\lambda},\boldsymbol{\theta } ) } - \underbrace{\frac{w(\boldsymbol{\lambda})}{\sigma^{4}(\boldsymbol{\theta})}\left[\frac{\partial}{\partial \theta_{i } } \sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta})}_{s_{2}(\boldsymbol{\lambda},\boldsymbol{\theta})}\ ] ] and apply lemma 1 . the same symbol @xmath3 is used for different nonessential constants appearing in the calculations below . by ( [ spectral_density ] ) and the choice of the weight function we obtain @xmath161^{2}\times \theta}\big|\log\left|2\cos\lambda_{i}-2u_{i}\right|\big|\,f(\boldsymbol{\lambda } , \boldsymbol{\theta } ) w(\boldsymbol{\lambda } ) < + \infty.\ ] ] therefore , by equation ( [ deriv_f_d ] ) and lemma [ lem2 ] : @xmath162^{2}).\ ] ] now , by ( [ fw1 ] ) , ( [ s2 ] ) and lemma [ lem2 ] we can estimate @xmath163 as @xmath164 f(\boldsymbol{\lambda},\boldsymbol{\theta } ) } { \sigma^{4}(\boldsymbol{\theta})}\right|\leq c\ , \frac{w(\boldsymbol{\lambda})f(\boldsymbol{\lambda},\boldsymbol{\theta } ) } { \min_{\theta}\sigma^{4}(\boldsymbol{\theta})}\le c.\ ] ] finally , * a4 * follows from ( [ s1up ] ) , ( [ s2up ] ) , and lemma [ lem1 ] with @xmath165 note that @xmath166^{2})\cap l_{2}([-\pi,\pi]^{2})= l_{2}([-\pi,\pi]^{2}).$ ] to verify the condition * a5 * for the weight function @xmath167 we have to show that @xmath168 @xmath169^{2}),$ ] for all @xmath170 by ( [ spectral_density ] ) and ( [ weightfunction ] ) the product @xmath171 @xmath172 is bounded for all @xmath173 except @xmath174 let @xmath175 then , for @xmath176 @xmath134 @xmath177^{2}\leq c\prod_{i=1}^{2}|{\lambda_i}\pm { \nu_i}|^{2a_i-4\tilde{d}_i}\log|\lambda_{i}\pm \nu_{i}|\in l_{1}([-\pi,\pi]^{2}).\ ] ] therefore , combining the above results we conclude that * a5 * holds . to verify the condition * a6 * we use the following function @xmath178 @xmath179 note that @xmath180 when @xmath157 thus , by the choice of @xmath181 and properties of @xmath182 the function @xmath183 is uniformly continuous on @xmath79^{2}\times \theta .$ ] also , it holds @xmath184^{2}).\ ] ] since the conditions * a1 * -*a6 * are satisfied theorem 1 follows from theorem 3 in @xcite.@xmath185 _ of theorem _ 2 . to prove the asymptotic normality of the mce in theorem [ theorem_2 ] we will show that the conditions * a7 * -*a9 * of the appendix hold . we begin by proving the condition * a7*. first , to verify the twice differentiability of the function @xmath186 on @xmath120 we formally compute the second - order derivatives of @xmath187 @xmath188 \sigma^{2}(\boldsymbol{\theta } ) - \left[\frac{\partial } { \partial \theta_{i } } \sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta})}{\sigma^{4}(\boldsymbol{\theta } ) } \right ] \nonumber\\ & = & \frac{1}{\sigma^{4}(\boldsymbol{\theta } ) } \left\ { \left[\frac{\partial^{2 } } { \partial \theta_{j}\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right ] \sigma^{2}(\boldsymbol{\theta } ) + \left[\frac{\partial } { \partial \theta_{j } } \sigma^{2}(\boldsymbol{\theta } ) \right ] \left[\frac{\partial } { \partial \theta}_{i } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right ] \right\ } \nonumber\\ & -&\frac{1}{\sigma^{4}(\boldsymbol{\theta } ) } \left\ { \left[\frac{\partial^{2 } } { \partial \theta_{j}\partial \theta_{i } } \sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta})- \left[\frac{\partial } { \partial \theta_{i } } \sigma^{2}(\boldsymbol{\theta } ) \right ] \left[\frac{\partial } { \partial \theta_{j } } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right ] \right\}\nonumber \\ & -&\frac{1}{\sigma^{8}(\boldsymbol{\theta } ) } \left\ { \frac{\partial } { \partial \theta_{j } } \sigma^{4}(\boldsymbol{\theta } ) \left ( \left[\frac{\partial}{\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right]\sigma^{2}(\boldsymbol{\theta } ) - \left[\frac{\partial } { \partial \theta_{i } } \sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right ) \right\}.\nonumber\end{aligned}\ ] ] note that in lemma [ lem2 ] we proved that the derivatives @xmath189 @xmath190 @xmath191 and @xmath192 exist . hence , by the above computations and lemma [ lem2 ] the function @xmath193 is twice differentiable on @xmath119 in addition , by estimates ( [ dsup ] ) , ( [ df2 ] ) , ( [ d2sigma ] ) , lemma [ lem2 ] , and the above representation for @xmath194 the product @xmath195 is bounded on @xmath79^{2}\times \theta.$ ] hence , for @xmath196 @xmath197 @xmath198^{2})\cap l_{2}([-\pi,\pi]^{2}).\ ] ] to prove part 2 of the condition * a7 * , we first note that by ( [ equation_spectraldensity ] ) and ( [ deriv_f_d ] ) : @xmath199 by lemma [ lem2 ] the second term is bounded . hence , it follows from ( [ fw1 ] ) and ( [ fw2 ] ) that the product @xmath200 is bounded on @xmath79^{2}\times \theta,$ ] that implies @xmath201^{2}),\quad k\geq 1,\ \ i=1,2 , \ \ \boldsymbol{\theta } \in \theta .\ ] ] to verify the condition * a8 * we first check the positive definiteness of the matrices @xmath202 and @xmath203 the entries of @xmath202 can be rewritten as @xmath204^{2}}\psi_w(\boldsymbol{\lambda},\boldsymbol{\theta } ) \frac{\partial^{2}}{\partial \theta_{i}\partial \theta_{j}}\log \psi_w(\boldsymbol{\lambda},\boldsymbol{\theta})d\boldsymbol{\lambda},\ ] ] where @xmath205 by ( [ spectral_density ] ) , ( [ weightfunction ] ) , and lemma [ lem2 ] the function @xmath206 is integrable , i.e. there is a constant @xmath3 such that @xmath207 is a density . hence , @xmath208 where @xmath209 is the fisher information matrix of the random vector @xmath210 with the density @xmath211^{2}}\psi_w(\boldsymbol{\lambda},\boldsymbol{\theta } ) d\boldsymbol{\lambda}.$ ] therefore , @xmath202 is non - negative definite . note that @xmath212 where @xmath213 the random variables @xmath214 and @xmath215 are not a.s . linearly related which implies positive definiteness of @xmath202 . the entries of @xmath216 can be rewritten as @xmath217^{2 } } w^{2}(\boldsymbol{\lambda } ) \frac{\partial}{\partial \theta_{i } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \frac{\partial}{\partial \theta_{j } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \ , d\boldsymbol{\lambda } = c\sigma^{4}(\boldsymbol{\theta})\,e\left(q_iq_j\right ) , \ ] ] where @xmath218 and the random vector @xmath219 has the density @xmath220^{2 } } w^{2}(\boldsymbol{\lambda } ) \ , d\boldsymbol{\lambda}}.$ ] as @xmath221 is a non - negative definite matrix , @xmath216 is non - negative definite too . moreover , it is positive definite , because the random variables @xmath222 and @xmath223 are not a.s . linearly related . now we compute elements of the matrix @xmath224 by ( [ sij ] ) @xmath225^{2}}w(\boldsymbol{\lambda } ) \left [ \frac{\partial^{2}}{\partial \theta_{i}\partial \theta_{j } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) - \frac{1}{\psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) } \frac{\partial}{\partial \theta_{i } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \frac{\partial}{\partial \theta_{j } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right]d\boldsymbol{\lambda}\nonumber\\ & = & \sigma^{2}(\boldsymbol{\theta } ) \int_{[-\pi,\pi]^{2}}\left(w(\boldsymbol{\lambda } ) \frac{\partial } { \partial \theta_{j } } \left [ \frac { \left[\frac{\partial } { \partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right ] \sigma^{2}(\boldsymbol{\theta } ) - \left[\frac{\partial } { \partial \theta_{i } } \sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta } ) } { \sigma^{4}(\boldsymbol{\theta } ) } \right]\right.\nonumber\\ & & - \frac{w(\boldsymbol{\lambda})}{f(\boldsymbol{\lambda},\boldsymbol{\theta})\sigma^{6}(\boldsymbol{\theta } ) } \left ( \left[\frac{\partial}{\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right ] \sigma^{2}(\boldsymbol{\theta } ) - \left[\frac{\partial}{\partial \theta_{i } } \sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right)\nonumber\\ & & \left . \times \left ( \left[\frac{\partial}{\partial \theta_{j } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right ] \sigma^{2}(\boldsymbol{\theta } ) - \left[\frac{\partial}{\partial \theta_{j } } \sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right)\right)d\boldsymbol{\lambda } \nonumber \end{aligned}\ ] ] @xmath226^{2}}\left(\frac{w(\boldsymbol{\lambda})}{\sigma^{2}(\boldsymbol{\theta } ) } \left ( \left[\frac{\partial^{2 } } { \partial \theta_{j}\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right ] \sigma^{2}(\boldsymbol{\theta } ) + \left[\frac{\partial } { \partial \theta_{j } } \sigma^{2}(\boldsymbol{\theta } ) \right ] \left[\frac{\partial } { \partial \theta}_{i } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right ] \right)\right . \nonumber\\ & & -\frac{w(\boldsymbol{\lambda})}{\sigma^{2}(\boldsymbol{\theta } ) } \left ( \left[\frac{\partial^{2 } } { \partial \theta_{j}\partial \theta_{i } } \sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta})- \left[\frac{\partial } { \partial \theta_{i } } \sigma^{2}(\boldsymbol{\theta } ) \right ] \left[\frac{\partial } { \partial \theta_{j } } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right ] \right)\nonumber \\ & & -2\frac{w(\boldsymbol{\lambda})}{\sigma^{6}(\boldsymbol{\theta } ) } \left ( \frac{\partial } { \partial \theta_{j } } \sigma^{2}(\boldsymbol{\theta } ) \left ( \left[\frac{\partial } { \partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right]\sigma^{2}(\boldsymbol{\theta } ) - \left[\frac{\partial } { \partial \theta_{i } } \sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right ) \right)\nonumber \\ & & - \frac{w(\boldsymbol{\lambda})}{\sigma^{4}(\boldsymbol{\theta } ) f(\boldsymbol{\lambda},\boldsymbol{\theta } ) } \left(\left[\frac{\partial}{\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right ] \sigma^{2}(\boldsymbol{\theta } ) - \left[\frac{\partial}{\partial \theta_{i}}\sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta})\right ) \nonumber\\ & & \times \left(\left[\frac{\partial}{\partial \theta_{j } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right ] \sigma^{2}(\boldsymbol{\theta } ) - \left[\frac{\partial}{\partial \theta_{j } } \sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta})\right)d\boldsymbol{\lambda}. \nonumber\end{aligned}\ ] ] by ( [ eq_sigma_factorization ] ) , ( [ dtheta ] ) , ( [ d2theta ] ) , and ( [ deriv_f_d ] ) we obtain @xmath227^{2}}\frac{w(\boldsymbol{\lambda } ) } { \sigma^{2}(\boldsymbol{\theta } ) } \left[\frac{\partial } { \partial \theta_{j } } \sigma^{2}(\boldsymbol{\theta } ) \right ] \left[\frac{\partial } { \partial \theta}_{i } f(\boldsymbol{\lambda},\boldsymbol{\theta})\right]d\boldsymbol{\lambda}\ ] ] @xmath228^{2}}\frac{w(\boldsymbol{\lambda})}{f(\boldsymbol{\lambda},\boldsymbol{\theta } ) } \left[\frac{\partial}{\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right ] \left[\frac{\partial}{\partial \theta_{j } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right]d\boldsymbol{\lambda}= \frac{3}{\sigma^{2}(\boldsymbol{\theta } ) } \left[\frac{\partial } { \partial \theta_{j } } \sigma^{2}(\boldsymbol{\theta } ) \right ] \left[\frac{\partial } { \partial \theta_{i } } \sigma^{2}(\boldsymbol{\theta } ) \right]\ ] ] @xmath229^{2 } } \log\left|2\cos\lambda_{i}-2u_{i}\right|\log\left|2\cos\lambda_{j}-2u_{j}\right|\ , w(\boldsymbol{\lambda } ) f(\boldsymbol{\lambda},\boldsymbol{\theta})d\boldsymbol{\lambda}.\ ] ] by ( [ aij ] ) the elements of the matrix @xmath216 are @xmath230^{2}}w^2(\boldsymbol{\lambda } ) \frac{\partial}{\partial \theta_{i } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \frac{\partial}{\partial \theta_{j } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \,d\boldsymbol{\lambda}\nonumber\\ & = & 8\pi^2 \int_{[-\pi,\pi]^{2}}w^2(\boldsymbol{\lambda } ) \left(\left[\frac{\partial}{\partial \theta_{i } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right ] \sigma^{2}(\boldsymbol{\theta } ) - \left[\frac{\partial}{\partial \theta_{i}}\sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta})\right ) \nonumber\\ & & \times \left(\left[\frac{\partial}{\partial \theta_{j } } f(\boldsymbol{\lambda},\boldsymbol{\theta } ) \right ] \sigma^{2}(\boldsymbol{\theta } ) - \left[\frac{\partial}{\partial \theta_{j } } \sigma^{2}(\boldsymbol{\theta } ) \right ] f(\boldsymbol{\lambda},\boldsymbol{\theta})\right)\,d\boldsymbol{\lambda}. \nonumber\end{aligned}\ ] ] hence , by ( [ deriv_f_d ] ) we get @xmath231 where @xmath232^{2}}\log\left|2\cos\lambda_{i}-2u_{i}\right|\log\left|2\cos\lambda_{j}-2u_{j}\right|\ , w^2(\boldsymbol{\lambda } ) f^2(\boldsymbol{\lambda},\boldsymbol{\theta})d\boldsymbol{\lambda},\vspace{-4mm}\ ] ] @xmath233\int_{[-\pi,\pi]^{2}}\log\left|2\cos\lambda_{i}-2u_{i}\right|\ , w^2(\boldsymbol{\lambda } ) f^2(\boldsymbol{\lambda},\boldsymbol{\theta})d\boldsymbol{\lambda } , \nonumber\\ s_3 & = & 8\pi^2\left[\frac{\partial}{\partial \theta_{j } } \sigma^{2}(\boldsymbol{\theta } ) \right]\left[\frac{\partial}{\partial \theta_{j } } \sigma^{2}(\boldsymbol{\theta } ) \right]\int_{[-\pi,\pi]^{2 } } w^2(\boldsymbol{\lambda } ) f^2(\boldsymbol{\lambda},\boldsymbol{\theta})d\boldsymbol{\lambda}.\nonumber\end{aligned}\ ] ] the proof of the condition * a9 * is based on the approach in @xcite . notice , that by ( [ weightfunction ] ) there exists a factorization @xmath234 of @xmath54 such that both @xmath235 and @xmath236 are bounded functions of @xmath237 for example , one can select the function @xmath238 to be equal a product of @xmath239 and a positive smooth function on @xmath79^{2}.$ ] let us denote @xmath240 then , @xmath241 ^ 2 } ( ei^{*}_{t}(\boldsymbol{\lambda})-f(\boldsymbol{\lambda},\boldsymbol{\theta}_0 ) ) w(\boldsymbol{\lambda } ) \frac{\partial}{\partial \theta_{i } } \log \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \ , d\boldsymbol{\lambda } = t\displaystyle\int_{[-\pi,\pi]^2 } ( ei^{*}_{t}(\boldsymbol{\lambda})w_1(\boldsymbol{\lambda})\ ] ] @xmath242 ^ 2 } ( ei^{*}_{t}(\boldsymbol{\lambda})w_1(\boldsymbol{\lambda})-e\tilde{i}^{*}_{t}(\boldsymbol{\lambda } ) ) \tilde{w}_2(\boldsymbol{\lambda},\boldsymbol{\theta})\ , d\boldsymbol{\lambda}\ ] ] @xmath243 ^ 2 } ( e\tilde{i}^{*}_{t}(\boldsymbol{\lambda})-\tilde{f}(\boldsymbol{\lambda},\boldsymbol{\theta}_0 ) ) \tilde{w}_2(\boldsymbol{\lambda},\boldsymbol{\theta})\ , d\boldsymbol{\lambda},\ ] ] where @xmath244 denotes the unbiased periodogram of the random field with the spectral density @xmath245 notice that @xmath246 is bounded . hence , by the first statement of proposition 2 in @xcite the last integral in ( [ integ ] ) is @xmath247 and the second term in ( [ integ ] ) vanishes when @xmath248 therefore , to prove the condition * a9 * , it is enough to show that the first term in ( [ integ ] ) vanishes too . let @xmath249 denote the auto - covariance function of the random field with the spectral density @xmath245 by multidimensional parseval s theorem , see @xcite , we get @xmath250 ^ 2 } ( ei^{*}_{t}(\boldsymbol{\lambda})w_1(\boldsymbol{\lambda})-e\tilde{i}^{*}_{t}(\boldsymbol{\lambda } ) ) \tilde{w}_2(\boldsymbol{\lambda},\boldsymbol{\theta})\ , d\boldsymbol{\lambda}\ ] ] @xmath251 ^ 2 } \bigl(w_1(\boldsymbol{\lambda } ) \sum_{t_{1}=1-t}^{t-1}\sum_{t_{2}=1-t}^{t-1 } e^{- i(\lambda_{1}t_{1}+\lambda_{2}t_{2 } ) } { \gamma}(\boldsymbol{t},\boldsymbol{\theta}_0)\ ] ] @xmath252 @xmath253 ^ 2 } \bigl(w_1(\boldsymbol{\lambda } ) \sum_{(t_{1},t_{2})\in \mathbb{z}^2 } e^{- i(\lambda_{1}t_{1}+\lambda_{2}t_{2 } ) } { \gamma}(\boldsymbol{t},\boldsymbol{\theta}_0)i_{[1-t , t-1]^2}(t_{1},t_{2})\]]@xmath254 ^ 2}(t_{1},t_{2})\bigr)\tilde{w}_2(\boldsymbol{\lambda},\boldsymbol{\theta})\ , d\boldsymbol{\lambda}\ ] ] @xmath255 ^ 2 } \bigl(w_1(\boldsymbol{\lambda})\displaystyle\int_{[-\pi,\pi]^2 } { f}(\boldsymbol{x},\boldsymbol{\theta}_0)\phi_{t-1}(\boldsymbol{\lambda}-\boldsymbol{x})\,d\boldsymbol{x}\ ] ] @xmath228 ^ 2 } \tilde{f}(\boldsymbol{x},\boldsymbol{\theta}_0)\phi_{t-1}(\boldsymbol{\lambda}-\boldsymbol{x})\,d\boldsymbol{x}\bigr)\tilde{w}_2(\boldsymbol{\lambda},\boldsymbol{\theta})\ , d\boldsymbol{\lambda}\ ] ] @xmath256 ^ 2 } \bigl(\displaystyle\int_{[-\pi,\pi]^2 } { f}(\boldsymbol{x}+\boldsymbol{\lambda},\boldsymbol{\theta}_0)\phi_{t-1}(\boldsymbol{x})(w_1(\boldsymbol{\lambda})-w_1(\boldsymbol{\lambda}+\boldsymbol{x}))\,d\boldsymbol{x}\bigr)\tilde{w}_2(\boldsymbol{\lambda},\boldsymbol{\theta})\ , d\boldsymbol{\lambda},\ ] ] where @xmath257 is the fejr kernel , @xmath258 ^ 2}(t_{1},t_{2})$ ] is the indicator function of the cube @xmath259 ^ 2.$ ] let @xmath260 ^ 2),$ ] i.e. @xmath261 then , one can refine the approach in theorems 2.1 and 2.2 by @xcite for the two - dimensional case with @xmath262 namely , let us split the inner integral in ( [ secint ] ) into two parts : the first integral is over the region @xmath263 and the second integral is over @xmath264\setminus a_{\alpha}.$ ] for @xmath265 , by the choice of @xmath238 and @xmath266\setminus a_\alpha}\phi_{t-1}(\boldsymbol{x})\,d\boldsymbol{x}=\left(\frac{1}{{\pi(t-1)}}\int_{t^{-\alpha}}^\pi\left(\frac{\sin ( ( t-1)x_1/2)}{\sin ( x_1/2)}\right)^2 dx_1\right)^2\ ] ] @xmath267 the both above integrals are bounded by @xmath268 where @xmath269 when @xmath248 it implies that first term in ( [ integ ] ) vanishes and completes the proof of the condition * a9*.@xmath185 in this section we present some numerical results to confirm the theoretical findings . figure [ fig3a ] demonstrates a series of box plots to characterize the sample distribution of mces of the parameters @xmath270 @xmath17 as a function of @xmath271 to compute it monte carlo simulations of the gegenbauer field with 100 replications for each @xmath272 were performed . for the parameters @xmath44 @xmath45 @xmath46 @xmath273 and @xmath274 realizations of @xmath275 were simulated using the truncated sum @xmath276 in ( [ gegenb ] ) . for example , a realization of the gegenbauer random field on a @xmath1 grid is shown in figure [ fig1 ] . we set the parameter values of the weight function in ( [ weightfunction ] ) to @xmath277 and @xmath278 the periodogram @xmath71 was computed and the minimizing argument @xmath279 of the functional @xmath73 was found numerically for each simulation . figure [ fig3a ] demonstrates that the sample distribution of @xmath279 converges to @xmath280 as @xmath281 increases . the plot of the sample probabilities @xmath282 in figure [ fig3b ] also confirms convergence in probability of @xmath279 to @xmath283 ] ] for each generated realization we also computed the value of @xmath88 using @xmath284 analogously to figures [ fig3a ] , [ fig3b ] , plots in figures [ fig4a ] , [ fig4b ] support convergence in probability of @xmath88 to @xmath285 when @xmath281 increases . notice that by ( [ eq_sigma_factorization ] ) we get @xmath286 for the selected parameters . the larger values of @xmath287 in figure [ fig4b ] comparing to figure [ fig3b ] are due to the difference in the scales for the parameters ( small values measured in decimals ) and variances ( large values measured in tens ) . ] ] to verify the result of theorem [ theorem_2 ] we used sample values @xmath288 which minimized the functional @xmath289 for each simulation . to avoid possible negative values the modified periodogram @xmath290 was truncated at zero by the r program . bearing in mind the edge effect and modified periodogram s correction , figures [ fig5a ] , [ fig5b ] demonstrate that the results are close to the expected ones even for the relatively small @xmath291 the normal q - q plot of each component of @xmath292 in figures [ fig5b ] matches with the theoretical normal distribution . to test the bivariate normality hypothesis about @xmath292 we used the shapiro - wilk , energy , and kurtosis tests of multivariate normality from the r packages mvnormtest , energy , and ics . in all the tests , p - values ( @xmath293 @xmath294 and @xmath295 ) confirmed that @xmath296 asymptotically follows a bivariate normal distribution . simulations for other values of the parameters were run , with similar results . ] ] hence , we conclude that the mces are consistent estimators and the distributions of @xmath296 converge to the bivariate normal law . note that the simulation studies not only comply with the obtained results for @xmath98 but also indicate that the theoretical results may be extended to all possible values of @xmath297 in @xmath298 the estimation methodology based on the unbiased periodogram was introduced in @xcite , see also @xcite . recently , the paper by @xcite and the references therein provided a detailed discussion on the topic . it studied mainly the difficulties arising in the application of the methodology in high dimensions . in particular , they investigated problems arising in relation to non - uniformly increasing domain asymptotics associated with different expansion rates of the studied domain in each spatial direction . in this paper , we considered the case @xmath299 and restricted out attention to the case of uniformly increasing domain asymptotics . the case of non - uniformly increasing domain asymptotics is left for future investigations . an extended version of the derived results can be obtained for more general formulations of the unbiased periodogram . in particular , different growing rates can be allowed for each spatial dimension in the definition of the sampling area . for example , one can consider the following generalized version of the two - dimensional unbiased periodogram ( see , for example , @xcite ) @xmath300 where the functions @xmath301 @xmath302 satisfy some suitable conditions ( for example , @xmath303 @xmath304 and @xmath305 @xmath306 for @xmath17 and sufficiently large @xmath281 ) . an important area for future explorations is to extend the results of @xcite and simultaneously estimate the locations of singularities and long - range dependence parameters using the mce methodology . a feasible way to approach this problem would be relaxing the @xmath307-integrability assumptions in conditions * a5-a8 . * it also would be interesting to extend the methodology by @xcite to prove the condition * a9 * for all @xmath297 in @xmath298 note that our simulation results show that the proposed minimum contrast estimation methodology works in the case of uniformly increasing domain asymptotics . * supplementary materials * the codes used for simulations in this paper are available from the site ` https://googledrive.com/host/ ` ` 0b7uxm8o_bnbxdg9znu9mdhfoquu / mce%20gegenbauer%20fields/ ` the authors are grateful for the referee s careful reading of the paper and many detailed comments and suggestions , which helped to improve the paper . also , this work has been supported in part by projects mtm2012 - 32674 of the dgi , mec , and p09-fqm-5052 of the andalousian cice , spain . a.olenko was partially supported by the 2013 ltu research grant . the conditions for consistency and asymptotic normality of the mce for parameters of stationary fractional riesz - bessel type random fields given in @xcite are specified below for random fields on @xmath6 let @xmath308 be a real - valued measurable stationary gaussian random field with zero mean and a spectral density @xmath309 where @xmath310^{2},$ ] @xmath69 and @xmath120 is a compact set . assume that @xmath311 where @xmath151 is the true value of the parameter vector @xmath85 the derivatives @xmath318 exist and it is legitimate to differentiate under the integral sign in equation ( [ equation_psi ] ) , i.e. @xmath159^{2 } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) w(\boldsymbol{\lambda})\,\ , d\boldsymbol{\lambda } = \displaystyle\int_{[-\pi,\pi]^{2 } } \nabla_{\boldsymbol{\theta } } \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) w(\boldsymbol{\lambda})\,\ , d\boldsymbol{\lambda } = 0 . \nonumber\ ] ] the spectral density @xmath309 the weight function @xmath331 and the function @xmath332 are such that for all @xmath333 and @xmath334 @xmath241 ^ 2 } ( ei^{*}_{t}(\boldsymbol{\lambda})-f(\boldsymbol{\lambda},\boldsymbol{\theta}_0 ) ) w(\boldsymbol{\lambda } ) \frac{\partial}{\partial \theta_{i } } \log \psi(\boldsymbol{\lambda},\boldsymbol{\theta } ) \,\ , d\boldsymbol{\lambda } \longrightarrow 0 , \quad\text{as } \quad t\longrightarrow \infty.\ ] ] ferrara l , gugan d ( 2001 ) comparison of parameter estimation methods in cyclical long memory time series . in : junis c , moody j , timmermann a ( eds . ) development in forecast combination and portfolio choice , chapter 8 , wiley , new york

in - network function computation is one of the fundamental paradigms that increases the efficiency of sensor networks vs - a - vs conventional data networks . sensor nodes , in addition to sensing and communication capabilities , are often equipped with basic computational capabilities . depending on the task for which they are deployed , a sensor network can be viewed as a distributed platform for collecting and computing a specific function of the sensor data . for example , a sensor network for environment monitoring may only be concerned with keeping track of the average temperature and humidity in a region . similarly ` alarm ' networks , such as those for detecting forest fires , require only the maximum temperature . the baseline approach for performing such tasks is to aggregate all the data at a central node and then perform offline computations ; the premise of in - network computation is that distributed computation schemes can provide sizable improvement in the performance of the network . however , from the perspective of designing network algorithms , in - network function computation poses a greater challenge than data networks as the freedom to combine and compress packets , as long as the desired information is preserved , destroys the flow conservation laws central to data networks . the network has a lot more flexibility , so much so as to make quantifying its performance much more challenging @xcite . our focus in this paper is to develop a queue - based framework for such systems , and use it to design and analyze network algorithms . by network algorithms , we refer to cross layer algorithms that jointly perform the following tasks : 1 . _ aggregating _ the data at nodes via in - network computation , 2 . _ routing _ packets between nodes , and 3 . _ scheduling _ links between nodes for packet transmission . cross - layer algorithms for data networks , although very successful in both theory and increasingly in real - system implementation , are concerned only with the scheduling and routing aspects . hence , there is a need for a new framework and new algorithms for in - network function computation in sensor networks . keeping in mind the lessons learnt from the success of data networks , our aim is to design network algorithms that are _ dynamic _ ( i.e. , the algorithm should not be designed assuming static network parameters , but rather , use the network state to adaptively learn the network parameters ) , _ robust _ ( i.e. , the algorithm adapts to temporal changes in traffic and network topology ) , _ capable of dealing with a large class of functions _ ( i.e. , if the function being computed by the network changes , then one should only need to make minor changes to the scheduling and routing algorithms ) , and _ generalizable to all network topologies_. due to the wide range of potential applications , there are many existing models for such networks , and many different perspectives from which they are analyzed . some representative works in this regard are as follows : * the pioneering work of giridhar and kumar @xcite considers the function computation problem from the point of view of the capacity scaling framework of gupta and kumar @xcite . in particular , they quantify scaling bounds for certain classes of functions ( symmetric , divisible , type - sensitive , and type - threshold ) under the protocol model of wireless communications and for collocated graphs and random geometric graphs . * other papers consider the function computation problem from the point of view of information theory @xcite and communication complexity @xcite , characterizing various metrics ( refresh rate , number of messages , etc . ) for different functions in terms of certain properties of the graph , the function to be computed , and the underlying data correlation . all the above works take an idealized ` bottom - up ' approach to determine the fundamental limits of the problem , and hence are not directly suitable for designing practical network algorithms . * similar in spirit to the above papers , another approach is to study function computation from the perspective of source coding@xcite . these works characterize bounds and show the existence of coding schemes for noiseless , wireline networks . as with the previous algorithms , these policies tend to be idealized , using more complex coding - based schemes instead of simple routing and aggregation ( we will later show that such simple strategies suffice for optimal in - network computation of a number of functions of interest ) ; further , these papers do not have explicitly defined policies but rather existence results for such policies . * in contrast to the ` bottom - up ' approach of all the above works , krishnamachari et al . @xcite adopt a more ` top - down ' approach whereby they formulate network models that abstract out some of the complexity while allowing quantification of performance gains ( in their case , energy and delay ) . their models do not , however , achieve the optimal throughput and also do not allow for the design of dynamic network algorithms . * an alternate model of sensor networks is to assume that nodes are capable of in - network _ compression _ , wherein only the compression ( and not merging ) of flows is permitted . for example , baek et al . @xcite consider routing algorithms for power savings in hierarchical sensor networks . similarly , sharma et al . @xcite design energy - efficient queue - based algorithms under the assumption that the only operation allowed in addition to routing and scheduling is compression of packets at the source node . the queue - based model for data networks has proved to be an essential tool in designing provably - efficient algorithms for such systems . this model has provided a common framework for understanding various aspects of data network performance such as throughput @xcite , delay @xcite , flow utility maximization @xcite , network utility maximization @xcite , distributed algorithms @xcite , among others ( for an overview , refer to @xcite ) . in addition , these algorithms have been implemented in real systems @xcite , including in sensor networks @xcite , with good results . however , these algorithms are designed for data networks , and can not exploit any potential benefit from in - network computation . more recently , this framework has been extended to fork - and - join networks with fixed routing@xcite , and resource allocation in processing networks@xcite . using fixed routing in a network usually leads to suboptimal operations as the routes may not be designed to optimize the network performance ; in general , even choosing the single best fixed route can perform arbitrarily worse than with dynamic routing ( see example in section [ sec : refrate ] ) . further , static routing is not robust to temporal changes in the network . however , introducing dynamic routing with in - network computation destroys the flow conservation equations that exist in data networks and networks with fixed flows , as the flow out of a node depends both on inflow as well as ( dynamic ) packet aggregation at that node . thus , there is a need to come up with a new queue - based framework and algorithms for efficient function computation in sensor networks , and our paper is a step in this direction . our main contributions in this paper are as follows : * we identify a class of functions , the fully - multiplexible or fmux functions , for which we provide a tight characterization of the maximum refresh rate with in - network computation , i.e. , the maximum rate at which the sensors can generate data such that the computation can be performed in a stable manner . more formally , we show that for these functions , if the refresh rate exceeds a certain graph parameter ( the stochastic min - min - cut , which we define formally in section [ sec : refrate ] ) , then the system is transient under _ any _ algorithm , whereas for any rate lower than this parameter , we construct a policy that can stabilize the system . * leveraging the results of massouli et al . @xcite on _ broadcasting _ , we obtain a wireline routing algorithm for _ aggregation _ via in - network computation of fmux functions in directed acyclic graphs . our approach is based on the observation that broadcasting and aggregation are in some sense , duals , of each other . more technically , the duality occurs between isolation of packets in aggregation ( i.e. , a packet does not have neighboring packets to aggregate with ) and that of multiple receptions of the same packet ( from different neighbors ) in broadcasting . by suitably modifying the approach in @xcite , we are able to develop an in - network aggregation algorithm for which routing is completely decentralized , and simple random packet forwarding and aggregation suffices for throughput - optimality . * for general wireless networks we develop dynamic algorithms based on a centralized allocation of routes ( dynamic flow splitting ) and maxweight - type scheduling . in particular , we show that loading rounds on trees in a greedy manner ( whereby an incoming round is loaded on the least weighted aggregation tree ) , coupled with an appropriate scheduling rule , is throughput - optimal for computing fmux functions . the analysis of this algorithm is unique in that in addition to an appropriate lyapunov function , it requires the construction of appropriate tree - packings of the network graph in order to show the throughput - optimality of this routing scheme . notation : throughout the paper , we use calligraphic fonts ( @xmath1 , etc . ) to denote sets and the corresponding capital letter ( @xmath2 , etc . ) to denote their cardinality . we interchangeably use @xmath3 or @xmath4 for adding elements to sets , and @xmath5 for deleting elements from sets , and sometimes for brevity of exposition , use the element @xmath6 to denote the singleton set @xmath7 when the meaning is clear from the context ( in particular , for a set @xmath8 and element @xmath6 , @xmath9 ) . we also use the shorthand notation @xmath10\triangleq\{1,2,\ldots , n\}$ ] . in this section we describe the system model we study in the rest of the paper . at a high level , the system consists of a network of @xmath11 nodes , one of which is the data aggregator and the rest are sensors . sensor nodes are capable of three tasks : sensing the environment , transmitting to and receiving data from other nodes , and performing computations on the data . the sensors are assumed to _ sense the environment in a synchronous manner _ , and the overall purpose of the system is to compute a specific function of the synchronously generated sensor data and forward it to the aggregator . further , the function computation is assumed to be done in a repeated manner , and the metric used to quantify the efficacy of an algorithm is the maximum synchronous rate at which the sensors can generate data such that the required function of the data can be forwarded to the aggregator in a stable manner . this rate is henceforth referred to as the _ maximum refresh rate _ of the network . before we describe the queueing framework for function computation , we first outline the general communication model that we consider in this work . this model is the same as that considered for studying data networks @xcite . in the next section , we will outline the modifications we make in order to capture the in - network computation aspect of a sensor network . * communication graph : * we model the topology of the sensor network as a directed graph @xmath12 , where @xmath13 is a set of @xmath11 nodes , and @xmath14 is a set of @xmath15 directed links which determine the connectivity between nodes . there is a special node , @xmath16 , referred to as the _ aggregator _ , and the rest of the nodes in @xmath13 are sensor nodes . directed link @xmath17 represents that there exists a communication channel from node @xmath6 to node @xmath18 ( in wireline this corresponds to a physical channel , while in wireless it represents the fact that the nodes are in radio range ) . * transmission model : * following the convention in literature @xcite , we consider a continuous time system in case of wireline systems , whereas in the case of wireless networks , we assume that time is slotted , and state all rates in bits per slot . in wireline networks , we define a vector of link rates @xmath19 ; one bit is assumed to traverse a link @xmath20 with a random transit time with distribution _ _ exponential__@xmath21 . the transit times are independent across links and across packets crossing the same link . for wireless networks , we make the following assumptions / definitions @xcite : * we assume that the channels between nodes are constant ( but can extend to time varying channels with added notation , see @xcite ) . the wireless nature of the network is reflected in the interference constraints . * for transmission schedule @xmath22 , @xmath23 denotes the link - rate vector of transmission rates over the links under the chosen schedule . * @xmath24 is the set of valid schedules that obey the interference constraints ( henceforth referred to as independent sets ) . @xmath25 is said to be _ admissible _ if the link - rates can be achieved simultaneously in a time slot . @xmath26 is the set of all admissible @xmath25 and is assumed to be time invariant as stated above . further , we assume that @xmath27 . * @xmath28 is said to be _ obtainable _ if @xmath29 , the convex hull of @xmath30 . an obtainable link - rate vector can be achieved by time sharing over admissible link - rate vectors . * from the definition of the convex hull , we have that for every obtainable rate vector @xmath29 , there exists a probability measure @xmath31 over @xmath32 such that @xmath33 . the vectors @xmath34 are called static service split ( or sss ) rules@xcite , and represent time sharing fractions between different independent sets in order to achieve the rate @xmath28 . up to this point , the system is identical to one considered for data networks . to highlight the unique features of a physical sensor network performing function computation ( and how they affect the modeling of such a system ) , we consider the following example . in the process , we also indicate the gains achievable via in - network computation versus data download and processing at the aggregator . [ cols= " < " , ] finally we present the complete dynamic algorithm for fmux computation . the algorithm separates the routing and scheduling components as follows : when a round of packets arrives in the network , we first ` load ' all packets of the round on an aggregation tree ( thereby fixing the routing ) ; next , in each time slot , scheduling is done according to a modified maxweight policy . the routing is performed using a _ greedy tree - loading policy _ , wherein all incoming rounds in a time slot are loaded on the tree with smallest sum useful - queue , i.e. , least number of useful packets . formally , we have : [ algo : treechoosealgo ] calculate @xmath35\right)$ ] for all @xmath36 . + find the minimum loaded tree @xmath37 $ ] as : @xmath38=\arg\min_{\tau\in\mathcal{t } } w_{\tau}[t].\ ] ] assign all incoming rounds to aggregation tree @xmath37 $ ] . the scheduling algorithm is similar to the maxweight policy@xcite , in that it picks a maximum independent set with weights given by the product of the rate and the maximum queue across an edge . formally we have the following algorithm : [ algo : compbpalgo1 ] place packets arriving on tree @xmath39 at node @xmath40 in @xmath41 $ ] for non - leaf nodes , and @xmath42 $ ] for leaf nodes . + calculate @xmath43=\max_{\tau\in\mathcal{t}:(i , j)\in\tau}q_i^{\tau , u}[t]$ ] . also define @xmath44 $ ] as the tree which maximizes @xmath43 $ ] . + compute schedule @xmath45 $ ] as : @xmath46=\arg\max_{\mathbf{c}\in\gamma}\sum_{(i , j)\in\mathcal{l}}p_{ij}[t]c_{ij}[t].\ ] ] + consider link @xmath47 . if @xmath48>0 $ ] , then transmit the first @xmath49 , q_i^{\tau^*(i , j)[t],u}[t])$ ] packets from @xmath50,nu}[t]$ ] . for the sake of completeness , we note that in all the above algorithms , tie - breaking rules as well as the service discipline ( i.e. , among a set of multiple packets suitable for transmission , which one gets priority ) are assumed to be random ; this is done for the sake of convenience , and we note that there are many possible tie - breaking rules and service disciplines which would suffice . now we can state and prove the throughput optimality of this algorithm . [ thm : treesss ] the dynamic queue based policy consisting of greedy tree loading ( algorithm [ algo : treechoosealgo ] ) and maxweight scheduling ( algorithm [ algo : compbpalgo1 ] ) stabilizes the system for any refresh rate @xmath51 that is less than the maximum refresh rate @xmath52 . before proceeding further , we point out a particular novel aspect of the proof of this theorem . similar to previous papers @xcite , we use a quadratic lyapunov function for showing stability ; however our technique for bounding the lyapunov drift is quite different from those used for point - to - point data . the difficulty arises from the fact that although edmonds theorem guarantees the existence of an optimal tree - packing for the network , the trees in this optimal packing are unknown to the algorithm ; consequently it is unclear whether routing over all trees could lead to instability via packet accumulation on trees not involved in the optimal packing . we circumvent this by showing the existence of some intermediate tree packings between the optimal and the desired refresh rates , which allow uniform bounding of the lyapunov drift . we now present the complete proof . we define a candidate lyapunov function @xmath53 $ ] as @xmath54=\sum_{i\in\mathcal{n}}\sum_{\tau\in\mathcal{t}}q_i^{\tau , u}[t]^2,\end{aligned}\ ] ] with corresponding lyapunov drift given by @xmath55=\mathbb{e}[v[t+1]-v[t]|\mathbf{q}[t]].\ ] ] similar to before , we have that @xmath53\geq 0 $ ] for all states of the system , and that @xmath56<\infty$ ] . we now need to show that given @xmath57 , there exists @xmath58 such that if @xmath59>q_{\max}$ ] for some @xmath40 , then @xmath56<-\delta$ ] . now we have @xmath55=\sum_{i\in\mathcal{n}}\sum_{\tau\in\mathcal{t}}\mathbb{e}\left[(\delta q_i^{\tau , u}[t])^2 + 2q_i^{\tau , u}[t]\delta q_i^{\tau , u}[t]|\mathbf{q}[t]\right],\ ] ] and defining @xmath60 $ ] to be arrival of useful packets on tree @xmath39 to node @xmath40 , we have @xmath61=a_i^{\tau , u}[t]-d_{(i , p^{\tau}(i))}^{\tau}[t],\ ] ] and thus @xmath62)^2\leq m_a + ( lc_{\max})^2 $ ] ( due to external arrivals plus inter - node transmissions ) . let @xmath63 . then we have @xmath55\leq m_2 + 2\sum_{i\in\mathcal{n}}\sum_{\tau\in\mathcal{t}}q_i^{\tau , u}[t]\mathbb{e}\left[a_i^{\tau , u}[t]-d_{(i , p^{\tau}(i))}^{\tau}[t]|\mathbf{q}[t]\right].\ ] ] from the definition of @xmath52 , we know that there exists an optimal rate point @xmath64 and the corresponding optimal sss rule @xmath65 that maximizes the min - mincut . consider now a refresh rate @xmath51 less than the @xmath52 , such that @xmath66 . note that the algorithm can potentially split the incoming flow @xmath51 over _ every spanning tree _ of the network , in order to dynamically arrive at the optimal packing . to uniformly bound the lyapunov drift , we first need to construct two tree packings : an ` achievable ' packing @xmath67 such that @xmath68 which serves as a proxy for the flow - splitting , and a ` near - optimal ' packing @xmath69 such that @xmath70 and further which has the property that @xmath71 _ uniformly over all spanning trees _ ( for some @xmath72 which we define below ) . we do so as follows . assume that there exists @xmath73 such that if any edge @xmath20 is scheduled alone ( i.e. @xmath74 ) , then @xmath75 ( this is simply a formal definition of existence of a link ) . we can now perturb the optimal sss rule to get a new rate point @xmath76 with the following two properties : 1 . every edge @xmath20 has capacity @xmath77 . 2 . the min - mincut of the network at the rate point @xmath78 is @xmath79 . this helps ensure that the ` near - optimal ' tree packing @xmath80 can have some mass on each edge of the graph . to construct the perturbed sss rule @xmath81 , consider the optimal sss rule @xmath82 . we define @xmath83 ( i.e. , the set of independent sets that have some mass under @xmath65 ) and @xmath84 ( which is @xmath85 as the cardinality of @xmath86 is finite ) . now we reduce each @xmath87 by @xmath88 . this reduces the min - mincut by at most @xmath89 . to see this , note that the capacity of any edge @xmath90 reduces from @xmath91 to @xmath92 where : @xmath93 further , the maximum number of edges across the min - mincut is bounded by @xmath15 . thus the min - mincut of the network at the rate point @xmath78 is @xmath79 . next , suppose @xmath94 is the set of edges with zero flow under @xmath65 . we now complete the definition of the perturbed sss rule @xmath81 ( using the fact that singleton edges are valid independent sets ) as follows : @xmath95 to see that this is a valid sss rule , note that @xmath96 , which is the weight we have distributed equally over all links in @xmath94 . the rate point under this sss rule is henceforth denoted as @xmath97 . then for edges in @xmath94 we have @xmath98 . now since there are only @xmath15 edges , each with positive capacity @xmath92 , therefore there exists some @xmath99 such that every edge @xmath20 has capacity @xmath100 under sss rule @xmath81 . finally , applying edmonds theorem ( theorem [ thm : edmonds ] ) on the network under @xmath81 , we get a packing @xmath101 such that we have @xmath102 before proceeding further , we need the following definitions : * @xmath103 . * @xmath104 . * @xmath105 . * @xmath106 . note that @xmath107 as @xmath108 and the packing is not tight on the finite set @xmath109 . similarly , @xmath110 . finally we can construct the tree packings ( on the network under sss rule @xmath81 ) that we need to bound the lyapunov drift : 1 . the ` achievable ' tree packing , @xmath111 is defined as : @xmath112 then clearly @xmath113 is a packing ( as we are only removing mass from a valid packing ) and further : @xmath114 2 . the ` near - optimal ' tree packing , @xmath115 is defined as : @xmath116 first we need to show that this is a valid tree packing . to see this , note that the maximum load added on any edge is bounded by @xmath117 ( since in the worst case , all the trees in @xmath118 can contain some edge ) . for any edge in @xmath109 , this is less than the slack ( @xmath119 by definition ) that was already present . for an edge in @xmath120 , we know at least one tree in @xmath121 contained it ( as every edge in the graph has positive capacity under the sss rule @xmath81 ) , and hence we subtract a load of at least @xmath122 , which is again greater than the amount of load we add . thus @xmath115 is a valid packing . + further we have that @xmath123 . + in addition , defining @xmath124 we get that @xmath125 . thus we have constructed the two tree packings we need . we now return to bounding the lyapunov drift . from above , we have @xmath55\leq m_2 + 2\sum_{i\in\mathcal{n}}\sum_{\tau\in\mathcal{t}}q_i^{\tau , u}[t]\mathbb{e}\left[a_i^{\tau , u}[t]-d_{(i , p^{\tau}(i))}^{\tau}[t]|\mathbf{q}[t]\right].\ ] ] now , let @xmath126 $ ] be the rate for packets on aggregation tree @xmath39 on link @xmath127 allocated by the policy in time slot @xmath128 ( thus @xmath129=c_{(i , p^{\tau}(i))}[t]$ ] ) . then we have @xmath130\mathbb{e}&\left[d_{(i , p^{\tau}(i))}^{\tau}[t]|\mathbf{q}[t]\right]\\ & = \sum_{i\in\mathcal{n}}\sum_{\tau\in\mathcal{t}}q_i^{\tau , u}[t]\mathbb{e}\left[c_{ip^{\tau}(i)}^{\tau}[t]-\max\{c_{ip^{\tau}(i)}^{\tau}[t]-q^{\tau , u}_i[t],0\}|\mathbf{q}[t]\right]\\ & \geq\sum_{i\in\mathcal{n}}\sum_{\tau\in\mathcal{t}}q_i^{\tau , u}[t]\mathbb{e}\left[c_{ip^{\tau}(i)}^{\tau}[t]-\max\{c_{\max}-q^{\tau , u}_i[t],0\}|\mathbf{q}[t]\right]\\ & \geq\sum_{i\in\mathcal{n}}\sum_{\tau\in\mathcal{t}}q_i^{\tau , u}[t]\mathbb{e}\left[c_{(i , p^{\tau}(i))}^{\tau}[t]\right]-nl|\mathcal{t}|c_{\max}^2.\end{aligned}\ ] ] further , from the definition of the policy , we know that @xmath130\mathbb{e}&\left[c_{(i , p^{\tau}(i))}^{\tau}[t]|\mathbf{q}[t]\right]\\ & = \mathbb{e}\left[\max_{\mathbf{c}\in\gamma}\sum_{(i , j)\in\mathcal{l}}\max_{\tau\in\mathcal{t}:(i , j)\in\tau}\left\{q_i^{\tau , u}[t]\right\}c_{ij}[t]|\mathbf{q}[t]\right]\\ & \geq\max_{\mathbf{c}\in\gamma}\sum_{(i , j)\in\mathcal{l}}\max_{\tau\in\mathcal{t}:(i , j)\in\tau}\left\{q_i^{\tau , u}[t]\right\}\mathbb{e}\left[c_{ij}[t]|\mathbf{q}[t]\right]\\ & \geq\max_{\mathbf{c}\in\mathcal{ch}(\gamma)}\sum_{(i , j)\in\mathcal{l}}\max_{\tau\in\mathcal{t}:(i , j)\in\tau}\left\{q_i^{\tau , u}[t]\right\}\mathbb{e}\left[c_{ij}[t]|\mathbf{q}[t]\right]\\ & \geq\max_{\mathbf{c}\in\mathcal{ch}(\gamma)}\sum_{(i , j)\in\mathcal{l}}\sum_{\tau\in\mathcal{t}:(i , j)\in\tau}q_i^{\tau , u}[t]\mathbb{e}\left[c_{ij}^{\tau}[t]|\mathbf{q}[t]\right]\\ & \mbox{(where $ c_{ij}^{\tau}[t]$ is any tree - packing of a given $ \mathbf{c}\in\mathcal{ch}(\gamma)$)}\\ & \geq\sum_{(i , j)\in\mathcal{l}}\sum_{\tau\in\mathcal{t}:(i , j)\in\tau}q_i^{\tau , u}[t]\hat{c}_{ij}^{\tau},\end{aligned}\ ] ] where for any edge @xmath47 , @xmath131 represents any valid split of @xmath132 between trees lying on that edge , i.e. , @xmath133 . in particular , therefore , we can use the tree packing @xmath115 to get @xmath134\mathbb{e}\left[c_{(i , p^{\tau}(i))}^{\tau}[t]|\mathbf{q}[t]\right ] \geq\sum_{i\in\mathcal{n}}\sum_{\tau\in\mathcal{t}}q_i^{\tau , u}[t]\hat{\lambda}^{\tau}.\end{aligned}\ ] ] combining inequalities , and defining @xmath135 we get @xmath136\leq m_3 + 2\sum_{i\in\mathcal{n}}\sum_{\tau\in\mathcal{t}}q_i^{\tau , u}[t]\left(\mathbb{e}\left[a_i^{\tau}[t]|\mathbf{q}[t]\right]-\hat{\lambda}^{\tau}\right).\end{aligned}\ ] ] finally , define @xmath137 $ ] to be the rate of rounds arriving on tree @xmath39 . then from the greedy round - tree assignment algorithm , and using the fact that each round results in exactly one useful packet at each node , we get _ i_q_i^,u[t ] & + & = _ i_q_i^,u[t ] + & = + & _ \{a^[t]}_i_q_i^,u[t ] + & _ i_q_i^,u[t]_. thus we get @xmath136&\leq m_3 - 2\sum_{i\in\mathcal{n}}\sum_{\tau\in\mathcal{t}}q_i^{\tau , u}[t](\hat{\lambda}_{\tau}-\lambda_{\tau}')\\ & \leq m_3 - 2\epsilon_4\sum_{i\in\mathcal{n}}\sum_{\tau\in\mathcal{t}}q_i^{\tau , u}[t].\end{aligned}\ ] ] in order to have @xmath56<-\delta$ ] if @xmath138\geq q_i^{\tau , u}[t]>q_{\max}$ ] for some @xmath139 , we can choose @xmath140 . thus @xmath53 $ ] is a valid lyapunov function and by foster s theorem , our policy is stabilizing for any @xmath141 . we have presented a queueing - based framework for in - network function computation . we have used this framework to gain insights into designing dynamic and distributed algorithms for in - network function computation and to quantify the performance gains over data - download . we have focused on a class of functions , the fmux functions , which exhibit maximum compression on aggregation , and for which we have used the parity and max functions as representative examples . for such functions we have developed scheduling and routing algorithms under different settings . for wireline networks , we have extended the random routing scheme of massouli et al . @xcite for aggregation . for wireless networks , we have provided a fixed - routing via dynamic flow splitting along with maxweight - like scheduling , which is shown to be throughput - optimal . the wireless algorithm , as presented , requires routing on all aggregation trees in order to achieve throughput optimality ; this may not be practical in many networks due to the potentially exponential number of trees . however , as we showed in the example with the complete graph , one can obtain optimal tree packings with a much smaller number of trees ( of the order of @xmath15 ) and one direction of future work is to show how such trees can be selected using simple rules in different networks . generalizing these algorithms to deal with a broader class of functions , as well as studying the performance of the algorithms with respect to other metrics ( delay , energy consumption , among others ) are other topics for future work . this work was supported in part by afosr under grant fa9550 - 09 - 1 - 0317 , nsf grants cns-0519535/0519401 ( collaborative grant ) and cns-0964391 . 10 a. giridhar and p. r. kumar , `` computing and communicating functions over sensor networks , '' _ ieee journal on selected areas in communications _ , 23 , no . 4 , pp . 755764 , 2005 . p. gupta and p. r. kumar , `` the capacity of wireless networks , '' _ ieee transactions on information theory _ , vol . 46 , no . 2 , pp . 388404 , 2000 . a. orlitsky and j. r. roche , `` coding for computing , '' _ ieee transactions on information theory _ , vol . 47 , no . 3 , pp . 903917 , 2001 . n. ma , p. ishwar , and p. gupta , `` information - theoretic bounds for multiround function computation in collocated networks , '' _ corr _ , vol . abs/0901.2356 , 2009 . h. kowshik and p. r. kumar , `` optimal computation of symmetric boolean functions in tree networks , '' in _ ieee international symposium on information theory isit 2010 _ , ieee , july 2010 . r. appuswamy , m. franceschetti , n. karamchandani , and k. zeger , `` network coding for computing part i : cut - set bounds , '' _ corr _ , vol . abs/0912.2820 , 2009 . t. moscibroda , `` the worst - case capacity of wireless sensor networks , '' in _ internat . workshop on inform . processing in sensor networks ( ipsn ) _ , pp . 110 , 2007 . b. krishnamachari , d. estrin , and s. b. wicker , `` the impact of data aggregation in wireless sensor networks , '' in _ proc . the 26th ieee internat . conf . distributed computing systems ( icdcs ) _ , pp . 575578 , 2002 . n. karamchandani , l. keller , c. fragouli , and m. franceschetti , `` function computation via subspace coding , '' in _ proc . information theory ( isit ) _ , ieee , july 2010 . s. j. baek , g. de veciana , and x. su , `` minimizing energy consumption in large - scale sensor networks through distributed data compression and hierarchical aggregation , '' _ ieee journal on selected areas in communications _ , 22 , no . 6 , pp . 11301140 , 2004 . a. sharma , l. golubchik , r. govindan , and m. j. neely , `` dynamic data compression in multi - hop wireless networks , '' in _ proc . acm sigmetrics / performance conf . _ , pp . 145156 , 2009 . l. tassiulas and a. ephremides , `` stability properties of constrained queueing systems and scheduling policies for maximum throughput in multihop radio networks , '' _ ieee transactions on automatic control _ , vol . 4 , pp . 19361948 , december 1992 . m. andrews , k. kumaran , k. ramanan , a. l. stolyar , r. vijayakumar , and p. whiting , `` scheduling in a queueing system with asynchronously varying service rates , '' _ probability in engineering and informational sciences _ , vol . 14 , pp . 191217 , 2004 . l. bui , r. srikant , and a. l. stolyar , `` novel architectures and algorithms for delay reduction in back - pressure scheduling and routing , '' in _ proc . ieee infocom . _ , pp . 29362940 , 2009 . m. j. neely , `` delay analysis for maximal scheduling with flow control in wireless networks with bursty traffic , '' _ ieee / acm trans . _ , vol . 17 , no . 4 , pp . 11461159 , 2009 . x. lin , n. shroff , and r. srikant , `` a tutorial on cross - layer optimization in wireless networks , '' _ ieee journal on selected areas in communications _ , pp . 14521463 , august 2006 . l. bui , r. srikant , and a. l. stolyar , `` optimal resource allocation for multicast flows in multihop wireless networks , '' _ proc . acm sigmetrics / performance evaluation review conf . _ , vol . 35 , no . 3 , p. 43 , 2007 . l. lin , n. b. shroff , and r. srikant , `` energy - aware routing in sensor networks : a large system approach , '' _ ad hoc netw . _ , vol . 5 , no . 6 , pp . 818831 , 2007 . l. ying , s. shakkottai , and a. reddy , `` on combining shortest - path and back - pressure routing over multihop wireless networks , '' in _ proc . ieee infocom . _ , pp . 16741682 , 2009 . a. l. stolyar , `` maximizing queueing network utility subject to stability : greedy primal - dual algorithm , '' _ queueing syst . _ , vol . 50 , no . 4 , pp . 401457 , 2005 . l. jiang , d. shah , j. shin , and j. c. walrand , `` distributed random access algorithm : scheduling and congesion control , '' _ corr _ , vol . abs/0907.1266 , 2009 . l. georgiadis , m. j. neely , and l. tassiulas , `` resource allocation and cross - layer control in wireless networks , '' _ foundations and trends in networking _ , vol . 1 , pp . 1144 , 2006 . u. akyol , m. andrews , p. gupta , j. d. hobby , i. saniee , and a. l. stolyar , `` joint scheduling and congestion control in mobile ad - hoc networks , '' in _ proc . ieee infocom . _ , pp . 619627 , 2008 . j. ryu , v. bhargava , n. paine , and s. shakkottai , `` design , implementation and evaluation of back - pressure routing / rate control for intermittently connected networks , '' in _ acm int . conf . on mobile computing and networking ( mobicom ) _ , s. moeller , a. sridharan , b. krishnamachari , and o. gnawali , `` routing without routes : the backpressure collection protocol , '' in _ internat . workshop on inform . processing in sensor networks ( ipsn ) _ , pp . 279290 , 2010 . h. zhao , c. h. xia , z. liu , and d. towsley , `` a unified modeling framework for distributed resource allocation of general fork and join processing networks , '' in _ proc . acm sigmetrics conf . _ , 2010 . l. jiang , and j. c. walrand , `` stable and utility - maximizing scheduling for stochastic processing networks , '' _ proc . allerton conf . communication , control , and computing _ , 2009 . l. massouli , a. twigg , c. gkantsidis , and p. rodriguez , `` randomized decentralized broadcasting algorithms , '' in _ proc . ieee infocom . _ , pp . 10731081 , 2007 . h. kowshik and p. r. kumar , `` zero - error function computation in sensor networks , '' in _ proc . conf . on decision and control _ , pp . 37873792 , 2009 . s. madden , m. j. franklin , j. m. hellerstein , and w. hong , `` tag : a tiny aggregation service for ad - hoc sensor networks , '' in _ sigops oper . _ , vol . 36 , pp . 131146 , 2002 . j. edmonds , `` edge - disjoint branchings . '' in _ combinatorial algorithms _ , 1972 . we now present the complete proof for the throughput - optimality of algorithm [ algo : wlinerpfalgo ] in directed acyclic graphs . since the proof closely follows the proof of massouli et al.@xcite , we do not go into complete details , but try mainly to highlight the modifications we make in order to perform aggregation rather than broadcast . first we need a lemma that ensures that under the useful packet transmission rule , each round of packets follows a spanning tree . recall that the footprint of a round of packets is defined as the set of nodes in which the packets of that round is present . further , recall that a set @xmath8 is said to be a valid footprint set if each node in @xmath8 has a path to @xmath142 in the subgraph induced by @xmath8 ; the collection of such sets is denoted by @xmath143 . we assume throughout that @xmath144 , for otherwise the min - mincut is @xmath145 . note that since we operate in continuous time , only one packet transmission ocurs at a given time with probability @xmath146 ; further , we require that the local state information is available at the time of making routing decision . now we have the following lemma : since the underlying graph is directed acyclic , we re - label the nodes as @xmath149 according to their topological ordering , where node @xmath145 is the aggregator @xmath142 , and all edges are from a higher numbered node to a lower numbered node . further , given a round of packets on a valid footprint set @xmath8 , we have that each node @xmath150 has at least one route to @xmath142 using only nodes in @xmath8 ; for short , we refer to such a route as a _ path from @xmath151 to @xmath142 in @xmath8_. since we are operating in continuous time , with probability @xmath146 only one packet transmission occurs at a given time . now suppose a useful packet is transmitted on edge @xmath152 , where @xmath153 , resulting in a new footprint set @xmath154 . for @xmath148 to be a valid footprint , we need that even after the transmission , each node @xmath155 has a path to @xmath142 in @xmath148 . to do this , we need to consider a partition of the nodes in @xmath148 into @xmath156 classes : * node @xmath155 such that @xmath157 in the topological order : due to the topological ordering property , a path from @xmath151 to @xmath142 in @xmath8 is clearly unaffected by the packet transmission from @xmath158 to @xmath40 . * node @xmath159 such that there exists a path from @xmath151 to @xmath142 in @xmath8 which does not include @xmath158 : such a path is also unaffected by the packet transmission from @xmath158 to @xmath40 and hence is still present in @xmath148 . * node @xmath159 such that _ all paths _ from @xmath151 to @xmath142 in @xmath8 pass through @xmath158 : we show by contradiction that this case is impossible under the rules of useful packet transmission . for any path from @xmath151 to @xmath142 in @xmath8 , let @xmath160 be the node immediately before @xmath158 ( i.e. , the path is @xmath161 ) . then @xmath162 has no path to @xmath142 in @xmath8 that does not pass through @xmath158 , for otherwise we have a path from @xmath151 to @xmath162 , and then to @xmath142 , which does not pass through @xmath158 . this means that @xmath162 becomes isolated upon transmission of packet from @xmath158 to @xmath40 , which violates the non - isolation condition of useful packet forwarding . the main idea behind the proof in @xcite was to define the ` footprint counter ' variables to represent the state of the system , and considering an appropriate function of these that allowed translating the local decisions of the nodes in terms of global graph parameters . in order to modify the proof for broadcast , we defined a similar collection of counter variables in section [ sec : algo1 ] , and now define their associated dynamics as follows . * arrival of new round : @xmath163 ( this corresponds to adding a packet to the queue with footprint @xmath13 , as a packet of the new round is simultaneously generated at all the nodes ) . * completion of packet transfer : this is only for active packets , i.e. , those currently under transmission . for active packet @xmath164 with corresponding @xmath165 and @xmath166 , we have : @xmath167 ( the first equation corresponds to removing the edge over which packet transmission was completed , and also updating the footprint of the packet to include the new node . the second updates the list of idle packets in case there is no other instance of this packet being transmitted . ) . * initiation of a new transfer at an idle link . the new packet is selected uniformly at random among the set of useful packets at the node . if @xmath168 , then a new packet transfer is formally described as follows : * * select a useful packet of an idle round with footprint @xmath169 , with probability @xmath170 select a useful packet of an active round @xmath164 with @xmath171 with probability @xmath172 * * if idle packet with footprint @xmath8 is selected : @xmath173 , with @xmath174 . if packet of active round @xmath175 is selected , then @xmath176 we note here that the node itself does not need to know these global counters to perform packet selection ; rather , this emerges from the use of the random useful packet forwarding rule . the idea of relating the local packet selection rule to the global counters is crucial in proving the optimality of the algorithm . the local rules for checking whether a packet is useful or not corresponds to selecting packets whose global footprint obeys certain properties ; picking a useful packet uniformly at random therefore corresponds to picking a packet from such a useful global footprint with a probability proportional to the corresponding counter variable . observe that in order to determine the flow into a footprint set @xmath8 , we need to consider the collection of sets which include @xmath8 and have one extra node . we now define the fluid limits of the system . this is similar in spirit to the fluid limit of the system in @xcite , so we try to use similar notation . the existence of the limit also follows immediately from their convergence results , so we omit it due to lack of space and refer interested readers to @xcite for technical details . * @xmath178 @xmath179 * work conservation : at almost every @xmath180 is differentiable and if @xmath181 ( where @xmath182 is the fluid trajectory associated with @xmath183 ) , then we have @xmath184 * @xmath185 are non - decreasing , lipschitz continuous , with lipschitz constant @xmath186 , and @xmath187 is @xmath186-lipschitz . for any @xmath188 set of all fluid trajectories with initial condition @xmath189 , and further , we define @xmath190 as the state of the mc with initial conditions @xmath191 , @xmath192 . now , as in @xcite , for a sequence of initial conditions @xmath193 s.t . for a sequence of positive numbers @xmath194 and the limit @xmath195 exists in @xmath196 , we have that @xmath197 : @xmath198}||y^n(t)-f(t)||\geq\epsilon]=0.\ ] ] next we define the candidate lyapunov function that we use to analyze the stability of the system . in @xcite , the function was defined in terms of queues ( or counters ) that counted all the packets whose footprint was contained inside a set @xmath8 . the advantage of these queues for studying broadcast was that their rate of increase was controlled by external arrivals to the system , while they were drained due to transfers across the cut defined by the set @xmath8 . for the purpose of studying aggregation , we need to identify an equivalent set of queues to reflect the unique dynamics of the system . in particular , we consider for each set @xmath8 a queue of all rounds whose footprints _ are not entirely contained within @xmath8_. these queues ( counters ) exhibit similar properties to the ones considered for broadcast in that every incoming round is counted by all these queues ( as every node in the network generates a packet ) , while the drain of these queues is controlled by flow across the cut defined by the set @xmath8 . formally , we have the following theorem : as in @xcite , before proving this theorem we first we need a combinatorial lemma . this lemma and its proof parallels a corresponding lemma in @xcite , with modifications to deal with aggregation and the @xmath206 counter variables we have defined above . note that lemma [ lem : counting ] does not depend on the algorithm , or the fluid model in any way . it is a pure combinatorial property of the way that the quantities are defined . in other words , any function mapping the sets @xmath147 to @xmath217 obeys the lemma _ for any @xmath218_. later we use the ability to control @xmath219 to obtain uniform bounds on the lyapunov drift . for the first condition , consider @xmath169 such that @xmath220 then we have @xmath221 and thus @xmath222 however , from the definition of the @xmath223 , we have that @xmath224 for all @xmath225 . hence we have that @xmath226 for the second condition , consider @xmath147 such that@xmath227 . further , consider set @xmath148 such that @xmath228 and satisfying @xmath229 then we have @xmath230 thus for our condition , we need @xmath231 and noting the fact that @xmath223 are increasing with @xmath40 , it is sufficient to ensure @xmath232 this in fact holds with equality because of our choice of @xmath223 . thus , given any @xmath218 , we can construct @xmath223 such that the two conditions hold . ( proof of theorem [ thm : lyap ] ) given @xmath218 , we define @xmath223 as in lemma [ lem : counting ] . then , or any @xmath233 , if @xmath234 is a set which belongs to arg - max of @xmath235 , then @xmath236 ( unless all the fluid sample paths are identically @xmath145 ) . next we use the optimality of @xmath234 to obtain some relations between @xmath237 and the weight across its cut - edges . @xmath238 such that @xmath17 , we have from the contrapositive of the first condition of lemma [ lem : counting ] ( as @xmath234 is in the arg - max ) that @xmath239 similarly from condition @xmath240 , @xmath241 such that @xmath242 , we have that @xmath243 now we have @xmath244\\ = & \lambda-\sum_{v\in s^ * , u\notin s^*}c_{uv}\left[1-\sum_{s'\nsubseteq s^ * , v\in s ' , u\notin s'}\frac{x_{s'+u}}{x_{+u - v}}\right],\\ & \mbox{(from defn of fluid trajectories)}\\ \leq & \lambda-\sum_{u\in s^ * , v\notin s^*}c_{uv}+\sum_{u\in s^ * , v\notin s^*}c_{uv}\sum_{s'\nsubseteq s^ * , v\in s ' , u\notin s'}\alpha\\ & \mbox{(from previous observation)}\\ \leq & \lambda-\sum_{v\in s^ * , u\notin s^*}c_{uv}+\max_{(u , v)\in \mathcal{l}}c_{uv}|\mathcal{l}|2^k\alpha.\end{aligned}\ ] ] if we choose @xmath219 and @xmath245 as follows : @xmath246 then we get that , for all @xmath247 , @xmath248 to argue that this implies negative drift of the lyapunov function , i.e. @xmath249 , we observe that by definition @xmath250 . finally , using the lipschitz continuity of the trajectories , it is sufficient to show this property holds for the sets @xmath247 .

infrared luminous galaxies ( irlgs ) , most of whose huge luminosities are radiated in the far - infrared ( 40500 @xmath0 m ) as dust emission ( @xmath2 @xmath4 10@xmath5 ergs s@xmath6 ) , @xmath7 75 km s@xmath6 mpc@xmath6 and @xmath8 @xmath7 0.5 are used throughout this paper . ] are a significant population at the bright end of galaxy luminosity function ( soifer et al . two types of activity , active galactic nucleus ( agn ) and starburst , are thought to contribute to the far - infrared emission of irlgs . the estimate of the relative energetic contributions of these kinds of activity is an important issue for understanding the nature of irlgs . furthermore , since the bulk of the cosmic submillimeter background emission can be explained by irlgs at high - redshift ( blain et al . 1999 ) , understanding the energy sources of nearby irlgs is closely related to the origin of the background emission . both agn and starburst activity produces strong emission lines . since many emission lines , such as narrow emission lines of hydrogen or h@xmath9 molecules , can be caused by both kinds of activity , it is not easy to distinguish their origin . the polycyclic aromatic hydrocarbon ( pah ) emission features can be a powerful tool for separating these kinds of activity and estimating their relative importance because the features are associated only with starburst activity and not with agn activity ( moorwood 1986 ) . using the pah emission feature at 7.7 @xmath0 m , systematic studies of the energy sources of irlgs have been reported based on the _ infrared space observatory _ ( _ iso _ ) spectra at 5.811.6 @xmath0 m ( lutz et al . 1998 ; genzel et al . 1998 ; rigopoulou et al . the observed ratios between the 7.7 @xmath0 m pah peak flux and 7.7 @xmath0 m continuum flux of irlgs were estimated to be as high as those of starburst galaxies with lower far - infrared luminosities , leading these authors to suggest that starburst activity is a dominant energy source for most of the irlgs they observed . however , the 7.7 @xmath0 m pah to far - infrared luminosity ratios of irlgs are roughly half of those of starburst galaxies ( genzel & cesarsky 2000 ) . this indicates that observed starburst activity can account for only half of the far - infrared luminosities for irlgs , although higher dust extinction toward starburst activity in irlgs could make the intrinsic contribution of starburst activity higher . in the 5.811.6 @xmath0 m spectra of irlgs , particularly obscured ones , not only the pah emission , but also broad silicate dust absorption , is likely to exist at 813 @xmath0 m ( mathis 1990 ) . limited wavelength coverage of most _ iso _ spectra ( 5.811.6 @xmath0 m ) does not allow the determination of a secure continuum level for many irlgs and thus the estimate of the pah flux could be quantitatively uncertain , as mentioned in rigopoulou et al . ( 1999 ) . in fact , the 7.7 @xmath0 m pah emission flux of arp 220 , an irlg for which an _ iso _ spectrum with wider wavelength coverage ( 516 @xmath0 m ) is available , could be considerably smaller than that estimated by rigopoulou et al . , if the continuum level is determined in a different way ( elbaz et al . 1998 ; dudley 1999 ) . furthermore , in the case of another irlg , iras 05189@xmath102524 , the reported equivalent widths of the 7.7 @xmath0 m pah emission feature based on the _ iso _ spectrum differ by a factor of @xmath113 according to different authors ( clavel et al . 2000 ; laureijs et al . 2000 ) . spectroscopy in the range 34 @xmath0 m can be a powerful method for investigating the energy source of irlgs , by avoiding the above uncertainty . first , since dust extinction at 34 @xmath0 m is estimated to be as small as that at 78 @xmath0 m ( @xmath12/@xmath13 @xmath3 0.8 ; lutz et al . 1996 ) , we can detect signs of obscured energy sources . next , we can diagnose the relative importance of agn and starburst activity using spectral features in the range 34 @xmath0 m . if an irlg is powered by starburst activity , pah emission should be detected at 3.3 @xmath0 m and sometimes at 3.4 @xmath0 m ( tokunaga et al . if a source is powered by obscured agn activity , a 3.4 @xmath0 m absorption feature of carbonaceous dust grains should be detected ( pendleton et al . 1994 ) , as found in some sources ( e.g. , ngc 1068 ; bridger , wright & geballe 1994 ; imanishi et al . if a source is powered by unobscured agn activity , a 34 @xmath0 m spectrum should be nearly featureless . figure 1 summarizes the expected spectral shapes for the above three cases and for the cases of composites of agn and starburst activity . finally , since a continuum level both longward and shortward of these features at 3.33.4 @xmath0 m is observable in the @xmath14-band ( 2.84.2 @xmath0 m ) atmospheric window of earth for nearby ( _ z _ @xmath15 0.18 ) sources , there is no serious uncertainty in determining a continuum level . ground - based spectroscopy usually makes use of small slit widths , typically less than a few arcsec . one arcsec corresponds to greater than a few 100 pc in irlgs at z @xmath11 0.02 . there is currently an increasing amount of evidence based on high spatial resolution imaging studies of infrared 225 @xmath0 m continuum emission that the bulk of the huge dust emission luminosities of irlgs come from compact ( @xmath15 a few 100 pc ) regions powered either by agn and/or compact starburst activity , with little contribution from extended regions ( surace & sanders 1999 ; surace , sanders , & evans 2000 ; scoville et al . 2000 ; soifer et al . thus , ground - based 34 @xmath0 m spectroscopy can fully diagnose the properties of dust emission from the energetically important compact regions in most irlgs . the present study reports ground - based spectroscopy in the wavelength range 34 @xmath0 m of nine irlgs to investigate their energy sources . to achieve signal - to - noise ratios higher than 1020 with a few hours of on - source integration time using the 3.8-m united kingdom infrared telescope ( ukirt ) , irlgs brighter than 11 mag at @xmath14 or @xmath16 were selected . the targets are summarized in table 1 . a total of nine irlgs were observed . far - infrared luminosities are higher than @xmath310@xmath5 ergs s@xmath6 . in addition to the irlgs , ngc 253 , a nearby ( @xmath33 mpc ; tully 1988 ) galaxy , was also observed . the emission properties of ngc 253 can be explained by starburst activity only ( e.g. , rieke et al . 1980 ; engelbracht et al . 1998 ) , and no strong agn activity has been found . although its far - infrared luminosity ( table 1 ) is more than an order of magnitude smaller than those of the irlgs , this source is powered by compact ( @xmath3100 pc ) starburst activity ( kalas & wynn - williams 1994 ) . ngc 253 is thus an important nearby source for studying the nature of compact starburst activity found in irlgs , and its 34 @xmath0 m spectrum can be used as a template of compact starburst activity . we used the cooled grating spectrometer ( cgs4 ; mountain et al . 1990 ) to obtain 34 @xmath0 m spectra of the irlgs and ngc 253 with ukirt on mauna kea , hawaii . an observing log is summarized in table 2 . sky conditions were photometric throughout the observations . the detector was a 256 @xmath17 256 insb array . the 40 l mm@xmath6 grating with 2 pixels wide slit (= 1@xmath18 ) was used . the resulting spectral resolution was @xmath3750 at 3.5 @xmath0 m . spectra were obtained toward the flux peak at 34 @xmath0 m . a standard telescope nodding technique was employed along the slit to subtract the signal from the sky . the nodding amplitude was @xmath312@xmath19 except for ngc 253 , for which it was @xmath324@xmath19 . the position angles of the slit used are summarized in table 2 . for irlgs that have clear double nuclei with a small separation ( @xmath15 several arcsec ; arp 220 , ngc 6240 , mrk 273 , and iras 08572 + 3915 ; scoville et al . 2000 ) , the position angles were set so that signals from both the nuclei were observed simultaneously . a- to g - type standard stars ( table 2 ) were observed with almost the same airmass as the target sources to correct for the transmission of earth s atmosphere . for target sources whose standard stars are a - type , data points at 3.74 @xmath0 m and 4.05 @xmath0 m were removed from final spectra because these data points are affected by , respectively , strong pf@xmath20 and br@xmath21 absorption lines in the spectra of a - type stars . the @xmath14- or @xmath16-band magnitudes of standard stars were estimated from their @xmath22-band magnitudes , by adopting the @xmath23 or @xmath24 colors of the stellar types of individual standard stars ( tokunaga 2000 ) . standard data analysis procedures were employed using iraf . first , data of bad pixels were replaced with the interpolated signals of the surrounding pixels . next , bias was subtracted from the obtained frames and the frames were divided by a flat image . then , the spectra of the targets and the standard stars were extracted . since a secondary nucleus was not clearly recognizable as a distinct nucleus in all cases , spectra were extracted by integrating signals over 8@xmath1920@xmath19 along the slit ( table 3 ) . wavelength calibration was performed using an argon or a krypton lamp , and is believed to be accurate within 0.005 @xmath0 m . the galaxy spectra were divided by those of the standard stars , and were multiplied by the spectra of blackbodies with temperatures corresponding to individual standard stars ( table 2 ) . after flux calibration based on the adopted standard star fluxes , the final spectra were produced . the flux - calibrated spectra are shown in figure 2 . spectra for some irlgs were available in the literature , and the new spectra generally agree with previously published spectra . detailed comparisons of our spectra with previous ones are found in appendix a. the comparison of the magnitudes of the cgs4 data with those in the literature is summarized in table 3 . the magnitudes of irlgs based on the cgs4 data roughly agree with photometric magnitudes measured with larger apertures , suggesting that nearly all of previously measured 34 @xmath0 m emission is also detected in the cgs4 data . the spectra of ic 694 , arp 220 , ngc 6240 , and mrk 273 are dominated by the 3.3 @xmath0 m pah emission and no detectable 3.4 @xmath0 m carbonaceous dust absorption , as in the case of ngc 253 . the spectrum of mrk 231 shows the 3.3 @xmath0 m pah emission and possible 3.4 @xmath0 m carbonaceous dust absorption , but the reality of the latter feature is currently doubtful . the spectrum of iras 05189@xmath102524 shows both 3.3 @xmath0 m pah emission and detectable 3.4 @xmath0 m carbonaceous absorption with an optical depth of @xmath25 @xmath3 0.04 . the spectrum of iras 08572 + 3915 is dominated by the 3.4 @xmath0 m carbonaceous dust absorption feature ( @xmath25 @xmath3 0.9 ) . neither detectable 3.3 @xmath0 m pah emission nor the 3.4 @xmath0 m carbonaceous dust absorption is found in the spectra of iras 20460 + 1925 and iras 23060 + 0505 . the fluxes and luminosities of the 3.3 @xmath0 m pah emission feature are summarized in table 4 . in addition to the above features , pf@xmath20 emission lines ( rest wavelength = 3.74 @xmath0 m ) are clearly detected in the spectra of ic 694 and ngc 253 . the observed fluxes are 2 @xmath17 10@xmath26 ergs s@xmath6 @xmath27 and 6 @xmath17 10@xmath26 ergs s@xmath6 @xmath27 for ic 694 and ngc 253 , respectively . in the spectrum of ngc 6240 , we attribute the flux excess at 3.71 @xmath0 m as h@xmath9 00 s(15 ) emission line ( rest wavelength = 3.626 @xmath0 m ) . its flux is 1 @xmath17 10@xmath26 ergs s@xmath6 @xmath27 . ngc 6240 is known to show strong h@xmath9 emission lines whose strength ratios can be approximated by thermal excitation with the temperature of @xmath32000 k ( sugai et al . the comparison with the h@xmath9 10 s(0 ) flux ( 1.8 @xmath17 10@xmath28 ergs s@xmath6 @xmath27 ) measured with a 3@xmath19 @xmath17 6@xmath19 aperture ( goldader et al . 1995 ) provides the 00 s(15 ) to 10 s(1 ) flux ratio of 0.06 . this ratio is similar to that found in shock gas in the orion molecular outflow where the h@xmath9 line ratios are roughly fitted by 2000 k thermal emission ( brand et al . in irlgs , agn activity and compact starburst activity are likely to coexist spatially close to each other . if energetic ionizing photons from agn activity penetrate the compact starburst regions and destroy pahs ( voit 1992 ) that otherwise should produce the strong 3.3 @xmath0 m emission feature , the 3.3 @xmath0 m pah luminosity could be small in spite of the presence of strong compact starburst activity . however , in irlgs , the destruction of pahs in the starburst regions by energetic photons from agn activity probably does not occur because of the large column density and high pressures in the interstellar medium ( maloney 1999 ) . in the following discussion , it is assumed that compact starburst activity in the irlgs produces the 3.3 @xmath0 m pah emission intrinsically as strongly as the star - formation activity in starburst galaxies with lower far - infrared luminosities , namely , in proportion to the far - infrared luminnosities . for starburst - dominated galaxies , the rest - frame equivalent widths of the 3.3 @xmath0 m pah emission feature ( ew@xmath29 ) do not change significantly as a result of dust extinction because flux attenuation of the 3.3 @xmath0 m pah emission and 34 @xmath0 m continuum is similar . thus very small ew@xmath30 values are most likely caused by the contribution of agn activity to the 34 @xmath0 m continuum emission . the rest - frame ew@xmath29 values of the nine irlgs are summarized in table 4 . the value for ngc 253 is 120 nm , which is nearly a median value for starburst - dominated galaxies ( moorwood 1986 ) . if the scatter of the values is taken into account and irlgs with ew@xmath29 = 60240 nm are regarded as candidate starburst - dominated galaxies , then only ic 694 , arp 220 , and ngc 6240 have ew@xmath29 values as high as those of starburst galaxies , and are thus candidate starburst - dominated galaxies . mrk 273 shows a factor of @xmath312 smaller value than those of starburst galaxies and thus agn activity is estimated to contribute @xmath3150% of the 34 @xmath0 m continuum flux . the remaining five irlgs ( mrk 231 , iras 05189@xmath102524 , iras 08572 + 3915 , iras 23060 + 0505 , and iras 20460 + 1925 ) have ew@xmath29 ratios more than an order of magnitude smaller , indicating that @xmath1190% of the 34 @xmath0 m continuum flux originates in agn activity . since the far - infrared to 34 @xmath0 m flux ratio of dust emission heated by agn activity through a radiative transfer mechanism depends strongly on dust radial density distribution ( ivezic , nenkova & elitzur 1999 ) , conversion from the agn contribution at 34 @xmath0 m to that at far - infrared wavelengths is not certain . however , the ew@xmath29 values for these six irlgs ( mrk 273 and the five irlgs ) give no evidence that they are powered predominantly by compact starburst activity . in @xmath325.1 , three irlgs have ew@xmath29 values as high as those of starburst galaxies . however , large ew@xmath29 values of the irlgs do not necessarily mean that their far - infrared luminosities are powered predominantly by starburst activity . although highly obscured agn activity can contribute to the observed far - infrared luminosity because of negligible extinction in this wavelength range , the contribution of the highly obscured agn to the 34 @xmath0 m flux can be highly attenuated . as shown in figure 1@xmath33 , the apparent spectral shapes of galaxies powered both by starburst and highly obscured agn activity could be nearly the same as those powered only by starburst activity ( fig . 1@xmath34 ) . a straightforward means of estimating the energy contribution of compact starburst activity is to estimate the absolute 3.3 @xmath0 m pah luminosity and compare it with the far - infrared luminosity . the 3.3 @xmath0 m pah emission to far - infrared luminosity ratios ( @xmath1/@xmath2 ) of the irlgs are summarized in table 4 . the ratios of nearby starburst galaxies with lower far - infrared luminosity are estimated to be @xmath31 @xmath17 10@xmath35 ( mouri et al . 1990 ) , although the scatter in this ratio is difficult to estimate because , in the case of nearby starburst galaxies , the ratios are affected by different aperture sizes between 34 @xmath0 m and far - infrared data ( mouri et al . the observed @xmath1/@xmath2 ratios for all nine irlgs are less than 3 @xmath17 10@xmath36 , a factor of @xmath113 smaller than those of starburst galaxies . the smaller ratios for the six irlgs with small ew@xmath29 values can be explained naturally by a dominant contribution of agn activity , while a few mechanisms can explain the smaller ratios in the three irlgs with large ew@xmath29 values ( ic 694 , arp 220 , and ngc 6240 ) . if highly obscured agn activity contributes significantly to the far - infrared luminosity but little to the 34 @xmath0 m flux due to high flux attenuation , then the large ew@xmath29 values and the small @xmath1/@xmath2 ratios can be explained . this is the first possible mechanism . a second mechanism is the effect of possible signal loss in the cgs4 data . the possible signal loss for ngc 6240 could be as high as 0.5 mag or a factor of 1.6 , while that for ic 694 and arp 220 is a factor of @xmath371.2 . provided that spectral shape outside the cgs4 slit is the same as that inside it , the possible slit loss could decrease the @xmath1/@xmath2 ratios only slightly . a third possible mechanism is the effect of dust extinction . the @xmath1/@xmath2 ratios for starburst - dominated irlgs could be smaller than those of starburst galaxies with lower far - infrared luminosities if dust extinction toward compact starburst activity in the irlgs is much higher . for ic 694 , arp 220 , and ngc 6240 , dust extinction toward starburst regions in the case of a foreground dust screen model could be as high as @xmath38 @xmath3 20 mag ( sugai et al . 1999 ; alonso - herrero et al . 2000 ) , @xmath38 @xmath3 45 mag ( rigopoulou et al . 1999 , fig . 7 ) , and @xmath38 @xmath3 7 mag ( @xmath39 = 0.8 mag ; sugai et al . 1997 ) , respectively . by adopting a standard extinction curve ( @xmath40 @xmath3 0.060.07 @xmath17 @xmath38 ; rieke & lebofsky 1985 ; lutz et al . 1996 ) and taking possible signal loss into account , the _ intrinsic _ @xmath1/@xmath2 ratios after correction for dust extinction could be as high as 47 @xmath17 10@xmath36 for all three . provided that the @xmath1/@xmath2 ratios of control starburst galaxy samples suffer the effect of dust extinction with @xmath38 @xmath3 10 mag ( rigopoulou et al . 1999 , fig . 7 ) , the intrinsic @xmath1/@xmath2 ratios of starburst - dominated galaxies are estimated to be 1.71.9 @xmath17 10@xmath35 . the intrinsic @xmath1/@xmath2 ratios for the three irlgs are a factor of @xmath33 smaller . for reference , the 3.3 @xmath0 m pah flux of ngc 253 within our slit is measured to be 2.8 @xmath17 10@xmath41 ergs s@xmath6 @xmath27 . the @xmath14-band magnitude inside our slit is 2.7 mag fainter than the total magnitude of this galaxy ( table 3 ) . if the 34 @xmath0 m spectrum shape outside the slit is the same as that inside it , and if @xmath38 = 14 mag is adopted as dust extinction toward starburst regions in this galaxy ( rigopoulou et al . 1999 , fig . 7 ) , then the intrinsic @xmath1/@xmath2 ratio is 1.31.5 @xmath17 10@xmath35 , consistent with the value of 1.71.9 @xmath17 10@xmath35 within @xmath325% . although dust extinction for arp 220 was derived based on mid - infrared data , the dust extinction for ic 694 and ngc 6240 was estimated based on near - infrared data . the dust extinction estimate based on near - infrared data could be smaller than that based on mid - infrared data ; the latter estimate is thought to be more reliable , particularly in the case of high dust extinction ( genzel et al . 1998 ) . from this viewpoint , it is possible that the intrinsic @xmath1/@xmath2 ratios for ic 694 and ngc 6240 could be higher than our estimate . further , when dust is mixed with emission sources , dust extinction correction factor could be higher than that for a foreground screen dust model ( e.g. , genzel et al . since the scatter of the intrinsic @xmath1/@xmath2 ratios for starburst galaxies is also poorly known , the energy source of these three irlgs can not be determined confidently from our diagnoses . the energy sources of the nine irlgs derived from the 34 @xmath0 m spectra is summarized in table 5 . for ic 694 , arp 220 , and ngc 6240 , which show strong 3.3 @xmath0 m pah emission in equivalent width , it is not clear whether compact starburst activity is the only important energy source or a significant fraction of the luminosity is powered by obscured agn activity ( figs . 1@xmath34 and 1@xmath33 ) . mrk 231 and mrk 273 , which show weak 3.3 @xmath0 m pah emission and no detectable 3.4 @xmath0 m carbonaceous dust absorption , are suggested to be powered by either unobscured or obscured agn activity ( figs . 1@xmath42 and 1@xmath33 ) . iras 05189@xmath102524 , which shows weak 3.3 @xmath0 m pah emission and detectable ( but weak ) 3.4 @xmath0 m carbonaceous dust absorption , is thought to be powered by moderately obscured agn activity ( fig . 1@xmath33 ) . the 34 @xmath0 m spectrum of iras 08572 + 3915 is typical of highly dust obscured agns , and thus this source is most likely powered by highly embedded agn activity ( fig . 1@xmath43 ) , as further supported by more detail study based on 320 @xmath0 m spectroscopy ( dudley & wynn - williams 1997 ; imanishi & ueno 2000 ) . the featureless 34 @xmath0 m spectra of iras 23060 + 0505 and iras 20460 + 1925 suggest that these sources are powered by less obscured agn activity ( fig . 1@xmath44 ) . for arp 220 , ngc 6240 , mrk 273 , mrk 231 , iras 05189@xmath102524 , and iras 23060 + 0505 , the 7.7 @xmath0 m pah peak flux to 7.7 @xmath0 m continuum flux ratios were used to diagnose their energy source ( rigopoulou et al . 1999 ; laureijs et al . 2000 ) . these ratios and the equivalent widths are a similar ( but not identical ) concept , and both provide the pah strength relative to the continuum emission . the results of the 34 @xmath0 m spectra and the _ iso _ spectra are compared in table 5 . as mentioned in @xmath32 1 , a continuum determination based on the _ iso _ 5.811.6 @xmath0 m spectra is not easy . when data at 11 @xmath0 m were used to determine a continuum level ( rigopoulou et al . 1999 ) , the 7.7 @xmath0 m pah flux for sources with a strong silicate dust absorption feature could be overestimated . since a different amount of effect of dust extinction at different wavelengths does not change the relative pah strengths significantly ( @xmath32 5.1 ) , the comparison of the strength of the 3.3 @xmath0 m and 7.7 @xmath0 m pah emission relative to the continuum could provide useful information on the effect of the silicate dust absorption in the _ iso _ spectra . the relative pah strengths ( ew@xmath29 and 7.7 @xmath0 m pah peak to continuum flux ratio ) for mrk 231 , iras 05189@xmath102524 , and iras 23060 + 0505 are much smaller than those of starburst galaxies in both wavelengths . the relative pah strength for ngc 6240 is the same within 30% as the typical value of starburst - dominated galaxies at both wavelengths . in fact , for ngc 6240 , no indication of strong silicate dust absorption was found in ground - based 813 @xmath0 m spectra ( roche et al . 1991 ; dudley 1999 ) . however , for arp 220 , the 34 @xmath0 m spectrum provides the ew@xmath29 value of @xmath380 nm , slightly smaller than those of starburst - dominated galaxies ( @xmath3120 nm ) , while the _ iso _ spectrum provides the 7.7 @xmath0 m pah peak - to - continuum flux ratio of @xmath34.2 , significantly higher than the average value of starburst galaxies ( @xmath33.0 ) . also for mrk 273 , the 34 @xmath0 m spectrum gives ew@xmath29 @xmath335 nm , a factor of @xmath33 smaller than starburst galaxies , while the _ iso _ spectrum gives the 7.7 @xmath0 m pah peak - to - continuum flux ratio @xmath31.9 , only @xmath340% smaller . for arp 220 and mrk 273 , dudley ( 1999 ) argued that silicate dust absorption plays an important role based on ground - based 813 @xmath0 m spectra . the discrepancy may be caused by the effect of the silicate dust absorption . the above comparison suggests that , for at least a some fraction of irlgs , the 7.7 @xmath0 m pah flux estimated by rigopoulou et al . ( 1999 ) could be too large because of the use of data affected by silicate dust absorption as a continuum level . in this case , the presence of the silicate dust absorption is underestimated . highly obscured agns show spectra characterized by weak pah emission and strong silicate dust absorption . virtually no irlgs powered by highly obscured agns were found based on the _ iso _ data ( genzel et al . 1998 ; rigopoulou et al . 1999 ) , which may be the result of the determination of the continuum level . finally , for iras 05189@xmath102524 , the flux and equivalent width of the 3.3 @xmath0 m pah emission estimated by an _ iso _ spectrum obtained using a larger ( 24@xmath19 @xmath17 24@xmath19 ) aperture ( clavel et al . 2000 ) are a factor of @xmath36 larger than the estimate in this paper . if the larger values are the case , the bulk of the 3.3 @xmath0 m pah emission detected in the _ iso _ spectrum must originate in extended star - forming regions outside the cgs4 slit . the extended star - formation regions could explain nearly all of the far - infrared luminosity of this source , if the value of @xmath1/@xmath2 = 1.71.9 @xmath17 10@xmath35 is adopted . however , this source shows no extended continuum emission at 225 @xmath0 m ( scoville et al . 2000 ; soifer et al . if the extended regions are fully responsible for the far - infrared emission at 60 @xmath0 m but contribute to , say , less than 5% of the 25 @xmath0 m flux ( soifer et al . 2000 ) , then the 60 @xmath0 m to 25 @xmath0 m flux ratio in jy is @xmath1180 , much larger than any nearby star - forming galaxies ( rice et al . 1988 ) . since the physical conditions in the extended regions of iras 05189@xmath102524 ( not nuclear compact regions that could have an extreme environment in density and so on ) are expected not to be quite different from those in nearby star - forming galaxies , such a sudden flux increase from the mid - infrared ( @xmath1525 @xmath0 m ) to far - infrared ( @xmath1140 @xmath0 m ) only in the case of iras 05189@xmath102524 is very unlikely . while the 3.3 @xmath0 m pah flux in this paper is estimated by tracing the spectral profile ( see the caption of table 4 for more details ) , clavel et al . simply sum up all the flux above the adopted continuum level in the pre - defined wavelength range ( j. clavel , private communication ) . the _ iso _ spectrum provided by dr . clavel ( private communication ) shows a flux excess above an adopted continuum level at @xmath33.3 @xmath0 m , but it has lower spectral resolution ( r @xmath3 100 ) and larger scatter in data points at 34 @xmath0 m than the cgs4 spectrum . furthermore , although the wavelength of the 3.3 @xmath0 m pah emission feature usually extends from 3.24 @xmath0 m to 3.34 @xmath0 m in the rest - frame ( tokunaga et al . 1991 ; this work ) , the flux excess in the _ iso _ spectrum is found from 3.25 @xmath0 m to 3.40 @xmath0 m ( not 3.34 @xmath0 m ) . if this excess is actually due to the 3.3 @xmath0 m pah emission feature , its spectral shape in the extended regions of iras 05189@xmath102524 is quite different from typical ones . more sensitive and higher spectral resolution spectroscopy with a wide aperture is necessary to address this issue . we tentatively consider that the extended regions contribute little to the total far - infrared luminosity of iras 05189@xmath102524 . soifer et al . ( 2000 ) suggested based on their 825 @xmath0 m imaging data that agn activity is the dominant energy source of mrk 231 , iras 05189@xmath102524 , and iras 08572 + 3915 . our suggestions are the same for these three sources . soifer et al . were unable to distinguish the energy source of mrk 273 and arp 220 . we argue that agn activity is the main energy source of mrk 273 , while the energy source of arp 220 is not clear . as a whole , the suggestions in this paper are consistent with those by soifer et al . in the @xmath152 @xmath0 m spectra of mrk 231 , iras 05189@xmath102524 , iras 23060 + 0505 , and iras 20460 + 1925 , a broad ( @xmath112000 km s@xmath6 in full width at half - maximum ) component of neutral hydrogen recombination line emission was detected , and their intrinsic luminosities are high enough to account for the far - infrared luminosities resulting from agn activity alone ( veilleux , sanders , & kim 1999b ) . our suggestions about the energy source for these four irlgs are consistent with theirs . among the nine irlgs , five ( mrk 231 , iras 05189@xmath102524 , iras 08572 + 3915 , iras 23060 + 0505 , and iras 20460 + 1925 ) have warm ( _ iras _ 25 @xmath0 m to 60 @xmath0 m flux ratio @xmath46/@xmath47 @xmath11 0.2 ) far - infrared color ( table 1 ) . the above three infrared diagnoses have provided a growing amount of evidence that these warm irlgs are powered predominantly by agn activity ( table 5 ) . however , warm irlgs constitute only @xmath320% of total irlgs , and most irlgs have cool ( @xmath46/@xmath47 @xmath15 0.2 ) far - infrared color ( kim & sanders 1998 ) . it was argued based on the estimate of the 7.7 @xmath0 m pah strengths in the _ iso _ spectra that these cool irlgs are powered predominantly by starburst activity , with little agn contribution ( genzel & cesarsky 2000 ) . among the nine irlgs , three ( arp 220 , ngc 6240 , and mrk 273 ) are cool irlgs and are classified as starburst - dominated irlgs based on the 7.7 @xmath0 m pah strength ( table 5 ) . however , for arp 220 and mrk 273 , our 34 @xmath0 m spectra suggest that the 7.7 @xmath0 m pah strengths , and thereby the magnitudes of starburst activity , may be overestimated based on the _ iso _ spectra . in fact , for mrk 273 , although the relative 7.7 @xmath0 m pah emission strength suggests this source is powered by starburst activity ( rigopoulou et al . 1999 ) , a different diagnosis based on high excitation lines in the _ iso _ spectra indicates that agn activity is a dominant energy source ( genzel et al . 1998 ) . for ngc 6240 , x - ray data suggest that highly embedded and highly luminous agn activity is an important energy source ( vignati et al . the contribution of agn activity in the cool irlgs may be higher than the estimate based on the 7.7 @xmath0 m pah emission strength in the _ iso _ spectra . high spatial resolution radio images have been used to investigate the relative importance between agn and starburst activity in irlgs . a compact radio core , which was thought to be strong evidence for agn activity , is no longer definitive evidence for agn activity because both agn and compact starburst activity could explain observed properties of the compact radio core ( smith , lonsdale , & lonsdale 1998 ) . however , irlgs whose radio emission morphology is extended without a compact core are thought to be powered by starburst activity . compact radio cores were detected in ic 694 , arp 220 , ngc 6240 , mrk 273 , mrk 231 , and iras 20460 + 1925 ( lonsdale , lonsdale , & smith 1992 ; lonsdale , smith , & lonsdale 1993 ; heisler et al . 1998 ; kewley et al . 2000 ) , but not in iras 05189 + 2524 , iras 08572 + 3915 , and iras 23060 + 0505 ( lonsdale et al . 1993 ; heisler et al . it is puzzling that no radio core emission was detected in the three warm irlgs , for which more than one infrared diagnoses consistently suggest that agn activity is a dominant energy source ( table 5 ) . a high fraction ( 6/9 ) of irlgs in this sample are suggested to be powered by agn activity . this is thought to be caused by the selection effect of picking irlgs that are bright at 34 @xmath0 m . at a given far - infrared flux , the 34 @xmath0 m flux of less obscured agns is generally higher than that of starburst - dominated galaxies ( schmitt et al . 1997 ) . among the nine irlgs , 5/9 ( 55% ) are optically classified as seyfert types ( table 1 ) and 4/9 ( 44% ; mrk 231 , iras 05189@xmath102524 , iras 23060 + 1925 , and iras 23060 + 0505 ) show a detectable broad emission line at @xmath152 @xmath0 m ( that is , agn activity is less obscured ) . these fractions are significantly higher than those in a larger sample of irlgs ( @xmath320% ; veilleux , kim , & sanders 1999a ; veilleux et al . furthermore , the fraction of warm irlgs in this sample ( 5/9 ; 55% ) is much higher than that in a larger sample of irlgs ( @xmath320% ; kim & sanders 1998 ) . observations of a larger number of fainter irlgs at @xmath14 are clearly necessary to evaluate the relative role between agn and starburst activity in irlgs as a whole . the energy sources of nine irlgs were diagnosed using the ew@xmath29 values and the @xmath1/@xmath2 ratios . the following main results were found : 1 . mrk 231 , mrk 273 , and iras 05189@xmath102524 show detectable 3.3 @xmath0 m pah emission in their 34 @xmath0 m spectra . however , the pah emission strengths are too small to account for the bulk of their far - infrared luminosities by starburst activity , indicating that these sources are powered by agn activity . no pah emission was detected in the spectra of iras 08572 + 3915 , iras 23060 + 0505 , and iras 20460 + 1925 . it was suggested that iras 08572 + 3915 is powered by highly obscured agn activity , while iras 23060 + 0505 and iras 20460 + 1925 are powered by less - obscured agn activity . 3 . for ic 694 , arp 220 , and ngc 6240 , while the ew@xmath29 values are as large as those of starburst galaxies , the @xmath1/@xmath2 ratios after correction for screen dust extinction are a factor of @xmath33 smaller . whether these three irlgs are powered by starburst activity only or obscured agn activity contributes significantly is not clear because actual dust extinction correction factor could be higher than our estimate and the scatter of the intrinsic @xmath1/@xmath2 ratios for starburst activity is poorly known . our energy diagnoses reinforced the previous suggestions that agn activity is the dominant energy source in warm irlgs . we thank dr . t. kerr , dr . j. davies , t. carroll , and t. wold for their support during the ukirt observing runs . we are grateful to dr . d. b. sanders and the anonymous referee for their useful comments . a. t. tokunaga and h. ando give mi the opportunity to work at the university of hawaii . l. good kindly proofread this paper , and k. teramura modified fig . the united kingdom infrared telescope is operated by the joint astronomy centre on behalf of the u.k . particle physics and astronomy research council . mi is financially supported by the japan society for the promotion of science during his stay at the university of hawaii . ccd gratefully acknowledges the national research council ( usa ) , the naval research laboratory , and the office of naval research for their generous support , and the sponsorship of dr . j. fischer . our new spectra ( mostly higher signal - to - noise ratios and spectral resolution ) are compared with old spectra in the literature . a flux - calibrated 34 @xmath0 m spectrum taken with a 2@xmath487 aperture was presented by dennefeld & desert ( 1990 ) . the overall spectral shape and flux level in the new spectrum are consistent with the previous spectrum . using narrow band filters , satyapal et al . ( 1999 ) estimated the flux of the 3.3 @xmath0 m pah emission to be 8.5 @xmath17 10@xmath28 ergs s@xmath6 @xmath27 within a 7@xmath19 aperture . this is @xmath340% higher than our estimate , but this difference does not affect our discussion significantly . low - resolution spectra taken using a circularly variable filter with a 8@xmath487 aperture were presented by rieke et al . they showed strong 3.3 @xmath0 m pah emission . the new spectra show slightly smaller continuum flux levels but similar spectral shapes . a flux - calibrated 34 @xmath0 m spectrum taken with a 3@xmath488 aperture was presented by imanishi et al . the new spectrum provides much higher signal - to - noise ratios . the spectral shape and flux level are consistent with the previous spectrum . a flux - uncalibrated 34 @xmath0 m spectrum was presented by wright et al . it showed weak 3.3 @xmath0 m pah emission and weak 3.4 @xmath0 m carbonaceous dust absorption with an optical depth of @xmath30.05 . the new spectrum is consistent with the previous spectrum in the spectral shape and the optical depth of the carbonaceous dust absorption . a flux - uncalibrated 34 @xmath0 m spectrum was presented by wright et al . it was dominated by the 3.4 @xmath0 m carbonaceous dust absorption feature , and its optical depth was estimated to be @xmath30.9 ( pendleton 1996 ) . both the spectral shape and the optical depth in the new spectrum are consistent with the previous spectrum . a low - resolution spectrum taken using a circularly variable filter with a 7@xmath485 aperture was presented by moorwood ( 1986 ) . both the new and old spectra show the pah emission at 3.3 @xmath0 m and 3.4 @xmath0 m . the flux level in our new spectrum is smaller than the old spectrum because of smaller aperture size . alonso - herrero , a. , rieke , g. h. , rieke , m. j. , & scoville , n. z. 2000 , apj , 532 , 845 armus , l. , heckman , t. m. , & miley , g. k. 1989 , apj , 347 , 727 blain , a. w. , kneib , j. -p . , ivison , r. j. , & smail , i. 1999 , apj , 512 , l87 brand , p. w. j. l. , moorhouse , a. , burton , m. g. , geballe , t. r. , bird , m. , & wade , r. 1988 , apj , 334 , l103 bridger , a. , wright , c. s. , & geballe , t. r. 1994 , in mclean i. , ed . , infrared astronomy with arrays : the next generation , kluwer , dordrecht , p.537 clavel , j. , et al . 2000 , a&a , 357 , 839 dennefeld , m. , & desert , f. x. 1990 , a&a , 227 , 379 dudley , c. c. 1999 , mnras , 307 , 553 dudley , c. c. , wynn - williams c. g. 1997 , apj , 488 , 720 elbaz , d. , et al . 1998 , proceedings 34th liege international astrophysics colloquium ` the next generation space telescope : science drivers and technological challenges ' liege , belgium , p.47 , ( astro - ph/9807209 ) engelbracht , c. w. , rieke , m. j. , rieke , g. h. , kelly , d. m. , & achtermann , j. m. 1998 , apj , 505 , 639 frogel , j. a. , gillett , f. c. , terndrup , d. m. , & vader , j. p. 1989 , apj , 343 , 672 genzel , r. , & cesarsky , c. j. 2000 , ara&a , in press ( astro - ph/0002184 ) genzel , r. , weitzel , l. , tacconi - garman , l. e. , blietz , m. , cameron , m. , krabbe , a. , & lutz , d. 1995 , apj , 444 , 129 genzel , r. , et al . 1998 , apj , 498 , 579 glass , i. s. , & moorwood , a. f. m. 1985 , mnras , 214 , 429 goldader , j. d. , joseph , r. d. , doyon , r. , & sanders , d. b. 1995 , apj , 444 , 97 heisler , c. a. , norris , r. p. , jauncey . d. l. , reynolds , j. e. , & king , e. a. 1998 , mnras , 300 , 1111 hill , g. j. , wynn - williams , c. g. , & becklin , e. e. 1987 , apj , 316 , l11 imanishi , m. , & ueno , s. 2000 , apj , 535 , 626 imanishi , m. , terada , h. , goto , m. , & maihara , t. 1998 , pasj , 50 , 399 imanishi , m. , terada , h. , sugiyama , k. , motohara , k. , goto , m. , & maihara , t. 1997 , pasj , 49 , 69 ivezic , z. , nenkova , m. , & elitzur , m. 1999 , user manual for dusty , university of kentucky internal report , accessible at http://www.pa.uky.edu/@xmath49moche/dusty/ ( astro - ph/9910475 ) joy , m. , telesco , c. m. , decher , r. , lester , d. f. , harvey , p. m. , rickard , l. j. , & bushouse , h. 1989 , apj , 339 , 100 kalas , p. , & wynn - williams , c. g. 1994 , apj , 434 , 546 kewley , l. j. , heisler , c. a. , dopita , m. a. , sutherland , r. , norris , r. p. , reynolds , j. , & lumsden , s. 2000 , apj , 530 , 704 kim , d. -c . , & sanders , d. b. 1998 , apjs , 119 , 41 laureijs , r. j. , et al . 2000 , a&a , in press ( astro - ph/0005076 ) lonsdale , c. j. , lonsdale , c. j. , & smith , h. e. 1992 , apj , 391 , 629 lonsdale , c. j. , smith , h. e. , & lonsdale , c. j. 1993 , apj , 405 , l9 lutz , d. , et al . 1996 , a&a , 315 , l269 lutz , d. , spoon , h. w. w. , rigopoulou , d. , moorwood , a. f. m. , genzel , r. 1998 , apj , 505 , l103 maloney , p. r. 1999 , ap&ss , in press , the ringberg workshop , `` ultraluminous galaxies : monsters or babies '' ( astro - ph/9903275 ) mathis , j. s. 1990 , ara&a , 28 , 37 moorwood , a. f. m. 1986 , a&a , 166 , 4 mountain , c. m. , robertson , d. j. , lee , t. j. , & wade , r. 1990 , proc . spie , 1235 , 25 mouri , h. , kawara , k. , taniguchi , y. , & nishida , m. 1990 , apj , 356 , l39 pendleton , y. 1996 , in the cosmic dust connection , greenberg , j. m. , ed . , kluwer academic publishers , p.71 pendleton , y. j. , sandford , s. a. , allamandola , l. j. , tielens , a. g. g. m. , & sellgren , k. 1994 , apj , 437 , 683 rice , w. , lonsdale , c. j. , soifer , b. t. , neugebauer , g. , kopan , e. l. , lloyd , l. a. , de jong , t. , & habing , h. j. 1988 , apjs , 68 , 91 rieke , g. h. , & lebofsky , m. j. 1985 , apj , 288 , 618 rieke , g. h. , curti , r. m. , black , j. h. , kailey , w. f. , mcalary , c. w. , lebofsky , m. j. , & elston , r. 1985 , apj , 290 , 116 rieke , g. h. , lebofsky , m. j. , thompson , r. i. , low , f. j. , & tokunaga , a. t. 1980 , apj , 238 , 24 rigopoulou , d. , spoon , h. w. w. , genzel , r. , lutz , d. , moorwood , a. f. m. , & tran , q. d. 1999 , aj , 118 , 2625 roche , p. f. , aitken , d. k. , smith , c. h. , & ward , m. j. 1991 , mnras , 248 , 606 rudy , r. j. , levan , p. d. , & rodriguez - espinosa , j. m. 1982 , aj , 87 , 598 sanders , d. b. , & mirabel , i. f. 1996 , ara&a , 34 , 749 satyapal , s. , watson , d. m. , forrest , w. j. , pipher , j. l. , fischer , j. , greenhouse , m. a. , smith , h. a. , & woodward , c. e. 1999 , apj , 516 , 704 schmitt , h. r. , kinney , a. l. , calzetti , d. , & bergmann , t. s. 1997 , aj , 114 , 592 scoville , n. z. , et al . 2000 , aj , 119 , 991 smith , h. e. , lonsdale , c. j. , & lonsdale , c. j. 1998 , apj , 492 , 137 soifer , b. t. , sanders , d. b. , madore , b. f. , neugebauer , g. , danielson , g. e. , elias , j. h. , lonsdale , c. j. , & rice , w. l. 1987 , apj , 320 , 238 soifer , b. t. , et al . 2000 , aj , 119 , 509 sugai , h. , malkan , m. a. , ward , m. j. , davies , r. i. , & mclean , i. s. 1997 , apj , 481 , 186 sugai , h. , davies , r. i. , malkan , m. a. , mclean , i. s. , usuda , t. , & ward , m. j. 1999 , apj , 527 , 778 surace , j. a. , & sanders , d. b. 1999 , apj , 512 , 162 surace , j. a. , sanders , d. b. , & evans , a. s. 2000 , apj , 529 , 170 tokunaga , a. t. , sellgren , k. , smith , r. g. , nagata , t. , sakata , a. , & nakada , y. 1991 , apj , 380 , 452 tokunaga , a. t. 2000 , in allen s astrophysical quantities , ed . a. n. cox ( 4th ed : aip press : springer ) , chapter 7 , p.143 tully , r. b. 1988 , nearby galaxies catalog , cambridge university press , cambridge , p.10 vader , j. p. , frogel , j. a. , terndrup , d. m. , & heisler , c. a. 1993 , aj , 106 , 1743 veilleux , s. , kim , d. -c . , & sanders , d. b. 1999a , apj , 522 , 113 veilleux , s. , kim , d. -c . , sanders , d. b. , mazzarella , j. m. , & soifer , b. t. 1995 , apjs , 98 , 171 veilleux , s. , sanders , d. b. , & kim , d. -c 1999b , apj , 522 , 139 vignati , p. , et al . 1999 , a&a , 349 , l57 voit , g. m. 1992 , mnras , 258 , 841 wright , g. s. , bridger , a. , geballe , t. r. , & pendleton , y. 1996 , new extragalactic perspectives in the new south africa , kluwer academic publishers , block , d. l. , & greenberg , j. m. , ( eds ) , p.143 zhou , s. , wynn - williams , c. g. , & sanders , d. b. 1993 , apj , 409 , 149 note . ( 1 ) : object name . col . ( 2 ) : energy source derived by this work . `` agn '' and `` sb '' mean agn - dominated and starburst - dominated , respectively . `` sb+agn '' means both kinds of activity contribute roughly equally . ( 3 ) : energy source derived based on the ratio between the 7.7 @xmath0 m pah peak flux and 7.7 @xmath0 m continuum flux . the ratios are shown in parentheses . adopted from rigopoulou et al . ( 1999 ) except for iras 05189@xmath102524 for which the ratio is adopted from laureijs et al . the average ratios for starburst galaxies and agns are @xmath33.0 and @xmath30.2 , respectively ( rigopoulou et al . 1999 ) . those with the ratios of @xmath111.0 and @xmath151.0 are classified as `` agn '' and `` sb '' , respectively ( rigopoulou et al . 1999 ) . ( 4 ) : energy source derived from high spatial resolution 825 @xmath0 m imaging ( soifer et al . ( 5 ) : energy source derived from 2 @xmath0 m spectroscopy ( veilleux et al . 1999b ) . those with a detected broad emission line whose luminosity is high enough to account for the far - infrared luminosity by agn activity are classified as `` agn ''

work belongs to the most basic notions of classical mechanics and also presents one of the cornerstones of thermodynamics . with the recent experimental progress in the fields of cold atoms@xcite , nano - mechanics , and opto - mechanics , @xcite to name just a few , there is an urgent need of a theoretical foundation of what work means in quantum mechanics and how it can be defined in an operational way . one reason why work does not simply fall into the category of observables like position , linear and angular momentum , and energy , comes from the fact that it is meant to characterize a process rather than an instantaneous state of a system @xcite . even in the simple case of a thermally isolated system the determination of work requires an interference of the system with a measurement device at two times . for a classical system , the interaction between the system and the measurement device can be made arbitrarily weak and , hence , the back - action can be arbitrarily small , without implying a principle limitation of the precision of the measurement . however , for a quantum systems , the back - action of the measurement device modifies the state of the system ; it consequently influences the outcome of a subsequent contact , therefore influencing the final value of the work @xcite . the paper is organized as follows . in section [ classical ] we survey various definitions of work that classically are equivalent to each other . we review in more detail the operational definition in terms of two projective energy measurements in section [ tema ] . in section [ twoenergy ] we discuss a gaussian energy measurement scheme which then is employed in a two gaussian energy measurement device . a recently proposed approach to determine work by a single , necessarily generalized measurement @xcite is discussed in section[meter ] for a gaussian pointer . a work - meter that functions in that way operates with a precision determined by a parameter combining the strength of the measurement and the covariance matrix of the gaussian initial pointer state . implementing a high precision the result of two projective energy measurements is recovered while the use of low precision leads to broad distributions yet with a finite average . a spin in a suddenly changing magnetic field serves as an example in section [ example ] . section [ con ] concludes the paper . restricting ourselves to thermally isolated systems we begin with the discussion of work in classical systems . when a classical system stays at a phase space point @xmath0 its energy is determined by the value of a conveniently chosen hamiltonian @xmath1 @xcite which also governs the dynamics of the system . a change of the parameter @xmath2 according to a protocol @xmath3 alters the energy and hence work will be applied to or taken from the system . we shall adopt the inclusive definition of work @xcite according to which the work is given by the energy - difference at the end and the beginning of the force protocol , @xmath4 with @xmath5 being the phase space point that evolves from @xmath6 according to the hamiltonian dynamics , @xmath7 where @xmath8 denotes the poisson bracket . as already mentioned , within the time - range @xmath9 $ ] the parameter @xmath10 changes according to a prescribed protocol @xmath11 . for classical systems an equivalent definition of work is based on the fact that the energy difference can be expressed as an integral of the total time - derivative of the hamiltonian , which coincides with the respective partial derivative yielding the work as integral of the supplied power , i.e. , @xmath12 where @xmath13 denotes the time - derivative of the force parameter @xmath10 . any adaptation of the two work definitions ( [ whhcl ] ) and ( [ wpower ] ) to quantum systems must take into account the back - actions of those measurements that have to be performed in order to determine the work . it is this inherently quantum - mechanical requirement that renders the two classically equivalent definitions ( [ whhcl ] ) and ( [ wpower ] ) inequivalent @xcite . based on the energy difference definition ( [ whhcl ] ) work can be determined by the _ two energy measurement approach _ @xcite which employs two projective measurements of energy at the end and the beginning of the force protocol . due to the projective nature of these measurements , two energy - eigenvalues @xmath14 and @xmath15 are obtained . from these eigenvalues the work in a single realization of the protocol is determined as the difference @xmath16 . with this definition the statistics of work complies with the fluctuation relations of jarzynski @xcite and crooks @xcite for systems initially staying in a canonical initial state @xcite . a quite different picture emerges if one tries to implement the power based approach @xcite . this , in principle , requires a continuous observation of the system in such a way that the time - evolution of the power can be inferred . even if this monitoring is not performed by means of projective measurements but in a weaker way , a continuous observation will freeze the dynamics of the system due to the quantum zeno effect @xcite . for relatively slowly varying protocols one might try to circumvent the zeno effect by restricting the observations to a discrete set of times with a suitably chosen strength of observation . yet the resulting statistics of work turn out to be essentially determined by the number and strength of the observations and only to a lesser extent by the force protocol . there are hardly any relations between the statistics obtained by the two energy measurement and the power approach . in particular , the fluctuation relations of crooks and jarzynski are only satisfied for the power approach if the hamiltonians at all different parameter values commute with each other @xcite . this is not in contradiction with the fact that measurements during the force protocol leave these fluctuation relations unchanged as long as the work is determined by two projective measurements at the beginning and the end of the force protocol @xcite . we finally note that in various attempts to formulate a `` quantum thermodynamics '' @xcite work is understood in a way such that its mean - value coincides with the difference of the mean - values of the energies at the end and the beginning of the force - protocol , however , without taking into account the back - action of the first onto the second measurement . in order to distinguish this notion of a work - like quantity from operational definitions that include the back - action of measurements we call it `` untouched work '' @xcite . in this sense , work is always untouched in classical mechanics . the defining property of untouched work hence postulates that its average @xmath17 can be expressed as @xmath18 where the density matrix @xmath19 describes the initial state and @xmath20 denotes the unitary time - evolution of the system over the extension of the force protocol , following as the solution of the schrdinger equation @xmath21 where @xmath22 denotes the unit operator on the hilbert space of the system . for later use , we introduce the spectral resolution of the time - dependent hamiltonian @xmath23 in terms of its eigenvalues @xmath24 and corresponding projection operators @xmath25 satisfying @xmath26 , @xmath27 and @xmath28 . the hamiltonian can then be represented as @xmath29 we will make frequent use of this representation in the sequel for the initial and final times @xmath30 , respectively . a way to obtain the average untouched work would be to prepare two copies of an ensemble representing the initial state of the system . the first ensemble is then used to determine its average energy in the standard way by projective measurements of the initial energy . the ensemble which is generated by this measurement need not coincide with the one prior to the measurement and is discarded after the measurement . the force protocol is applied on the members of the second ensemble and upon its completion the energy is measured and its average calculated . the difference of the two average energies coincides with the average of the untouched energy . it only agrees with the average work obtained by the above mentioned two energy measurement approach if the discarded ensemble was identical with the original one . this is the case for an initial state that is stationary with respect to the initial hamiltonian . the given prescription though is incomplete in that it only determines an average value but does not allow to determine fluctuations in a meaningful way @xcite . attempts to define the work as an observable in terms of the difference of the initial and final hamiltonians in the heisenberg - picture @xcite disregard the inherent process - dependence of work @xcite . also for a canonical initial state , the jarzynski and crooks relations do not hold with this definition apart from the trivial , yet untypical case of commuting hamiltonians , @xmath31=0 $ ] for all @xmath32 . in a recent effort , allahverdyan @xcite suggested a characterization of untouched work as a fluctuating quantity which indeed yields ( [ varw ] ) for the average . the statistics of this hypothetical object is determined in @xcite by a pdf - like function @xmath33 the weight @xmath34 is a quasi - probability for the eigenvalues of the two hamiltonians @xmath35 and @xmath36 based on the terletsky - margenau - hill distribution which is given by @xmath37 with @xmath38 denoting the anti - commutator of two operators @xmath39 and @xmath40 . the fact that , as a quasi - probability , the terletsky - margenau - hill distribution assumes negative values for non - commuting hamiltonians may also translate to @xmath41 which hence is not a probability density function and consequently fails to characterize a proper random variable . the joint probability @xmath42 of observing the system at an eigenvalue @xmath15 of the hamiltonian @xmath35 by projective measurements at the beginning of the force protocol @xmath11 and the eigenvalue @xmath14 of the hamiltonian @xmath36 at its end is given by @xcite @xmath43 where the projection operators @xmath25 , at times @xmath30 , are defined by the spectral representations of the respective hamiltonians , see ( [ hspec ] ) . it is important to note that due to the first energy measurement the initial density matrix @xmath19 is projected onto the eigenbasis of the initial hamiltonian @xmath35 . in other words , all non - diagonal elements of the initial density matrix in the energy eigenbasis are erased by the first measurement . this will influence the subsequent time evolution unless the initial density matrix and hamiltonian commute with each other . moreover , this projection excludes the use of correlations between energy eigenstates as a `` resource '' to extract work out of a system by a unitary process , as it has been recently suggested @xcite . the probability density function ( pdf ) @xmath44 of finding the work @xmath45 by means of two projective energy measurements ( pem ) can be expressed in terms of the joint probability ( [ pmn ] ) as @xcite @xmath46 in a realization of the force - protocol for which the work @xmath45 was obtained the state @xmath47 , immediately after the second energy measurement , is proportional to a linear , positive map @xmath48 acting on the initial density matrix @xmath19 with a proportionality constant inversely proportional to the work pdf @xmath49 . hence , the post - measurement state can be written as @xmath50 where the so - called operation @xmath48 @xcite is given by @xmath51 if the energies are measured but the work is not registered the system is found in the so - called non - selective post - measurement state @xmath52 , given by @xmath53 leading with ( [ rphir ] ) to @xmath54 this result immediately follows by a two - fold application of the lders - von - neumann rule @xcite . with the requirement of a normalized selective post - measurement state ( [ rphip ] ) , the work pdf follows from the operation acting on the initial system density matrix as @xmath55 hence , the operation @xmath48 can be considered as a complete characterization of the work measurement . it is a linear map on hilbert space operators with finite trace that preserves the positivity , i.e. , @xmath56 implies @xmath57 and is contractive , i.e. , @xmath58 for all @xmath56 . also more general measurement strategies than the projective one are characterized by operations with these mathematical properties of being linear , positivity preserving and contractive maps on the space of linear hilbert space operators with finite trace . coming back to projective measurements we finally want to note the crooks relation @xcite . it connects the pdfs @xmath59 and @xmath60 for a protocol @xmath61 and the time reversed protocol @xmath62@xcite : @xmath63 this requires that the initial density matrices of the forward and the backward processes must have boltzmann weights as diagonal elements in the energy eigenbasis , i.e. @xmath64 , where @xmath65 is the degeneracy of the energy @xmath24 and @xmath66 the partition function for @xmath30 . the free energy difference @xmath67 is defined in terms of the partition functions of the initial states of the forward and the backward processes , @xmath68 . we note that most often , canonical density matrices are considered as initial states of the forward and backward processes . we point out , however that for the crooks relation to hold , _ only the diagonal elements _ of these density matrices need to have the form of boltzmann factors whereas the non - diagonal elements may be arbitrary , as long the density matrix is positive . instead of the projective energy measurements one may employ generalized measurements of energy for determining work . as demonstrated by venkatesh _ @xcite work distributions obtained by means of generalized measurements typically do not satisfy fluctuation relations . however , for a whole class of generalized energy measurements , in particular for gaussian measurements , modified fluctuation theorems exist for which the modifications are solely determined by the properties of the measurement apparatus and hence are independent of the force protocol @xcite . we here choose a less formal approach compared to @xcite by adapting von neumann s model @xcite to energy measurements . in von neumann s model a measurement apparatus , here called the `` pointer '' , is coupled to the system during a short time @xmath69 . within this time the value of the observable to be measured is transcribed onto the scale of the pointer while ideally @xmath69 is small enough that the unitary dynamics of the system is negligible @xcite . the information transfer is formally achieved by a unitary operator @xmath70 which when acting on the state of the pointer shifts the pointer position by an amount proportional to the @xmath23 . here , @xmath71 is the momentum operator which is canonically conjugate to the position @xmath72 of the pointer where @xmath73 is the projection operator on the eigenstate @xmath74 with eigenvalue @xmath75 . the unitary operator @xmath76 describes the time evolution of system and pointer governed within the time span @xmath69 by the interaction hamiltonian @xmath77 where @xmath78 describes the coupling strength . the effective interaction strength @xmath79 between system and pointer therefore is given by @xmath80 . system and pointer are supposed to be uncorrelated before the interaction has taken place . then the total density matrix is given by the direct product @xmath81 of the system density matrix @xmath82 and the pointer density matrix @xmath83 immediately before the interaction . the system pointer interaction is followed by a projective measurement of the pointer position @xmath75 . the state of the system immediately after the measurement can be expressed as @xmath84 where the operation @xmath85 is composed of the interaction and the projective measurement , yielding @xmath86 here , @xmath87 denotes the trace over the hilbert space of the pointer , @xmath88 is the density matrix of the pointer in the position representation , and @xmath89 gives the probability of observing the pointer in the eigenstate @xmath74 . it is determined by the normalization of the post - measurement state @xmath90 and therefore becomes @xmath91 we now assume that the initial state of the pointer is given by a gaussian density matrix with vanishing averages of position and momentum , i.e. , @xmath92 and @xmath93 . the density matrix @xmath83 is then uniquely determined by the second moments @xmath94 , @xmath95 and @xmath96 . note that the positivity of the density matrix is equivalently expressed by the inequality @xmath97 where @xmath98 denotes the anti - commutator . in position representation a gaussian density matrix is given by @xcite @xmath99 where @xmath100 is a quadratic form in the position variables @xmath75 and @xmath101 defined as @xmath102 ^ 2 \right \}\\ & = \frac{\langle p^2 \rangle}{2 \hbar^2 } \left [ x - y \right ] ^2 + \frac{1}{2\langle x^2 \rangle } \\ & \quad\times \left [ \frac{1}{2 } ( x+y ) - i \frac{\langle \ { x , p \}\rangle}{2 \hbar } ( x - y ) \right ] ^2\ : . \label{mu } \end{split}\ ] ] due to the vanishing average pointer position , @xmath92 , the pointer position @xmath75 presents an unbiased estimate of the system energy having the value @xmath103 . hence , the pdf @xmath104 to find the system at energy @xmath105 is given by @xmath106 where @xmath107 is defined by @xmath108 for a state with sharp energy @xmath109 , i.e. @xmath110 one obtains from ( [ pe ] ) with @xmath111 the conditional probability @xmath112 hence , @xmath107 specifies the inaccuracy of the energy measurement . even though the energy pdf ( [ pe ] ) agrees with the one obtained in @xcite the operation @xmath113 characterizing the measurement generally differs from the form assumed in @xcite . the latter is defined as @xmath114 in terms of a single gaussian kraus operator @xmath115 . one obtains this particular form from ( [ phix ] ) only if the initial pointer density matrix corresponds to a pure gaussian state in position representation , given by @xmath116 where @xmath117 in this case @xmath118 and consequently @xmath119 holds . for a general gaussian initial state specified by ( [ sigmag ] ) and ( [ mu ] ) the operation @xmath120 becomes @xmath121 ^ 2 } \\ & \quad \times \pi_n(t ) \rho \pi_{n'}(t)\ : , \label{phig1 } \end{split}\ ] ] where @xmath122 controls the contribution of non - diagonal elements on the post - measurement state . for comparison , the operation @xmath123 as defined above in terms of a single gaussian kraus operator @xmath124 becomes @xmath125 } \\ & \quad \times \pi_n(t ) \rho \pi_{n'}(t)\ : . \label{phime } \end{split}\ ] ] the subsequent action of @xmath126 followed by the unitary time evolution @xmath127 and the final operation @xmath128 results in an operation characterizing the joint occurrence of the energies @xmath105 and @xmath129 at the beginning and the end of the force protocol . it is used in the following subsection for characterizing a work measurement composed of two gaussian energy measurements . the occurrence of energies @xmath105 and @xmath129 by two gaussian measurements separated by the time span of the unitary time evolution under the force protocol is characterized by the composed operation @xmath130 which acts on a system density matrix @xmath82 as @xmath131 considering only the resulting work @xmath132 one introduces the according operation @xmath133 @xcite which becomes @xmath134 } \\ & \quad \times e^{-\frac{1}{4 \sigma^2_e } \left [ w- ( w_{m , n } + w_{m',n'})/2 + i \frac{\langle \ { x , p \}\rangle}{2 \hbar } ( w_{m , n } - w_{m'n ' } ) \right ] ^2 } \pi_m(\tau ) u_\lambda \pi_n(0 ) \rho \pi_{n'}(0 ) u^\dagger_\lambda \pi_{m'}(\tau)\ : , \label{phiw2 } \end{split}\ ] ] where @xmath135 would be a possible work value if the energies were measured projectively . the non - selective post - measurements state @xmath136 is given by the integral of @xmath137 over @xmath45 according to ( [ rhous ] ) yielding @xmath138 } \\ & \quad \times \pi_m(\tau ) u_\lambda \pi_n(0 ) \rho \pi_{n'}(0 ) u^\dagger_\lambda \pi_{m'}(\tau)\ : , \end{split } \label{run2}\ ] ] the pdf @xmath139 to find the work @xmath45 with two gaussian energy measurements is given by the trace of ( [ phiw2 ] ) giving @xmath140 ^ 2 } e^{-\frac{1}{4 \sigma^2_e } \left [ w - e_m(\tau ) + \frac{1}{2 } ( e_n(0 ) + e_{n'}(0 ) ) - i \frac{\langle \ { x , p \ } \rangle}{2 \hbar } ( e_n(0 ) - e_{n'}(0 ) ) \right ] ^2}p_\lambda(m , n , n')\ : , \label{p2ew } \end{split}\ ] ] where @xmath141 note that the diagonal elements of @xmath142 with @xmath143 coincide with the joint probability @xmath42 obtained by two projective energy measurements , see ( [ pmn ] ) . in analogy to ( [ rphip ] ) the selective post - measurement state is given by @xmath144 . a detailed discussion of the results ( [ phiw2][p2ew ] ) is presented in the context of the similar expressions for the work meter discussed next . based on a recent idea put forward by roncaglia , cerisola and paz @xcite , recently , de chiara , roncaglia and paz @xcite suggested a method to determine work by means of a single generalized measurement . for the sake of completeness we describe the principle of operation of such a work meter following ref . @xcite and discuss its functioning in comparison with two gaussian energy measurements . based thereon , the limiting cases of most accurate and maximally imprecise measurements are investigated as well . the work meter consists of an auxiliary system with a pointer . this pointer interacts twice with the system . the first interaction takes place immediately before the force protocol sets in . it leads to a shift of the pointer proportional to the momentary energy of the system . as for a generalized energy measurement described above the interaction is given by a unitary transformation @xmath145 transferring the negative value of the system energy to the pointer . after the first interaction has ended the position of the pointer is not read out but rather left unchanged until the force protocol has ended . then , a second interaction between system and the pointer moves the pointer by an amount that is proportional to the final energy . accordingly , the interaction is described by the unitary operator @xmath146 . only after the completion of this interaction , the pointer position is measured projectively . the whole process consisting of the two unitary pointer system interactions , interrupted by the unitary system evolution under the force protocol and the final position measurement can be expressed by a total operation @xmath147 characterizing the non - normalized post - measurement state of the system @xmath148 as for the generalized energy measurement , @xmath83 denotes the density matrix of the pointer before the first interaction has taken place . using for @xmath83 the same gaussian form ( [ sigmag ] ) , ( [ mu ] ) together with the pointer calibration @xmath149 , the respective operation becomes @xmath150 ^ 2}\\ & \quad \times \pi_m(\tau ) u_\lambda \pi_n(0 ) \rho(0 ) \pi_{n'}(0 ) u^\dagger_\lambda \pi_{m'}(\tau)\ : . \label{phiw } \end{split}\ ] ] this expression is similar in its structure to the operation ( [ phiw2 ] ) but contains two differences . as a first difference one notices the way how the non - diagonal terms with @xmath151 or @xmath152 are suppressed by the first exponential terms in the sums on the right sides of ( [ phiw2 ] ) and ( [ phiw ] ) . the exponent is proportional to @xmath153 for two energy measurements and @xmath154 for the work - meter . the second difference stems from the fact that , in the two measurement approach , each energy measurement contributes to the variance of the work and therefore the width of the second exponential factor is twice as large for the work - meter . the non - selective post - measurement state @xmath155 imposed by the work meter ( wm ) is given by @xmath156 it also differs by the weights of the non - diagonal terms from @xmath136 , cf . ( [ run2 ] ) . the work pdf @xmath157 becomes @xmath158 ^ 2}%\\&\quad \times e^{-\frac{1}{2 \sigma^2_e } \left [ w - e_m(\tau ) + \frac{1}{2}(e_n(0 ) + e_{n'}(0 ) ) - i \frac{\langle \ { x , p \}\rangle}{2 \hbar}(e_n(0)-e_{n'}(0 ) ) \right ] ^2 } p_\lambda(m , n , n')\ : , \label{pwp } % \end{split}\ ] ] due to the trace the non - diagonal terms @xmath159 , which add to the operation @xmath48 , do not enter the work pdf . those non - diagonal terms with @xmath160 contribute with the same exponential weight as for the work pdf resulting from the gaussian two energy measurement approach . these terms vanish only for an initially stationary state , i.e. if @xmath161 = 0 $ ] . then all non - diagonal elements @xmath142 with @xmath160 vanish and the work probability reduces to the following expression : @xmath162 ^ 2 } p_\lambda(m , n , n)\ : . \label{pwd}\ ] ] here , the superscript d stands for `` diagonal '' indicating the restriction to initial density matrices that commute with the initial hamiltonian and hence are diagonal in the respective eigen - basis . because , as already mentioned , the diagonal elements @xmath163 agree with the joint probabilities ( [ pmn ] ) , which determine the work pdf ( [ pw ] ) for the two projective energy measurement scheme , the work pdf ( [ pwd ] ) becomes a smeared variant of the discrete work distribution obtained by projective energy measurements . the amount of broadening is uniformly determined by the variance @xmath164 , see ( [ se ] ) . the work probability ( [ pwd ] ) can now be expressed as the convolution of the projective work pdf and a gaussian which is solely determined by the properties of the pointer yielding @xmath165 the pointer properties only enter this expression in the combination of @xmath166 ; in particular , the work pdf does not depend on the pointer momentum variance @xmath96 nor on the symmetrized pointer position momentum cross - correlation @xmath167 . it is also worth noticing that the gaussian factor in the convolution is independent of the force protocol . for a canonical initial state @xmath168 the following modified crooks relation exists @xcite @xmath169 where @xmath170 denotes the work pdf for the time - reversed protocol . as in the original crooks relation ( [ cr ] ) , the free energy difference @xmath67 refers to the two initial equilibrium states of the backward and the forward processes . from the modified crooks relation one immediately obtains the modified jarzynski equality @xcite @xmath171 the correction factor solely depends on the pointer variance , but is independent of the force protocol . for a pure gaussian position state of the pointer ( pgp ) , specified by ( [ spure ] , [ ps ] ) , the work pdf ( [ pwp ] ) simplifies to read @xmath172 ^ 2}\\ & \quad \times e^{-\frac{1}{2 \sigma^2_e } \left [ w - e_m(\tau ) + \frac{1}{2}(e_n(0 ) + e_{n'}(0 ) ) \right ] ^2 } p_\lambda(m , n , n')\ : . \label{mpw } \end{split}\ ] ] in the section [ example ] this expression will be illustrated with an example . in the limit of large values of the pointer - system coupling strength @xmath79 , both variances @xmath173 and @xmath174 defined in ( [ se ] ) and ( [ snd ] ) , respectively , become arbitrarily small . this suppresses all non - diagonal contributions from terms with @xmath160 and further renders the range of deviating work values about each discrete energy difference @xmath175 very narrow . in the limit @xmath176 the work pdf of the work - meter approaches the two - projective energy measurement result ( [ pw ] ) , i.e. @xmath177 in order to sufficiently suppress all non - diagonal contributions to the sum in ( [ pwp ] ) , the variance @xmath178 needs to be substantially smaller than the smallest squared eigenenergy difference of the initial hamiltonian @xmath35 , i.e. , @xmath179 once the non - diagonal elements are sufficiently suppressed the variance @xmath107 must be chosen smaller than the minimum work difference in order to achieve their unambiguous resolution . hence , the following inequality must be satisfied @xmath180 additionally , as mentioned above , the inequality @xmath181 must be fulfilled . an illustrative example is presented in section [ example ] . the opposite limit of maximally imprecise measurements is reached for a vanishingly small measurement strength @xmath79 . one expects that in this limit the back - action of the measurement apparatus on the system will be strongly suppressed , however , most likely at the cost of a large measurement error . before confirming this qualitative picture on the basis of the work pdf ( [ pwp ] ) we first consider the mean value of the work @xmath182 in this particular limit . using ( [ pwp ] ) we obtain for the average work @xmath183 p_\lambda(m , n , n ' ) \ : . % \end{split } \label{mw}\ ] ] using ( [ pmnn ] ) one can further simplify this average by performing the sum on @xmath184 and splitting the remaining sum into its diagonal and non - diagonal contributions leading to @xmath185 where @xmath186 denotes the projection of the initial density matrix onto the eigen - basis of the initial hamiltonian @xmath187 . the expression ( [ aw ] ) for the average work is still exact . the first two terms on the right hand side coincide with the average work obtained from two projective energy measurements ; the sum contains contributions from non - diagonal elements of the initial density matrix . note that the imaginary term on the right hand side of ( [ mw ] ) , which is caused by a finite symmetrized pointer position momentum correlation , does not contribute to the average . in the imprecise limit @xmath188 the exponential term under the sum on the right hand side approaches unity . then , the sums in ( [ aw ] ) can be performed to yield the work as the difference of the energy expectation values at the end and the beginning of the force protocol , leading to @xmath189 as expected , this result does not contain any back - action of the first energy measurement and coincides with the untouched work average ( [ varw ] ) . hence , one might be tempted to expect that the imprecise limit could provide a physical access to untouched work . in order to study this limit in more detail , we consider the characteristic function of work @xcite , which is defined as the fourier transform of the work pdf . hence it becomes @xmath190 where @xmath191 is determined by the fourier transformed pointer density matrix @xmath192 with @xmath193 and @xmath194 the expression ( [ g ] ) with ( [ chi ] ) and ( [ cmnn ] ) is still exact . it can be further evaluated in the imprecise limit in which the first , u - independent exponential factor on the right hand side of ( [ chi ] ) can be replaced by unity . the triple sum in ( [ g ] ) can then be performed to yield the characteristic function in the imprecise limit , @xmath195 , reading @xmath196 where @xmath197 further , @xmath198 denotes the final hamiltonian in the heisenberg picture . the respective work pdf @xmath199 results as a sum of wide gaussians , i.e. , @xmath200 ^ 2}\\ & \quad \times p_\lambda(m , n , n ' ) \label{pilw } \end{split}\ ] ] provided the inverse fourier transform of @xmath201 exists the work pdf can also be expressed as a convolution with a gaussian of width @xmath202 , i.e. @xmath203 where @xmath204 is the inverse fourier transform of @xmath205 . before we further discuss the condition of existence and the properties of @xmath206 we note that the result given in the first line of ( [ pilw ] ) can also be directly obtained from the work pdf ( [ pwp ] ) by replacing the first exponential on the right hand side of ( [ pwp ] ) by unity . in this way all non - diagonal elements contribute equally . the expression ( [ glu ] ) coincides with the characteristic function of work obtained by two projective energy measurements @xcite only if the initial density matrix @xmath19 commutes with the hamiltonian @xmath35 at the beginning of the protocol . then the two exponents containing @xmath207 can be combined into a single term such that @xmath208 and @xmath209 hold . in general , though , the inverse fourier transform of @xmath201 leads to a divergent result due to the presence of the imaginary part of the coefficient @xmath210 . only if the symmetrized pointer position momentum cross - correlation vanishes , @xmath211 , @xmath210 is real and @xmath206 becomes a sum of the projective work pdf @xmath44 and further delta - contributions at work values that correspond to the differences @xmath212 between the final energy eigenvalues and the arithmetic averages of all pairs of initial eigenvalues . hence , for @xmath213 , the inverse fourier transform of the trace expression becomes @xmath214 the integral of @xmath206 over all possible work values is unity as follows from ( [ pilw ] ) with @xmath215 . because the projective work pdf @xmath44 itself is normalized the integral over the sum expression on the right hand side must vanish . this implies that either this sum vanishes for all values of @xmath45 or that it assumes both positive and negative values . the first case happens only if @xmath216 for all @xmath160 . one can show that for this to hold the initial density matrix must be stationary , i.e. @xmath217=0 $ ] . we conclude that in all other cases @xmath206 assumes both positive and negative values and therefore does not qualify as a proper pdf . the positivity of @xmath218 is only guaranteed by the convolution with a sufficiently broad gaussian which smooths out any negative contributions . in agreement with the expectation formulated in the beginning of this subsection the non - selective post - measurement state ( [ runwm ] ) converges to the state that is reached solely by the force protocol without the inference of the work meter , i.e. one obtains @xmath219 provided @xmath206 exists , the convolution in the second line of ( [ pilw ] ) can be evaluated in the limit @xmath220 yielding @xmath221 we conclude that the imprecise limit suppresses any influence of the interactions of the work measurement device and in this sense yields the untouched work but on the other hand does not carry any system - specific information on the work beyond its average . finally , we note that the limiting form ( [ pilw ] ) presents a valid approximation of the work pdf ( [ pwp ] ) if @xmath222 where both @xmath223 and @xmath224 are elements of the set @xmath225 that exhaust the total probability obtained from the diagonal elements @xmath226 up to a negligibly small fraction @xmath227 , such that @xmath228 . here , @xmath229 and @xmath230 are the minimum and maximum energies , respectively , out of this set of states contributing to the energy distribution in the initial state . the criterion ( [ snde ] ) is obtained from the fact that the non - diagonal elements of @xmath142 are proportional to @xmath231 , cf . ( [ pmnn ] ) , @xcite . due to the positivity of the density matrix @xmath19 , the absolute values of these non - diagonal elements of the density matrix are dominated by the diagonal elements as @xmath232 . to @xmath233 given by ( [ hi ] ) and ( [ hf ] ) , respectively , is displayed as it is generated by a work meter with a pure gaussian state in position representation . the level separation of the hamiltonian after the switch is given by @xmath234 . both panels refer to an initial density matrix ( [ rho0 ] ) with ground - state probability @xmath235 . in panel ( a ) the work pdf is displayed for vanishing non - diagonal elements of the initial density ( @xmath236 ) and different values of the work variance @xmath107 which are @xmath237 ( red ) , @xmath238 ( blue ) , @xmath239 ( black ) . at the smallest variance , the work pdf consists of clearly separated peaks , which overlap at the intermediate variance and are completely washed out at the large variance . in those cases where peaks can be identified they are always exactly located at the positions of the work values ( [ w12 ] ) obtained by projective measurements . these values are indicated by the four thin vertical black lines . panel ( b ) exemplifies the effect of a non - diagonal initial density matrix at the intermediate variance @xmath240 . as a reference , the solid blue line refers to the diagonal case ( @xmath236 ) whereas the red dashed curve corresponds to @xmath241 and the black dash - dotted line to @xmath242 . the non - diagonal elements of the initial density matrix @xmath19 lead to a substantial distortion of the work pdf . in particular , the positions of the maxima do no longer coincide with those of the projective work values.,width=302 ] ( [ varw ] ) as a function of the variance @xmath107 , see ( [ mw ] ) . all curves are for a density matrix of the type ( [ rho0 ] ) with @xmath235 . the average work is displayed for @xmath241 by the upper red curve , for @xmath243 , @xmath244 , @xmath245 and @xmath246 by the blue , green und lower red curves , respectively . the according asymptotic values given by ( [ mw ] ) are indicated by thin horizontal lines of respective color . the final energy is chosen as @xmath247 . the change in the average work is largest for the two limiting values @xmath248 and vanishes for @xmath236 . , width=302 ] ( narrow blue curve ) and @xmath249 ( broad blue curve ) is compared to the limiting result of the imprecise measurement limit for two values of the variance @xmath107 ( thin red curves ) . at the smaller value @xmath250 the gaussian factor of the convolution integral on the right hand side of ( [ pilw ] ) is not yet broad enough to suppress the negative contributions around @xmath251 . at the larger variance , the approximate work pdf only insignificantly differs from the exact one . the other parameters , @xmath235 and @xmath252 , are the same for all curves.,width=302 ] as a simple example we consider a spin 1/2 which undergoes a sudden change of the hamiltonian from its initial form @xmath253 to the final form @xmath254 where @xmath255 and @xmath256 are pauli - spin matrices . the central quantity @xmath257 can then be expressed in terms of the eigenfunctions @xmath258 and @xmath259 of @xmath260 and @xmath233 , respectively as @xmath261 the indices @xmath262 taking the values 1 and 2 refer to the ground and the exited state , respectively , of each hamiltonian . the scalar products are readily determined to yield @xmath263 . we specify the density matrix @xmath19 in the basis of @xmath260 as @xmath264 with @xmath265 and @xmath266 . choosing for the initial pointer state @xmath83 a pure gaussian state as defined in ( [ spure ] , [ ps ] ) , we obtain @xmath267\ : , \label{pw2l } \end{split}\ ] ] where the diagonal elements of the initial density matrix contribute with @xmath268 p \right . \\ & \quad \left . + \left [ e^{-\frac{1}{2 \sigma^2_e}(w - w_{1,2})^2 } + e^{-\frac{1}{2 \sigma^2_e}(w - w_{2,2})^2 } \right ] ( 1-p ) \right \}\ : . \end{split } \label{pdiag2l}\ ] ] the allowed projective work - values @xmath269 are given by @xmath270 figure [ f1 ] displays the work pdf ( [ pw2l ] ) for different variances @xmath107 and different initial system density matrices . while an increase of @xmath107 leads , as expected , to a broadening of the peaks , the presence of non - diagonal terms @xmath271 in the system s density matrix gives rise to further deformations and , in particular , also to shifts of these peaks . figure [ f2 ] demonstrates the transition of the average work from the accurate limit yielding the results of projective energy measurements to the limit of maximally imprecise measurements for different values of the non - diagonal elements of the initial system s density matrix . in the latter limit the approach to the average untouched work ( [ varw ] ) is confirmed . finally , in figure [ f3 ] the asymptotic work pdf ( [ pilw ] ) is compared with the full result ( [ pw2l ] ) for two different values of the variance @xmath107 . for the smaller one , a negative contribution resulting from negative parts of @xmath206 is visible whereas for the larger one the gaussian factor in ( [ pilw ] ) is sufficiently broad to suppress any negative part . as a result , the limiting distribution coincides fairly well with the exact one , even though with @xmath272 and @xmath273 the inequality ( [ snde ] ) is only marginally satisfied . we reviewed several attempts to specify work . in particular , we considered three different setups for which we compared the state of the considered system immediately after the completion of selective and non - selective measurements of work as well as the respective probabilities of finding a particular value of work . in the first scenario , work is determined in the standard way by two projective measurements of energy ; in the second scenario , also two energy measurements are performed in each of which the system is coupled to a gaussian pointer state and read out by projective measurements of the pointer state ; the third scenario , proposed by de chiara , roncaglia and paz in @xcite , the result of the first pointer - system interaction encoding the negative initial energy is stored in the pointer until the force protocol is completed . the final energy is transferred to the same pointer and added to the stored value . only then the pointer state is read out by a projective measurement . in all three cases the operations characterizing the selective as well as the non - selective post - measurement states and the distributions of work values are different . while the projective energy measurements select the elements of the according density matrix that are diagonal with respect to the measured hamiltonians , non - diagonal elements contribute to the post - measurements states generated by both gaussian strategies , however with different weights . moreover , when considering the work pdfs one finds broadened peaks instead of the sharp lines in the case of projective energy measurements . the widths of these peaks are governed by an effective measurement strength , @xmath274 , which is determined by the variance of the pointer position @xmath94 , in its initial state and a parameter @xmath80 combining the strength @xmath78 and duration @xmath275 of the interaction between system and pointer . a second variance - like quantity ( @xmath276 ) governs the influence of non - diagonal terms of the initial system density matrix . these terms not only add to the broadening but also cause shifts of the peaks of the work pdf relative to the possible work values obtained by projective energy measurements . the two variances are restricted by an inequality which suppresses the influence of non - diagonal elements of the initial density matrix if the broadening of the lines is small . on the other hand it entails huge broadening if the non - diagonal elements are only little suppressed . the disjoint parameter regimes of small @xmath202 and large @xmath277 correspond to the limiting cases of accurate and imprecise measurements . in the accurate limit the work statistics based on projective energy measurements is approached , whereas in the imprecise limit the work statistics is governed by an extremely broad gaussian distribution with an average of the untouched work @xmath17 and width @xmath278 . in this limit , the fluctuations are completely determined by the imprecision of the measurement device . therefore this limit is of minor practical use ; to achieve a reliable estimate of the average untouched work an extremely large number of realizations would be required . finally , we note that the fluctuation relations of jarzynski and crooks remain valid in its modified form only if the system initially stays in a canonical state . in contrast to the projective two energy measurement approach it is not sufficient if only the diagonal elements in the initial energy basis are given by boltzmann factors but additional finite non - diagonal elements are present . pt thanks the foundation for polish science ( fnp ) for providing him with an alexander von humboldt polish honorary research fellowship . 99 i. bloch , j. dalibard and w. zwerger , rev . phys . * 80 * , 885 - 964 ( 2008 ) . kippenberg and k.j . vahala , kerry j. , optics express * 15 * , 17172 ( 2007 ) . m. aspelmeyer , t.j . kippenberg and f. marquardt , rev . phys . * 86 * , 1391 ( 2014 ) . p. talkner , e. lutz and p. hnggi , phys . e * 75 * , 050102 ( 2007 ) . p. talkner , p. hnggi and m. morillo , phys . e * 77 * , 051131 ( 2008 ) . roncaglia , f. cerisola and j.p . paz , phys . * 113 * , 250601 ( 2014 ) . g. de chiara , a.j . roncaglia and j.p . paz , new j. phys . * 17 * , 035004 ( 2015 ) . note that for non - autonomous hamiltonian systems not any hamiltonian rendering the correct equations of motion gives a proper expression for the energy . however , with a convenient gauge transformation it can always be achived that the value of the hamiltonian corresponds to the instantaneous energy of the system @xcite . c. jarzynski , c. r. physique * 8 * , 495 ( 2007 ) . venkatesh , g. watanabe and p. talkner , new j. phys . * 17 * , 075018 ( 2015 ) . j. kurchan , arxiv : cond - mat/0007360 ( 2000 ) . h. tasaki , arxiv : cond - mat/0009233 ( 2000 ) . c. jarzynski , phys . lett . * 78 * , 2690 ( 1997 ) . crooks , phys . e. * 60 * , 2721 ( 1999 ) . for slightly less restrictive conditions on the initial state see the paragraph following ( [ cr ] ) below . m. campisi , p. talkner and p. hnggi , phys . lett . * 105 * , 140601 ( 2010 ) . g. watanabe , b.p . venkatesh , p. talkner , m. campisi and p. hnggi , phys . e * 89 * , 032114 ( 2014 ) . j. gemmer , m. michel and g. mahler , _ quantum thermodynamics _ , springer , berlin 2009 . we choose the name `` untouched work '' because the first registration of energy is assumed to leave the subsequent dynamics untouched . for the two - ensemble implementation of untouched work the fluctuations of work are given by the pdf @xmath279 . this expression , though does not have the correct classical limit and therefore can not be considered as a useful definition . allahverdyan and th.m . nieuwenhuizen , phys . e * 71 * , 066102 ( 2005 ) . a. engel and r. nolte , europhys . lett * 79 * , 10003 ( 2007 ) . m. campisi , p. hnggi and p. talkner , rev . phys . * 83 * , 771 ( 2011 ) . m. campisi , p. hnggi and p. talkner , rev . mod . phys . * 83 * , 1653 ( 2011 ) . p. hnggi and p. talkner , nat . phys . * 11 * , 108 ( 2015 ) . allahverdyan , phys . e * 90 * , 032137 ( 2014 ) . r. alicki and m. fannes , phys . e * 87 * , 042123 ( 2013 ) . m. perarnau - llobet , k.v . hovhannisyan , m. huber , p. skrzypczyk , n. brunner and a. acn , phys . x * 5 * , 041011 ( 2015 ) . k. kraus , _ states , effects and operations _ , springer , berlin 1983 . g. lders , ann . ( leipzig ) * 8 * , 322 ( 1951 ) ; english translation : ann . phys . ( leipzig ) * 15 * , 663 ( 2006 ) . j. von neumann , _ mathematische grundlagen der quantenmechanik _ , springer verlag , berlin , 1932 ; english translation : _ mathematical foundation of quantum mechanics _ princeton , 1955 . p. talkner and p. hnggi , j. phys . a * 40 * , f569 ( 2007 ) . in order that the time - reversed process can be realized , the hamiltonian of the system must satisfy the time - reversal symmetry relation @xmath280 for all times @xmath281 $ ] with @xmath282 being the time - reversal operator and @xmath283 the parity of the parameter @xmath2 under time - reversal . for the sake of simplicity , we consider a parameter @xmath10 that transforms evenly under time - reversal . for odd parameters , its sign must be changed . venkatesh , g. watanabe and p. talkner , new j. phys . * 16 * , 015032 ( 2014 ) g. watanabe , b.p . venkatesh and p. talkner , phys . e * 89 * , 052116 ( 2014 ) . the assumptions of a short interaction time @xmath69 during which the system dynamics is negligible is made for the sake of simplicity . one may allow for larger interaction times during which the system dynamics is no longer negligible @xcite without affecting the general conclusions . p. talkner , z. phys . b * 41 * , 365 ( 1981 ) . the superscript @xmath284 is meant to distinguish a work operation based on two gaussian energy measurements in contrast to the operation characterizing the work meter discussed below , see section [ meter ] . for the sake of simplicity , we assume that the energy eigenvalues @xmath15 of @xmath285 with corresponding eigenstates @xmath286 are non - degenerate . a. prinz - zwick , g .- ingold and p. talkner , phys . scr . * t165 * , 014014 ( 2015 ) .

observations of accretion flows around black holes in galactic systems and in the centers of galaxies show strong evidence for the presence of long - lived , global modes at characteristic frequencies of the black - hole spacetimes ( see * ? ? ? * ; * ? ? ? * for reviews ) . the origin of these modes is still a matter of debate ( see , e.g. , * ? ? ? * ; * ? ? ? it is anticipated , however , that modeling accurately their origin and physical properties will provide the first measurements of the spins of black holes in the universe as well as direct tests of strong - field general relativity @xcite . systematic studies of modes in accretion disks around black holes have so far been mostly analytical , performed in the linear regime for flows with an artificial viscosity law ( see * ? ? ? * ; * ? ? ? * and reference therein ; see , however , gammie & balbus 1994 and menou 2003 for studies of mhd disks ) . these studies have shown the presence of three types of trapped global modes , namely inertial - gravity , corrugation , and inertial - pressure modes , with properties that appear to agree with observations ( see * ? ? ? despite their success , analytical models have several limitations . first , they rely on a number of approximations , such as the wkb approximation in the radial direction . second , because only linear terms in perturbations are taken into account , these studies are unable to reproduce resonances and couplings between various modes . however , the detected qpos often have fractional amplitudes as large as 40% of the total x - ray flux , making the linear approximation inaccurate ( see , e.g. , * ? ? ? furthermore , pairs of qpos are typically observed ( see * ? ? ? * ; * ? ? ? * ) with frequency ratios equal to ratios of small integers , strongly arguing for the presence of resonances between modes @xcite . third , analytical and numerical studies of oscillations in mhd disks cast doubt on the presence of oscillations at the radial epicyclic frequencies , as required by all diskoseismic calculations ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? answering all the above questions necessitates a systematic , numerical study of accretion disk oscillations . numerical studies of accretion disk modes have been very few and have not addressed in detail the above issues ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * for simulations of viscous disks ; see also rezzolla 2004 for simulations of hydrodynamic tori ) . in fact , no numerical simulation of hydrodynamic disks to date has resulted in solutions that show all three types of accretion - disk modes that were discovered analytically . in this series of papers , we perform a systematic study of the oscillatory behavior of hydrodynamic and mhd accretion disks in two and three dimensions . to this end , we have developed a pseudo - spectral numerical algorithm , which we present here . even though spectral methods have been used extensively in the study of hydrodynamic modes in various planetary atmospheres , including the earth s ( see , e.g. , * ? ? ? * ) , their application to the studies of accretion disks has been very limited @xcite . spectral algorithms are high order numerical methods and are perfectly suited to studying modes of hydrodynamic flows for a number of reasons . their major advantage is accuracy and economy in the number of degrees of freedom @xcite . in spectral methods , only @xmath0 collocation points are needed to resolve one wavelength @xcite , whereas in finite difference methods a lot more grid points are required to reach the same accuracy . the solution is already expressed as a set of orthogonal modes ; this property improves the accuracy of studies of mode coupling and resonance . moreover , the dimensionality of the problem can be altered very easily , for investigating the effects of the number of spatial dimensions on the properties of hydrodynamic modes and turbulence . in simulating mhd flows , the high order of spectral methods allows for the equations to be solved in terms of the vector potential and not of the magnetic field itself ; this guarantees that the resulting magnetic field is always divergence free . also , super spectral viscosity @xcite , which is one kind of artificial viscosity , is trivial to implement in spectral methods . it resolves shocks and yet preserves the high spectral accuracy of the solution . finally , incorporating the self - gravity of the flows is trivial and does not increase significantly the computational time ( see , e.g. , * ? ? ? * ) . in this first paper , we discuss our numerical algorithm for evolving two - dimensional viscous , hydrodynamic accretion disks . in the following section , we present our assumptions and equations . in 3 , we discuss the details of our numerical method . in 4 , we present a series of tests to verify our algorithm . in 5 , we apply our algorithm to non - axisymmetric instabilities in viscous spreading rings . we confirm the result of @xcite , which shows that the standard solution of a viscous spreading ring @xcite is unstable to non - axisymmetric perturbation . in 6 , we present our conclusions . we consider two - dimensional , viscous , compressible flows . in this first paper , we neglect self - gravity and the magnetic fields of the flows . the hydrodynamic equations , therefore , contain the continuity equation @xmath1 the navier - stokes equation @xmath2 and the energy equation @xmath3 we denote by @xmath4 the height - integrated density , by @xmath5 the velocity , and by @xmath6 the thermal energy . in the navier - stokes equation , @xmath7 is the height - integrated pressure , @xmath8 is the viscosity tensor , and @xmath9 is the gravitational acceleration . we use @xmath10 to denote the viscous dissipation rate , @xmath11 to denote the heat flux vector , and @xmath12 to denote the radiation flux on the @xmath13-@xmath14 plane . the last term in the heat equation , @xmath15 , takes into account the radiation losses in the vertical direction . we allow for large flexibility in the physical assumptions in our algorithm , i.e. , we can change the functional form of @xmath7 , @xmath9 , @xmath11 , @xmath12 , and @xmath16 easily and apply our algorithm to various problems . we also write the viscosity tensor in the general form @xmath17 here @xmath18 , @xmath19 , and @xmath20 are the coefficients of radiative , bulk , and shear viscosity , respectively . these coefficients can also be changed easily to satisfy different physical assumptions . the strain - rate tensor @xmath21 is @xmath22 the viscous dissipation rate is @xmath23 in appendix [ app : equations ] , we write down explicitly the equations used in the algorithm . the default setup for our calculations is that of a geometrically thin disk , in which radiation pressure is assumed to be negligible . using the ideal gas law , the thermal energy density and pressure become @xmath24 where @xmath25 is the boltzmann constant , @xmath26 is the central temperature , @xmath27 is the mean molecular weight , and @xmath28 is the mass of the hydrogen atom . the viscosity coefficients @xmath18 , @xmath19 , and @xmath20 are chosen based on the physical assumptions of the specific problems . in order to approximate the effects of general relativity in the vicinity of compact objects , we use the pseudo - newtonian approximation for the gravitational acceleration @xcite @xmath29 ^ 2\left[\frac{r^2 - 2(a / c)\sqrt{gmr / c^2 } + ( a / c)^2 } { \sqrt{gmr / c^2}(r - r_\mathrm{g } ) + a / c}\right]^2\mathbf{\hat r}.\ ] ] here , @xmath30 is speed of light , @xmath31 is the gravitational constant , @xmath32 is the mass of the central object , @xmath33 is its specific angular momentum , @xmath34 is its schwarzschild radius , and @xmath35 is the unit vector in the radial direction . for a non - rotating star , we set @xmath36 and the gravitational acceleration reduces to the standard pseudo - newtonian acceleration @xmath37 . when @xmath38 , it recovers newtonian gravity . regarding radiative cooling , we follow @xcite and @xcite to take into account the energy loss in the vertical direction by the standard form @xmath39 where we use @xmath40 to denote the stefan - boltzmann constant . the optical depth @xmath41 is again chosen based on the physical problem . we finally set to zero the other two terms , @xmath42 and @xmath43 , in the energy equation . in our numerical algorithm , we use pseudo - spectral methods to evaluate the spatial derivatives in the partial differential equations and an explicit runge - kutta method to advance the solution in time . spectral methods are based on the idea that any function can be expanded in a series of orthogonal basis functions . when a well - behaved basis is chosen , this series converges exponentially in the absence of discontinuities and the numerical partial derivatives can be easily evaluated . there exist different ways to approximate functions by a series with a finite number of terms , such as the galerkin method , the tau method , and the pseudo - spectral method ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in pseudo - spectral methods , as the one we use here , we choose a set of collocation points and then evaluate the expansion ( spectral ) coefficients such that the truncated series agree exactly with the original functions at the collocation points up to the machine accuracy . in the next subsection , we discuss our implementation of the pseudo - spectral method . the fourier basis is the natural choice for expanding functions with periodic boundary conditions , as is the case along the azimuthal direction in our problem . any physical quantity @xmath44 can be expanded in this basis as @xmath45 we use a total of @xmath32 evenly spaced , discrete collocation points in the domain @xmath46 , @xmath47 so that the above fourier series can be approximated by the ( truncated ) discrete fourier series @xmath48 the derivative of @xmath49 with respect to @xmath14 is , therefore , @xmath50 in order to evaluate equation ( [ eq : fourier_derivative ] ) , we use a fast fourier transform algorithm to find @xmath51 , multiply the fourier coefficients by @xmath52 , and then take the inverse fast fourier transform of the series . the computational order of the azimuthal derivative is therefore @xmath53 . when non - periodic boundary conditions are present , as in the case along the radial direction , the fourier basis can no longer be used . popular choices of bases are the legendre and chebyshev polynomials , which are both defined on the domain @xmath54 $ ] . between these two , the chebyshev polynomials are usually preferred because of their relation to the cosine function , which makes them easier to compute . suppose the domain of @xmath13 is @xmath55 $ ] . in order to map the standardized radial coordinate @xmath56 $ ] to the physical radial coordinate @xmath57 $ ] , we introduce a mapping function @xmath58 which is strictly increasing and satisfies both @xmath59 and @xmath60 . let @xmath61 be the @xmath62-th order chebyshev polynomial , i.e. , @xmath63 the chebyshev - gauss - lobatto collocation points are defined by @xmath64 where the collocation points in the physical radial coordinate are given by @xmath65 . note that there are totally @xmath66 collocation points including both boundary points of the domain . for any physical quantity @xmath67 , we approximate its chebyshev - fourier series by the double sum @xmath68 \nonumber\\ & = & \sum^{m/2 - 1}_{m =- m/2}\sum^n_{n=0}\check f_{n , m}t_n(\bar r_k)e^{i2\pi mj / m } \nonumber\\ & = & \sum^{m/2 - 1}_{m =- m/2}\sum^n_{n=0}\check f_{n , m}\cos\left(\frac{\pi nk}{n}\right)e^{i2\pi mj / m}.\end{aligned}\ ] ] we then use a fast discrete cosine transform to obtain the spectral coefficient @xmath69 from @xmath70 $ ] . the radial derivative can be found using the chain rule @xmath71 expressing the derivative with respect to the standardized coordinate as @xmath72 we can employ the following three - term recursive relation to obtain @xmath73 from @xmath69 , @xmath74 where @xmath75 , and @xmath76 , for @xmath77 . this recursion relation allows the use of an @xmath78 algorithm in calculating the numerical derivatives in spectral space along the radial direction . the transformation between physical and spectral space can then be done by a fast cosine transform , the order of which is @xmath79 . the most trivial map @xmath58 is linear , i.e. , @xmath80 in this case , @xmath81 is just a constant . however , the grid spacings @xmath82 , for @xmath83 , scale with the number of collocation points @xmath84 as @xmath85 , when @xmath86 is close to @xmath87 or @xmath84 . this requires the stability condition for time - stepping to be @xmath88 due to viscous time scale ( see [ sec : time_stepping ] ) . the grid spacings for fourier collocation are @xmath89 . they are uniform and scale as @xmath90 which gives the stability condition @xmath91 the stability condition in the radial direction is too expensive numerically compared to the one required by the azimuthal direction . to overcome this restriction , @xcite proposed the mapping @xmath92 - \frac{r_{\min}}{2}\left[\frac{\arcsin(\alpha \bar r)}{\arcsin(\alpha ) } - 1\right ] , \label{eq : modified_chebyshev_collocation}\ ] ] which makes the spacing @xmath93 around the boundaries . in the above transformation , @xmath94 is a parameter . let @xmath95 be the machine accuracy . @xcite showed that the choice @xmath96 minimizes the errors in the numerical derivatives . we therefore refer to the mapping ( [ eq : modified_chebyshev_collocation ] ) with the optimized choice ( [ eq : optimized_alpha ] ) as the modified chebyshev collocation and employ it in our algorithm . the numerical derivatives for the modified chebyshev collocation can be easily calculated by @xmath97 where @xmath98 is obtained by the usual chebyshev method ( [ eq : chebyshev_derivative ] ) . the constant scaling in ( [ eq : chebysehv_derivative_linear_scaling ] ) and the scaling in equation ( [ eq : chebysehv_derivative_scaling ] ) can be done together with the normalization , so neither of them have extra computational cost . in low order numerical methods , one uses the conservative form of the hydrodynamic equations to ensure conservation of mass and momentum @xcite . this is not necessarily the ideal way for pseudo - spectral methods because the conservative form is sometime numerically unstable ( see * ? ? ? * ) . to illustrate this , consider two arbitrary functions @xmath99 and @xmath100 the derivative of their product has two different but equivalent forms @xmath101 the pseudo - spectral fourier coefficients ( with @xmath32 collocation points ) of @xmath102 and @xmath103 are @xmath104 , \\ \hat d_m & = & im\sum_{p+q = m}\hat u_p\hat v_q + i\left[\sum_{p+q = m+m}(m+m)\hat u_p\hat v_q \right.\nonumber\\ & & \left.+ \sum_{p+q = m - m}(m - m)\hat u_p\hat v_q\right].\end{aligned}\ ] ] the first term in each equation corresponds to the truncated fourier series . the other terms in the square brackets are therefore due to the aliasing error of pseudo - spectral methods ( see * ? ? ? * ) , which may cause numerical instability . one can show that these aliasing terms have opposite signs . the skew symmetric form @xmath105 , therefore , has a much smaller aliasing error . unfortunately , calculating the skew symmetric form increases the number of numerical derivatives and lowers the algorithmic performance . a similar argument is valid for the chebyshev collocation . however , in that case , the problem becomes more complicated due to the boundary conditions . @xcite carried out numerical experiments and reported that , without aliasing removal , the convective form is stable for the two different boundary method they tested ; the conservative form is unstable for one of their methods . we carried out some numerical experiments for our algorithm . when the initial conditions are smooth , both conservative and convective forms are stable however , when an initial perturbation is introduced , the conservative form sometimes become unstable . in these cases , the spectral coefficients at high wavenumbers in the radial direction grow exponentially even with spectral filtering ( see the next subsection ) . the convective form , on the other hand , is able to reproduce various analytical stability criteria , as shown in [ sec : tests ] . because the convective form gives an approximation that is of the same order as the conservative form , in the present implementations of the numerical algorithm , we use the convective form for the non - linear terms in navier - stokes equation to ensure numerical stability . the continuity and energy equations , on the other hand , are written in conservative form to conserve mass and energy . the exact equations we solve in the algorithm can be found in appendix [ app : equations ] . in our numerical algorithm , we use a combination of spectral filters in order to resolve the two principle drawbacks of spectral methods . first , because of the non - linear character of the navier - stokes equations described above , we need to filter out the high - frequency modes in each time step to reduce the aliasing error . the spectral filters ensure long - time stability of the algorithm . second , when shocks are present in the solutions , the spectral coefficients do not converge exponentially and oscillations appear around the discontinuities ( gibbs phenomenon ) . introducing an additional spectral filter increases the converge rate of the solutions . indeed , it has been proven that the filtered solution converges to the correct entry solution and hence gives the correct shock properties . ( for more information see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * and reference therein . ) suppose we want to apply a filter in the @xmath14-direction . we first find the fourier coefficients @xmath106 according to equation ( [ eq : fourier_expansion ] ) . we then denote the filtered sum by @xmath107 where we use the exponential filter @xmath108 the parameter @xmath95 here is again the machine accuracy . we choose @xmath109 so that the filtered sum approximates the original function very well and does not reduce the accuracy of the numerical solution . the same filter is used in the radial direction , i.e. , @xmath110 where @xmath111 is chosen close to @xmath84 so the double filtered sum is @xmath112 spectral methods have been used successfully to study non - randomized perturbation , such as the tidal effects presented by @xcite . later , @xcite described a method to add random perturbations to their initial conditions . in their simulation , they generated 100 gaussian profiles randomly within the computation domain . this method ensure that the spectral coefficients of the perturbed function converge exponentially , and avoids changing the direction of the characteristic at the boundary . in our calculations , we require the initial perturbations to be smooth , i.e. , we require the spectrum of the perturbations to converge exponentially . one simple method to obtain such a perturbation is to generate random noise in spectral space , multiply the noise by an exponential filter , and transform the noise back to physical space . to formulate this method , we first generate a flat - noise spectrum , @xmath113 , which is totally random . the perturbation is added to the original function by the following equation @xmath114f_0(r,\phi),\ ] ] where @xmath115 is the perturbed function , @xmath116 is the size of the perturbation , @xmath117 is the filtered sum @xmath118 as defined in the pervious subsection , and @xmath119 is the unperturbed function . the order of filter , @xmath111 , controls the spacial scale of the perturbation . because @xmath120 is periodic in @xmath14 , we can simply use a @xmath14-independent function @xmath121 . however , in the @xmath13-direction we do not want the perturbation to extend the boundaries . we , therefore , set @xmath122 which is pre - computed by equation ( [ eq : chebysehv_derivative_scaling ] ) , when we use the modified chebyshev collocation method . in figure [ fig : perturbation ] , we illustrate this method of introducing perturbations to an one - dimensional constant function @xmath123 . the left plot uses a fourier basis with 256 collocation points . where the right plot uses a chebyshev basis with 257 collocation points . all the lines have @xmath124 but different orders @xmath111 . note that the extra rescaling for chebyshev collocation lowers the amplitude of the perturbations around the boundaries . solving partial differential equations in a finite domain usually requires dirichlet or / and neumann boundary conditions . when using finite difference methods , one can simply change the values of physical quantities at the boundary points to achieve dirichlet boundary conditions alternatively , one can adjust the `` ghost points '' to achieve neumann boundary conditions . unfortunately , these methods do not generally work for spectral methods . indeed , changing the boundary points ( or ghost points ) effectively introduces step functions . when discontinuities are present , the spectral coefficients do not converge exponentially and the spectral methods fail to produce stable solutions . perturbations around the boundaries can also change the characteristic directions of the flow . a naive dirichlet or neumann boundary treatment fails to take care of these oscillations between inflowing and outgoing characteristics . when the inflowing boundary conditions are not given , or when the outgoing boundary conditions are not consistent with the solutions in the computation domain , the system of differential equations is ill - posed and the algorithm diverges . although we generate the initial perturbations in a way that they are small around the boundaries ( see the previous subsection ) , they can still propagate and reach the boundaries as time evolves . in our numerical algorithm , we introduce a new boundary treatment in the radial direction , which is a spatial filter . this method forces each dynamic variable to approach smoothly to its boundary value so the characteristics directions at the boundaries are well - defined . this ensures that any instabilities of the solution are not due to boundary conditions . for example , consider a one - dimensional problem @xmath125\ ] ] for any functional form @xmath126 . if we want to apply a dirichlet boundary condition at the outer boundary , we choose some smooth monotonic function @xmath127 such that @xmath128 for interior points of the computational domain and @xmath129 at the boundary . at each time step , we impose the boundary condition by the spatial filter @xmath130 where @xmath131 denotes the numerical approximation of our function at time @xmath132 and radial collocation point @xmath86 . this filter makes the function @xmath49 approach its boundary value @xmath133 and produces a numerical boundary layer . rearranging , the above step is equivalent to setting @xmath134(f_k^i - f_0).\ ] ] hence , it is equivalent to adding an extra source / sink term @xmath135 in the original equation @xmath136 - \frac{1 - h(r)}{\delta t}(f - f_0 ) \equiv f[f ] + \dot f_\mathrm{s},\ ] ] where @xmath137 is the time step ( see [ sec : time_stepping ] ) . omitting @xmath126 in the above equation , it becomes @xmath138 . the value of @xmath139/\delta t$ ] controls the converge rate of @xmath49 to @xmath133 , which is automatically proportional to the velocity , because @xmath140 . we use an exponential filter that is similar to the spectral filter discussed in 3.3 . to impose an outer boundary condition , we need the monotonically decreasing filter @xmath141.\ ] ] on the other hand , for the inner boundary , we use the monotonically increasing filter @xmath142.\ ] ] we can change the thickness of the numerical boundary layer by changing the order of the filter , @xmath111 . this boundary treatment turns out to be very convenient in modeling non - reflective boundary conditions . when we choose the parameter @xmath111 to be equal to the number of collocation points , the boundary layer is thick enough to discard any outgoing waves but thin enough so that the interior solution is not affected significantly . for applying neumann boundary conditions , the addition of a ghost zone is not practical in spectral methods . this is because chebyshev polynomials approximate the derivatives based on all the interior points . @xcite presented a method which involves solving for the boundary values of physical quantities , so that their derivatives agree with the boundary conditions . this is effectively a dirichlet boundary treatment with the boundary values depending on all the interior values at each time step . to avoid numerical instabilities due to the neumann boundary conditions , @xmath143 , we choose another method that involves applying a spatial filter to each variable after taking the radial derivatives . this is equivalent to replacing the radial derivatives @xmath144 by @xmath145 \left(\left.\frac{\partial f}{\partial r}\right|_{r_k}^{\ i } - f'_0\right)\ ] ] for the inner boundary , and @xmath146 \left(\left.\frac{\partial f}{\partial r}\right|_{r_k}^{\ i } - f'_0\right)\ ] ] for the outer boundary . this ensures that the first derivatives around the boundaries are smooth . we integrate the hydrodynamics equations with low - storage , explicit runge - kutta methods . let @xmath147 be the right hand sides of hydrodynamic equations ( [ eq : hydro_begin ] ) ( [ eq : hydro_end ] ) and let @xmath148 be the dynamic variable ; the superscript @xmath149 here stands for transpose . the low - storage explicit runge - kutta methods have the general from @xmath150 where @xmath137 is the time step and @xmath151 . all other @xmath152 s and @xmath153 s are coefficients that characterize the scheme and @xmath154 are the intermediate stages of @xmath155 . the method to obtain these coefficients can be found in @xcite . depending on the complexity of the problem , we choose between second order or third order runge - kutta , and third order or fourth order carpenter - kennedy runge - kutta methods . the different algorithms are compared during the verification tests ( see [ sec : tests_free_fall ] and fig . [ fig : error_time ] below ) . the third order runge - kutta scheme , proposed by williamson ( see * ? ? ? * ) , turns out to be the most popular in spectral methods because of its efficiency and accuracy . for this reason , we present here this scheme explicitly : @xmath156 recalling the definition of @xmath157 and @xmath158 in [ sec : collocation ] , we define the courant - friedrich - lewy ( cfl ) time step in different directions independently , @xmath159 where @xmath160 is the sound speed . let @xmath161 be the minimum grid separation . we also define the viscously restricted time step as @xmath162 where @xmath163 is the kinematic viscosity coefficient . using these definitions , the maximum allowed time step @xmath137 is chosen adaptively by @xmath164 here @xmath165 is a constant that depends on the physical problem . for example , for a free falling dust ring , the flow is very stable and @xmath165 could be chosen as large as 29 . for most of accretion problems with @xmath166 , @xmath165 is of order unity . we have verified our numerical algorithm by comparing our numerical results to a number of test problems with analytical solutions . these problems are designed to test the implementation of the different terms in the equations , such as the ones that describe gravity , pressure , and viscosity . in this section , we describe these tests in details . + as a first test , we study the free - fall of an axisymmetric distribution of matter with no angular momentum , pressure , or viscosity . this problem tests the implementation of gravity , the radial boundary conditions , and the conservation of mass in the radial direction . also , we use this problem to evaluate the efficiency and stability of the different time - stepping schemes . we set @xmath167 , @xmath168 , @xmath169 , and consider only newtonian gravity with @xmath170 . in order to verify our numerical result , we also solve the problem analytically . using dimensionless quantities so that @xmath171 , we write the conservation of energy for a fluid element initially at @xmath172 as @xmath173 integrating over time , we obtain the trajectory @xmath174 of each fluid element in the implicit form @xmath175 we now consider conservation of mass , i.e. , at any time we require that @xmath176 evaluating @xmath177 by using equation ( [ eq : r_0 ] ) , i.e. , @xmath178^{-1},\ ] ] the analytical solution for the density becomes @xmath179^{-1},\ ] ] where @xmath172 can be obtained implicitly using equation ( [ eq : r_0 ] ) . although this is an one dimensional problem , we simulate it in the two - dimensional domain @xmath180\times[-\pi , \pi]$ ] with @xmath181 collocation points . the initial condition is a gaussian ring @xmath182\ ] ] with zero initial velocity . at the outer boundary , we use the standard neumann boundary conditions , i.e. , we set @xmath183 and @xmath184 for all times . moreover , because the characteristics at the inner boundary point towards the negative @xmath13-direction and we have neglected viscosity and pressure , we do not need to impose any explicit boundary conditions there . because filtering stabilizes the flow , we do not apply any filters in this test in order to compare how different time - stepping methods affect the stability of the algorithm . the top panels of figure [ fig : free_fall ] compare the results of the simulations to the analytical solutions ; the open circles denote the numerical results using the third order runge - kutta method and the solid lines are the analytical solution . the numerical results are indistinguishable from the analytical solution . the bottom panels of figure [ fig : free_fall ] show the error between the numerical and analytical solution for the maximum stable time step . we test four different time - stepping schemes : second order runge - kutta ( rk2 ) , third order runge - kutta ( rk3 ) , third order carpenter - kennedy ( rk3ck ) , and four order carpenter - kennedy ( rk4ck ) . figure [ fig : error_time ] shows the result of this test . in the left plot , the vertical axis shows the computation time . the horizontal axis shows the dimensionless time - stepping factor @xmath185 ( see [ sec : time_stepping ] ) . the right plot shows the numerical errors , @xmath186 , at @xmath187 , for the same problem . for @xmath188 in rk2 , @xmath189 in rk3 , @xmath190 in rk3ck , and @xmath191 in rk4ck , the solutions diverge so neither the computation time nor the error are plotted in the graphs . this test demonstrates the ability of our algorithm to conserve mass to an accuracy better than a few parts in a million . it also demonstrates the advantage of the third order runge - kutta method over other algorithms , when solutions of high accuracy are necessary . + we now study the ability of our algorithm to model correctly instabilities in hydrodynamic flows . in the absence of viscosity , the necessary and sufficient condition for a rotating flow with angular velocity @xmath192 to be stable is @xmath193 which is known as rayleigh s criterion @xcite . although rayleigh s criterion usually describes the stability condition in a couette flow , i.e. , for an incompressible fluid between two rotating cylinders , the criterion is also valid for rotating compressible fluids , when the unperturbed radial velocity vanishes everywhere in the flow . this can be achieved by balancing the centrifugal and centripetal accelerations . let @xmath194 be the square of the epicyclic frequency . rayleigh s criterion is equivalent to requiring @xmath195 , because @xmath196 . we define a modified newtonian gravity with a negative gravitational index @xmath197 such that the gravitational acceleration is @xmath198 setting @xmath171 , the corresponding `` keplerian '' velocity @xmath199 is given by @xmath200 substituting this into rayleigh s criterion , i.e. , equation ( [ eq : rayleigh ] ) , we find that @xmath201 is the critical stability condition . the growth rate of the instability is @xmath202 of course , the same result can be obtained by observing that the effective potential @xmath203 has no local minimum when @xmath204 ; here @xmath205 is the angular momentum of a fluid element with a unit mass . we use a spatial resolution of @xmath181 collocation points to test the rayleigh criterion . the initial conditions are with a uniform density , @xmath206 , and the velocity profile ( [ eq : velocity ] ) . random perturbations that are of order @xmath207 are added to all physical quantities and the boundary conditions are set to @xmath208 . because rayleigh s criterion refers to an one - dimensional problem , we discuss here explicitly the evolution of the @xmath209 mode . this is equivalent to averaging over the azimuthal direction . figure [ fig : rayleigh ] shows the evolution of power at @xmath210 , @xmath211 , @xmath212 , and @xmath213 ( and @xmath214 ) of the density , i.e. , it is the plot of @xmath215 against time . when the flow is unstable , the perturbations grow very fast and diverge . as required by rayleigh s criterion , flows with @xmath216 are stable to perturbations . cccc 1.0 & 1.011210 & 1.000000 & 1.12095% + 1.1 & 1.062727 & 1.048808 & 1.32702% + 1.2 & 1.107763 & 1.095445 & 1.12449% + 1.3 & 1.153139 & 1.140175 & 1.13695% + 1.4 & 1.193408 & 1.183216 & 0.86140% + 1.5 & 1.235294 & 1.224744 & 0.86136% in order to test our implementation of the thermal pressure and the numerical spreading of wavefronts , we neglect gravity , set all velocities equal to zero , and simulate the propagation of a sound wave . we use the equation of state @xmath217 with @xmath218 and vary the polytropic index @xmath219 from @xmath87 to @xmath220 . this pressure term decouples the energy equation from the navier - stokes equation and the sound speed is simply @xmath221 . the initial density is @xmath222.\ ] ] we use @xmath181 collocation points and calculate the solution from @xmath223 to @xmath224 . figure [ fig : density_contour ] shows six gray - scale snapshots of the density for @xmath225 . the resulting sound wave propagates outward as a circle . note that , because of our boundary treatment , the density wave is not reflected back as it propagates close to the boundaries . to measure the sound speed , we take the density profile at @xmath226 and trace the wavefronts . the collocation points that correspond to the peak density are then used to calculate , via @xmath227-fitting , the sound speed . table [ tab : error_speed ] shows the @xmath227-fitted sound speed of the simulation . compared to the analytical sound speed , the error is of order 1% . the spreading of an axisymmetric viscous ring with small kinematic viscosity @xmath228 @xcite has become a standard test for numerical methods including both smoothed particle hydrodynamics ( sph ) and grid - based codes . it has also often been used to guide studies of many other astrophysical problems , such as the fall - back disk after a supernova explosion ( see , e.g. , * ? ? ? * ; * ? ? ? * ) . in the standard solution , it is assumed that @xmath229 , so the viscosity tensor can be approximated by @xmath230 using conservation of mass , conservation of angular momentum , the fact that @xmath199 is always close to the keplerian value , and assuming that the kinematic viscosity coefficient @xmath231 is a constant , the analytical solution for the density , in our units , becomes @xmath232 where @xmath233 is a modified bessel function . we perform our simulations with axisymmetric initial conditions . we follow @xcite and use a typical value @xmath234 . the energy equation is decoupled from the hydrodynamic equations as in the previous subsection . we set @xmath225 , and choose @xmath235 such that @xmath236 . the spatial resolution is @xmath181 and the time step factor @xmath237 . we start with an initial density given by equation ( [ eq : pringle ] ) at @xmath238 , which corresponds to @xmath239 . the boundary conditions are @xmath240 and @xmath241 , which are the ones assumed in the standard solution . figure [ fig : pringle ] compares the analytical solution to the numerical solution . the simulation shows that some concentric rings appear at the beginning . they come from the relaxation effect as our initial velocities approach standard solution . in fact , these additional structures have been found in other simulations ( for more information see * ? ? ? * ) , and decay away after about 24 rotations . + here we consider the thin disk problem and suppose that the disk is able to settle to a steady - state structure . by assuming an @xmath94-viscosity law , @xcite solved the local disk structure analytically . their work has been highly cited for studies in accretion disks . ( see frank , king , & raine 2002 and reference therein ) . in our simulation , we solve the hydrodynamic equations ( [ eq : continuity ] ) ( [ eq : energy ] ) including the energy equation . the kinematic viscosity coefficient is given by @xmath242 the sound speed is given by @xmath243 and the disk scale height is @xmath244 . we , also , assume the vertical optical depth of the disk to be @xmath245 , where the rosseland mean opacity @xmath246 is given by kramers law @xmath247 based on the above assumption and using the fact that @xmath228 is small , the shakura - sunyaev disk solution is given by @xmath248 where @xmath249 , @xmath250 , @xmath251 , and @xmath252 , with @xmath253 being the radius of the central object . considering the flow around a standard neutron star with a mass of @xmath254 , and a radius of @xmath255 , we setup the simulation with @xmath256 and @xmath257 . the initial conditions for density and energy are as the shakura - sunyaev disk solution with @xmath258 , @xmath259 , and @xmath260 . we set the initial radial velocity to zero and the initial azimuthal velocity to be keplerian . perturbations were added to all variables with amplitude @xmath261 and order @xmath262 . the simulation runs with @xmath263 collocation points up to @xmath264 , which is approximately @xmath265 viscous time scales . in figure [ fig : shakura_sunyaev ] , we show our numerical steady solution for @xmath266 , at @xmath267 ; note that we use the magnitude of the radial velocity in order to plot it in log - log scale . the numerical values are different from the shakura - sunyaev solution by a few percent . we point out explicitly that the pressure term in the navier - stokes equation is neglected in obtaining the analytical solution . although the pressure term does not transport angular momentum , the effect of pressure compared to those of viscosity is proportional to @xmath268 . since @xmath269 in our simulation , the pressure effects can account for the difference between the analytical and numerical solution . + as we mentioned in [ sec : tests_pringle ] , the viscous spreading ring has become a standard test for numerical simulations . in these simulations , additional structures always appear , such as the concentric rings in figure [ fig : pringle ] . we attributed the presence of these concentric rings to relaxation effects of the initial conditions . @xcite , on the other hand , showed that the concentric rings that appear in sph simulations using the artificial viscosity given by @xcite are numerical artifacts . later , @xcite used perturbation methods to show that viscous spreading rings are unstable to non - axisymmetric perturbations . assuming a compressible , pressureless , and viscous fluid , the equations used in @xcite are the continuity equation ( [ eq : continuity ] ) and the navier - stokes equation ( [ eq : navier_stokes ] ) , with zero pressure . by setting @xmath163 to be constant and @xmath270 , the first order instability is an one - armed spiral , whereas the second order instability is a superposition of an one - armed and a two - armed spirals . if we define a function @xmath271 where @xmath272 is the keplerian angular velocity , then @xmath273 measures the first order perturbation amplitude . its dispersion relation is @xmath274 where @xmath86 is the radial wavenumber and @xmath275 is the standard solution ( [ eq : pringle ] ) . assuming @xmath276 , it follows that the condition for instability is @xmath277 and the growth rate is @xmath278 in order to test this claim , we carried out a test with the same condition as speith and kley s rh2d simulation ( a radiative hydrodynamic code ; see * ? ? ? the only differences are in the initial perturbations and a small constant background term in the density , i.e. , the setup is the same as in [ sec : tests_pringle ] but we add initial perturbations by the method described in [ sec : perturbation ] with order 3 and magnitude 0.001 . the additional constant background term in the density varies from 0.1 to 0.001 depending upon the resolution , which keeps the solution stable . we perform the simulation with resolution @xmath279 , @xmath181 , @xmath263 , and @xmath280 . figure [ fig : speith_kley_compare ] shows the density contours for different resolutions at time @xmath281 . it is clear that the direction of the spiral in low resolution is different then in high resolution . we confirm this result with speith and kley ( private communication ) , i.e. , that the transition radius between leading and trailing spirals vary as the resolution changes . figure [ fig : speith_kley ] is the density contour of the simulation with @xmath263 collocation points and 0.001 background density , at different times . at @xmath282 , some concentric rings due to relaxation effects are seen . at @xmath283 , a one - armed spiral starts to form at the inner disk . the spiral is fully developed by @xmath284 . in order to find out the unstable wave number , we plot in figure [ fig : speith_kley_evolve ] the contour of density along @xmath226 against time region i shows the relaxation effect , where the relaxation waves propagate outwards . in region ii , the solution has relaxed to the standard solution . later on , the speith - kley instability starts to develop , which is shown in region iii . we can estimate the wavelength of the fully developed spirals , which varies from @xmath261 to @xmath285 . for region iv , the unstable wavelengths are too short to be resolved in our numerical method . finally , in figure [ fig : speith_kley_modes_evolve ] we plot the first 6 modes , i.e. , @xmath286 , of @xmath287 . the late time behavior of our solution agrees with figure 8 in @xcite . to compare our numerical result with the analytical growth rate ( [ eq : growth_rate ] ) , we @xmath227-fit the slope of the @xmath288 mode in figure [ fig : speith_kley_modes_evolve ] between @xmath289 and @xmath290 , which gives @xmath291 . that corresponds to unstable wavenumber @xmath292 . it agrees with the width of the spiral @xmath293 as shown in figures [ fig : speith_kley ] and [ fig : speith_kley_evolve ] . physically , the viscous spreading ring problem is an important guide to understanding the transport of angular momentum in accretion disks . we have independently verified the speith - kley instability : a cold viscous disk is unstable to non - axisymmetric perturbations . we described a numerical method based on a pseudo - spectral algorithm for studying the rapid variability properties of two - dimensional , viscous , hydrodynamic accretion disks . we demonstrated the ability of the spectral methods to handle correctly non - reflective boundary conditions and different stability conditions . we verified the implementation of the algorithm using various test problems . also , we confirmed the non - axisymmetric instability of viscous spreading rings discovered by speith and kley . spectral methods calculate the time evolution of a solution as an interaction of different waves . from a mathematical point of view , this approach naturally agrees with existing methods to study non - linear equations . it helps us to confirm directly the result of linear mode analysis as well as turbulence theories . from a computational point of view , these methods are high order and produce accurate result in most problems . it is straightforward to extend the current algorithm to simulate 3d mhd accretion disks . in particular , because of the high order of the spectral methods , we can solve directly for the vector potential and not for the magnetic field , in order to guarantee that the resulting field will be free of divergence at each time step . also , spectral methods can be used to solve self - gravity problem without increasing significantly the computation time . finally , and most important , form a physical point of view , our algorithm provides another tool to set up numerical experiments and test astrophysical models . it is useful to confirm independently the results from other numerical simulations as well as test new models . as we describe in [ sec : non - linear ] , we employ a convective - conservative mixed method . the non - linear term @xmath294 in the euler equation is implemented in convective from , and all other terms are in conservation form . @xmath295 here , the notation @xmath296 denote partial derivative with respect to time . the other notations @xmath297 and @xmath298 denote the numerical derivative operators . hence , the notation @xmath299 means , we first take a numerical derivative of the variable @xmath300 and multiply the result with @xmath300 , which is in convective form . on the other hand , @xmath301 means , we first use a temporary variable to store the produce @xmath302 , apply the procedure to calculate the numerical derivative on it , and then divide it by @xmath13 . this is the conservative form . the viscosity tensor @xmath303 in the above equation has the following general form @xmath17 as we describe before , @xmath18 , @xmath19 , and @xmath20 are coefficient of radiative , bulk , and shearing viscosity . the strain rate tensor @xmath21 written in cylindrical coordinate becomes @xmath304 and @xmath305 in cylindrical coordinate is @xmath306 for completeness , we provide again the viscous dissipation rate @xmath307 the functional forms of @xmath7 , @xmath18 , @xmath19 , @xmath20 , @xmath11 , @xmath12 , and @xmath16 can be change easily depend upon the physical assumption ( where the defaults are described in [ sec : equations ] ) . c. and d. p. acknowledge the support from a nasa atp grant nag-513374 . f. . acknowledges support by nasa through hubble fellowship grant hf-01156 from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 .

ltude de la distribution des points rationnels sur les varits algbriques est un sujet classique de la gomtrie diophantienne . beaucoup de varits do nt le fibr anticanonique est gros vrifient le principe de batyrev - manin @xcite avec la constante de peyre @xcite , y compris les varits toriques lisses projectives ( voir @xcite pour des exemples ) . plus prcisment , soit @xmath1 une telle varit dfinie sur un corps de nombres @xmath2 munie dune hauteur de weil @xmath3 associe au fibr anticanonique @xmath4 . on dit que @xmath1 vrifie le principe de batyrev - manin sil existe un ouvert dense @xmath5 de @xmath1 tel que @xmath6 o la constante @xmath7 a reu une interprtation gomtrique ( cf . @xcite,@xcite ) . la raison pour laquelle on ne considre pas tous les points de la varit est que souvent il existe des sous - varits , que lon appelle ici _ globalement accumulatrice _ , do nt la croissance du nombre de points rationnels domine celle dun ouvert dense . on peut dfinir certaines constantes ( par exemple , la constante @xmath8 dans @xcite ) pour dtecter de telles varits . soit @xmath5 louvert priv de toutes les sous - varits globalement accumulatrices ( si un tel ouvert existe ) . une extension naturelle de ce problme est de considrer la convergence au sens faible de la suite de mesures de probabilit @xmath9 si cette suite de mesures converge , elle reflte un phnomne dquidistribution , et en regardant la densit de la mesure limite on peut obtenir une caractrisation de la concentration des points rationnels . peyre @xcite a conjectur une mesure limite , appele mesure de tamagawa . ce texte est concern par un autre aspect , savoir le problme de la distribution _ locale _ des points rationnels autour dun point rationnel fix . le terme local est au sens de lapproximation diophantienne systmatiquement formul en premier par d. mckinnon et m. roth ( @xcite , @xcite ) . une diffrence majeure par rapport aux problmes globaux est que mme sil nexiste pas de sous - varits globalement accumulatrices , il peut y avoir des sous - varits qui sont _ localement _ accumulatrices . moralement , les meilleurs approximants ne peuvent tre choisis que sur telles sous - varits . comme dans le problme global , on peut aussi dfinir certaines constantes , appeles maintenant _ les constantes dapproximation _ , pour les caractriser . lune de des constantes , appele _ la constante dapproximation _ @xmath10 ( dfinition [ pd : appconstant ] ) , caractrise la meilleure approximation . la deuxime constante , appele _ la constante essentielle _ @xmath11 ( dfinition [ df : essential ] ) , dcrit des approximations gnriques . au niveau de la distribution plus fine , s. pagelot @xcite a tudi en premier _ la distribution locale _ des points rationnels sur certaines varits algbriques via une opration appele zoom . le but est de dnombrer et localiser asymptotiquement les points le plus proches dun point rationnel fix . elle nous permet dobtenir plus dinformations sur la concentration locale des points rationnels que les constantes dapproximation car il existe des sous - varits ayant la mme constante dapproximation , sur lesquelles le nombre dapproximants qui se trouvent tant dordre de grandeur diffrente . on propose une dfinition ( dfinition [ df : weaklocal ] ) base sur ce fait pour les caractriser . dans ce texte on se placera sur une surface torique dfinie sur @xmath12 que lon notera toujours @xmath13 . elle est obtenue en clatant les @xmath14 points invariants de @xmath0 , qui correspond aux ventails ] on choisit le relev du point @xmath15\times[1:1]$ ] de lorbite ouverte que lon va approcher . comme conjectur par d. mckinnon ( conjecture [ conj : mckinnon ] _ infra _ ) , les courbes rationnelles donnent les meilleurs approximants . en effet , nous dmontrons quil existe @xmath14 courbes rationnelles @xmath16 passant par @xmath17 de degr anticanonique minimal qui donnent des meilleures approximations . alors quen dehors des @xmath18 , les approximants optimaux sont peu nombreux et se situent sur une famille de courbes nodales en @xmath17 de degr bas do nt la runion est dense dans @xmath13 . elles sont obtenues en considrant les courbes nodales do nt les tangentes au point @xmath17 ont les pentes irrationnelles . autrement dit , les points au - dessus de la dsingularisation en @xmath17 nest pas dfinies sur @xmath12 ( en fait ils sont dfinis sur une extension quadratique de @xmath12 ) . avant desquisser la formulation de la distribution locale afin pour noncer notre thorme principal , nous considrons dabord un exemple particulier @xmath19 . on fixe le point @xmath20 $ ] et la hauteur de weil comme @xmath21)=\max(|x|,|y|,|z|),\quad \operatorname{\operatorname{pgcd}}(x , y , z)=1.\ ] ] on sintresse lensemble des points rationnels de hauteur borne @xmath22\in{\mathbb{p}}^2({\mathbb{q}}):h(p)\leqslant b\}.\ ] ] un calcul nous dit que les points le plus proches de @xmath17 sont une distance de lordre de @xmath23 . nous invitons les lecteurs consulter ( * ? ? ? * figure 1 ) pour un dessin de la rpartition des points de hauteur borne autour de @xmath17 , o nous observons que les points se rpartissent apparemment de prfrence sur les droites rationnelles . lide est quaprs une manipulation de zoom , on peut dfinir une suite de mesures dcrivant la distribution asymptotique des points proches de @xmath17 . soit @xmath1 une bonne varit munie dune hauteur de weil @xmath3 . on fixe un point @xmath24 approcher et un diffomorphisme local @xmath25 de la varit en @xmath17 sur le plan tangent @xmath26 qui envoie @xmath17 sur lorigine . pour @xmath27 et @xmath28 , on dfinit la suite de mesures de dirac par la formule suivante : @xmath29 en gnral on ne prend que les @xmath30 tels que @xmath31 parce que sinon on est ramen des sous - varits strictes ( proposition [ po : weakzoom ] ) . on fixe une distance euclidienne @xmath32 sur @xmath33 ( qui est quivalente localement toute distance projective dfinie sur @xmath1 ) . on note @xmath34 la boule centre en lorigine de rayon @xmath35 et @xmath36 la fonction caractristique associe . sil existe @xmath37 tels que @xmath38 pour une infinit de @xmath39 , ils dcrivent la grandeur daccumulation des points rationnels autour de @xmath17 . on renormalise en considrant la suite de mesures @xmath40 si elles convergent ( au sens faible ) vers une mesure non - nulle , on lappellera _ la mesure limite_. le phnomne gnral serait que quand @xmath41 , ( le zoom est _ sous - critique _ , dfinition [ def : criticalzoom ] ) , @xmath42 aurait lordre de grandeur attendu et la rpartition des points rationnels serait plus uniforme . le cas o @xmath43 ( le zoom est _ critique _ ) est particulirement intressant . s. pagelot a tudi dans @xcite la distribution locale dun @xmath44-point sur la droite projective et sur des surfaces de del pezzo ( toriques ) de degr @xmath45 . en particulier pour @xmath19 il a affirm ( pour le zoom critique ) lexistence de la mesure limite attendue de masse concentre sur les droites ( ceci est redmontr dans @xcite ) . il est naturel de se poser la mme question pour les points algbriques et pour les varits de dimension suprieure . lauteur a tudi dans @xcite ce problme pour une surface de del pezzo torique de degr @xmath46 et il a obtenu pour le cas @xmath43 une mesure limite . [ notationcqbx ] on note @xmath47 lespace vectoriel des fonctions continues de support compact dfinie sur @xmath48 valeurs relles . on nonce le thorme principal du texte . [ th : maintheorem ] on a @xmath49 . soit @xmath50 . soit @xmath30 le facteur de zoom . alors 1 . si @xmath51 , on a que pour toute @xmath52 , @xmath53 il nexiste pas de mesure limite dans ce cas . si @xmath54 , alors il existe une constante @xmath550,1\mathclose[$ ] telle que pour toute positive @xmath56 , @xmath57 o @xmath58 est une mesure absolument continue par rapport la mesure de lebesgue planaire . en combinant avec le rsultat de pagelot ( thorme [ th : pagelot ] ) , on trouve que @xmath59 pour toute positive @xmath60 . ceci implique le nombre dapproximants sur les @xmath18 domine celui dans le complmentaire bien que leur constantes dapproximation concident . remarquons que daprs lheuristique nave du principe de batyrev - manin ( cf . ) , on souhaiterait que la puisse de @xmath61 soit @xmath62 . cest - - dire que lon voudrait pourvoir prendre @xmath63 . nous dmontrons tous les @xmath64 suffisamment petit nous permettent de garantir la minoration ( cf . thorme [ th : lowerbound ] ) . lobservation cruciale dans la dmonstration de ce rsultat est que les meilleures approximations dans @xmath5 sont obtenues sur certaines courbes rationnelles qui sont nodales en @xmath17 . quand on restreint ces courbes nodales , lapproximation du point @xmath17 revient lapproximation dun point _ algbrique quadratique _ , ce qui nous amne tudier plus soigneusement la distribution locale des points algbriques sur la droite projective . notons quen dimension @xmath65 , il ny a pas de sous - varits localement accumulatrices . [ th : secondtheorem ] on munit le fibr @xmath66 sur @xmath67 une hauteur de weil ( cf . ) . soit @xmath68 un nombre irrationnel algbrique . on identifie @xmath68 comme un point rationnel @xmath69 $ ] de @xmath67 . alors @xmath70 . * et @xmath68 et un nombre quadratique , alors pour toute fonction @xmath72 , @xmath73 et il existe certaines fonctions @xmath74 telles que @xmath75 en dautres termes , il nexiste pas de mesure limite pour le zoom critique . * si @xmath76 . alors pour toute @xmath72 , @xmath77 en particulier , la mesure limite est proportionnelle la mesure de lebesgue . la constante dapproximation de @xmath68 est dduite des thormes de dirichlet et roth ( ou de lingalit de liouville ) . pour les nombres quadratiques , le zoom critique est directement li aux quations de pell - fermat . trouver des points dans le zoom quivaut plus ou moins trouver des solutions _ primitives _ ( dfinition [ df : primitive ] ) de ces quations . la thorie algbrique des nombres pour les corps @xmath78 o @xmath79 est un entier naturel sans facteur carr nous donne une bonne comprhension de la structure des solution des quations @xmath80 pour @xmath81 . pour les quations de la forme @xmath82 , grce la rsolubilit de lquation @xmath83 on peut toujours localiser certaines solutions , qui suffisent pour lutilisation que nous en faisons . on remarque aussi que la constante implicite dans le premier nonc est explicitement calculable . voir les dtails et les noncs prcis dans la section [ se : theoryalgebraic ] . le deuxime nonc nous dit quil existe beaucoup dapproximants ayant la constante dapproximation proche de @xmath84 . mais il semble difficile de construire des suites de points rationnels ayant cette proprit mme pour un seul nombre cubique ( voir @xcite ) . en fait dans la dmonstration , on utilise le thorme de roth sous la forme ineffective : pour tout @xmath85 , il existe une constante @xmath86 telle que pour tout rationnel @xmath87 , on ait @xmath88 mais la constante @xmath89 nest pas explicite . en terme de lexpansion en fraction continue , on ne sait pas comment contrler la croissance des quotients partiels sauf pour les nombres quadratiques . la consquence est lineffectivit du terme derreur . la difficult majeure de comptage est que ( quand on prend une fonction caractristique dun intervalle par exemple ) les points que lon veut dnombrer sont des points entiers primitifs dans une rgion trs fine . la formule empirique ( le terme principal est le volume et le terme derreur est de grandeur de la longueur du bord ) nest plus applicable . lide que lon va adapter est la thorie dquirpartition modulo @xmath65 . nous allons dmontrer cet nonc pour des nombres rels ayant la mesure dirrationalit @xmath90 ( dfinition [ def : irrationality ] ) , et pour certains nombres quadratiques particulier avec la constante implicite prcise ( cf . la section [ se : zoomalgebraic ] ) . la dmonstration du thorme [ th : maintheorem ] utilise le paramtrage par des courbes nodales . la premire partie se dduit de celle du thorme [ th : secondtheorem ] en remarquant que dans un voisinage fix du point @xmath17 , il ny a quun nombre fini de courbes nodales qui interviennent pour le zoom critique . pour le zoom sous - critique , sur certaines courbes nodales on trouve une quidistribution au sens du deuxime nonc du thorme [ th : secondtheorem ] ( cf . proposition [ po : zoompointsoncab ] _ infra _ ) . chacune de ces courbes donne des points dordre de grandeur gale une puissance de @xmath61 . le problme est de les sommer . pour cela on utilise la technique dveloppe par r. de la bretche et t. d. browning dans @xcite , @xcite , @xcite pour traiter lordre moyen des diviseurs pour des forme binaires , qui nous entrane une formule asymptotique pour le dnombrement de points sur les courbes do nt le paramtre est petit . la mesure @xmath58 qui minore le zoom peut tre vue comme une saturation de ces courbes nodales , chacune tant quipe dune mesure uniforme . ceci est analogue lexemple de la surface del pezzo torique de degr @xmath46 tudie dans @xcite , o la mesure limite existe pour le zoom critique , qui est obtenue en utilisant le paramtrage par des droites . avec les techniques utilises dans ce texte , on nest pas capable de trouver une formule asymptotique pour le zoom sous - critique de la surface @xmath13 ( cest - - dire `` @xmath91 '' remplac par `` @xmath92 '' ) . toutefois le deuxime nonc assure que lordre de croissance de la distribution locale est minor par une puissance positive de @xmath61 avec une mesure uniforme . esquissons lorganisation du texte . dans la section [ se : approximationconstants ] on introduit la notion de constante dapproximation et lon prcise lopration de zoom . la section [ se : theoryalgebraic ] est consacre aux proprits fondamentales des corps quadratiques rels et lapplication aux quations de pell - fermat . en les utilisant , on dmontre le thorme [ th : secondtheorem ] en tudiant la distribution locale critique des nombres algbrique quadratiques dans la premire partie de la section [ se : zoomalgebraic ] , et dans la deuxime partie on dmontre la formule asymptotique pour le zoom sous - critique de certains nombres rels bas sur les thormes classiques sur lquirpartition modulo @xmath65 . ltude de la distribution locale de la surface torique @xmath13 ( tout niveau ) est faite dans la dernire section [ se : toricsurface ] , o dmontrons - nous le thorme [ th : maintheorem ] . nous rfrons le lecteur aux appendices pour une prsentation du rsultat de s. pagelot et celui de r. de la bretche et t. d. browning . * notations : * on note @xmath93 la fonction indicatrice deuler , @xmath94 le fonction arithmtique donnant le nombre total de diviseurs donne par : @xmath95 et @xmath96 la fonction de mbius . on dfinit , pour @xmath97 , @xmath98 alors on a @xmath99 pour un nombre rel @xmath100 , on note @xmath101 _ la partie entire _ de @xmath100 qui est le plus grand entier @xmath102 , et @xmath103 _ la partie fractionnaire _ de @xmath100 . fixons @xmath104 un ensemble . pour @xmath105 , on note @xmath106 le sous - ensemble de @xmath104 dfini par @xmath107 pour @xmath108 un sous - ensemble de @xmath109 , on note @xmath110 lensemble des points primitifs de @xmath108 . on dfinit plusieurs _ constantes dapproximation _ dun point rationnel sur une varit algbrique gnralisant la notion de _ mesure dirrationalit _ venant de lapproximation diophantienne classique . dans cet article , nous considrons uniquement les distances archmidiennes relles . une dfinition analogue pourrait sappliquer aux places non - archmidiennes . soit @xmath1 une varit projective dfinie sur un corps de nombres @xmath2 et quipe dune distance projective @xmath32 ( cf . un fibr en droites sur @xmath1 muni dune hauteur de weil absolue @xmath112 . on fixe un point rationnel @xmath24 . pour @xmath113 une partie constructible de @xmath1 , on considre les ensembles @xmath114 @xmath115 chacun est un intervalle : si @xmath116 , tout @xmath117 appartient @xmath118 ; de mme , si @xmath119 , tout @xmath120 appartient @xmath121 . soit @xmath5 un ouvert de @xmath1 . on dit que @xmath111 vrifie _ la proprit de northcott _ pour @xmath5 si pour tout @xmath122 , @xmath123 [ pd : appconstant ] soient @xmath24 et @xmath113 une partie constructible de @xmath1 . supposons que @xmath111 vrifie la proprit de northcott pour un ouvert de zariski non - vide @xmath5 contenant @xmath17 . alors on a @xmath124 cette quantit est appele _ constante dapproximation _ en @xmath17 dans @xmath113 , note @xmath125 sil le fibr @xmath111 est fix dans la considration . sil existe une sous - varit ferme @xmath126 contenue dans @xmath113 tel que @xmath127 , on dit que @xmath128 peut tre calcule sur @xmath126 . cette dfinition ne dpend pas du choix du choix de @xmath5 . dans la suite , lorsque lon parle de la constante dapproximation , on admet que @xmath111 vrifie la proprit de northcott pour un ouvert dense contenant @xmath17 . [ df : essential ] on dfinit _ constante essentielle _ en @xmath17 comme la quantit @xmath129 o @xmath113 parcourt tous les ouverts denses de @xmath1 . sil existe une sous - varit @xmath126 de @xmath1 de sorte que pour tout ouvert dense @xmath130 , @xmath131 , on dit que @xmath126 est _ localement accumulatrice_. ces notions ne dpendent ni du choix de la hauteur associe fibr @xmath111 , ni du choix de la distance projective . puisque le problme que lon tudiera est local et que le point que lon va approcher est dans lorbite ouverte , on peut se restreindre des ouverts de @xmath1 . dans la suite on suppose que le fibr en droites @xmath111 est gros , cest - - dire sa classe est dans lintrieur du cne pseudoeffectif de @xmath1 . alors il existe un entier @xmath132 tel que lapplication rationnelle @xmath133 est birationnelle vers son image . en particulier on peut choisir un ouvert @xmath5 tel que @xmath134 et donc @xmath135 grce au thorme de northcott . dans ce qui va suivre , on suppose toujours que @xmath111 vrifie cette proprit , ce qui est le cas pour le fibr anticanonique des varits toriques compltes lisses ( parce que le polytope associ @xmath136 est le reflt de lventail de @xmath1 contenant lorigine , cf . 3.4 p. 66 ) . en outre le fibr @xmath136 est sans lieu de base sur lorbite ouverte et donc la proprit de northcott est vrifie . toutes ces constantes sont connues en dimension @xmath65 . [ th : approximationtheorem1 ] soient @xmath137 et @xmath138 . on fixe une hauteur de weil absolue associe @xmath139 . alors @xmath140 [ rmk : thmofmckinnon ] la constante dapproximation est sensible aux singularits . prenons une courbe rationnelle @xmath89 dfinie sur @xmath2 et @xmath111 un faisceau inversible ample sur @xmath89 . on fixe @xmath141 . soit @xmath142 le morphisme de normalisation . alors @xmath143 o @xmath144 , @xmath145 est la multiplicit de la branche de @xmath89 passant par @xmath17 correspondant @xmath146 et @xmath147 ici @xmath148 veut dire que @xmath149 , qui arrive par exemple pour @xmath150 et @xmath17 un point imaginaire @xmath151 et donc il est impossible dtre approch par des nombres rationnels . nous ne considrons pas ce cas dans ce texte . cest un fait empirique que les approximants proviennent essentiellement des courbes rationnelles . en effet d. mckinnon a propos la conjecture suivante ( * ? ? ? * conjecture 2.7 ) : [ conj : mckinnon ] soient @xmath111 un fibr ample sur @xmath1 avec une hauteur de weil choisie et @xmath152 . supposons quil y a une courbe rationnelle passant par @xmath17 ( ceci implique en particulier que @xmath153 . alors il existe une courbe rationnelle @xmath89 telle que @xmath154 cette conjecture prdit que @xmath10 peut tre calcule sur des courbes rationnelles passant par @xmath17 , qui est accord avec notre cas . on utilise souvent la procdure suivante pour dterminer la constante essentielle . elle consiste essentiellement en deux tapes : dmontrer une borne infrieure uniforme valide dans un ouvert pour le produit dune certaine puissance de la distance avec la hauteur et trouver une famille de courbes rationnelles do nt la runion dans la varit est dense pour la topologie de zariski et chacune atteint la constante dapproximation souhaite . il convient de remarquer que cette procdure donne seulement une condition suffisante pour trouver @xmath155 . en gnral il nest pas clair que @xmath155 puisse toujours tre atteinte sur une famille de courbes rationnelles . maintenant nous allons dcrire lopration de zoom en dtail . on identifie localement @xmath156 avec lespace tangent @xmath26 en envoyant @xmath17 sur lorigine , en utilisant un systme de coordonnes analytiques . dans @xmath26 , on dfinit une contraction par une homothtie de rapport une puissance de @xmath61 ( la lettre @xmath61 dsigne ici la borne de la hauteur ) . plus prcisment , si lon note le diffomorphisme entre @xmath156 et @xmath26 par @xmath25 , alors _ loprateur de zoom _ @xmath157 avec _ le facteur _ @xmath30 est dfini par @xmath158 fixons @xmath5 une sous - varit de @xmath1 . rappelons la notation @xmath47 ( notation [ notationcqbx ] ) . soit @xmath56 , on introduit la mesure @xmath159 dfinie par @xmath160 cette dfinition dpend de la diffrentielle en lorigine du diffomorphisme choisi . existe @xmath161 tels que soit vrifie pour une infinit de @xmath61 . on peut donner , laide de la conjecture de batyrev - manin et celle de peyre sur lquidistribution des points rationnels , une prdiction nave de cet ordre de grandeur . soit @xmath162 . quand on calcul @xmath163 , on est en train de compter des points rationnels dans la boule @xmath164 do nt le volume est de grandeur @xmath165 . supposons la validit de la conjecture de batyrev - manin pour louvert @xmath5 , cest - - dire ( on note @xmath166 ) @xmath167 et la validit dune forme forte de lquidistribution , cest - - dire pour des voisinages analytiques do nt la taille dpendent de @xmath61 , alors on devrait avoir dans , @xmath168 autrement dit on devrait avoir @xmath169 en fait , dans tous les ( rares ) exemples connus , la valeur de @xmath8 est accord avec cette prdiction , mais celle de @xmath170 ne lest pas . il arrive parfois que @xmath171 , comme pour la varit considre dans ce texte . le comportement de la suite dcrit la _ distribution locale _ autour du point @xmath17 sur @xmath1 . on appelle _ mesure limite _ comme la limite de la suite ( si elle existe ) . prenant du recul , une minoration du type @xmath172 pour une mesure @xmath173 dfinie sur @xmath26 veut dire que la distribution locale est uniformment minore par une mesure , ce qui assure lexistence de beaucoup de points dans le zoom . cest ce que nous allons dmontrer pour certains zooms de la surface @xmath13 . inspir par le fait que certaines courbes rationnelles contiennent beaucoup plus de points que les autres dans une mme varit dans lopration de zoom , bien quelles aient la mme constante dapproximation , nous proposons la dfinition suivante pour les distinguer . [ df : weaklocal ] supposons que @xmath174 . on dit quune sous - varit ferme stricte @xmath175 de @xmath1 est _ localement faiblement accumulatrice _ si @xmath176 et pour tout @xmath35 suffisamment grand , pour tout ouvert @xmath5 dense pour la topologie de zariski satisfaisant @xmath177 et @xmath178 dense dans @xmath175 , @xmath179 autrement dit , les sous - varits localement faiblement accumulatrices ne sont pas localement accumulatrices mais dominent leur complment dans le zoom critique . le zoom dun certain facteur va mettre en vidence les sous - varits do nt la constante dapproximation atteignent ce facteur et ignorer ceux do nt la constante dapproximation est plus grande . donc le zoom dun facteur plus petit que @xmath155 va nous ramener des sous - varits fermes localement accumulatrices . cela plus la proposition suivante explique la raison pour laquelle on sintresse au zoom avec le facteur @xmath180 . [ po : weakzoom ] soit @xmath5 une partie analytique de @xmath1 pour laquelle @xmath111 vrifie la proprit de northcott telle que @xmath177 . alors pour tout @xmath181 et @xmath56 , on a 1 . @xmath182 si @xmath183 ; 2 . @xmath184 sinon , pour @xmath185 . on peut supposer que @xmath60 est une fonction caractristique @xmath36 pour @xmath35 . on choisit une distance @xmath32 sur @xmath26 . daprs la proposition - dfinition [ pd : appconstant ] , pour @xmath186 tel que @xmath187 , il existe une @xmath188 telle que pour tout @xmath189 diffrent de @xmath17 , @xmath190 . supposons que @xmath191 . cela implique que @xmath192 les points aprs zoom doivent situer dans le support de @xmath60 , donc @xmath193 do @xmath194 donc un tel @xmath146 nexiste pas quand @xmath195 . [ def : criticalzoom ] on dit que le zoom est _ critique _ ( resp . _ sous - critique _ ) si son facteur @xmath43 ( resp . @xmath196 ) . dans cet article on sintresse aux varits toriques lisses et projectives do nt le fibr en droites anticanonique est engendr pas ses sections globales et on utilise une hauteur de weil associe . les travaux @xcite @xcite @xcite confirment la conjecture de manin pour les varits toriques munies dune hauteur de weil associe au fibr anticanonique . ceci est en faveur de la prdiction prcdente sur lordre de grandeur ( au mois pour @xmath8 ) . on rappelle des faits classiques sur les corps quadratiques rels . pour les dtails , voir par exemple ( @xcite 2.5 , 4.6 , 5.4 ) . on fixe un entier @xmath197 sans facteur carr . soit @xmath198 . on note @xmath199 lunit fondamentale de @xmath200 telle que @xmath201 . alors le groupe dunits @xmath202 . on introduit les sous - groupes de @xmath203 : @xmath204 @xmath205 on note @xmath206 ( resp . @xmath207 ) llment de @xmath208 ( resp . @xmath209 ) do nt la classe engendre le groupe @xmath210 ( resp . @xmath211 ) . [ le : samuel ] @xmath212 pour @xmath81 , on dsigne par @xmath213 lensemble des solutions entires de lquation pell - fermat @xmath214 @xmath215 on note aussi @xmath216 la notation @xmath217 signifie ici que @xmath218 est un idal de lanneau @xmath219 . on utilise @xmath220 pour dsigner la norme des idaux et @xmath221 pour dsigner la norme des lments . on dfinit @xmath222 on note in ( resp . de , ra ) lensemble des nombres premiers qui sont inertes ( resp . sont dcomposs , se ramifient ) dans lextension @xmath223 . pour tout @xmath224 , sa conjugaison est dfinie par @xmath225 on a besoin dune estimation explicite du cardinal de lensemble @xmath226 ainsi que celui de @xmath227 . on fixe @xmath81 . soient les entiers @xmath228 dfinis par la factorisation @xmath229 o @xmath94 dsigne la fonction arithmtique de nombre du diviseur ( cf . ) . on observe que lensemble @xmath226 est parfois vide . [ co : upperboundforimm ] @xmath230 quelque soit @xmath231 , on considre sa dcomposition @xmath232 alors comme lapplication @xmath220 est multiplicative , @xmath233 la quantit droite est en fait la factorisation de @xmath132 ( [ factorisationofm ] ) . on en conclut que @xmath234 en particulier @xmath235 est pair pour @xmath236 . rciproquement , pour tout @xmath236 , on choisit @xmath237 pair et pour tout @xmath238 , donnons - nous un couple @xmath239 tel que @xmath240 . le produit @xmath241 est un idal de norme @xmath242 . pour chaque @xmath238 , il y a @xmath243 tels couples dentiers . do lnonc . le groupe @xmath208 agit sur @xmath244 librement et lensemble des orbites est en bijection avec @xmath245 [ co : structuretheorem1 ] le groupe @xmath209 agit sur lensemble @xmath213 librement et on a la majoration suivante pour le cardinal de lensemble des orbites : @xmath246 daprs le lemme [ le : samuel ] on a @xmath247\leqslant 3 $ ] . le cardinal des orbites de laction de @xmath209 sur @xmath213 , qui est aussi libre , est major par @xmath248 la conclusion dcoule du corollaire [ co : upperboundforimm ] . [ df : primitive ] une solution @xmath249 de lquation est dite _ primitive _ si @xmath250 . toute solution de vrifie @xmath251 . donc lorsque @xmath252 , toutes les solutions sont automatiquement primitives . un avantage de la dcomposition de @xmath213 par laction de @xmath209 est le suivant . [ le : pgcddonotchange ] supposons que @xmath253 . alors @xmath254 ne dpend pas des lments @xmath255 qui sont dans une mme orbite . soient @xmath256 tels que @xmath257 . alors @xmath258 comme idaux . donc pour tout nombre premier @xmath259 , on a @xmath260 on appelle _ famille de solutions _ de lensemble des solutions correspondant une orbite de laction de @xmath261 dans @xmath262 . sil existe une solution @xmath263 qui est primitive , alors lorbite par laction de @xmath209 donne une famille de solutions primitives . si la solution nest pas primitive , on note @xmath264 et on a @xmath265 . alors @xmath266 est une famille de solutions _ primitives _ de lquation @xmath267 do nt le coefficient est plus petit . donc en lidentifiant avec une famille de solutions primitives dans @xmath268 , on peut dfinir @xmath269 comme lensemble des familles des solutions primitives des quations pour toutes les valeurs de @xmath132 . alors @xmath270 puisque tout couple @xmath271 engendre une famille de solutions primitives de pour @xmath132 arbitrairement grand . cela est suffisant pour lutilisation ultrieure , et lon na pas besoin de dcrire pour quels @xmath132 , on a @xmath272 . maintenant on considre les quations diophantiennes @xmath273 dabord on dmontre que les solutions de sont contenues dans celles des quations de type @xmath80 , do nt la structure est connue daprs la discussion prcdente . deuximement on dmontre que quand @xmath274 , lensemble des solutions de lquation ( modulo @xmath275 ) est engendr par une puissance de lunit fondamentale @xmath276 . cela nous permet de construire une suite de solutions primitives de partir de nimporte quelle solution primitive de pour @xmath277 fix . on factorise @xmath278 avec @xmath279 sans facteur carr . ce qui nous ramne aux quations considres prcdemment est lobservation simple suivante . [ le : passtogeneralequations ] on note @xmath280 lensemble des solutions @xmath281 de . alors @xmath282 . plus prcisment , on a @xmath283 @xmath284 maintenant comme toute solution @xmath281 de lquation @xmath285 vrifie @xmath286 , en crivant @xmath287 et en utilisant les notations , on obtient lquivalence suivante @xmath288 notons quici @xmath289 est un entier sans facteur carr . ainsi chercher des solutions de revient rsoudre une quation de pell sous certaine condition de divisibilit , do nt la rsolubilit est assure grce un thorme de dirichlet . on donne une preuve courte pour faciliter la lecture . soit @xmath79 en entier positif qui nest pas un carr . on note @xmath290 alors lensemble des solutions @xmath291 de lquation @xmath292 est un sous - groupe de @xmath293 vrifiant que @xmath294 est cyclique engendr par la classe dune puissance de lunit fondamentale @xmath295 du corps @xmath296 . lensemble @xmath297 est un sous - groupe ( multiplicatif ) de @xmath293 . puisque @xmath291 est le noyau du morphisme @xmath298 dans le diagramme commutatif suivant , @xmath299 \ar[r ] & \mathcal{b}_d^*=\{z\in{\mathbb{z}}+{\mathbb{z}}\sqrt{d}:n_{{\mathbb{q}}(\sqrt{d})/{\mathbb{q}}}(z)=1\ } \ar@{^{(}->}[d]^j \ar[rd]^-{i\circ j } \\ \ker ( i)\ar@{^{(}->}[r]&({\mathbb{z}}+{\mathbb{z}}\sqrt{d})^ * \ar[r]^-i & ( { \mathbb{z}}+{\mathbb{z}}\sqrt{d}/(s))^ * , } \ ] ] qui part de lensemble @xmath300 de cardinal infini vers limage @xmath301 de cardinal fini . donc le noyau @xmath291 doit tre un ensemble infini ( mme cyclique ) . [ po : pgcddonotchange ] le groupe @xmath302 agit sur lensemble des solutions de de la manire suivante . pour @xmath303 , et @xmath249 une solution de , on note @xmath304 alors @xmath305 dfini par @xmath306 est une solutions de . de plus , @xmath307 . on vrifie que @xmath308 soit @xmath309 , alors @xmath310 do @xmath311 on a videmment @xmath312 puisque ( comme dans la dmonstration du lemme [ le : pgcddonotchange ] ) @xmath313 on en dduit que @xmath314 do lnonc . [ rmk : inftysol ] soient @xmath315 qui engendrent le groupe @xmath316 . on fixe @xmath317 une solution de et soient @xmath318 dfinis comme prcdemment . comme @xmath319 , on en conclut que les nombres rationnels @xmath320 forment une suite dapproximants de @xmath321 . comme lon a vu prcdemment , @xmath322 puisque tout couple de types @xmath323 ou @xmath324 appartiennent cet ensemble . ainsi on peut construire une infinit de suites de solutions primitives laide de la procdure ci - dessus . si @xmath281 est primitif , on obtient alors une suite dapproximants _ primitifs_. dans cette section on tudie la distribution locale dun point @xmath325 . on identifie localement @xmath67 avec @xmath326 via lapplication dfinie pour @xmath327 , @xmath328\longmapsto \frac{u}{v}\quad ( [ \theta:1]\longmapsto \theta).\ ] ] on utilise la hauteur de weil canonique associe @xmath66 : @xmath329)=\max(|u|,|v| ) , \quad(u , v)\in{\mathbb{z}}^2,\quad \operatorname{\operatorname{pgcd}}(u , v)=1.\ ] ] on utilise la valeur absolue comme la distance . il convient de choisir , pour tout @xmath330 , les fonction caractristiques @xmath36 et @xmath331 dfinies par @xmath332 pour tester la rpartition asymptotique des rationnels autour du point @xmath68 . on remarque que @xmath331 est en fait la fonction caractristique de la runion des intervalles @xmath333\theta+\eta,\theta+\varepsilon\mathclose],\quad \mathopen[\theta-\varepsilon,\theta-\eta\mathclose[.\ ] ] ( mme pour @xmath36 . ) on fixe dans la suite @xmath334 et lon considre la suite de mesures @xmath335 . on voit que @xmath336 vaut le nombre de point @xmath337\in{\mathbb{p}}^1({\mathbb{q}})$ ] qui vrifient les conditions suivantes @xmath338 @xmath339 si @xmath340=2 $ ] , i.e. @xmath68 est quadratique , un tel nombre est en gnral reprsent de faon unique sous la forme @xmath341 en vertu de , il suffit de considrer les nombres de la forme @xmath342 avec @xmath343 . daprs les thormes de roth et dirichlet , on a @xmath344 . donc on prend @xmath71 correspondant au zoom critique . on rcrit linquation comme @xmath345 on a aussi lingalit de liouville @xmath346 o la constante @xmath347 est calculable ( cf . la dmonstration de la proposition [ po : effectiveliouville ] ) . le rsultat principal de cette partie est le suivant , qui dmontre la partie @xmath71 du thorme [ th : secondtheorem ] . [ th : criticzoomforquadratics ] pour @xmath348 , il existe @xmath349 , @xmath350 tels que @xmath351 par consquent , il nexiste pas de mesure limite pour le zoom critique . ce thorme est une consquence des propositions [ po : controlofepsilon ] , [ po : avoidance ] et [ po : trackpoints ] . dans @xcite s. lang a considr aussi le dnombrement dapproximants de nombres quadratiques mais sans lopration de zoom ( cest - - dire seulement la condition de borne de la hauteur est impos ) . il a obtenu une formule asymptotique du type @xmath352 , ce qui est une consquence directe des conclusions sur la structure des solutions des quation de pell - fermat . nous trouvons dabord le lien avec des quations de pell - fermat . les quations et nous donnent @xmath353 on en conclut que @xmath354 ne prend quun nombre fini de valeurs entires et ce nombre est indpendant de @xmath61 . cela nous permet de dcomposer la quantit @xmath355 de la faon suivante @xmath356 o @xmath357 de plus , on a , daprs , @xmath358 do @xmath359 . cest - - dire quand le diamtre du voisinage auquel la fonction caractristique correspond est suffisamment petit , il ny a pas de points rationnels pour tout @xmath61 . on ne cherche pas trouver la valeur maximale possible de @xmath360 puisque lintrt de cette ingalit est quil rvle , comme dans le cas de la distribution locale dun point rationnel , un phnomne de trou . voir a.1 pour un analogue du cas o @xmath68 est un nombre rationnel . on suppose comme prcdemment que @xmath342 . le but de cette section est double . sappuyant sur les rsultats de la section [ pell - fermatsimple ] , on donne une borne effective pour et on construit des suites de bornes telles que @xmath361 pour toute fonction caractristique @xmath331 suffisamment petite . on conserve les notations dans . [ po : controlofepsilon ] pour tout @xmath362 ( cf . ) , on a , pour tout @xmath363 , @xmath364 o @xmath279 sont dfinis par . [ po : avoidance ] il existe une fonction dtape @xmath365 qui est dcroissante et continue infrieurement telle que pour tout @xmath35 et tout @xmath55\frac{\varepsilon}{g(\varepsilon)},\varepsilon\mathclose[$ ] , on puisse choisir une suite de bornes @xmath366 qui tend vers @xmath367 de manire que @xmath368 le lemme [ le : passtogeneralequations ] nous permet de se ramener aux cas o @xmath369 o @xmath370 est sans facteur carr . plus prcisment , pour @xmath330 et pour chaque entier @xmath132 , on note @xmath371 et @xmath372 lensemble des points @xmath373 vrifiant les conditions @xmath374 @xmath375 il suffit de dmontrer quil existe @xmath366 telle que @xmath376 puisque @xmath377 on est ramen un problme concernant des solutions des quations de pell - fermat . on extrait un encadrement pour @xmath378 . dune part , @xmath379 do @xmath380 dautre part , @xmath381 do @xmath382 en rsum , on a @xmath383 en reportant dans , on a @xmath384 donc @xmath385 de la mme manire @xmath386 il en rsulte que @xmath387 daprs , on a que pour @xmath388 , @xmath389 donc entrane une borne pour le nombre de solutions dans toute famille de solutions de lquation @xmath390 lapplication @xmath391 est dfinie par @xmath392 limage dune famille de solutions @xmath393 @xmath394 est un rseau do nt la priode est @xmath395 dans @xmath396 . prenons une famille de solutions @xmath397 , on a ( compte - tenu du signe ) pour @xmath398 , @xmath399 la proposition [ co : structuretheorem1 ] donne une majoration du nombre dorbites , on arrive la borne dans la proposition [ po : controlofepsilon ] . donc pour tout @xmath400 fix , lensemble des images de toutes les solutions de ( [ eq : familyofpellequation ] ) , not @xmath401 , est une runion finie de rseaux de mme priode dans @xmath396 par lapplication @xmath391 . donc la fonction @xmath402 dfinie par @xmath403 est videmment dcroissante et continue infrieurement . elle est constante sur tout intervalle @xmath404 si lon prend @xmath405 assez proches tels que @xmath406 savoir @xmath407 on peut choisir @xmath408 de sorte quil existe une constant @xmath409 tel que lintervalle @xmath410\ ] ] vrifie @xmath411 daprs ( [ boundforloguv1 ] ) , ceci dmontre que @xmath412 pour tout @xmath413 . la fonction @xmath414 vrifie les mme proprits que celles de @xmath108 , do la proposition [ po : avoidance ] . en imitant cette mthode , quand la fonction caractristique choisie permet davoir une solution dune quation pell - fermat gnralise , on peut choisir la suite @xmath366 de la faon quelle donne au moins une famille de solutions . [ po : trackpoints ] soit @xmath415 avec @xmath343 . pour tout couple @xmath416 , @xmath330 , tel que lintervalle @xmath4172\sqrt{ab}\theta^{-2}\eta,2\sqrt{ab}\theta^{-2}\varepsilon\mathclose[$ ] contienne la valeur absolue dun entier @xmath132 de sorte que @xmath418 contienne une solution primitive ( cf . les notations du lemme ) . alors il existe une suite de bornes @xmath408 tel que @xmath419 on choisit @xmath4200,1\mathclose[$ ] tel que @xmath421 2\sqrt{ab}\theta^{-2}\eta,2\sqrt{ab}\theta^{-2}\varepsilon \lambda^2\mathclose[$ ] . on suppose dans la suite que @xmath422 , la dmonstration tant analogue lorsque @xmath423 . soit @xmath122 tel que @xmath424 on dfinit lensemble @xmath425 des @xmath426 primitifs vrifiant @xmath427 pour @xmath428 , on a , daprs , @xmath429 en revanche , grce la deuxime ingalit de , @xmath430 on note @xmath431 . alors la positivit de @xmath132 implique celle de @xmath1 . notons que @xmath432 , on obtient , daprs lhypothse sur @xmath132 , @xmath433 cela entrane aussi la majoration @xmath434 ainsi que la minoration pour @xmath1 : @xmath435 par la premire ingalit de . cela montre que @xmath436 on prend @xmath437 une solution primitive de lquation @xmath438 on dfinit @xmath439 comme dans la remarque . comme @xmath440 , on peut choisir @xmath408 de la manire que @xmath441 donc pour tout @xmath400 suffisamment grand , on a @xmath442 , do @xmath443 pour @xmath348 fixs , on peut choisir @xmath444 tel que @xmath445 et que @xmath418 contienne une solution primitive ( remarque [ rmk : inftysol ] ) . rappelons les proprits des fonctions @xmath108 et @xmath414 dans la dmonstration de la proposition [ po : avoidance ] . il existe alors @xmath446 tel que @xmath447 on dfinit @xmath448 maintenant on prend @xmath449\frac{m\theta^2}{2\sqrt{ab}}-\frac{\delta_1}{2|\varepsilon_d^*|^2},\frac{m\theta^2}{2\sqrt{ab}}\right[.\ ] ] puisque @xmath450 le couple @xmath416 vrifie toutes les hypothses des propositions [ po : avoidance ] , [ po : trackpoints ] . comte - tenu de la majoration fournie par la proposition [ po : controlofepsilon ] , il en dcoule donc que @xmath451 on va prouver un rsultat sur la distribution locale en un point rel vrifiant des ingalits de type analogue celle de liouville , qui _ a fortiori _ sapplique aux nombres algbriques . pour cela on rappelle des notions standard dapproximation diophantienne des nombres rels . [ def : irrationality ] soit @xmath452 . on dit quun nombre rel @xmath68 a _ un ordre dirrationalit _ @xmath453 sil existe une constante @xmath454 telle que @xmath455 _ la mesure dirrationalit _ @xmath456 est la borne infrieure de lensemble des ordres dirrationalit de @xmath68 . de manire quivalente , elle est gale au maximum des nombres @xmath457 tels que @xmath458 si @xmath68 est algbrique , @xmath456 est rien dautre que la constante dapproximation @xmath459 associe au fibr @xmath66 . le nombre @xmath90 est un ordre dirrationalit et la mesure dirrationalit dun nombre quadratique . pour tout @xmath35 , le nombre @xmath460 est un ordre dirrationalit dun nombre algbrique de degr @xmath461 do nt la mesure dirrationalit est @xmath90 . le rsultat principal de cette section est le suivant , qui est valide pour tout nombre algbrique de degr @xmath462 . pour ceux do nt le degr est @xmath90 , on a une meilleur contrle du terme derreur . [ th : localdistributionofrealnumbers ] soient @xmath463 un nombre rel et @xmath76 . alors pour tout @xmath464 et @xmath465 , * si @xmath466 , supposons que @xmath467 , on a @xmath468 si de plus @xmath90 est un ordre dirrationalit de @xmath463 , savoir @xmath4690,1\mathclose[$ ] tel que pour tout nombre rationnel @xmath470 on ait @xmath471 on a , en notant @xmath472 @xmath473 * si @xmath474 , on a @xmath475 la dpendance du terme derreur dans sur @xmath463 et @xmath94 est intrinsque au sens similaire au thorme de roth . en effet , elle dpende de les constante @xmath476 dans avec @xmath453 proche de @xmath90 ( cf . aussi _ infra _ ) . alors que le terme derreur dans ne dpende que de la taille de @xmath463 . tout dabord on reformule ce problme de zoom en un problme de dnombrement . pour @xmath477 , on dfinit @xmath478 lensemble des point @xmath337 $ ] vrifiant @xmath479 graphiquement , si lon prend @xmath480 , cest - - dire on prend la fonction de test @xmath36 et on calcule @xmath481 , on compte des points entiers primitifs lintrieur du triangle do nt laire est dordre de grandeur @xmath482 et celle de la longueur du bord est @xmath61 . donc la comparaison classique avec laire du triangle nest utilisable que pour les cas o @xmath474 . ceux o @xmath483 ncessitent un travail supplmentaire . notre approche sappuie sur la thorie dquirpartition modulo @xmath65 utilise dans @xcite . [ de : discrpance ] soit @xmath484 une suite de nombres rels . elle est dite _ quirpartie modulo @xmath65 _ si pour tout @xmath485 , on a @xmath486 soit @xmath487 une suite de nombres rels quirpartie modulo @xmath65 . soit @xmath488 . ( @xmath489 ) @xmath490 de cette suite est dfinie par @xmath491}\left|\frac{\sharp\{n\leqslant n:\{x_n\}\in\mathopen[0,\lambda\mathclose]\}}{n}-\lambda\right|.\ ] ] si la suite @xmath492 pour un nombre @xmath463 rel fix , on notera @xmath493 . on rappelle lingalit de koksma - denjoy pour les suites quirparties modulo @xmath65 . ( voir , par exemple , @xcite p. 143 ) [ th : koksma - denjoy ] soient @xmath487 une suite quirpartie modulo @xmath65 et @xmath494 . soit @xmath495 une fonction mesurable variation borne dfinie sur @xmath496 $ ] ( on note @xmath497 la variation totale de @xmath495 ) . alors @xmath498 lingalit de erds - turn donne un contrle de la discrpance ( cf . @xcite p. 122 - 123 ) pour les nombres irrationnels do nt la mesure dirrationalit est finie . [ co : controlforthediscrepancy ] soit @xmath68 un nombre irrationnel ayant un ordre dirrationalit @xmath452 . alors pour tout @xmath499 , on a @xmath500 si les quotients partiels dans lexpansion en fraction continue dun nombre irrationnel sont born , de manire quivalent , lingalit de liouville tant vrifie , alors une meilleure majoration de la discrpance existe ( cf . * theorem 3.4 ) , p. 125 ) . soient @xmath501 $ ] un nombre irrationnel . supposons que les quotients partiels de @xmath502 sont borns , i.e. il existe @xmath503 tel que @xmath504 . alors @xmath505 o @xmath506 . [ le : boundforboundquotients ] soit @xmath502 un nombre irrationnel . existe @xmath5070,1\mathclose[$ ] tel que pour tout nombre rationnel @xmath470 , on ait @xmath508 soit @xmath501 $ ] son expansion en fraction continue . alors pour tout @xmath509 , @xmath510 par consquent,@xmath511 soient @xmath512 les convergents principaux de @xmath502 . alors ils vrifient , pour tout @xmath513 , la rgle de rcurrence ( voir ( * ? ? ? * theorem 1 ) ) @xmath514 @xmath515 et ( voir ( * ? ? ? * theorem 9 ) ) pour tout @xmath516 , @xmath517 il en dcoule que @xmath518 et donc @xmath519 en reportant dans , on obtient que pour @xmath516 , @xmath520 on peut supposer que @xmath521 . on fixe @xmath522 . + * cas @xmath523 . * on suppose que @xmath524 . la premire tape est de comparer le cardinal de @xmath478 avec celui de @xmath525 qui consiste en les @xmath526 de sorte que @xmath527 on prend @xmath528 . alors implique que @xmath529 donc pour un tel @xmath378 choisi , il ny a quun nombre fini de choix pour @xmath530 . de plus on a @xmath531 on en conclut que @xmath532 rciproquement , si lon prend @xmath533 , implique que @xmath534 on en conclut de faon analogue que @xmath535 et ainsi que @xmath536 on est ramen calculer le cardinal de lensemble @xmath525 . laide de linversion de mbius , on dfinit pour @xmath138 , @xmath537 o @xmath538 consiste en les @xmath526 tels que @xmath539 et @xmath538 consiste en les @xmath526 tels que @xmath540 on obtient une borne sur @xmath32 de la faon suivante . dune part , @xmath541 daprs lhypothse sur @xmath463 , pour tout @xmath85 , il existe une constante @xmath542 telle que pour tout @xmath543 , @xmath544 donc @xmath545 en combinant les deux ingalits on obtient @xmath546 donc on peut restreindre la somme en @xmath32 ceux vrifiant @xmath547 si de plus est valide , alors @xmath548 dans ce cas en combinant on obtient @xmath549 on tablit maintenant les formules asymptotiques suivantes un peu plus gnrales faisant intervenir une constante fixe @xmath550 , qui permet de lappliquer une hauteur quivalente celle donne . pour une utilisation ultrieure , on prcise la dpendance en @xmath550 du terme derreur . [ po : smallepsilon ] fixons @xmath551 et @xmath552 . supposons que @xmath553 on dfinit lensemble @xmath554 des @xmath526 satisfaisant @xmath555 ainsi que @xmath556 alors pour tout @xmath499 et pour tout @xmath557 , @xmath558 si de plus est vrifie , on a @xmath559 les formules ci - dessus restent valides si lon remplace par @xmath560 soit @xmath561 fix dans la suite . on peut supposer que @xmath489 est un entier ( quitte rajouter des constantes absolues dans les termes derreur ) . on considre le dcoupage de lintervalle @xmath5620,\frac{k}{\alpha d}b \mathclose ] $ ] en les intervalles @xmath563\frac{(k-1)kb}{\alpha d n},\frac{kkb}{\alpha d n}\right],\quad ( 1\leqslant k\leqslant n).\ ] ] soit @xmath564 tel que @xmath565 on a @xmath566 et donc @xmath567 daprs la condition et le choix de @xmath489 , @xmath568 on en conclut que pour tout @xmath569 et pour tout entier @xmath378 fix , il existe au plus un entier @xmath530 tel que @xmath570 sont uniques . fixons dans la suite @xmath2 et considrons lintervalle @xmath571 on cherche maintenant une quivalence de lhypothse suivante : * ( h ) * pour @xmath572 fix , il existe un ( unique ) @xmath573 tel que @xmath574\frac{\eta ( k-1)k}{\alpha d n}b^{1-\frac{1}{r}},\frac{\varepsilon k k}{\alpha d n}b^{1-\frac{1}{r}}\right [ $ ] . la difficult de la dmonstration qui suit vient du cas @xmath575 , pour lequel lintervalle @xmath576 peut contenir un entier , y compris pour des grandes valeurs de @xmath61 . on a les deux possibilit suivante : + * cas i * : lintervalle @xmath576 contient un entier @xmath577 . alors on a @xmath578 alors * ( h ) * revient dire que @xmath579 @xmath580 rciproquement , si @xmath378 vrifie lune des conditions ci - dessus , lentier @xmath530 tel que * ( h ) * soit vrifie existe et vaut @xmath581 ou @xmath582 selon lappartenance de @xmath583 . notons que @xmath584 . en dsignant @xmath5850,1\mathclose[$ ] , qui est soit un intervalle ( si lun des @xmath586 est vide ) , soit la runion disjointe de deux intervalles , on conclut que la condition * ( h ) * est quivalente @xmath587 . + * cas ii * : lintervalle @xmath576 ne contient aucun entier . suppose que @xmath588u_{k}-1,u_k\mathclose[$ ] o @xmath589 . alors dans ce cas * ( h ) * est quivalente @xmath5900,1\mathclose[.\ ] ] on vrifie que lon a une quivalence analogue si lon suppose . ayant tabli cette quivalence , on peut rduire le comptage en deux variables @xmath591 en une seule @xmath378 . par labuse de notation , pour une proprit @xmath592 , on note @xmath593 la fonction qui vaut @xmath65 si @xmath592 est vrifie et vaut @xmath594 sinon . nous avons la dcomposition suivante @xmath595 o le terme derreur @xmath596 on traite dabord le terme principal en appliquant les thormes [ co : controlforthediscrepancy ] et [ th : koksma - denjoy ] . on dfinit la fonction @xmath597\to\mathbb{r}$ ] comme @xmath598 on dcompose @xmath599 comme @xmath600 daprs le thorme [ th : koksma - denjoy ] et le thorme [ co : controlforthediscrepancy ] ( avec @xmath601 ) , @xmath602 o la constante implicite dpend de @xmath603 et @xmath463 . de faon similaire on obtient @xmath604 donc @xmath605 on somme sur tous les @xmath2 . faisons - le dabord pour le terme principal de @xmath599 : @xmath606 ensuite pour le terme derreur de @xmath607 : @xmath608 on obtient @xmath609 o la constante implicite dpend de @xmath603 et @xmath463 . ensuite on estime le terme @xmath610 en utilisant la mme mthode . en imitant la discussion pour @xmath599 , on en conclut de lexistence des ensembles @xmath611 qui sont la runion de au plus @xmath90 intervalles disjoints de @xmath612 et @xmath613 tels que @xmath614 \longleftrightarrow \{\alpha v\}\in l_{1,k}\end{aligned}\ ] ] @xmath615 \longleftrightarrow \{\alpha v\}\in l_{2,k}\end{aligned}\ ] ] on dcompose @xmath616 comme @xmath617 on estime les @xmath618 en appliquant une autre fois les thormes [ th : koksma - denjoy ] et [ co : controlforthediscrepancy ] . @xmath619 donc @xmath620 on obtient aussi une formule analogue pour @xmath621 . ainsi @xmath622 en rsum de et , on a dmontr que pour @xmath463 gnral ayant la mesure dirrationalit @xmath90 , @xmath623 maintenant , supposons la validit de lingalit . il suffit de modifier tous les termes derreur concernant la discrpance , savoir , les termes qui contiennent @xmath603 . plus prcisment , en utilisant le lemme [ le : boundforboundquotients ] , on a @xmath624 et donc , @xmath625 _ suite de la dmonstration du thorme [ th : localdistributionofrealnumbers]_. on dcompose @xmath538 en des parties @xmath626 o @xmath627 avec @xmath628 choisies de sorte que @xmath629 en appliquant la proposition [ po : smallepsilon ] ces ensembles ( avec @xmath630 ) on dduit que pour tout @xmath499 et @xmath631 , @xmath632 ce qui est valide aussi pour @xmath633 . il reste sommer sur tous les @xmath32 . que lon a pour tout @xmath85 , @xmath634 on choisit @xmath635 on vrifie que avec ce choix @xmath636 et donc @xmath637 pour @xmath638 . on rcrit comme @xmath639 en utilisant lestimation lmentaire @xmath640 la somme du terme principal de est gale @xmath641 et celle du terme derreur est dordre de grandeur gal @xmath642 finalement en reportant dans on en conclut que @xmath643 il ne reste qu prendre , pour @xmath465 donn , @xmath499 tel que @xmath644 . en reportant dans , cela dmontre finalement que @xmath645 maintenant , supposons la validit de . rappelons la borne pour @xmath32 . dans ce cas on prend @xmath646 comme lon suppose que @xmath647 , on dduit de @xmath648 on somme sur tous les @xmath32 dans le terme principal et le terme derreur respectivement . @xmath650 on en conclut que @xmath651 cela dmontre la formule . + si @xmath652 , alors il suffit de modifier la dfinition de lensemble @xmath525 en remplaant @xmath653 par @xmath654 . + * cas @xmath474 . * une inversion de mbius nous fournit que @xmath655 o @xmath656 on a , par la mthode de comparaison classique avec laire du domaine rel , @xmath657 puis on somme sur tous les @xmath32 possible et lon obtient @xmath658 les rsultats sur lapproximation dun nombre rel quon a dmontrs prcdemment peuvent tre vu comme le fait dapprocher une droite par des points primitifs de pente donne . pour une utilisation ultrieure , nous allons prsenter une version plus gnrales du thorme [ th : localdistributionofrealnumbers ] . cest - - dire , on compte des points primitifs sur un rseau proche dune droite do nt la pente est un nombre quadratique donn . le but est dessayer de prciser la dpendance des constantes implicites , qui fait intervenir le dterminant du rseau . lide est qu laide dune transformation linaire lie au rseau , lapproximation sur le rseau de la droite quivaut lapproximation sur @xmath109 dune autre droite de pente gale un autre nombre quadratique . comme lon a vu , lun des inconvnients de la dmonstration ci - dessus est que la constante @xmath659 qui apparat dans est en gnral difficile prciser , cause de labsence deffectivit du thorme de roth ( ou bien de lingalit de liouville ) . cela mne aussi une ineffectivit de la discrpance , car dans la dmonstration du corollaire [ co : controlforthediscrepancy ] , on a utilis le thorme de roth ( voir @xcite , p 123 ) . pour les nombres quadratiques , cette difficult disparat parce que lon peut prciser facilement les constantes dans lingalit de liouville et dans la majoration de la discrpance . on remarque que la technique ci - dessous fonctionne aussi pour les rels do nt les quotients partiels dans lexpansion en fraction continue sont borns ( ceux qui vrifient lingalit de liouville puissance @xmath90 ) . on ne rentrera pas dans les dtails . * notations et conventions * : on fixe un rseau @xmath660 et @xmath661 . on suppose que @xmath662 contient un point primitif , donc @xmath663 pour tout @xmath664 . les nombres quadratiques auxquels on sintresse sont de la forme @xmath665 avec @xmath666 , que lon notera toujours comme @xmath463 . en pratique , les constantes @xmath667 sont bornes et lon na pas besoin dindiquer leur rles dans les constante implicites . on note @xmath668 . et lon dfinit @xmath669 il existe @xmath670 une base de @xmath109 telle que @xmath671 puisque @xmath662 contient un point primitif , donc @xmath672}\\ & = \prod_{p\nmid \det(\lambda)}\left(1-\frac{1}{p^2}\right)\prod_{p|\det(\lambda)}\left(1-\frac{1}{p}\right)\frac{1}{p^{v_p(\det(\lambda))}}\\ & = \frac{1}{\det(\lambda)}\frac{6}{\pi^2}\prod_{p|\det(\lambda)}\frac{1-p^{-1}}{1-p^{-2}}=\frac{6}{\pi^2}\frac{\psi_1(\det(\lambda))}{\det(\lambda ) } \end{split}\ ] ] o @xmath673 on voudrait estimer le cardinal de lensemble suivant @xmath674 pour rduire la difficult technique et pour obtenir une formule asymptotique , nous allons imposer plusieurs conditions techniques . on ne prtend pas que les coefficients dans et soient optimaux mais ils suffisent pour lutilisation ultrieure . @xmath675 @xmath676 @xmath677 en gros , le but dimposer la condition est de rduire le problme de dnombrement lquidistribution modulo @xmath65 , et celui de la condition est dobtenir un meilleur terme derreur . [ th : zoomingonlattices ] avec les hypothses ci - dessus , on a @xmath678 rappelons que lon peut aussi interprter @xmath679 , dans lesprit de lquidistribution , comme @xmath680 qui correspond la probabilit pour un point de @xmath681 provenant dun lment de @xmath662 primitif dans @xmath109 . le deuxime terme derreur peut facilement dpasser le terme principal quand @xmath682 est trop petit et @xmath683 est trop grand , ce qui perd lintrt de la formule . toutefois pour notre utilisation ultrieure il y aura des relations troites entre @xmath684 . il savre quen fait ce sera le premier terme derreur qui contribuera plus que le deuxime . lide de la dmonstration est que les points que lon veut dnombrer sont dans un triangle . toute transformation dfinie par une matrice lenvoie sur un autre triangle do nt laire est celle du triangle initial divis par le dterminant et la pente de lune des bords correspond un nouveau nombre quadratique approcher , que lon notera @xmath685 , ce qui nous permet de se ramener au cas trait prcdemment et appliquer la mme technique . commenons par quelques prparations . on prend un rseau @xmath686 . on choisit une base @xmath687 engendrant @xmath688 telle que ( cf . @xcite p. 135 ) @xmath689 o @xmath690 dsigne le @xmath691-me minima successif de @xmath688 par rapport la norme symtrique et convexe @xmath692 : @xmath693 le thorme de minkowski ( cf . * theorem v , viii 4.3 ) ) dit que @xmath694 on dfinit @xmath695 quitte remplacer @xmath696 par son oppos , on peut supposer que @xmath697 on tablit un thorme de liouville effectif , savoir , avec des constantes explicites . [ po : effectiveliouville ] pour tout @xmath698 , on a @xmath699 o on peut prendre @xmath700 on va reprendre la dmonstration de lingalit de liouville respectivement pour @xmath463 et @xmath68 . on note @xmath701 le polynme minimal sur @xmath395 du nombre quadratique @xmath702 . alors @xmath703 est un polynme entier qui annule @xmath68 . dune part comme @xmath704 , on a @xmath705 dautre part , daprs le thorme de valeur moyenne , pour @xmath706\alpha-1,\alpha+1\mathclose[$ ] et @xmath707\theta-1,\theta+1\mathclose[$ ] , @xmath708 puisque @xmath709 @xmath710 daprs les thormes de minkowski , en rappelant le choix de @xmath711 et @xmath712 , il en dcoule que @xmath713 @xmath714 do @xmath715 les ingalits sont triviales si @xmath716 dans le cas contraire les ingalits rsultent de , , , . on en dduit lencadrement suivant qui sera utilis frquemment . [ co : rangeoflambdamu2 ] on a @xmath717 premirement , @xmath718 pour la minoration on utilise les ingalits : @xmath719 [ co : discrepancyfortheta ] pour @xmath720 , on a @xmath721 ceci rsulte du lemme [ le : boundforboundquotients ] et de la proposition [ po : effectiveliouville ] . avant de commencer de dnombrer lensemble , on prsente deux conditions supplmentaires qui sont des consquences des conditions et pour @xmath722 , @xmath723 @xmath724 puisque @xmath725 premirement , comme toujours , on utilise linversion de mbius pour liminer la coprimalit : @xmath726 o @xmath727 pour dduire une borne prcise pour @xmath32 nous suivons . pour tout @xmath728 , on note @xmath729 . en utilisant la proposition [ po : effectiveliouville ] , on a @xmath730 do @xmath731 notons que @xmath732 ltape suivante consiste liminer la condition de rseau sur @xmath733 pour les @xmath32 petits . pour contrler la contribution des @xmath32 grands on peut utiliser la majoration donne par . on utilisera les notations , , pour le rseau @xmath734 , cest - - dire @xmath735 . on crit @xmath736 avec @xmath737 . on voudrait se ramener au dnombrement sur @xmath738 puisquil ny a plus de contraints sur la divisibilit de @xmath739 . avec les notations ci - dessus lensemble scrit @xmath740 * cas i. @xmath32 est petit . * on va estimer le cardinal de @xmath741 quand @xmath32 vrifie la condition @xmath742 la condition assure que tels @xmath32 existent . dabord on dtermine le signe de @xmath739 . en utilisant le corollaire [ co : rangeoflambdamu2 ] et en rappelant la condition , on a @xmath743 cela implique non seulement que pour @xmath744 fix , il y a au plus un @xmath745 tel que @xmath746 , mais aussi que , pour un rseau @xmath747 fix , on a @xmath748 . on ne peut avoir quune seule possibilit parce que le signe de @xmath739 dtermine aussi le signe de @xmath378 , que lon a suppos tre positif . donc dans la suite , sans perte de gnralit on suppose que @xmath739 sont positifs . en fait on a dans ce cas @xmath749 ( sinon on aurait @xmath750 et do @xmath751 ) . on dfinit lensemble @xmath752 des couples @xmath753 vrifiant @xmath754 et lensemble @xmath755 ( rappelons la notation @xmath756 . ) on a @xmath757 . soit @xmath746 . on a @xmath758 daprs et le corollaire [ co : rangeoflambdamu2 ] , on a @xmath759 notons aussi que , daprs le corollaire [ co : rangeoflambdamu2 ] , , et , @xmath760 donc @xmath761 deuximement , @xmath762 on va montrer que pour tout @xmath763 , @xmath764 en particulier , comme on a @xmath765 , cela implique que @xmath766 et donc le lemme . si @xmath746 , daprs lestimation et la proposition [ po : effectiveliouville ] , @xmath767 pour @xmath768 , on dduit directement du corollaire [ co : rangeoflambdamu2 ] que @xmath769 on en conclut @xmath770 et @xmath771 pour que soit vraie , il suffit que @xmath772 il suffit donc que @xmath773 daprs la condition et la condition sur @xmath32 , on a @xmath774 ce qui implique que @xmath775 donc lingalit qui fallait dmontrer dcoule de . maintenant on dcoupe lintervalle @xmath776\ ] ] en @xmath489 pices ( avec @xmath489 dterminer ) @xmath777,\quad 1\leqslant k\leqslant n.\ ] ] on applique le thorme [ th : koksma - denjoy ] . @xmath778 pour @xmath779 , on dfinit @xmath780 on a @xmath781 donc daprs le thorme [ th : koksma - denjoy ] avec le terme derreur prcis ( corollaire [ co : discrepancyfortheta ] ) , @xmath782 @xmath783 donc @xmath784 on calcule maintenant le terme derreur provenant du changement de @xmath744 en les valeurs du bord des intervalles . en appliquant le thorme [ th : koksma - denjoy ] la fonction auxiliaire @xmath785 on obtient @xmath786 on en conclut que @xmath787 grce la condition , @xmath788 en sommant sur tous les @xmath2 pour le premier terme ( le terme principal ) @xmath789 puis sur les termes derreur @xmath790 on obtient que @xmath791 il nous reste majorer le terme derreur venant du cardinal de @xmath752 . @xmath792 grce la condition . on en conclut que @xmath793 on choisit @xmath794 comme @xmath795 , la condition assure que @xmath561 . on conclut que @xmath796 maintenant on somme sur tous les @xmath32 petits . dabord faisons - le pour le terme principal . en rappelant la constante @xmath679 et en remarquant que @xmath797 , @xmath798 ensuite pour le terme derreur @xmath799 * cas ii . @xmath32 est grand . * pour les @xmath32 vrifiant @xmath800 on utilise la formule de la proposition [ po : smallepsilon ] avec @xmath801 , notant que la condition implique lhypothse de la proposition . on obtient @xmath802 donc la contribution totale provenant de ces @xmath32 est dordre de grandeur @xmath803 la formule est dduite de et . le thorme [ th : secondtheorem ] ne couvre pas le zoom critique des nombres algbriques de degr plus grand que @xmath90 . les experts semblent penser que lexpansion en fraction continue dun nombre algbrique @xmath463 de degr @xmath461 na pas de quotients partiels borns . cest - - dire que pour tout rel @xmath188 , il existe toujours des nombres rationnels @xmath470 vrifiant @xmath804 en terme de la distribution locale ( critique ) , on pourrait interprter cela comme : les nombres rationnels proches de @xmath463 seraient beaucoup plus nombreux que dans le cas quadratique et on naurait plus de phnomne de trou . la surface @xmath13 est une surface torique qui nest pas de fano . elle est construite comme lclatement des @xmath14 points invariants de @xmath0 . elle peut tre aussi obtenue en clatant plus de points invariants sous laction du tore sur la surface torique @xmath805 , qui est une surface de del pezzo torique de degr @xmath46 tudie dans @xcite . dans la suite on prend le premier point de vue . on considre la surface torique @xmath13 obtenue en clatant @xmath806 en @xmath14 points invariants @xmath807\times [ 1:0],[0:1]\times[1:0],[1:0]\times[0:1],[0:1]\times[0:1]$ ] . lventail de @xmath113 est le suivant . ] on note @xmath808 les tirs en arrire de sections hyperplans dans chaque @xmath809 , et @xmath810 les images inverses des points clats . en dehors des points clats on utilise encore les coordonnes @xmath811\times[s : t]$ ] de @xmath806 . les gnrateurs @xmath812 des raies de lventail de @xmath13 vrifient les relations @xmath813 ce sont toutes les _ collections primitives _ ( @xcite definition 2.6 , 2.7 , proposition 3.2 ) minimales centres pour @xmath13 . pour les varits toriques lisses complte , les collection primitives minimales centres reprsentent les courbes rationnelles de degr anticanonique minimal . dans notre cas ils correspondent @xmath14 courbes rationnelles de degr lisses @xmath90 relativement au fibr anticanonique passant par @xmath15\times[1:1]$ ] , que lon appellera les courbes spciales et lon notera @xmath814 dquations respectives @xmath815 elles appartiennent , respectivement , aux classes de @xmath816 dans le groupe de picard . pour chacun de ces fibrs il existe un unique diviseur irrductible passant par @xmath17 . il y a @xmath14 familles de courbes rationnelles de degr @xmath817 relativement au fibr anticanonique passant par @xmath17 qui sont lisses , do nt les classes sont @xmath818 le diviseur anticanonique @xmath819 do nt les sections globales sont des combinaisons linaires des monmes @xmath820 est engendr par ses sections globales mais il nest pas ample . sur louvert @xmath821 , en utilisant les coordonnes @xmath822 , on identifie localement lespace tangent de @xmath17 @xmath823 par le diffomorphisme @xmath824\times [ s : t]\longmapsto \left(\frac{y}{x}-1,\frac{t}{s}-1\right)=(w-1,z-1).\ ] ] avec cette identification , les @xmath14 courbes spciales scrivent comme @xmath817 droites et une hyperbole de degr anticanonique @xmath90 : @xmath825 maintenant on associe une hauteur de weil @xmath826.tout dabord pour un point @xmath827\times[s : t]$ ] avec @xmath828 nappartenant pas aux diviseurs exceptionnels , @xmath829 donc on peut prendre comme hauteur de weil @xmath830 par symtrie on peut se ramener la rgion @xmath831 on note @xmath832 pour @xmath827\times[s : t]\in v$ ] , la hauteur se calcule comme @xmath833 la distance que lon va utiliser est @xmath834 nous allons montrer que la meilleure constante dapproximation est @xmath90 par une estimation directe . [ po : lowerbound1 ] @xmath835 pour @xmath827\times[s : t]\neq q$ ] satisfaisant @xmath836 ( les cas o @xmath837 se dmontre de faon analogue ) , on a @xmath838 cela montre que @xmath839 . mais les courbes spciales @xmath18 donnent des sous - varits avec @xmath840 ( remarque [ rmk : thmofmckinnon ] ) . ceci clt la dmonstration . cette majoration uniforme indique un phnomne de trou en dimension @xmath90 dans le zoom critique . cest - - dire , il existe @xmath841 tel que pour tout @xmath842 , et pour tout @xmath61 , on ait @xmath843 . ceci peut aussi sinterprter comme une ingalit du type liouville : @xmath844 considrons les courbes @xmath845 avec les coordonnes @xmath846 , cette quation scrit @xmath847 elles forment donc une famille de courbes cubiques passant par les @xmath14 points invariants avec un point singulier @xmath17 . la classe de @xmath848 dans le groupe de picard est la mme que celle de @xmath826 et on a @xmath849 si @xmath850 , la courbe @xmath851 a en fait deux composantes irrductibles @xmath852 lorsque @xmath853 , la courbe @xmath854 est une courbe cubique gomtriquement intgre qui est nodale . si @xmath853 , le polynme @xmath855 est irrductible sur @xmath856 $ ] . une courbe cubique irrductible singulire ne peut pas avoir quun point singulier qui est nodal ou cuspidal . les tangentes au point @xmath17 sont de pente @xmath857 donc @xmath17 est un point nodal de @xmath854 . [ le : curvescoverallpoints ] pour tout point rationnel @xmath858 o @xmath859 , il existe une unique courbe @xmath854 passant par ce point . on crit ( de faon unique ) le quotient en une fraction positive rduite : @xmath860 alors @xmath861 . gomtriquement , on considre @xmath862 lclatement de @xmath13 en @xmath17 , et on note @xmath863 la transformation stricte de @xmath854 . alors @xmath863 est une normalisation de @xmath854 et @xmath864 . on peut voir @xmath865 comme lune des coordonnes de @xmath866 . * cas i. * @xmath867 sont deux carrs de nombres entiers . alors @xmath868 et @xmath869 . approcher @xmath17 sur la courbe revient approcher un nombre rationnel par des nombres rationnels . donc @xmath870 * cas ii . * lun des @xmath867 nest pas le carr dun entier , ( not @xmath871 ) comme @xmath872 et @xmath683 sont premiers entre eux , @xmath873 . aucun des points @xmath874 nest dfini sur @xmath44 . lapproximation du point @xmath17 le long une branche de @xmath854 est quivalente lapproximation du point _ irrationnel quadratique _ @xmath866 ( dpendant de la branche choisie ) le long @xmath863 . donc dans ce cas daprs le thorme [ th : approximationtheorem1 ] , @xmath875 cependant , pour les @xmath14 courbes spciales @xmath18 , lapproximation au point @xmath17 est vraiment une approximation dun point rationnel dfini sur @xmath12 sur @xmath67 , savoir @xmath876 . bien quelle aient la mme constante dapproximation , le nombre des points rationnels que lon trouve dans lopration de zoom sont de grandeur trs diffrente . cest cette diffrence qui explique les phnomnes radicalement diffrents entre les deux types de courbes ci - dessus . on en conclut [ th : approximationessential ] on a @xmath877 par consquent , il ny a pas de sous - varits localement accumulatrices ( dfinition [ df : essential ] ) . cela rsulte du fait que lensemble des @xmath854 vrifiant la condition ( h ) est dense ( mme pour la topologie analytique ) et de la borne infrieure que lon a tablies prcdemment ( proposition [ po : lowerbound1 ] ) donc la famille @xmath878 donnent une autre faon de paramtrer les points rationnels bien adapte notre problme car la constante dapproximation sur elles est @xmath90 , plus petite celle donne par les droites gnrales . nous utiliserons les courbes nodales @xmath854 pour paramtrer les points rationnels autour de @xmath17 . nous avons vu quelles nont pas la mme constante dapproximation et nous avons constat que lapproximation essentielle est prise sur celles vrifiant la condition @xmath871 . regardons dabord ce que la prdiction nave nous donne . on note @xmath30 le facteur de zoom et @xmath61 la borne de la hauteur . alors dans le zoom de facteur @xmath30 le nombre de points rationnels dans le voisinage de diamtre @xmath879 devrait tre @xmath880 si @xmath881 , on devrait pouvoir trouver beaucoup de points cest - - dire de cardinal au moins une puissance de @xmath61 . si @xmath51 , _ a priori _ cette heuristique ne prdit pas grande chose . nous allons montrer quen fait il y a au plus un nombre fini de points en dehors des courbes spciales dans ce dernier cas . nous nous bornons la rgion @xmath882 . puisque les droites @xmath883 forment le bord du domaine @xmath882 , il reste trouver la distribution dans lintrieur de @xmath882 . grce au lemme [ le : curvescoverallpoints ] , les courbes nodales donnent un paramtrage local des points rationnels . donc le dnombrement des points rationnels dans lintrieur de @xmath882 peut se dcomposer en des sommes des points sur chaque @xmath854 . on va calculer la formule de la hauteur restreinte @xmath854 en choisissant un paramtrage rationnel des coordonnes de @xmath67 pour la courbe nodale . par la formule , seulement les courbes @xmath854 avec @xmath666 intersectent @xmath882 . on fixe un couple @xmath884 vrifiant la condition @xmath666 . dabord on change les coordonnes @xmath885 en @xmath886 . alors lquation de @xmath854 devient @xmath887 daprs le principe de bzout , une droite gnrale intersecte @xmath854 en @xmath817 points ( comptant la multiplicit ) . la droite @xmath888 passe par @xmath889 la multiplicit dintersection @xmath890 en @xmath889 est @xmath90 . donc cette droite intersecte @xmath854 en un autre point rationnel @xmath891 diffrent de @xmath889 avec @xmath892 . un calcul nous donne @xmath893 revenons aux coordonnes @xmath885 . @xmath894 donc on obtient un morphisme de paramtrage @xmath895 @xmath328\longmapsto \left(\frac{\frac{u}{v } a(\frac{u}{v}-1)}{b-\frac{u}{v } a},\frac{b(\frac{u}{v}-1)}{\frac{u}{v}(b-\frac{u}{v } a)}\right)=\left(\frac{ua(u - v)}{v(bv - ua)},\frac{bv(u - v)}{u(bv - ua)}\right).\ ] ] on introduit la notation @xmath896 alors @xmath897 et de la mme faon , @xmath898 on trouve le paramtrage suivant pour les coordonnes primitives dans @xmath0 . @xmath899 on a alors @xmath900 @xmath901 @xmath902 @xmath903 enfin on obtient la formule de la hauteur restreinte @xmath854 induite par limage inverse sur @xmath67 dans la rgion @xmath882 : pour @xmath337\in { \mathbb{p}}^1 $ ] satisfaisant @xmath904 , @xmath905))&=\frac{t^2 xy}{\operatorname{\operatorname{pgcd}}(x , s)\operatorname{\operatorname{pgcd}}(x , t)\operatorname{\operatorname{pgcd}}(y , s)\operatorname{\operatorname{pgcd}}(y , t)}\\ & = \frac{b(ua(u - v))^2}{(\operatorname{\operatorname{pgcd}}(v , a)\operatorname{\operatorname{pgcd}}(u , b)\operatorname{\operatorname{pgcd}}(u - v , b - a))^2}\\ & = \frac{b(ua(u - v))^2}{(d_1d_2d_3)^2}.\end{aligned}\ ] ] pour @xmath522 fixs , on prend la fonction de test @xmath36 et lon regarde la distribution locale dans le voisinage de diamtre @xmath879 intersectant la rgion @xmath882 , cest - - dire la quantit @xmath906\times[s : t]\\ & x , y , s , t>0;\frac{t}{s}>\frac{x}{y}>1 \end{aligned } \left| \begin{aligned } & b^{\frac{1}{r}}{\operatorname{d}}(p)=b^{\frac{1}{r}}\left(\frac{t}{s}-1\right)\leqslant \varepsilon \\ & \operatorname{h}_{\omega_{y_4}^{-1}}(p)=\frac{t^2 xy}{\operatorname{\operatorname{pgcd}}(x , s)\operatorname{\operatorname{pgcd}}(x , t)\operatorname{\operatorname{pgcd}}(y , s)\operatorname{\operatorname{pgcd}}(y , t)}\leqslant b \end{aligned } \right\}\right.\ ] ] on le prcisera comme un problme de dnombrement . pour un couple @xmath907 satisfaisant @xmath666 , considrons lensemble @xmath908 des @xmath337\in{\mathbb{p}}^1 $ ] vrifiant ( cf . ) @xmath909 @xmath910)=\frac{b(ua(u - v))^2}{(\operatorname{\operatorname{pgcd}}(v , a)\operatorname{\operatorname{pgcd}}(u , b)\operatorname{\operatorname{pgcd}}(u - v , b - a))^2}\leqslant b .\ ] ] alors @xmath911 [ th : finitenessofcriticzoom ] pour tout @xmath35 , on a que , pour tout @xmath363 , @xmath912 par consquent , la sous - varit @xmath913 est celle localement faiblement accumulatrice ( dfinition [ df : weaklocal ] ) . il nexiste pas de mesure limite pour le zoom critique . on dmontre dabord que , pour @xmath35 fix , en utilisant le paramtrage ci - dessus , dans le cas critique ( @xmath51 ) , il ny a quun nombre fini de courbes @xmath854 qui interviennent . en factorisant @xmath378 , lingalit scrit @xmath914 donc @xmath915 donc en combinant avec , on en dduit @xmath916 do @xmath917 notons que @xmath918 sont premiers deux deux . une observation importante est que @xmath919 et donc @xmath920 cela nous donne la majoration du paramtre @xmath683 pour @xmath51 : @xmath921 ce qui signifie que dans un voisinage fix aprs zoom , le nombre de courbes nodales est uniformment major , savoir @xmath922 parmi ces courbes @xmath854 qui interviennent , si @xmath923 , on a @xmath924 . alors la proposition [ po : weakzoom ] nous donne que @xmath925 pour tout @xmath363 . pour tous les @xmath871 , sur @xmath67 ( limage inverse de @xmath926 ) on a une distance et une hauteur induites de celles sur @xmath13 , la hauteur tant quivalente la hauteur de weil canonique @xmath927 ( cf . , ) . comme , le zoom induit sur @xmath854 avec le facteur @xmath51 est exactement le zoom critique . la majoration dcoule donc de la proposition [ po : controlofepsilon ] . daprs le thorme [ th : criticzoomforquadratics ] , pour tout @xmath330 suffisamment proches , on a @xmath928 alors que pour tout @xmath871 , il existe certains couples @xmath416 tels que @xmath929 cela dmontre la non - existence de mesure limite . le fait que @xmath18 est localement faiblement accumulatrice dcoule du thorme [ th : pagelot ] puisque @xmath930 sur @xmath18 et donc @xmath931 on va dmontrer une borne infrieure du nombre de points rationnels dans le zoom sous - critique ( @xmath932 ) bas sur le fait que sur certaines courbes nodales on trouve beaucoup de points avec la hauteur et la distance induites ( thorme [ th : zoomingonlattices ] ) . fixons @xmath933 . nous nous bornons la rgion @xmath934 lnonc prcis est le suivant . rappelons que @xmath935 . [ th : lowerbound ] pour @xmath936on a @xmath937 o le symbole @xmath938 signifie que la somme est prise sur les @xmath939 et les ensembles @xmath940 sont dfinis de manire analogue @xmath908 en remplaant par @xmath941 et @xmath942 au vu de , on va appliquer le thorme [ th : zoomingonlattices ] pour compter le nombre de points dans le zoom de facteur @xmath30 induit sur chaque @xmath854 , savoir le cardinal de chaque @xmath940 , avec un terme derreur prcis . on rappelle la dfinition de la fonction @xmath943 . on dfinit une fonction multiplicative @xmath944 [ po : zoompointsoncab ] supposons que @xmath945 alors pour tout couple @xmath946 satisfaisant , @xmath947 on a que , pour tout @xmath948 , @xmath949 si lon identifie localement une branche de @xmath854 avec lespace tangent de @xmath67 en le point @xmath321 , la proposition [ po : zoompointsoncab ] indique quil existe une quidistribution sur chaque @xmath854 ( @xmath871 ) avec la hauteur et la distance induites ( cf . , ) . il suffit compter le cardinal de @xmath950 pour @xmath879 fix . la condition correspond la restriction de la hauteur @xmath951 sur @xmath854 , et une inversion de mbius conduit une condition de rseau . dans un premier temps on voudrait rduire la condition en un zoom avec dune distance induite sur limage inverse de @xmath952 pour que lon puisse appliquer le thorme [ th : zoomingonlattices ] . on va approcher le cardinal de lensemble @xmath950 par celui de lensemble @xmath953 des @xmath337 $ ] satisfaisant @xmath954 @xmath955 on compare maintenant ces deux ensembles . compte tenu lencadrement sur @xmath87 , la condition implique que @xmath956 en combinant avec la majoration de @xmath378 obtenue partir de la condition , on conclut que @xmath957 de plus en rappelant lhypothse sur @xmath946 , on a daprs , @xmath958 donc @xmath959 cela implique que @xmath960 o @xmath961 avec @xmath962 . ensuite on va borner le cardinal de @xmath963 . par une inversion de mbius , on a @xmath964 o @xmath965 le points dans cet ensemble se trouvent dans iintersection dun triangle do nt lorigine est lun de ses sommets et laire est de grandeur @xmath966 avec le rseau @xmath967 puisque @xmath968 pour @xmath969 , on vrifie que @xmath970 le lemme suivant nous permet de contrler le nombre des points entiers primitifs dans une rgion planaire autour de lorigine . pour une dmonstration , voir ( * ? ? ? * lemma 2 ) . [ le : countinglatticepointsinsmalldomains ] soit @xmath971 un rseau . soit @xmath972 convexe do nt le bord est lisse par morceau avec @xmath973 . alors @xmath974 en utilisant ce lemme , on obtient que pour tout @xmath975 , comme @xmath976 ( @xcite i.5 corollary 5.3 ) , @xmath977 ( * ? ? ? * theorem 5.7 ) , @xmath978 o @xmath979 peuvent tre arbitrairement petits . on en conclut que @xmath980 maintenant on compte le cardinal de @xmath981 . comme on a fait pour @xmath963 , et par une inversion de mbius on obtient @xmath982 o @xmath983 nous allons appliquer le thorme [ th : zoomingonlattices ] . cette fin on prend dans @xmath984 notons quici la signification de @xmath61 et @xmath30 sont diffrentes . on prend @xmath985 la condition sur @xmath30 et @xmath683 implique les condition - . la condition se traduit en @xmath986 en utilisant la majoration @xmath987 la condition sur @xmath683 entrane que @xmath988 do la condition . de plus , sur @xmath30 entrane que @xmath989 et pour un tel @xmath30 on a @xmath990 donc implique la condition suivante vrifies de la mme faon , qui elle - mme implique pour tout @xmath991 @xmath992 _ suite de la dmonstration de la proposition [ po : zoompointsoncab]_. donc on peut appliquer le thorme [ th : zoomingonlattices ] avec les quantits @xmath993 ci - dessus et on obtient @xmath994 en reportant dans , on calcule la somme du coefficient du terme principal @xmath995 rappelons la majoration pour la fonction @xmath996 , ( * ? ? ? * theorem 5.7 ) @xmath997 on peut majorer le terme derreur de la faon suivante . pour tout @xmath948 , @xmath998 et pour tout @xmath999 , @xmath1000 compte tenu le terme derreur provenant du cardinal de @xmath1001 , on en conclut la formule asymptotique sur chaque courbe @xmath854 . tout dabord rappelons le diffomorphisme local @xmath1002 et les notations de coordonnes de lespace tangent @xmath1003 on dduit un encadrement des @xmath946 tels que la courbe @xmath854 intervienne dans le dnombrement quand on prend la fonction caractristique @xmath1004 de la rgion @xmath1005 dans @xmath882 . la condition de zoom dit que @xmath1006 prenons un couple @xmath946 vrifiant la condition , daprs lquation dfinissant limage de la courbe @xmath854 , on a @xmath1007 on note @xmath1008 sil existe @xmath1009 , alors daprs on a @xmath1010 maintenant prenons un couple @xmath946 vrifiant et @xmath1011 puisque un point @xmath1012 vrifie @xmath1013 la condition de zoom implique que sur la courbe @xmath854 , si @xmath1014 , @xmath1015 donc @xmath1016 donne un point de @xmath1017 . on restreint le dnombrement sur les @xmath946 tels que @xmath871 car cest sur telles @xmath854 quon trouve une quidistribution de points pour le zoom sous - critique ( proposition [ po : zoompointsoncab ] ) . on en conclut que @xmath1018 comme ( cf . , ) @xmath1019 do @xmath1020 et leur diffrence est petite . le terme principal sur chaque courbe nodal fait apparatre la fonction @xmath1021 qui ressemble la fonction @xmath94 cest - - dire au nombre de diviseurs au sens de la convolution , do nt lordre moyen de ce type est connu grce une srie de travaux de r. de la bretche et t. d. browning ( @xcite,@xcite,@xcite ) . on donnera les dtails dans les appendices , o nous dcrivons les rsultats . on est en mesure de sommer les courbes @xmath854 sur les paramtres @xmath946 vrifiant @xmath1022 qui correspond . afin dappliquer la proposition [ po : asymptoticforpsiquotient ] sur lordre moyen des diviseurs de formes binaires , on tend la somme sur les @xmath946 vrifiant la condition @xmath1023 en rajoutant les @xmath946 vrifiant la condition @xmath923 . le terme derreur correspondant est de grandeur @xmath1024 pour tout @xmath10250,1\mathclose[$ ] . daprs la formule [ po : asymptoticforpsiquotient ] de la proposition [ po : problemofdivisor ] , la somme des termes principaux de donne @xmath1026\varepsilon_2,\varepsilon_1\mathclose]\\t\in\mathopen]\tau_2,\tau_1\mathclose[}}\frac{1}{t^{2 } } \operatorname{d}t\operatorname{d}y\right)b^{(1+\eta)\left(\frac{1}{2}-\frac{1}{r}\right)}(\log b)^3 + o ( b^{(1+\eta)\left(\frac{1}{2}-\frac{1}{r}\right)}(\log b)^2 ) \\ & = \left(c_2\int \chi(w(\varepsilon_1,\varepsilon_2,\tau_1,\tau_2))\frac{\operatorname{d } x\operatorname{d}y}{y}\right)b^{(1+\eta)\left(\frac{1}{2}-\frac{1}{r}\right)}(\log b)^3+o ( b^{(1+\eta)\left(\frac{1}{2}-\frac{1}{r}\right)}(\log b)^2 ) . \end{aligned}\ ] ] la contribution de la somme des termes derreur dans la proposition [ po : zoompointsoncab ] est majore par @xmath1027 grce la condition sur @xmath30 et pour @xmath1028 la contribution de la somme du terme derreur dans est , en utilisant la formule [ po : asymptoticforpsiquotient ] encore , @xmath1029 cela termine la preuve du thorme [ th : lowerbound ] . nous expliquons une raison pour laquelle nous ne choisissons pas tablir une formule asymptotique pour le zoom sous - critique . comme lon a vu , la prdiction nave affirme que la puissance de @xmath61 devrait tre @xmath1030 . en fait on a @xmath1031 daprs . le terme principal de la borne infrieure ( thorme [ th : lowerbound ] ) deviendrait @xmath1032 si lon pourrait prendre la constante @xmath63 . une difficult se cache sur lexactitude de la formule pour les paramtres @xmath946 grands ( cest - - dire @xmath1033 pour certain @xmath1034 ) . dans ce cas le zoom sur la courbe nodale @xmath854 compte au plus un point et la technique utilise dans ce texte ne permet pas de dterminer sil est non - nul ou pas . dans @xcite , nous avons surmont un obstacle similaire laide de la transformation de cremona . mais cette astuce nest plus applicable ici car la transformation de cremona prserve les courbes @xmath854 . le but de cette section est de signaler le fait que les points de @xmath13 sont en bijection avec un ensemble mince de @xmath0 . notons que , outre le morphisme dimplosion utilis jusqu maintenant , la surface @xmath13 possde un autre morphisme vers @xmath0 comme suit . le diviseur @xmath826 dfinit un morphisme birationnel @xmath1035 do nt limage @xmath113 est une surface torique de del pezzo de degr @xmath14 singulire de type @xmath1036 dfinie comme lintersection de deux quadriques dans @xmath1037 : @xmath1038 et lventail est un croix qui ressemble celui de @xmath0 . ] le morphisme @xmath60 est en fait la dsingularisation minimale torique de @xmath113 puisque les diviseurs @xmath1039 @xmath1040 ont le nombre dauto - intersection @xmath1041 . lquation @xmath1042 implique ( comme pour toute surface de del pezzo de degr @xmath14 fibre en conique ) que @xmath13 possde deux projections vers la droite projective , et donc un morphisme @xmath1043 vers @xmath0 qui est fini de degr gnrique @xmath90 . plus prcisment , @xmath1043 est dfini de la faon suivante . dabord la projection @xmath1044 depuis le point @xmath1045 $ ] est bien dfinie sur @xmath113 . la varit de produit sinjecte sur @xmath1046 comme une surface quadrique : @xmath328\times[s : t]\longmapsto [ us : vt : ut : vs].\ ] ] si lon utilise les coordonnes @xmath1047 $ ] venant de @xmath1037 , limage de @xmath0 est dfinie par lquation . en rappelons les quations , @xmath1048 induit un morphisme @xmath1049 de degr gnrique @xmath90 do nt limage est un ensemble _ mince _ de @xmath1050 : @xmath1051\times [ s : t]:uvst\in\square\}.\ ] ] en les composant , on voit que le morphisme @xmath1052 est un revtement lisse de degr gnrique @xmath90 . de plus , on a @xmath1053 le lieu de ramification est sur les diviseur du bord , donc autour du point @xmath1054\times[1:1]$ ] , @xmath60 est un diffomorphisme . par la fonctorialit de la hauteur on peut se ramener au cas de @xmath0 restreinte lensemble mince . dans cette section , on produira des varits toriques de dimension suprieure sur lesquelles en dehors un ferm de zariski les meilleures approximants se trouvent dans une famille de courbes nodales , comme pour @xmath13 . elles sont construites en fait comme produits de @xmath13 avec dautres varits toriques . pour tout @xmath1055 , considrons la varit @xmath1056 . on note @xmath1057 les morphismes de projection vers @xmath13 et @xmath1058 . pour @xmath1059 dans lorbite ouverte , on a @xmath1060 qui est atteinte sur une famille de courbes nodales couvrant un ouvert dense de @xmath1061 . on peut supposer que @xmath1062 a pour coordonnes @xmath1054\times [ 1:1]$ ] et @xmath1063 $ ] . tout dabord , comme @xmath1064 , on choisit la hauteur de weil associe @xmath1065 comme @xmath1066)=h_{\omega_{y_4}^{-1}}(x)h_{\omega_{{\mathbb{p}}^n}^{-1}}([y_0:\cdots : y_n ] ) , \quad h_{\omega_{{\mathbb{p}}^n}^{-1}}([y_0:\cdots : y_n])=\max_{0\leqslant i\leqslant n}(|y_i|)^{n+1},\ ] ] et la distance @xmath32 comme le maximum de celle @xmath1067 sur @xmath13 et @xmath1068 sur @xmath1058 dfinie par la formule ( pour @xmath1069 ) , @xmath1070)=\max_{1\leqslant i\leqslant n}\left|\frac{y_i}{y_0}-1\right|.\ ] ] par dfinition , si @xmath1071\neq q$ ] , @xmath1070)^{n+1}h_{\omega_{{\mathbb{p}}^n}^{-1}}([y_0:\cdots : y_n])\geqslant 1.\ ] ] donc pour un point gnral @xmath1072 tel que @xmath1073 , on a , daprs la proposition [ po : lowerbound1 ] , @xmath1074 cela dmontre que @xmath1075 . ensuite on construit des courbes nodales dans @xmath1061 partir de celles dans @xmath13 . avec les notations dans la section [ se : cab ] , pour @xmath1076 , on a construit une courbe nodale @xmath854 dans @xmath13 avec le morphisme de paramtrage @xmath926 . maintenant on prend une droite @xmath1077 dans @xmath1058 avec un paramtrage @xmath1078 qui envoie @xmath1079 $ ] sur @xmath1080 . on dfinit un morphisme de paramtrage comme suit @xmath1081 @xmath1082)=(\psi_{a , b}([u : v]),\psi_l\circ \theta_{a , b}([u : v])),\ ] ] o @xmath1083 est dfini comme le revtement double @xmath1084)=[au^2-bv^2:v^2].\ ] ] on voit que limage de @xmath1085 est une courbe rationnelle @xmath1086 nodale en @xmath1087)$ ] avec @xmath1088 comme @xmath1089 dfinit une droite double . on en dduit que @xmath1090 plus concrtement , comme lapproximation diophantienne des nombres quadratiques est quivalente la structure des quations de pell - fermat gnralises ( cf . section [ se : eqpellfermat ] ) , en choisissant une famille de solutions primitives @xmath320 de lquation @xmath1091 on obtient automatiquement une suite dapproximants @xmath1092)$ ] du point @xmath17 sur @xmath1061 . puisque lon a lquivalence de hauteurs @xmath1093)\sim y_n^{2n+6},\ ] ] et lquivalence de distances @xmath1094 on est ramen au problme de lapproximation diophantienne dun nombre quadratique particulier et la constante dapproximation se dduite des thormes de dirichlet , liouville , et roth . dans un souci de compltude , nous redmontrons le rsultat de s. pagelot concernant la distribution locale dun @xmath44-point @xmath17 sur la droite projective @xmath1095 . pour simplicit on suppose que @xmath1096 $ ] ; le rsultat pour un point gnral diffre par une constante . [ th : pagelot ] on a @xmath1097 . on fixe la hauteur de weil absolue associe au fibr @xmath66 dfinie par @xmath1098)=\max(|u|,|v|),\quad ( u , v)\in{\mathbb{z}},\quad\operatorname{\operatorname{pgcd}}(u , v)=1,\ ] ] et lon note @xmath1099 le facteur de zoom . soit @xmath60 une fonction intgrable support compact sur @xmath1100 . alors * si @xmath575 alors @xmath1101 o @xmath1102 * si @xmath474 , alors @xmath1103 on voit que das les deux cas la mesure asymptotique existe . pour le zoom critique , si @xmath1104 - 1,1\mathclose[$ ] , alors @xmath1105 . on observe donc quil y a un trou autour du point @xmath17 . pour un zoom sous - critique ( cest - - dire @xmath474 ) , la distribution est uniforme . rien dans cette section nest nouveau . le but est de rappeler comment traiter ce type de problme dans le cas le plus simple . une observation basique , qui est aussi celle que lon va suivre dans la suite , est quau lieu de considrer des fonctions intgrables gnrales , il suffit de regarder les fonctions simple , cest - - dire les fonctions caractristiques dintervalles , car ces fonctions suffisent pour caractriser la convergence faible de mesures de probabilit . on note les coordonnes de @xmath1106 par @xmath337 $ ] . sans perte de gnralit on peut supposer que @xmath1107 . puisque le point approcher est @xmath1096 $ ] , on utilise le diffomorphisme local @xmath25 dfini par @xmath328\longmapsto \frac{u}{v}\in{\mathbb{r}}.\ ] ] alors @xmath17 est envoy sur lorigine @xmath1108 . sur @xmath1100 on utilise la distance canonique @xmath1109,q)=\left|\frac{u}{v}\right|.\ ] ] par symtrie , il suffit dexaminer les fonctions caractristique des intervalles @xmath562\eta,\varepsilon\mathclose]\subset \mathopen]0,\infty[$ ] . on dsigne par @xmath1110 une telle fonction . nous remarquons ici que lusage de cette notation a un sens lgrement diffrent dans les parties prcdentes du texte . ltude de la convergence de la suite @xmath1111 revient lestimation du cardinal de lensemble @xmath1112\neq[0:1]\\ & \operatorname{\operatorname{pgcd}}(u , v)=1 \end{aligned } \left| \begin{aligned } & 0<b^{\frac{1}{r}}d(p , q)=b^{\frac{1}{r}}\frac{u}{v}\leqslant \varepsilon\\ & h(p)=\sup(u , v)\leqslant b \end{aligned } \right\}\right . , \ ] ] puisque @xmath1113 . les conditions sur @xmath591 implique que @xmath1114 cela implique quil ny a quun nombre fini de @xmath530 dans un voisinage fix et nous permet de fixer @xmath530 tout en comptant @xmath378 . on peut crire @xmath1115 o @xmath1116 on rappelle la dfinition de la fonction @xmath495 et lon en dduit , en utilisant ( * ? ? ? * exercise 5.2 ) @xmath1117 alors @xmath1118 cela clt la dmonstration du cas critique . comme prcdemment on a la dcomposition @xmath1119 o @xmath1120 fixons @xmath530 dans cette runion . pour que @xmath1121 soit non - vide , il faut que @xmath1122 avant de calculer le cardinal , citons dabord quelques formules asymptotique bien connues ( voir @xcite 3.2 , 3.4 ) : @xmath1123 @xmath1124 @xmath1125 en appliquant ces formules , on obtient , encore daprs ( * ? ? ? * exercise 5.2 ) @xmath1126 do @xmath1127 lexpression finale est acheve . quand @xmath474 , les points dnombrer se trouvent dans un triangle do nt la longueur du bord est ngligeable par rapport laire . on pourrait adapter la mthode utilise dans la dmonstration du mme problme pour les nombres irrationnels ( cf . thorme [ th : localdistributionofrealnumbers ] ) . le raisonnement dcrit ici donne un meilleur terme derreur au niveau de la puissance de @xmath1128 . lordre de grandeur des nombres de points dans les zooms pour @xmath67 tant en accord avec celui prvu , nous discutons maintenant quoi correspondent les mesures limites . prenons comme dhabitude la fonction caractristique @xmath331 . si lon identifie @xmath681 avec lensemble des points primitifs dans @xmath109 le comptage de points de hauteur borne @xmath1129 quivaut au comptage des points primitifs dans le rectangle @xmath1130 . lopration de zoom revient prendre des points primitifs dans lintrieur du triangle @xmath1131 do nt la taille dpend du facteur de zoom @xmath30 et de la borne @xmath61 . pour les zooms sous - critiques @xmath474 , lordre de grandeur de laire domine celui du bord , donc elle donne le terme principal . on a que @xmath1132 on voit que dans ce cas l les points primitifs sont quidistribus et lon obtient une mesure proportionnelle la mesure de lebesgue . cela concide avec celui du zoom sous - critique pour tous les nombres algbriques . pour le zoom critique @xmath575 , laire de @xmath1133 et son bord ont la mme dordre de grandeur @xmath61 . comme lon a vu dans la dmonstration , les points se trouvent en fait dans les droites horizontales dans lintrieur du triangle do nt la longeur est @xmath1134 puisque @xmath1135 o @xmath1136\frac{u}{\eta},\frac{u}{\varepsilon}\mathclose]\right\}.\ ] ] on en conclut que sur chaque droite on compte des nombres rationnels avec le numrateur fix . cela explique do viennent la fonction densit @xmath1137 et la fonction arithmtique @xmath603 . on remarque que le phnomne pour des nombres algbriques est radicalement diffrent ( pour eux dans le zoom de facteur @xmath65 on trouve aussi une quidistribution , voir la section [ se : weakzoomofalgebraicnumbers ] ) . dans @xcite , t. d. browning a tudi lordre moyen du nombre de diviseurs pour des formes binaires cubiques dployes , do nt la technique remonte @xcite , o les auteurs ont dduit une formule asymptotique de lordre moyen primitif de fonctions arithmtiques ressemblant la fonction donnant le nombres de diviseurs pour certaines formes binaires quartique . ils lutilisent pour dmontrer que le nombre de points de hauteur borne sur une surface de del pezzo de degr @xmath14 saccorde avec la prdiction de batyrev - manin - peyre . en les imitant , nous allons dduire une telle formule pour une forme binaire cubique particulire ( la mme dmonstration marche pour toutes les formes binaires cubiques dployes , ce do nt nous naurons pas besoin ici ) . on considre les fonctions arithmtiques @xmath1138 et la constante @xmath1139 : @xmath1140 [ po : problemofdivisor ] soient @xmath1141 , on a @xmath1142 @xmath1143 on fixe trois formes linaires coefficients entiers primitives @xmath1144 en deux variables deux deux linairement indpendantes et @xmath1145 un sous - ensemble convexe ferm born de @xmath1146 do nt les coordonnes sont notes @xmath1147 . pour @xmath1148 , on note @xmath1149 on suppose que @xmath1150 pour @xmath1151 . on note @xmath1152 o @xmath1153 dsigne le maximum des valeurs absolues des coefficients de @xmath1154 . on note aussi @xmath1155 @xmath1156 pour @xmath1157 , on note @xmath1158 et @xmath1159 la fonction @xmath1002 se calcule comme @xmath1160 elle est donc multiplicative en dimension @xmath817 : @xmath1161 pourvu que @xmath1162 . pour @xmath259 un nombre premier , on dfinit @xmath1163 on note @xmath1164 le plus grand entier tel que @xmath1165 . pour @xmath1166 tels que @xmath1167 , on a besoin de lordre moyen de la quantit suivante @xmath1168 [ th : browning ] soient @xmath35 , @xmath1169 . supposons que @xmath1170 . alors il existe un polynme @xmath1171 $ ] de degr @xmath817 tel que @xmath1172 o @xmath1173 et le coefficient du terme principal de @xmath146 est @xmath1174 avec @xmath1175 et pour @xmath1176 , @xmath1177 pour une utilisation ultrieure , nous voudrions en savoir un peu plus sur la constante @xmath1178 ainsi que sur la constante @xmath1179 . nous procdons comme dans @xcite 3 . on choisit une base @xmath1180 minimale pour @xmath1181 au sens suivant @xmath1182 et on dfinit les formes linaires binaires @xmath1183 via le changement de variables @xmath1184 comme @xmath1185 , les formes @xmath1183 nouvellement construites satisfont @xmath1186 soient @xmath1187 @xmath1188 les entiers tels que @xmath1189 soient des formes primitives . alors @xmath1190 puisque lon a suppos que les formes @xmath1154 sont primitives . on note @xmath1191 fixons un nombre premier @xmath259 et @xmath1192 . rappelons la dfinition de @xmath1193 et considrons le rseau @xmath1194 on a que @xmath1195 aprs ce changement de variables on rcrit le rseau @xmath1196 comme @xmath1197 en mme temps on dfinit un rseau @xmath1198 alors on obtient @xmath1199 do @xmath1200 on crit le produit @xmath1201 en une somme des fonctions multiplicatives en dimension @xmath817 : @xmath1202 o @xmath1203 lavantage de cette reprsentation est prsent dans le lemme suivant , qui sera utile pour la suite . [ le : convolutionrepresentation ] pour tout @xmath35 et @xmath1204 telle que @xmath1205 , on a @xmath1206 en particulier , on a @xmath1207 on dfinit une fonction arithmtique multiplicative @xmath1208 . rappelons la formule . un calcul donne @xmath1209 donc @xmath1210 on obtient que @xmath1211 en particulier @xmath1212 est _ petite _ dans le sens de ( 2.19 ) dans @xcite : il existe @xmath1213 petit tel que la srie @xmath1214 converge . cette proprit jouera un rle important dans le traitement des termes derreur . dans la suite on fixe nos formes @xmath1154 comme @xmath1215 et la rgion @xmath1216 le but est dobtenir la formule asymptotique pour la somme des points entiers do nt les coordonnes @xmath1217 sont premires entre elles . cela consiste en une application directe du thorme [ th : browning ] avec une inversion de mbius comme fait dans ( * ? ? ? * corollaire 1 ) pour certaines formes binaires de degr @xmath14 . malheureusement cela nest pas fait dans @xcite . nous suivons les techniques venant de @xcite pour dmontrer les formules asymptotiques . on dfinit pour @xmath1218 , @xmath1219 soient @xmath1220 comme prcdemment avec @xmath1221 . la sommation do nt la formule asymptotique cherche est @xmath1222 @xmath1223 o @xmath1224 avec @xmath1225 par une inversion de mbius , on a @xmath1226 on dfinit @xmath1227 , et on note @xmath1228 comme @xmath1229 la somme [ eq : sumofdivisorsprimitive ] scrit comme @xmath1230 on divise cette somme en deux parties , une pour les @xmath1231 petits et lautre pour les @xmath1231 grands . on applique le thorme [ th : browning ] @xmath1232 en notant que @xmath1233 et on obtient en notant @xmath1234 @xmath1235 en outre , on a une majoration ( @xcite , ( 2.5 ) ) @xmath1236 donc on peut majorer le terme derreur comme @xmath1237 @xmath1238 en reportant dans ( [ eq : mobiustransform ] ) , on obtient @xmath1239 on note pour @xmath1240 , @xmath1241 on calcule maintenant la constante , avec les notations et , @xmath1242 vrifions que les facteurs locaux @xmath1243 sont donns par . si @xmath1244 , on dfinit le rseau @xmath1245 alors on a @xmath1246 @xmath1247 si @xmath1248 , on a dans ce cas pour @xmath1249 et donc en remarquant que @xmath1250 on obtient @xmath1251 maintenant on est prt dduire la formule asymptotique pour la fonction @xmath1021 . on voit que , daprs @xmath1252 et @xmath1253 , @xmath1254 on fixe les trois formes linaires @xmath1255 et la rgion @xmath12560,1\mathclose]^2:\tau_2\leqslant\frac{x_2}{x_1}\leqslant\tau_1\right\}.\ ] ] alors @xmath1257 la quantit @xmath1258 tant nulle sil existe @xmath1259 tel que @xmath1260 . daprs , on a , en notant @xmath1261 , @xmath1262 le terme principal de la sommation est @xmath1263 o daprs , @xmath1264 grce notre choix des formes linaires , fixons un nombre premier @xmath259 , on a que pour @xmath1265 , @xmath1266 on calcule pour @xmath1267 , @xmath1268 de faon analogue on a @xmath1269 on en dduit les valeurs de @xmath1270 comme suit . @xmath1272 donc le facteur local se calcule , en rappelant que @xmath1273 , @xmath1274 donc @xmath1275 or on a aussi une majoration pour @xmath1258 ( @xcite , ( 3.1 ) ) : @xmath1276 qui dcoule des majorations standards pour le nombre de diviseurs et pour le dnombrement des points sur un rseau . soit @xmath1277 tel que @xmath1278 et @xmath1279 . comme prcdemment on dcompose la somme en deux parties @xmath1280 selon la taille de @xmath1281 . cela nous permet de contrler les termes derreur , compte - tenu de , @xmath1282 @xmath1283 @xmath1284 daprs le lemme [ le : convolutionrepresentation ] et la majoration , on a @xmath1285 do @xmath1286 on en conclut que peut se calculer comme @xmath1287 cela dmontre la formule [ po : asymptoticforpsi ] de la proposition [ po : problemofdivisor ] en remarquant que @xmath1288 tout dabord on traite la somme @xmath1289 comme @xmath1290 on obtient , daprs le thorme de fubini , @xmath1291 comme @xmath1292 lgalit ci - dessus scrit @xmath1293 finalement on arrive sommer @xmath1294 on dfinit @xmath1295 et on rappelle que @xmath1296 alors @xmath1297 il rsulte de que @xmath1298 il faut traiter le deuxime terme . daprs le thorme de fubini et en remplaant @xmath1299 par @xmath453 , @xmath1300 on obtient finalement que @xmath1301 je tiens remercier emmanuel peyre pour mavoir propos ce problme et pour des conseils tout au long de mon travail . rgis de la bretche , marc hindry mont donn des indications trs prcieuses . je les en remercie chaleureusement . des parties de ce travail ont t faites luniversit paris - diderot et beijing international center for mathematical research do nt latmosphre stimulante est sincrement apprci . lauteur est partiellement support par le projet anr gardio . victor v. batyrev , on the classification of smooth projective toric varieties . thoku math . 43 , no.4 , 569 - 585 ( 1991 ) . victor v. batyrev , yuri i. manin , sur le nombre des points rationnels de hauteur born des varits algbriques . 286 , no . 1 - 3 , 27 - 43 ( 1990 ) . victor v. batyrev , yuri tschinkel , rational points of bounded height on compactifications of anisotropic tori , intern . math . research notices , 12 , 591 - 635 , ( 1995 ) victor v. batyrev , yuri tschinkel , manin s conjecture for toric varieties . j. algebraic geom . 7 ( 1998 ) , no . 1 , 1553 timothy d. browning , the divisor problem for binary cubic forms . j. thor . nombres bordeaux 23 ( 2011 ) , no . 3 , 579602 . timothy d. browning , quantitative arithmetic of projective varieties . progress in mathematics 277 . birkhuser , basel . xi , 160 p. ( 2009 ) . rgis de la bretche , timothy d. browning , le problme des diviseurs pour des formes binaires de degr 4 . j. reine angew . math . 646 ( 2010 ) , 144 . rgis de la bretche , timothy d. browning , manin s conjecture for quartic del pezzo surfaces with a conic fibration . duke math . j. 160 ( 2011 ) , no . 1 , 169 . cassels , j.w.s . an introduction to the geometry of numbers . die grundlehren der mathematischen wissenschaften . berlin - gttingen - heidelberg : springer - verlag . viii , 344 pp . g. v. chudnovsky , on the method of thue - siegel . ann . of math . ( 2 ) 117 ( 1983 ) , no . 2 , 325382 . lejeune - dirichlet p .- sur une proprit des formes quadratiques dterminant positif . journal de mathmatiques pures et appliques . paris . 1 , ( 1856 ) 76 - 79 . william fulton , introduction to toric varieties . annals of mathematics studies . 131 . princeton , nj : princeton university press . xi , 157 p. ( 1993 ) . roger heath - brown , diophantine approximation with square - free numbers . z. 187 , 335 - 344 ( 1984 ) . zhizhong huang , distribution locale des points rationnels de hauteur borne sur une surface de del pezzo de degr 6 . apparatre dans int . j. number theory . khintchine , continued fractions . translated by peter wynn . p. noordhoff , ltd . , groningen 1963 iii+101 pp . l. kuipers , h. niederreiter , uniform distribution of sequences . pure and applied mathematics . wiley - interscience [ john wiley & sons ] , new york - london - sydney , 1974 . xiv+390 pp . serge lang , asymptotic approximations to quadratic irrationalities . _ * 87 * ( 1965 ) , 481 - 487 ; _ ibid . _ 488496 . david mckinnon , _ a conjecture on rational approximations to rational points _ , j. algebraic geom . * 16 * ( 2007 ) , 257 - 303 . david mckinnon , mike roth , _ seshadri constants , diophantine approximation , and roth s theorem for arbitrary varieties _ , invent . math . * 200 * ( 2 ) , ( 2015 ) 513 - 583 . s. pagelot , _ tude de la distribution asymptotique fine des points rationnels de hauteur borne _ , manuscrit non - publi . emmanuel peyre , hauteurs et mesures de tamagawa sur les varits de fano . duke math . j. 79 ( 1995 ) , no . 1 , 101218 . emmanuel peyre , points de hauteur borne , topologie adlique et mesures de tamagawa . nombres bordx . 15 , no . 1 , 319 - 349 ( 2003 ) . per salberger , tamagawa measures on universal torsors and points of bounded height on fano varieties . nombre et rpartition de points de hauteur borne . astrisque no . 251 ( 1998 ) , 91258 . pierre samuel , _ thorie algbrique des nombres . _ hermann , paris 1967 130 pp . grald tenenbaum , introduction la thorie analytique et probabiliste des nombres . cours spcialiss 1 . socit mathmatique de france , paris , 1995 . xv+457 pp .

observation of an atomic electric dipole moment would signal the violation of time - reversal ( t ) symmetry@xcite , which kaon decay tells us is present at some level@xcite . so far all measurements , whether on elementary particles or atoms and despite rather high sensitivity , have been statistically zero , but experiments continue to improve@xcite . the level at which dipole moments are finally seen will help decide among a number of candidates for the fundamental source of t violation . several theorists have proposed that the light actinides would be the best elements in which to detect a small dipole moment@xcite . most recently , the authors of refs . @xcite have argued that the existence of octupole ( pear - shaped ) deformation in the nuclei of these atoms enhances the sensitivity of atomic dipole moments to nuclear parity ( p ) and t violation by factors of 100 to 1000 ( typically about 400 in the later reference ) over the sensitivity in the atom with the best current experimental limit , @xmath0hg . this level of enhancement is due in large part to the existence of close - lying parity doublets and favorable atomic structure in the light actinides , but also to fact that it is not the dipole moment of the nucleus that induces a dipole moment in the surrounding electrons , but rather the `` schiff moment '' , a quantity that reflects the mean - square radius of the nuclear dipole distribution . asymmetric nuclei have large intrinsic schiff moments even though their intrinsic dipole moments are very small , in the same way that a neutral particle can have a finite charge radius . the arguments of refs . @xcite warrant careful investigation . in this paper , we give a pedagogical derivation of the schiff operator , explore its action on nuclear ground states , and address the role of octupole correlations in generating ground - state schiff moments . the discussion is organized as follows : section i contains a derivation of the nucleus - electron interaction responsible for atomic - dipole moments and introduces the schiff operator . the section concludes with an evaluation of the nuclear schiff moment under the assumption that the dominant source of time - reversal invariance is a nucleon electric dipole moment . section ii reveals important differences between the dipole and schiff operators , showing that in the goldhaber - teller model no schiff strength is produced by the giant dipole resonance . in section iii , we look at schiff moments in light nuclei , particularly @xmath1f , confirming the near absence of schiff strength in the giant resonance and pointing out a strong component of strength correlated with low - lying octupole excitations . the first excited state provides the largest contribution to the schiff moment . section iv takes up octupole correlations in heavy nuclei , focusing on octupole deformation . we find no flaw in the argument that moments in such nuclei are collective and enhanced , but point to physics that may make the enhancement less dramatic than claimed in refs . @xcite ( more detailed microscopic calculations of both the octupole - deformed nuclei and those that are currently used in experiments should resolve the uncertainty ) . in section v we argue that the collective schiff moments do not depend on the delicate and sometimes unanswerable question of whether a nucleus is octupole deformed . low - lying octupole vibrations generate them in the same way as static octupole deformation , increasing the number of atoms in which one can expect large effects . section vi summarizes our findings . we begin by deriving the nucleus - electron interaction responsible for generating atomic dipole moments . though the result is well known@xcite , a complete derivation has never appeared in one place , and our derivation differs in its details from the others . in 1939 feynman@xcite developed a quantum theory of molecular forces as part of what is now called the hellmann - feynman theorem , or sometimes the parameter theorem . this charming relative of the virial theorem@xcite allows insight into how forces in a complex quantum system are balanced against one another , just as they are in a classical system . in a molecule , for example , internal coulomb forces between electrons and nuclei counter each other so that there is no net force on the molecule ( or it would move ) . in the system we consider here , a neutral atom in a uniform electric field , there is no net charge so there is no net force on the system . to achieve this the electrons rearrange themselves @xcite so that there is no net electric field at the nucleus ( or it would move ) . this shielding effect has dramatic and unfortunate implications for experiments that would probe the atom - nucleus system . because nuclei are of finite extent , however , the shielding of the electrons varies over the nuclear volume , and this makes probing the nuclear interior possible , as shown originally by schiff@xcite . we present next a variant of schiff s derivation that uses modern effective - field - theory techniques@xcite to produce the simpler approximate result that has been obtained more recently@xcite . we assume a neutral ( nonrelativistic , for simplicity ) atom containing an extremely heavy nucleus with non - vanishing spin sitting in a uniform electric field , @xmath2 . the atom contains @xmath3 electrons , each with charge @xmath4 , while the nuclear charge is @xmath5 , where @xmath6 is the proton charge , and in our units the fine - structure constant is given by @xmath7 . the nucleus has both an electric monopole distribution and a tiny electric dipole distribution leading to an electric dipole moment @xmath8 ; other moments can be easily added . in electric - field gauge the atomic - plus - interacting - nucleus hamiltonian can be written in the form @xmath9 -e_p \vec{e}_0 \cdot \vec{d}_0 \ \ , \ ] ] where @xmath10 is the kinetic energy of the i@xmath11 electron ( @xmath12 in the nonrelativistic approximation ) , @xmath13 and @xmath14 are that electron s coordinate and momentum relative to the nuclear center - of - mass ( cm ) , @xmath15 is the electron mass , @xmath16 is the electron - electron coulomb interaction , and @xmath17 is the electrostatic potential due to the complete nuclear charge distribution @xmath18 ( with dimensions @xmath19 , and normalized to @xmath3 ) . to explore the physics of eq . ( [ eq : atom ] ) we employ a trick to remove the last term in that equation . because @xmath20 is tiny even on nuclear scales , it is sufficient to manipulate eq . ( 1 ) to first order in that quantity , ignoring all higher - order terms . for example , performing a first - order unitary transformation on @xmath21 produces @xmath22,$ ] where @xmath23 in our case is the hermitian operator @xmath24 performing the commutator generates two terms , one from the last bracketed term in eq . ( 1 ) that cancels the interaction of the nucleus and the external field ( @xmath25 was constructed to do this ) , and a second that takes its place : @xmath26 the difference , @xmath27 , between eq . ( 5 ) and eq . ( 1 ) , which can not lead to an energy shift to first order in @xmath28 , is given by @xmath29 this type of relationship , generated by obvious equalities such as @xmath30 \rangle \equiv 0 $ ] , is often called a hypervirial theorem@xcite and can be derived from the hellmann - feynman theorem@xcite . if one neglects the finite extent of the nucleus and replaces @xmath31 by @xmath32 , eq . ( 6 ) can be rearranged into the form @xmath33 where @xmath34 is the electric field at the nucleus caused by the electrons . this simple result states that exact screening holds in the point - nucleus approximation , and highlights how a nonzero nuclear volume leads to a small but significant breakdown of screening . screening is now directly incorporated into @xmath35 . we next separate @xmath35 into a part independent of dipole moments @xmath46 and @xmath45 and another time - reversal- and parity - violating part proportional to these moments : @xmath52 expanding @xmath18 as in eq . ( [ eq : expansion ] ) , we get @xmath53 @xmath54 @xmath55 writing out the explicit expansions for @xmath56 and @xmath57 leads to the general result for @xmath58 expressed in terms of the schiff moment@xcite , @xmath59 : @xmath60 \ \ . \end{aligned}\ ] ] thus , the coupling of the nuclear dipole distribution to the atomic electrons is through the schiff moment in the first of eqs . [ eq : def ] is often @xcite incorporated in the definition of @xmath59 . ] . the result ( [ eq : def ] ) depends in leading order on terms of order @xmath61 , because eq . ( [ eq : point ] ) mandates the cancellation of terms of order @xmath62 . one can make contact with schiff s paper@xcite by defining quantities @xmath63 and @xmath64 such that @xmath65 and @xmath66 . equation ( 10 ) of ref . @xcite then follows from eq . ( [ eq : del ] ) above : @xmath67 this result is exact but not particularly useful . expanding @xmath68 and @xmath69 in the form of eq . ( [ eq : expand ] ) leads to the approximate result @xmath70 from which we deduce his form should be @xmath28 times a typical nuclear size ; this need not be the case . ] of @xmath45 : @xmath71 finally , the hypervirial ( or hellmann - feynman ) quantity @xmath72 in eq . ( 6 ) can be written using eq . ( 12 ) in the form @xmath73 where @xmath34 is the electrons electric field at the nuclear cm , and the last term arises from averaging that electric field over the nuclear volume . because it is the averaged field that cancels the external field at the nuclear cm , we see explicitly how nuclear finite size affects screening . moreover , because that last term is equivalent to @xmath74 we can see that it produces the @xmath46 term in the schiff moment in eq.([eq : def ] ) ( via the second term in eq . ( [ eq : del ] ) ) . the @xmath45 term in eq . ( [ eq : def ] ) arises from the first term in eq . ( [ eq : del ] ) . having formulated expressions for nuclear schiff moments , we want to use them together with assumptions about the dominant source of p and t violation to evaluate schiff moments in real nuclei . in the rest of this paper we will assume that a p , t - violating component of the nucleon - nucleon interaction causes a schiff moment in the distribution of protons , but we conclude this section by briefly describing another possibility : that dipole moments of individual nucleons are responsible for the nuclear schiff moment . we introduce proton and neutron ( isotopic ) projection operators , @xmath75 and @xmath76 , for the i@xmath11 nucleon . the dipole moment must point along the spin , @xmath77 , of the nucleon , leading to the impulse - approximation result@xcite @xmath78 where @xmath79 and @xmath80 are the proton and neutron electric dipole densities ( normalized to 1 ) associated with @xmath81 and @xmath82 . this yields @xmath83 which depends only on the ground - state expectation value of nucleon spin and isospin operators . the schiff moment can be obtained by evaluating @xmath45 with eq . ( [ eq : nucleon ] ) , producing @xmath84 where @xmath85 and @xmath86 are the mean - square radii of the densities @xmath79 and @xmath87 and @xmath88 we expect @xmath89 to be comparable to nuclear sizes , and thus much larger than @xmath90 , which should be comparable to nucleon sizes . for our considerations , as we have said , the most important mechanism for a nuclear ground - state electric dipole or schiff moment is the action of a pseudoscalar t - violating nucleon - nucleon potential , @xmath91 . this induces a schiff moment given by second - order perturbation theory : @xmath92 where the state @xmath93 has @xmath94 and @xmath95 is the @xmath96-component of the schiff - moment operator @xmath59 . the distribution of schiff strength to the excited states @xmath97 is therefore a crucial ingredient in the ground - state moment . at first sight , one might think that this distribution should resemble that of the electric dipole operator @xmath98 , but the two are remarkably different . most of the electric dipole strength is in a broad resonance at 10 to 20 mev of excitation energy . almost none of the schiff strength , however , goes to the states in the giant resonance . in fact , in the simple but venerable goldhaber - teller ( gt ) model , in which the giant dipole resonance corresponds to the @xmath99 oscillation of all the protons with respect to all the neutrons , the schiff strength to the resonance is identically zero . to see this , we start by assuming@xcite rigid distributions for the @xmath3 protons and for the @xmath100 neutrons , which oscillate harmonically about their cm with frequency @xmath101 . the separation of the two rigid spherical ( for simplicity ) distributions is denoted @xmath102 and the distance of the proton cm from the overall cm is @xmath103 . the charge distribution is then given by @xmath104 where @xmath105 is the `` bare '' nuclear - ground - state charge distribution normalized to @xmath3 protons . expanding this equation , one finds components of @xmath106 proportional to @xmath107 , @xmath108 , @xmath109 , @xmath110 , or equivalently ( because @xmath102 oscillates harmonically ) , @xmath111 , @xmath112 and @xmath113 , @xmath114 and @xmath115 , etc . in particular one finds that the @xmath116-term renormalizes the ground - state charge distribution through `` vacuum fluctuations '' ( viz . , the _ net _ 0@xmath117 part , where the nucleus is excited 1 @xmath117 and then deexcited by the same amount ) , a result that we will neglect for the moment since it does nt affect the basic physics . ignoring all but the monopole and dipole parts in eq . ( [ eq : rhogt ] ) , we have @xmath118 we perform the usual decomposition of @xmath102 in terms of normalized ( cartesian ) creation and destruction operators @xmath119 with @xmath120 = \delta_{i j } \ \ , \ ] ] where @xmath121^{-1 } + [ n m]^{-1}$ ] is the inverse reduced mass of the ( rigid ) protons - neutrons system , and @xmath122 is the nucleon mass . we discuss the effects of vacuum fluctuations under ref . @xcite in the reference list . the only modification they produce is the replacement of @xmath105 in eq . ( [ eq : rhogtexp ] ) by the complete ground - state charge density of the model , @xmath123 . this `` renormalization '' ( removal of the vacuum fluctuations ) has for example shifted the 1@xmath117 component of the operator @xmath124 ( the last term in eq . ( [ eq : rhogtexp ] ) ) to the second term of that equation , and only true 3@xmath117 excitations remain from that operator . the modified eq . ( [ eq : rhogtexp ] ) then expresses the nuclear charge - density operator in terms of the ground - state charge density , the 1@xmath117 transition charge density , the 3@xmath117 ( dipole ) transition density , etc . the contribution of the 1@xmath117 excitations to the nuclear moments @xmath46 , and @xmath45 are now easy to obtain . to evaluate the matrix element of the schiff operator in eq . ( [ eq : sm ] ) we need these two moments of the transition charge density , defined by the second term in eq . ( [ eq : rhogtexp ] ) ( @xmath125\ , \vec{q } \cdot \vec{\nabla } \rho_{\rm ch } ( x)$ ] when vacuum fluctuations are included ) , expressed in terms of @xmath102 , which contains the nuclear raising and lowering operators . one determines the 1@xmath117 components of these moments from eqs . ( 8 - 11 ) : @xmath126 @xmath127 the electron - nucleus coupling is defined by @xmath128 in eq . ( [ eq : del ] ) . recalling that @xmath56 in that equation is just @xmath123 here , and using eq . ( [ eq : d ] ) in @xmath57 , we immediately obtain @xmath129 implying that @xmath130 the last conclusion also follows directly from eqs . ( [ eq : o ] ) and ( [ eq : def ] ) . equation ( [ eq : h=0 ] ) is more general than eq . ( [ eq : s=0 ] ) , however , because it is true to all orders in @xmath40 . thus , despite the fact that it contains all the dipole strength , the goldhaber - teller giant resonance generates no contribution to the schiff moment and therefore to the atomic dipole moment . in the next section we will see that the same is nearly true in real nuclei . to understand in greater detail the distribution of schiff strength and the resulting ground - state schiff moment , we first examine the situation in light nuclei . although the small radii and charges of light nuclei mean that their schiff moments will not be large compared to those of heavy nuclei , they have the advantage that their structure can be calculated at a detailed microscopic level . what kinds of excitations will carry the schiff strength ? we describe elementary excitations in terms of harmonic - oscillator shell - model quanta ( @xmath117 ) ; these are not exactly the same as the @xmath117 of the goldhaber - teller model , but the two are related . as shown by brink@xcite , the electric dipole operator @xmath46 can excite only those components of the harmonic - oscillator shell - model hamiltonian ( with no residual interactions ) corresponding to the giant resonance . that is , the simple 1@xmath117 electric dipole excitations in the harmonic oscillator are exactly the same as in the goldhaber - teller model . the operator @xmath45 , on the other hand , can excite other shell - model modes isoscalar 1@xmath117 and all kinds of 3@xmath117 that are not a part of the gt model . the gt part of the 1@xmath117 excitations will cancel in @xmath59 as shown above ; the isoscalar 1@xmath117 and all 3@xmath117 excitations , by contrast , will contribute to @xmath59 . octupole modes are important because the e3 and @xmath45 operators are in some sense the @xmath131 and @xmath132 angular - momentum projections of the same operator . the _ isoscalar _ @xmath133 and e3 strengths , which contain both 1@xmath117 and 3@xmath117 components , are pulled down into low - lying @xmath134 and @xmath135 states ( in even - even nuclei ) with similar structure . in combination with the suppression of the schiff strength in the giant - resonance region , this similarity in structure results in most of the available schiff strength being strongly correlated with octupole excitations . the correlation can be seen in the closed - shell nucleus @xmath136o even without much calculation . the lowest 3@xmath137(6.05 mev ; t=0 ) state in @xmath136o has an enhanced e3 transition to the ground state , b(e3)=13.5@xmath1380.7 weisskopf units ( w.u . ) . the lowest 1@xmath137(7.12 mev;t=0 ) state decays to this 3@xmath137 state through an enhanced e2 transition , b(e2)=@xmath139 w.u . , suggesting that the two states are of similar structure ( i.e. that the @xmath134 state is a quadrupole phonon coupled to the 3@xmath137 state ) . the @xmath134 state shows an enhanced isospin - forbidden e1 transition to the ground state , b(e1)=@xmath140 w.u . although isospin mixing obviously contributes to the transition , a significant portion of the isospin - forbidden e1 strength appears to come from a large isoscalar matrix element of @xmath45 , a part of the e1 operator that is normally masked@xcite . shell - model calculations reflect the strong correlation between the lowest isoscalar @xmath135 and @xmath134 states @xcite . furthermore , the @xmath141 ground - state correlations in such calculations have a large overlap with the state formed by acting on the closed shell with two successive e3 operators . these higher-@xmath142 correlations in the ground - state wave function enhance the excitation of octupole - like @xmath143 components in the 1@xmath137 and 3@xmath137 states , leading to larger low - lying isoscalar e3 and @xmath45 matrix elements . let us see how this physics works out in @xmath1f , which has odd a and is therefore able to have a ground - state schiff moment . the ground state of @xmath1f is the @xmath144 head of a @xmath145 rotational band , while the first excited state ( a 1/2@xmath137 state at 110 kev ) is the basis for a 4p-1h ( @xmath146 ) @xmath147 band . su(3)-basis shell - model calculations@xcite for these two bands indicate that the @xmath148 band is an octupole excitation of the ground - state band ( in fact @xmath1f is considered the closest thing among light nuclei@xcite to an octupole - deformed system , though a nonnegligible part of the @xmath148 state is an isovector excitation ) . though this relation in itself implies large e3 and schiff matrix elements between these two bands , we have to go beyond these old restricted calculations to get the full picture , for two reasons . first , the just - mentioned octupole ground - state correlations that further enhance the schiff strength were omitted . second , we need to ensure realistic behavior for the strength distribution of the p , t - violating nn interaction , which according to eq.([eq : sm ] ) is as important as the schiff - strength distribution in determining the ground - state schiff moment . fortunately , its behavior is much simpler . for the sake of pedagogy , we consider in this section a simple but often accurate@xcite one - body approximation to @xmath91 : @xmath149 in both isovector and isoscalar channels ( we will use a more sophisticated one - body approximation later ) . the @xmath150 e1 and @xmath91 operators then differ only in their effect on spin ( they are in the same su(4 ) multiplet ) . the strength from the isovector @xmath151 is therefore concentrated in an su(4 ) analog of the giant dipole resonance , and to first approximation lies at the same energy . the isoscalar strength is also mostly in a resonance at a similar energy , so that low - lying p , t - violating strength is depleted@xcite . there will be some strength at low energies where the schiff strength is concentrated , just as there is some e1 strength , but it will represent the tail of a resonance . this tail was not included in the calculations of ref . @xcite . figures 1 - 3 display the results of a complete ( 0 + 1)@xmath142 shell - model calculation for @xmath1f , with the center - of - mass motion fully eliminated . in these calculations contributions to the schiff moment from 2 and 3@xmath142 excitations have been included through the use of an effective charge and @xmath143 excitations consistently in the low - lying states . however , the concept of an effective charge for e3 transitions works well throughout this mass region . the results of very truncated ( 0 + 1 + 2 + 3)@xmath142 calculations suggest a similar prescription for the @xmath45 operator , and in our @xmath150 calculations we applied octupole effective charges to @xmath45 . reference @xcite makes a case for this same kind of renormalization . ] . figure 1 shows the isovector @xmath152 distribution , and above it the isovector @xmath153 strength . the extent of the cancellation suggested by the goldhaber - teller model is remarkable . also displayed is the isoscalar schiff strength , which is uncancelled and has a significant low - lying component ( again correlated with the e3 distribution ) . figure 2 displays the calculated @xmath151 strength , and the giant resonances ( including tails ) are evident in both the isovector and isoscalar channels . finally , in fig . 3 we graph the terms in eq . ( [ eq : sm ] ) as a function of excitation energy , assuming that the isoscalar and isovector potentials have equal strength ; from this it is clear that the lowest @xmath148 state almost completely determines the schiff moment , which is given by the sum of all the lines in the plot . we expect the gross features of the schiff and @xmath154 distributions to be general . the schiff strength will be correlated with the e3 strength and lie low in energy . nuclei with octupole deformation , where the e3 strength lies as low as @xmath15550 kev and is particularly concentrated , will show the most - enhanced schiff moments . nuclei with strong low - lying octupole vibrations should also show enhancement . as the e3 strength moves up in energy ( and/or the octupole collectivity is diluted ) , so should the dominant contribution to the schiff strength , and the schiff moments will become smaller . in the remainder of this paper we examine the extent to which these statements are true in heavy nuclei , where shell - model calculations are not possible . we begin our discussion with a short review of the arguments of ref.@xcite , the crux of which is related to what we have already discussed : the ground - state schiff moment need not be directly related to the dipole moment , and in the case of octupole - deformed nuclei is considerably more enhanced . the authors adopted the particle - rotor model , which is not fully microscopic and omits a certain amount of valence - space physics . the arguments are based on collective octupole correlations , however , and this model represents them clearly and efficiently . in the particle - rotor model the nucleus is described as a single particle coupled to a collective core , the shape of which can be specified through a function that describes the dependence of its radius on angle : @xmath156 where @xmath157 . in the intrinsic frame of a deformed nucleus three of the @xmath158 variables are no longer independent ; they are replaced by three euler angles @xmath159 ( @xmath160 ) that specify the orientation with respect to the laboratory . we will assume axial symmetry so that all intrinsic @xmath158 s vanish except for the @xmath161 s , which we denote by @xmath162 . the valence particle , in the `` strong - coupling '' version of the model , moves in a potential that is deformed to match the shape of the core . the full nuclear wave function ( for a state with angular momentum quantum numbers @xmath163 and @xmath164 , intrinsic magnetic quantum number @xmath165 , and parity @xmath166 ) thus depends on the euler angles , the intrinsic deformation parameters @xmath162 , and the intrinsic space ( @xmath167 ) and spin ( @xmath168 ) coordinates of the odd particle : @xmath169 d^{j*}_{mk}(\theta_i ) \left[1+p\hat{p}\right ] \langle \beta_l,\vec{r},s |\psi_{\rm int } \rangle ~,\ ] ] where the intrinsic particle - core state factorizes as of total angular momentum along the symmetry axis is equal to that of the particle s angular momentum ( sometimes denoted @xmath170 ) . ] @xmath171 here @xmath172 is a normalization constant , @xmath173 rotates the @xmath174-functions and the intrinsic wave function by 180 degrees around the @xmath175-axis , and @xmath176 changes the sign of @xmath167 and of the odd - multipole @xmath177 s in the intrinsic wave function . as long as we do not allow nonaxial core vibrations to be excited , we can , for the purposes of this paper , write the schiff operator @xmath178 as @xmath179 where @xmath180 is the intrinsic - frame operator , with only the @xmath181 component relevant because others do not generate collective excitations in the absence of nonaxial deformation or vibration . [ for @xmath91 , the transformation to the intrinsic frame is trivial because that operator is invariant under rotation . ] we have placed hats on the @xmath182 s and ( on @xmath183 ) because at the quantum level they are operators that act on the wave functions @xmath184 in eq . ( [ eq : intwf ] ) . all of this formalism can be justified at least in part through projected mean - field calculations , in which the state is a function of all @xmath185 nucleon coordinates ; we will argue shortly that such calculations are necessary to answer questions that arise in the simpler description . in the simple particle - rotor model , matrix elements of one - body operators like @xmath95 are straightforward to calculate . the coordinates of the individual nucleons in the core are integrated over in the intrinsic frame to give an operator that is a function of the collective coordinates and those of the odd particle , and can be applied to wave functions of these same coordinates . when one does the integration for the schiff operator , assuming a nuclear charge density proportional to the mass density ( so that the intrinsic dipole moment is zero ) , the result is @xmath186 where the last term is the contribution of the valence particle , which in this case can be neglected compared to the collective piece . without the hats , the first term in eq . ( [ eq : sop ] ) is just the classical schiff moment of a deformed drop . interesting things happen when a core is both quadrupole and octupole deformed ( i.e. when @xmath184 is peaked around nonzero values of @xmath187 and @xmath188 ) . [ for a comprehensive review of the subject , see ref . the reflection asymmetry implies a double - well potential in the coordinate @xmath188 , which in turn means that the wave functions with good parity will be linear combinations of functions peaked around some @xmath188 and its negative . if the barrier between the two wells is high enough , the result will be parity - doubling ; low - lying states will have partners nearby with the same angular momentum but opposite parity . this means , for one thing , that there will be an intermediate state in ( [ eq : sm ] ) that enters with a small energy denominator and therefore a large amplitude . in fact , it is often reasonable to ignore all other states in the sum ( see the shell - model result for @xmath1f in fig . 3 ) , mainly because of the energy denominator , but also because if the deformation is strong enough the matrix element of the schiff operator to that state is likely to be large . the reason for that is that the doublets can be viewed as the projection onto positive and negative parity of the same reflection - asymmetric intrinsic state @xmath189 . the existence of more than one state with the same intrinsic structure is exactly the same phenomenon as the rotational bands associated with ordinary reflection - symmetric deformation . the matrix element of an operator between the two states of the same doublet is proportional to the diagonal intrinsic - state matrix element of the intrinsic operator , just as it is for states within a rotational band . for the operator @xmath180 , this diagonal matrix element is large ; it is given roughly by the expression in eq . ( [ eq : sop ] ) with the operators @xmath190 replaced by the values around which the wave function @xmath184 is peaked ( i.e. by the classical schiff moment of the deformed asymmetric core ) , with coherent contributions from all the nucleons in it . thus , from eq.([eq : sm ] ) , the physical ground - state schiff moment @xmath153 that determines the atomic electric dipole moment should be approximately @xmath191 where @xmath192 is the opposite - parity partner of the ground state @xmath193 , @xmath194 is the energy difference between the two states , and @xmath183 is the intrinsic schiff moment ( the @xmath163-dependent factor is from the euler - angle integration ) . the conclusion from all this is that large intrinsic schiff moments and small energy denominators should make atoms with octupole - deformed nuclei especially sensitive tests of p , t - violation in the nucleon - nucleon interaction . just how sensitive the tests will be depends on the matrix element of the interaction @xmath91 , to which we now turn . we will supply more detail in addressing this subject because the treatment of ref . @xcite is incomplete and not entirely accurate . we assume that the pion is responsible for transmitting most of the t - violating force from one nucleon to another @xcite . the nucleon - nucleon interaction produced by one - pion exchange then has the general form@xcite @xmath195 \\ + c_1 ( \vec{\sigma_1}+\vec{\sigma_2 } ) \cdot ( \vec{r_1}-\vec{r_2 } ) ( \tau_{1z}-\tau_{2z } ) \end{array } \right\ } \nonumber \\ & \times & \frac{{\rm exp}(-m_{\pi } |\vec{r_1}-\vec{r_2}|)}{m_{\pi } \right]~,\end{aligned}\ ] ] where the @xmath196 s label isoscalar , isovector , and isotensor contributions . as long as we excite no particles out of the core , p , t - violating interactions between these particles sum to zero ( see ref . @xcite for discussion ) , and we need worry only about the interactions between the valence particle and those in the core . in what follows , we take the total mass density to be proportional to the charge density , and in fact _ from now on use the symbol @xmath51 to represent the mass density_. this assumption means that the terms with different isospin structure in eq . ( [ eq : pion ] ) enter in similar ways . taking the range of the pion to be very short ( a decent approximation@xcite ) , summing eq . ( [ eq : pion ] ) over the particles in the core , and assuming the neutron and proton densities to be equal gives the effective nucleon - core interaction @xmath197 where ( again ) @xmath198 is the core mass - density operator , @xmath199 is the fermi constant , and @xmath122 is the nucleon mass ( these factors are inserted to follow convention ) . the dimensionless parameter @xmath200 then depends on the coupling strengths @xmath196 of the two - body interactions and the isospin of the nucleus . despite its slight dependence on nuclear structure , this parameter is often taken as a `` heuristic '' fundamental quantity . the one - body approximation given by eq . ( [ eq : sppot ] ) is slightly different from the simpler and more phenomenological one we used in our discussion of light nuclei . the form of eq . ( [ eq : sppot ] ) makes the schiff moment unpleasantly sensitive to the distribution of spin near the nuclear surface , where @xmath201 is largest . the results of ref . @xcite sometimes differ by factors of several from those of ref . @xcite , primarily because of differences in the valence single - particle wave function , which carries all the nuclear spin @xmath202 in the particle - rotor model . only a significantly more sophisticated calculation ( which we advocate ) will reduce this uncertainty . we therefore do not present our own complete `` particle - rotor - model - with - octupole - deformation '' calculations in this section , but instead use that model in its simplest form , together with qualitative arguments , to identify a few systematic effects overlooked in the existing calculations . our estimate of the size of these effects is obviously uncertain , but indicates what can be expected in more sophisticated calculations . the new physics always tends to lessen the enhancement . to see the what was neglected in refs . @xcite we follow ref . @xcite and expand the density in the deformation parameters : @xmath203 where now @xmath204 is the bare ( spherical ) ground - state mass density of the core , a constant up to radius @xmath205 in the simplest version of the liquid - drop picture . the operator @xmath206 in @xmath207 therefore depends on the @xmath208 s , and , contrary to the statements in ref . @xcite , can not be broken up into pseudoscalar pieces that act separately on the core and particle . in fact , from eq . ( [ eq : density ] ) we have @xmath209^l_0 - \sqrt{\frac{l}{2l+1 } } \left ( \frac{d}{dr } + \frac{l+1}{r } \right ) \rho_0 ' [ y_{l-1 } \sigma]^l_0 \right ] \nonumber \\ & - & 1/2 r_0 ^ 2 \sum_{l , l',l } \hat{\beta}_l \hat{\beta}_{l ' } \sqrt{\frac{(2l+1)(2l'+1)}{4\pi ( 2l+1)^2}}\langle l0,l'0 | l0 \rangle \nonumber \\ & & \times \left[\sqrt{l+1 } \left(\frac{d}{dr}-\frac{l}{r}\right ) \rho_0 '' [ y_{l+1}\sigma]^l_0 -\sqrt{l } \left ( \frac{d}{dr}+\frac{l+1}{r } \right ) \rho_0 '' [ y_{l-1}\sigma]^l_0 \right ] ~,\end{aligned}\ ] ] where the small square brackets indicate angular - momentum coupling . reference @xcite considers only the first term in this expression , which changes the parity of the single particle and leaves the core alone . the terms with odd powers of @xmath210 can do the opposite , however . including these other terms is important because in the strong - coupling limit of the particle - rotor model they tend to cancel the first term . the reason is hinted at in ref . @xcite , where it is argued that schiff moments between close - lying states are suppressed when deformation is rigid and symmetric . following ref . @xcite , the authors note that to the extent that the density is proportional to the strong one - body potential @xmath211 felt by the odd particle and that the spin - orbit force is negligible , matrix elements of @xmath206 between two states should be proportional to the energy difference between those states , and therefore very small for close - lying doublets . the reason is that under these circumstances @xmath212 = i [ \vec{\sigma } \cdot \vec{p},\hat{h}(\rm s.p . ) ] ~,\ ] ] so that for two opposite - parity states labeled @xmath213 and @xmath214 with the same core structure , @xmath215 where the @xmath216 s are single - particle energies . the authors then argue that complications associated with asymmetric deformation eliminate this effect , but in the strong - coupling limit the situation is even worse because now the two states @xmath217 and @xmath218 have the same intrinsic structure . as mentioned above , the matrix element of any operator between two such states is proportional in the strong - coupling limit to the intrinsic - state expectation value of the operator . in nuclei with strong octupole deformation , we therefore have @xmath219 which , according to the argument above , should vanish . the estimates of refs . @xcite apparently neglect the terms in @xmath51 containing the @xmath182 s that make the shapes of the density distribution and the potential similar , and so do not take this effect into account . of course the spin - orbit force is not negligible and the intrinsic density is not exactly proportional to the mean field , so the cancellation will not be complete . to get a handle on how much the terms containing the @xmath182 s affect the matrix element of @xmath207 , we consider the ratio of the matrix element in eq . ( [ eq : spexp ] ) with the terms included to that without them ( the latter being the a simplified version of the quantity calculated in refs . @xcite ) for a large number of single - particle orbits . we use a deformed harmonic oscillator as a potential , @xmath220 with deformations @xmath187 , @xmath188 , and @xmath221 equal to those from table i of ref . @xcite ( we ignore higher multipoles and neglect pairing ) . we take the density to be constant inside the liquid drop and zero outside , a distribution that has the same angular shape as the potential , but a significantly different radial dependence . figure 4 shows the absolute value of the ratio of @xmath222 to the same matrix element without the @xmath182-dependent terms , for all @xmath223 and @xmath224 single - particle levels in @xmath225ra below 8 @xmath142 of single - particle energy . the new terms generally have the opposite sign from that of the @xmath182-independent term , and the sum is usually less than the first term by itself . the average cancellation is less for @xmath226 states than for states with larger spin . whether the decrease will be stronger or weaker in realistic calculations is an open question . in self - consistent mean - field calculations , though , there is obviously a correlation between the density and the spin - independent part of the field , and we could well see a significant effect . to @xmath227 ( see text ) for all @xmath223 ( boxes ) and 3/2 ( crosses ) states in a deformed asymmetric potential for @xmath225ra . the ratio is usually less than one.,width=529,height=453 ] other physics neglected in the earlier work will also have an effect . the first term in eq . ( [ eq : expansion ] ) ( @xmath228 ) is the spin - flip analog of the electric dipole operator ( with a difference only in radial form and isospin ) . we refer to fig . 2 , where the strength of the related operator @xmath229 in the isovector and isoscalar channels is plotted for @xmath1f . as noted in section iii , the strength is clearly concentrated in resonances at about the same energy as the giant dipole resonance . a simple argument@xcite with a schematic residual interaction in rpa shows that the existence of the giant dipole resonance of the core suppresses e1 transitions between low - lying single - particle states by a factor of 3 or 4 that depends only on the energy of the resonance and the energy at which strength would be centered if there were no resonance . the low - lying transitions induced by @xmath230 should be suppressed by roughly the same amount because the operator @xmath231 is so much like @xmath167 ( again , see fig . 2 ) . in many - body perturbation theory , this effect can be understood as core polarization : the residual strong interaction can create a collective @xmath232 p - h pair , at the same time changing the parity of the valence particle . the contribution to the schiff moment coming from the annihilation of the pair by @xmath207 nearly cancels the contribution coming from the direct action of @xmath207 on the valence particle itself . in any event , the resonances were completely neglected in refs.@xcite , and the single - particle matrix elements in those papers should therefore probably be three or four times smaller . the very large schiff moments may be saved , however , by the _ combination _ of the two new effects , even though each reduces schiff moments when added to the calculations of refs . @xcite in isolation . the spin - flip giant resonance should not affect those parts of @xmath207 that contain @xmath188 and operators like @xmath233 ^ 3_0 $ ] that do not change the parity of the valence particle . the suppression by the residual interaction of one part of the schiff moment by a factor of three without any effect on another part may upset any balance between the two at the mean - field level produced by the similarity between the density and the potential . we will need accurate microscopic calculations to test the existence of both effects and the extent to which they offset one another . such calculations are within the range of today s mean - field technology , and in even - even nuclei they have already been carried out@xcite , confirming the large intrinsic schiff moments in the radium isotopes@xcite . in self - consistent ( e.g. skyrme - hfb ) calculations the effects of the residual interaction on the ground state are minimized . therefore , not only is the relationship between the mean field and the density likely to be most accurate in this case , but it is also least likely to be vitiated by corrections to the single - particle picture . in odd nuclei , the first - order core polarization , which has to be treated as a correction to particle - rotor / nilsson models , is built into the mean - field ; 1-particle-1-hole excitations of the core do not mix with the ground state . this fact , together with a realistic two - body interaction that contains all multipole - multipole terms , has implications for corrections to low - lying transitions from resonances , as well as for the density . the incorporation of core polarization means that the interplay between collective excitations and low - lying states is already apparent at the mean - field level , or in other words that the usual first - order particle - phonon mixing that reduces low - lying single - particle transitions need not be treated by other means ( e.g. the rpa ) . we should therefore be able at the mean - field level to go a long way towards quantifying the influence of shape and resonances on schiff moments in octupole - deformed nuclei . the question of whether the light actinides are octupole deformed has a long history . in fact the question is not entirely physical it s really about the economy of one collective - model basis versus another and it should not matter so much whether the low - lying states in a nucleus are best described as the rotation of an octupole - deformed shape or as a strong low - lying octupole vibration around a rotating quadrupole shape . collective schiff moments arise in either scheme . this fact should not be surprising in light of our calculations in @xmath1f . to see how it falls out of the collective picture , we assume that the nuclear core has no equilibrium octupole deformation ( i.e. @xmath234 ) and write the operator @xmath210 in terms of creation and annihilation operators : @xmath235 where @xmath236 creates an octupole phonon with ( intrinsic ) magnetic quantum number @xmath237 . it is then clear from eq . ( [ eq : sop ] ) that the schiff operator acting on a quadrupole - deformed state with no octupole phonons will create an excited state with one phonon . the terms in @xmath207 ( see eq . ( [ eq : expansion ] ) ) that are proportional to @xmath210 can then destroy the phonon , reconnecting the one - phonon state to the ground state and generating a collective schiff moment through eq . ( [ eq : sm ] ) . to see how big such a moment would be we need to know the matrix element of @xmath210 between states with zero and one phonons . as can be seen from eq . ( [ eq : phonon ] ) , this quantity is just the zero - point root - mean - square deformation , ( @xmath238 ) , which we will call @xmath239 . [ in other words , @xmath239 measures the spread in @xmath188 of the intrinsic core wave function @xmath184 . ] this quantity can be estimated from the collective ( vibrational ) b(e3 ) transition in an even - even neighbor . using eq . ( [ eq : density ] ) to lowest order in @xmath239 one finds@xcite @xmath240 the important point is that if a collective vibration is soft the r.m.s.deformation @xmath239 can be as large as the value around which the wave - function is peaked in octupole - deformed nuclei , and the intrinsic schiff moment can therefore be just as large as well . in the laboratory ( physical ) schiff moment , there is an additional factor of @xmath239 coming from the annihilation of the phonon by @xmath207 , so that naively we expect the moment to depend on the deformation parameters in the combination @xmath241 , where an unbarred @xmath182 is the value around which the deformed wave function is peaked . the relevant quantity for octupole - deformed nuclei is @xmath242 ( see ref . @xcite for a discussion of why ) , so that if the r.m.s . octupole deformation @xmath239 in a vibrational nucleus is comparable to the static value @xmath188 of the deformation in an octupole - deformed nucleus , any differences in schiff moments come from the energy denominator , single - particle structure , or other core excitations , not from the difference between deformation and vibration . we will refine this statement shortly . first , however , we note that the terms in @xmath207 that do nt contain @xmath210 are usually even more important than those just discussed , even though they do nt alter the number of phonons , because the zero- and one - phonon states mix through the residual strong particle - core interaction . the approximate form of this coupling can be derived in many ways ; one is to examine the change in energy under a small deformation of the core . not surprisingly , for an oscillator single - particle potential this leads to the same interaction that appears in the octupole - deformed potential of the strong - coupling scheme ( see eq . ( [ eq : defho ] ) : @xmath243 where @xmath101 is the oscillator energy of the ( symmetric ) potential , @xmath244 acts on the core , and @xmath245 acts on the particle . denoting a state with @xmath246 phonons and a particle in orbit @xmath247 by @xmath248 , we have for the matrix element of the interaction @xmath249 between excited states with one phonon and the unperturbed ground state ( assuming just for illustration that the ground state has positive parity ) : @xmath250 with a value for @xmath239 from an appropriate @xmath251 , we can use the `` intermediate''-coupling scheme of ref . @xcite to diagonalize @xmath252 separately in positive- and negative - parity bases ( @xmath253 just contains the diagonal vibrational energies of the zero- and one - phonon states ) , so that the ground state has the form @xmath254 and the excited states of opposite parity have the form @xmath255 label single - particle states around the fermi surface , and we are still ignoring nonaxial vibrations . the terms in @xmath207 that are independent of @xmath210 connect the first terms in eq . ( [ eq : gs ] ) to the second in eq . ( [ eq : excited ] ) and vice versa . the schiff operator affects the core , connecting the first term in eq . ( [ eq : gs ] ) to the first in eq.([eq : excited ] ) , and the second to the second , effectively replacing @xmath210 in eq . ( [ eq : sop ] ) by @xmath239 . in this way the spherical @xmath210-independent part of @xmath207 ( the only part considered in refs . @xcite ) can also generate a collective schiff moment . it is possible to use the intermediate - coupling scheme even as the phonon energy goes to zero and octupole deformation sets in . in that case , because the single - particle hamiltonians in the two schemes are the same , energies and matrix elements should not depend strongly on which scheme is used . one implication ( which is a stronger version of a remark made above ) is that if the dynamic @xmath239 associated with the vibration is comparable to the static @xmath188 in an octupole - deformed nucleus , and if the energy of the octupole phonon is small compared to typical single - particle splittings or nonaxial core - excitation energies , the _ only _ major difference between schiff moments in the two cases is the energy denominator in eq . ( [ eq : sm ] ) . to see this , one can imagine treating the phonon as a `` decoupling '' perturbation ( along with the coriolis interaction ) in the strong - coupling scheme , as is done in ref . @xcite . although the diagonal matrix elements of the perturbation cause energy shifts , wave functions are only affected by the off - diagonal matrix elements with bands built on higher single - particle states or other kinds of vibrations . thus wave functions , transition amplitudes , etc . , will not undergo large changes until the energy of the phonon approaches those of other excitations . for the intermediate - coupling scheme , this means that the matrix elements of @xmath153 and @xmath91 connecting ground states to low - energy octupole phonons should not undergo radical change once the phonon is low enough in energy , and that nothing special will happen in the limit that the phonons have zero energy and the core develops static deformation . of course if the phonon lies high in the spectrum , the matrix elements can be very different from the static limit , and one must carry out the intermediate - coupling calculation to get a handle on the size of the schiff moment induced by vibrations . we have done just that in several quadrupole - deformed nuclei , taking vibrational @xmath239 s and phonon energies from tabulations of nearby even - even nuclei@xcite , and again neglecting pairing . in @xmath0hg , the most accurately measured isotope at present , we use @xmath239 = .09 , a phonon energy of 3 mev ( both taken from an e3 transition in @xmath256hg , which may have a larger @xmath239 than @xmath0hg ) , and quadrupole and hexadecupole parameters from ref . @xcite ( for this simple estimate we ignore the fact that this nucleus is probably very soft ) . the resulting schiff moment is @xmath257 e @xmath258 , about half of the estimate from ref . @xcite that includes no nuclear correlations of any kind . in a nucleus like this , moreover , with a relatively high - energy phonon , the nonaxial octupole vibrations will lie nearby in energy and can be expected to contribute comparable amounts . when all is said and done , vibrations may turn out to be the dominant contribution to the schiff moment in @xmath0hg , and they clearly should be included in any realistic calculation . such a calculation has never been done , but is crucial if we want a reliable assessment of the advantages offered by nuclei with strong octupole correlations . here we need a good microscopic treatment of all kinds of vibrations , including the very soft @xmath259 quadrupole mode , and must obviously go beyond mean - field theory . a shell - model calculation may be possible . an example of a large vibrational schiff moment is in the nucleus @xmath260pu . this isotope has several features that make it attractive for experiment , in particular a spin-1/2 ground state ( to eliminate quadrupole effects in a magnetic field ) and a long half - life compared to the light actinides [ the drawback is in its electronic structure , which is more complicated than that of radium ] . the collective e3 in @xmath261pu gives @xmath262 , and the nucleus has large quadrupole and hexadecupole deformations ( @xmath263 , @xmath264 ) . together with the high value of @xmath3 , this makes the intrinsic schiff moment very large . the phonon lies at 470 kev , about 8 times higher than the lowest state in @xmath225ra , but the large intrinsic schiff moment compensates in part . our calculations , with only the @xmath208-independent terms included in @xmath198 ( as in ref.@xcite ) give a laboratory schiff moment of @xmath265 , a value a few times smaller than the results of ref . @xcite for most of the light actinides . when we include the @xmath208-dependent terms , this number goes up to @xmath266 , which is 300 times the estimate for hg in ref . @xcite and comparable to the results of ref . @xcite for @xmath225ra . these calculations are far from perfect ; we had to push the energy of the octupole phonon well above the value from @xmath261pu to get the energy of the first excited state in @xmath260pu right . the uncertainty in the results is therefore quite large and we need microscopic calculations here too . but the intrinsic schiff moments will of nuclei with low - lying octupole vibrations will clearly be collective , and some may of these nuclides may be easier to investigate experimentally than the short - lived radium isotopes . the size of schiff moments in nuclei with octupole correlations is determined by three factors : intrinsic schiff moments , energy denominators , and the matrix elements of @xmath91 . in their discussion of the first two of these , refs . @xcite are on rather firm ground ; it is hard to imagine , for example , that the matrix elements between parity doublets of the schiff operator are radically different from those estimates , and as we have pointed out , even nuclei without asymmetrically deformed cores can benefit from the same mechanism . the third factor is far trickier , however . the particle - core calculations reported both here and in earlier work can only supply a gross estimate of the matrix element of @xmath91 . the mixing that that interaction induces depends sensitively on the valence single - particle wave function at the nuclear surface , where @xmath267 is largest . truly microscopic calculations will give better valence wave functions and , if they are self - consistent , will also better represent the correlation between density and mean field , and incorporate the effects of resonances caused by the residual interaction . in vibrational nuclei it will be necessary to go a little further , but even there mean - field calculations will shed light on the issues we ve discussed . finally , for experimentalists to draw strong conclusions about enhancements over @xmath268hg , better calculations in that nucleus must be done as well . it is conceivable that the atomic dipole moment of @xmath225ra is 400 or more times larger than that of @xmath0hg ( this is the figure reported in ref.@xcite ) , but we have pointed to physical effects that could make the schiff moment in @xmath268hg a few times larger than earlier calculations indicate and the schiff moments in the light actinides somewhat smaller than suggested by the calculations of refs . @xcite , even within the same model . the machinery of modern nuclear structure theory , which is powerful enough to provide reasonably accurate estimates of the moments in both kinds of nuclei , should be used as soon as possible to provide experimentalists firm predictions for the enhancement they can expect in difficult experiments with radioactive nuclei . we thank j.l . egido for useful calculations of schiff moments in the even - even light actinides . we are grateful to w. nazarewicz , i.b.khriplovich , p. herczeg , and s. lamoreaux for helpful comments about the manuscript . the work of j.e . was supported by the u.s . department of energy under grant de fg0297er41019 , and the work of j.l.f . and a.c.h . was performed under the auspices of the united states department of energy . feynman , phys . rev . * 56 * , 340 ( 1939 ) ; h. hellmann , _ einfuhrung in die quantumchemie _ ( franz deuticke , leipzig , 1937 ) , p. 285 . the usual statement of the theorem is that if a hamiltonian @xmath269 depends on a parameter @xmath270 , the energy shift @xmath271 for that hamiltonian satisfies @xmath272 . if one performs a unitary transformation so that @xmath273 , uses the theorem and then sets @xmath270 to zero , one finds @xmath274 \rangle$ ] , since there can be no energy shift from a unitary transformation . our quantity @xmath27 in eqs . ( [ eq : delta ] ) and ( [ eq : fh ] ) is given by this relationship . j.l . friar and s. fallieros , phys . * 114b * , 403 ( 1982 ) . although no consensus notation for retarded electric multipoles exists , we adopt the nomenclature of this reference , which has seen considerable use . wilcox , j. math . * 8 * , 962 ( 1967 ) derives and discusses various forms of the baker - campbell - hausdorff ( bch ) formula ( including eq . ( 4.15 ) that we require ) . fourier transforming eq.([eq : rhogt ] ) leads to @xmath275 , where @xmath276 and @xmath277 is the fourier transform of @xmath105 . because @xmath278 = n^2 \hbar k^2/ 2 \mu \omega a^2 $ ] is a c - number , one obtains the exact bch result , @xmath279 , where the exact ground - state charge form factor for the gt model is given by @xmath280 , since @xmath281 follows from latexmath:[$\xi

traditionally security in wireless networks has been mainly focused on higher layers using cryptographic methods @xcite . pioneered by aaron wyner s work @xcite , which introduced the wiretap channel and established fundamental results of creating perfectly secure communications without relying on private keys , physical - layer - based security has drawn increasing attention recently . the basic idea of physical layer security is to exploit the physical characteristics of the wireless channel to provide secure communications . the security is quantified by the _ secrecy capacity _ , which is defined as the maximum rate of reliable information sent from the source to the intended destination in the presence of eavesdroppers . wyner showed that when the eavesdropper channel is a degraded version of the main channel , the source and the destination can exchange secure messages at a non - zero rate . the following research work @xcite studied the secrecy capacity of the gaussian wiretap channel , and @xcite extended wyner s approach to the transmission of confidential messages over the broadcast channels . very recently , physical layer security have been generalized to investigate wireless fading channels @xcite@xcite , and various multiple access scenarios @xcite@xcite . note the fact that if the source - wiretapper channel is stronger than the source - destination channel , the perfect secrecy rate will be zero @xcite . some work @xcite@xcite has been proposed to overcome this limitation with the help of relay cooperation by _ cooperative relaying _ @xcite@xcite , and _ cooperative jamming _ for instance , in @xcite and @xcite , the authors proposed effective decode - and - forward ( df ) and amplify - and - forward ( af ) based cooperative relaying protocols for physical layer security , respectively . cooperative jamming is another approach to improve the secrecy rate by interfering the eavesdropper with codewords independent of the source messages . in yener and tekin s work @xcite , a scheme termed _ collaborative secrecy _ was proposed , in which a non - transmitting user was selected to help increase the secrecy capacity for a transmitting user by effectively `` jamming '' the eavesdropper . following similar idea as @xcite , they first proposed cooperative jamming in @xcite and @xcite in order to increase achievable rates in the scenarios where general gaussian multiple access wire - tap channel and two - way wire - tap channel were assumed , respectively . the authors of @xcite and @xcite investigated the effects of user cooperation on the secrecy of broadcast channels by considering a cooperative relay broadcast channel , and showed that user cooperation can increase the achievable secrecy region . the study of communicating through unauthenticated intermediate relays between a source - destination pair started from yenner and he s work @xcite@xcite . the relay channel with confidential messages was also investigated in @xcite@xcite , where the untrusted relay node acts both as an eavesdropper and a conventional assistant relay . two - way communication is a common scenario in which two nodes transmit information to each other simultaneously . recently , the two - way relay channel @xcite@xcite has attracted lots of interest from both academic and industrial communities due to its bandwidth efficiency and potential application to cellular networks and peer - to - peer networks . in @xcite and @xcite , both af and df protocols for one - way relay channels were extended to general full - duplex discrete two - way relay channel and half - duplex gaussian two - way relay channel , respectively . in @xcite , network and channel coding were used in two - way relay channel to increase the sum - rate of two sources . the work in @xcite introduced a two - way memoryless system with relays in which the signal transmitted by the relay was obtained by applying an instantaneous relay function to the previously received signal in order to optimize the symbol error rate performance . as for the secure communications , in @xcite , yener and he investigated the role of feedback in secrecy for two - way networks , and proved that the loss in secrecy rate when ignoring the feedback is very limited in a scenario with half - duplex gaussian two - way relay channels and an eavesdropper . it is well known that , in a cooperative communication network , proper relay / jammer selection can have a significant impact on the performance of the whole system . several relay selection techniques @xcite@xcite have been explored by far . the authors in @xcite proposed a non - jamming relay selection scheme for two - way networks with multiple af relays in an environment without eavesdroppers , which maximized the worse received signal - to - noise ratio ( snr ) of the two end users . in @xcite , several relay selection techniques were proposed in one - way cooperative networks with secrecy constraints . in @xcite , the authors investigated some relay selection techniques in a two - hop df cooperative communication system with no central processing unit to optimally select the relay . although cooperative networks have received much attention by far , the physical layer security issues with secrecy constraints in two - way schemes have not yet been well investigated . to this end , in this paper , we propose a scheme that can implement information exchange in the physical layer against eavesdroppers for two - way cooperative networks , consisting of two sources , a number of intermediate nodes , and one eavesdropper , with the constraints for physical layer security . unlike @xcite , in which the relay selection is operated in an environment with no security requirement , our work takes into account the secrecy constraints . in contrast to @xcite , where many relay selections based on the df strategy for one - way cooperative wireless networks were proposed and a safe broadcast phase was assumed , the problem we consider here involves a non - security broadcast phase , and the information is transferred bidirectionally . specifically , a node is selected from an intermediate node set to operate at a conventional relay mode , and then uses an af strategy in order to assist the sources to deliver data to the corresponding destination . meanwhile , another two intermediate nodes that perform as jammers are selected to transmit artificial interference in order to degrade the eavesdropper links in the first and second phases of signal transmissions , respectively . we assume that both destinations can not mitigate artificial interference , and thus , the jamming will also degrade the desired information channels . the principal question here is how to select the relay and the jamming nodes in order to increase information security , and meanwhile protect the source message against eavesdroppers . several selection algorithms are proposed , aiming at promoting the assistance to the sources as well as the interference to the eavesdropper . the theoretical analysis and simulation results reveal that the proposed jamming schemes can improve the secrecy rate of the system by a large scale , but only within a certain transmitted power range . in some particular scenarios , the proposed schemes become less efficient than the conventional ones . we then propose a hybrid scheme with an intelligent switching mechanism between jamming and non - jamming modes to solve this problem . the rest of this paper is organized as follows . in section slowromancap2@ , we describe the system model , and formulate the problem under consideration . section slowromancap3@ presents the proposed selection techniques , and introduces their hybrid implementations . in section slowromancap4@ , we provide both quantitative analysis and qualitative discussions of different selection schemes in some typical configurations . numerical results are shown in section slowromancap5@ , and in section slowromancap6@ , we draw the main conclusions . we assume a network configuration consisting of two sources @xmath0 and @xmath1 , one eavesdropper @xmath2 , and an intermediate node set @xmath3 with @xmath4 nodes . in fig . [ fig_1 ] it schematically shows the system model . as the intermediate nodes can not transmit and receive simultaneously ( half duplex constraint ) , the communication process is performed into two phases . during the broadcasting phase , @xmath0 and @xmath1 transmit their data to the intermediate nodes . in addition , according to the security protocol , one node @xmath5 is selected from @xmath6 to operate as a `` jammer '' and transmit intentional interference to degrade the eavesdropper links in this phase . since the jamming signal is unknown at the rest nodes of @xmath6 , the interference will also degrade the performance of the relay links , as shown in fig . [ fig_1 ] . in the second phase , according to the security protocol , an intermediate node @xmath7 is selected to operate as a conventional relay and forwards the source messages to the corresponding destinations . a second jammer @xmath8 is selected from @xmath6 , for the same reason as that for @xmath5 . note that the destinations @xmath0 and @xmath1 are not able to mitigate the artificial interference from the jamming node , either . and @xmath1.,width=460 ] in both two phases , a slow , flat , and block rayleigh fading environment is assumed , i.e. , the channel remains static for one coherence interval and changes independently in different coherence intervals with a variance @xmath9 , where @xmath10 denotes the euclidean distance between node @xmath11 and node @xmath12 , and @xmath13 represents the path - loss exponent . the channel between node @xmath11 and node @xmath12 is denoted as @xmath14 , which is modeled as a zero - mean , independent , circularly - symmetric complex gaussian random variable with variance @xmath15 . furthermore , additive white gaussian noise ( awgn ) with zero mean and unit variance is assumed . let @xmath16 , @xmath17 and @xmath18 denote the transmitted power for the source nodes , the relay node and the jamming nodes , respectively . in order to protect the destinations from severe artificial interference , the jamming nodes transmit with a lower power than the relay nodes @xcite , and thus their transmitted power can be defined as @xmath19 , where @xmath20 denotes the power ratio of the relay to the jammer . in the first phase , the two sources send information symbols @xmath21 and @xmath22 , respectively , which are mapped to a psk set . the intermediate node @xmath7 and eavesdropper @xmath2 thus receive @xmath23 where @xmath24 and @xmath25 denote the noise at @xmath7 and @xmath2 , respectively . in the second phase , the node @xmath7 is selected to amplify its received signal and forward it to @xmath0 and @xmath1 , i.e. , @xmath7 broadcasts @xmath26 where @xmath27 . since the destination @xmath28 knows @xmath29 ( for @xmath30 ) , it can cancel the self - interference . therefore , @xmath0 , @xmath1 , and the eavesdropper @xmath2 get @xmath31 where @xmath32,@xmath33 , and @xmath34 represent the noise terms at @xmath0 , @xmath1 , and @xmath2 , respectively . then , @xmath35 , defined as the overall signal to interference - plus - noise ratio ( sinr ) of the channel @xmath36 ( for @xmath37 , @xmath38 ) , can be calculated as @xmath39 where @xmath40 represents the instantaneous signal - to - noise ratio ( snr ) for the link @xmath41 : @xmath42 strictly speaking , in order to maximize the overall sinr of the eavesdropping links , the eavesdropper ( @xmath2 ) can perform whatever operations as it wishes with the signals received in the previous two phases . here in this paper , we take a simple case in which the eavesdropper applies maximal ratio combining ( mrc ) @xcite , so as to examine the efficiency of the proposed jamming schemes . according to mrc , @xmath2 combines the received signals by multiplying @xmath43 in ( [ e1 ] ) and @xmath44 in ( [ e2 ] ) with proper weighting factors @xmath45 and @xmath46 , respectively . without loss of generality , consider the scenario in which @xmath2 intends to optimize the sinr of eavesdropper link @xmath47 , for @xmath48 , the combined eavesdropping signal can be written as @xmath49 where @xmath50 with @xmath51 , and @xmath52 is the conjugate transpose . @xmath53 and @xmath54 represent the total interference and noise power terms in @xmath43 and @xmath44 , respectively : @xmath55 where @xmath56 in order to calculate the sinr of link @xmath47 , we assume two different channel knowledge sets : 1 . @xmath57 that denotes a global instantaneous knowledge for all the links , 2 . @xmath58 that denotes an average channel knowledge for the eavesdropper links . with the assumption of @xmath59 , we can get the instantaneous snr of any channel @xmath60 in the system . thus , the sinr of link @xmath61 can be calculated as @xmath62 in an environment where the instantaneous channel knowledge set @xmath59 is not available , we can use the expectation of snrs for the eavesdropper links @xmath63 $ ] , which is provided by the average channel knowledge @xmath64 , to get the sinrs : @xmath65}}{{\mathbb{e}\left [ { { \gamma _ { s_j , e } } } \right ] + \mathbb{e } \left [ { { \gamma_{j_1,e } } } \right ] + 1 } } + \frac{{\mathbb{e}\left [ { { \gamma _ { s_j , r , e } } } \right]}}{{\mathbb{e}\left [ { { \gamma _ { s_j , r , e } } } \right ] + \mathbb{e}\left [ { { \gamma _ { j_1,r , e } } } \right ] + \mathbb{e}\left [ { { \gamma _ { j_2,e } } } \right ] + \mathbb{e}\left [ { { \gamma _ { r , e } } } \right ] + 1 } } , \nonumber \\ & \mbox{s.t.}~\psi_1,\end{aligned}\ ] ] where @xmath66 $ ] stands for the expectation operator . the instantaneous secrecy rate for the node set @xmath6 for source @xmath28 can be expressed @xcite @xmath67^ + , \end{aligned}\ ] ] where @xmath30 , @xmath68 , @xmath69 , and @xmath70^ + \buildrel \delta \over = \max \left\ { { 0,x } \right\}$ ] . the overall secrecy performance of the system is characterized by the ergodic secrecy capacity that is the expectation of the sum of the two sources secrecy rates , @xmath71 $ ] , where @xmath72 our objective is to select appropriate nodes @xmath7 , @xmath5 , and @xmath8 in order to maximize the instantaneous secrecy rate subject to different types of channel feedback . the optimization problem can be formulated as @xmath73 where @xmath74 ; @xmath75 , @xmath76 and @xmath77 denote the selected relay and jamming nodes , respectively . note that here the selected jammers @xmath78 and @xmath79 in the two phases may be the same node , which is determined by the instantaneous secrecy rate . in a conventional cooperative network , the relay scheme does not have the help from jamming nodes . we derive the following solutions under this scenario . the conventional selection does not take the eavesdropper channels into account , and the relay node is selected according to the instantaneous snr of the links between node @xmath0 and node @xmath1 only . therefore , the sinr given in ( [ gamma_j ] ) becomes @xmath80 where @xmath81 represents the sinr of the channel @xmath36 ( for @xmath37 , @xmath38 ) without considering the eavesdropper . hence , the conventional selection algorithm can be expressed as @xmath82 with @xmath83 and @xmath84 for @xmath85 given by ( [ gamma_sisj ] ) and ( [ gamma_rsj ] ) , respectively . since ( [ cs ] ) shows that this selection does not consider the eavesdropping links , the cs algorithm may not able to support systems with the secrecy constraints even though it may effective in non - eavesdropper environments . this solution takes the eavesdropper into account and selects the relay node based on @xmath57 , which provides the instantaneous channel knowledge for all the links . then , the sinr of link @xmath47 in ( [ gamma_ei ] ) can be rewritten as @xmath86 the optimal selection is given as : @xmath87 where @xmath88 the suboptimal selection implements the relay selection based on the knowledge set @xmath58 , which gives the average estimate of the eavesdropper links . therefore , it avoids the difficulty of getting instantaneous estimate of the channel feedbacks . similar to the os algorithm in ( [ os ] ) , the suboptimal selection can be written as @xmath89 where @xmath90}}{{\mathbb{e}\left [ { \gamma _ { s_j , e } } \right]+1 } } + \frac{{\mathbb{e}\left [ { \gamma _ { s_i , r , e } } \right]}}{{\mathbb{e}\left [ { \gamma _ { s_j , r , e } } \right ] + \mathbb{e}\left [ { \gamma _ { r , e } } \right]+1}}.\end{aligned}\ ] ] note that in comparison of the os in ( [ os ] ) , the only difference of the ss algorithm in ( [ ss ] ) is that it requires the average channel state information , @xmath58 , which would be more useful in practice . in this section , we present several node selection techniques based on the optimization problem given by ( [ selecriterion ] ) in the two - way systems . unlike @xcite , where the selection techniques only concern about the secrecy performance in the second phase of transmission , here , our work takes into account both the two phases in order to select a set of relay and jammers that can maximize the overall expectation of secrecy rate . the optimal selection with maximum sum instantaneous secrecy rate assumes a knowledge set @xmath59 and ensures a maximization of the sum of instantaneous secrecy rates of node @xmath0 and node @xmath1 given in ( [ cs ] ) , which gives credit to @xmath91 where @xmath92 and @xmath93 are given by ( [ gamma_j ] ) and ( [ gamma_ei ] ) , respectively . the approach in ( [ os - msisr_selection ] ) reflects the basic idea of using both cooperative relaying and cooperative jamming in order to promote the system s secrecy performance . specifically , the os - msisr scheme here tends to select a set of relay and jammers that maximizes @xmath92 , which means promoting the assistance to the sources . meanwhile this relay and jammer set tends to minimize @xmath93 , which is equivalent to enhance the interference to the eavesdropper . although the os - msisr scheme seems to be a straightforward application for cooperative relaying and cooperative jamming , the actual selection procedure usually involves trade - offs . for instance , according to ( [ gamma_j ] ) and ( [ gamma_j1sj ] ) , we should select the relay and jammer set that minimizes @xmath94 in order to make @xmath92 as high as possible . considering ( [ gamma_sjre ] ) , ( [ gamma_j1re ] ) and ( [ gamma_ei ] ) , however , the lower @xmath94 is , the higher @xmath93 is , which is undesirable . thus , we have to make a trade - off between raising @xmath92 and inhibiting @xmath93 in order to optimize the right part of ( [ os - msisr_selection ] ) . it is obvious that the os - msisr in ( [ os - msisr_selection ] ) is complicated , in this subsection we propose a reduced - complexity algorithm . it is common that the sum secrecy rate of two sources , i.e. @xmath95 , may be driven down to a low level by the user with the lower secrecy rate . as a result , for low complexity , the intermediate nodes , which maximize the minimum secrecy rate of two users , can be selected to achieve the near - optimal performance . in addition , in some scenarios , the considered secrecy performance does not only take into account the total secrecy rate of all the source nodes , but also the individual secrecy rate of each node . if one source node has low secrecy rate , the whole system is regarded as secrecy inefficient . furthermore , assuring each individual source node a high secrecy rate is another perspective of increasing the whole system s secrecy performance . the os - mmisr selection maximizes the worse instantaneous secrecy rate of the two source nodes with the assumption of knowledge set @xmath59 , and we can get @xmath96 where @xmath92 and @xmath93 are given by ( [ gamma_j ] ) and ( [ gamma_ei ] ) , respectively . the original idea of using jamming nodes is to introduce interference on the eavesdropper links . however , there are two side - effects of using jamming . firstly , the jamming node in the second phase @xmath8 poses undesired interference directly onto the destinations . secondly , it degrades the links between the relay node @xmath7 and the destinations . given the assumption that the destinations can not mitigate this artificial interference , continuous jamming in both phases is not always beneficial for the whole system . in some specific situation ( e.g. , @xmath8 is close to one destination ) , the continuous jamming may decrease secrecy rate seriously , and act as a bottleneck for the system . in order to overcome this problem , we introduce the idea of intelligent switching between the os - msisr and the os scheme in order to reduce the impact of `` negative interference '' . the threshold for the involvement of the jammer nodes is @xmath97 where @xmath98^ + .\end{aligned}\ ] ] thus , ( [ osw_criterion ] ) can be further written as @xmath99 where @xmath92 , @xmath93 , @xmath100 and @xmath101 are given by ( [ gamma_j ] ) and ( [ gamma_ei ] ) , ( [ gamma_os_i ] ) and ( [ gamma_os_ei ] ) , respectively . for each time slot , if ( [ osw_criterion_approx ] ) is met , the os - msisr scheme provides higher instantaneous secrecy rate than os does and is preferred . otherwise the os scheme is more efficient in promoting the system s secrecy performance , which should be employed . because of the uncertainty of the channel coefficient @xmath14 for each channel @xmath102 , the osw should outperform either the continuous jamming scheme or the non - jamming one . with the assumption of @xmath59 , we can get some optimal selection metrics . however , its practical interest and potential implements are only limited to some special ( e.g. military ) applications , where the instantaneous quality of the eavesdropper links can be measured by some specific protocols . in practice , only an average knowledge of these links @xmath64 would be available from long term eavesdropper supervision . the selection metrics is modified as @xmath103 where @xmath92 and @xmath104 are given by ( [ gamma_j ] ) and ( [ a_gamma_ei ] ) , respectively . from ( [ ss - msisr_selection ] ) , we can predict that for a scenario in which the intermediate nodes are sparsely distributed across the considered area , the ss - msisr scheme can provide similar relay and jammer selection with the os - msisr scheme . this is because a slightly difference between @xmath105 $ ] provided by @xmath64 and @xmath106 provided by @xmath59 would not be enough for the scheme to select another far - away intermediate node . thus , under this condition , the average eavesdropper channel knowledge set @xmath64 may contain sufficient channel information as well for a quasi - optimal selection . this scheme refers to the practical application of the above selection with maximum worse instantaneous secrecy rate in ( [ os - mmisr_selection ] ) . the basic idea of considering @xmath64 as the average behavior of eavesdropper links is the same as ss - msisr , but aimed at looking for the maximum worse instantaneous secrecy rate , which is written as @xmath107 where @xmath92 and @xmath104 are given by ( [ gamma_j ] ) and ( [ a_gamma_ei ] ) , respectively . given the fact that jamming is not always a positive process for the performance of the system , the suboptimal switching refers to the practical application of the intelligent switching between the ss - msisr and the ss schemes . the basic idea is the same as osw , but the switching criterion uses the available knowledge set @xmath64 . more specifically , the required condition for switching from ss - msisr to ss mode is @xmath108 where @xmath92 , @xmath104 , @xmath109 and @xmath110 are given by ( [ gamma_j ] ) and ( [ a_gamma_ei ] ) , ( [ gamma_i_ss ] ) and ( [ gamma_ei_ss ] ) , respectively . the previous selection techniques are proposed based on the assumption that the jamming signal is unknown at both the two destinations . this assumption avoids the initialization period in which the jamming sequence is defined , and thus , it reduces the risk of giving out the artificial interference to the eavesdropper . for comparison reasons , here we propose a `` control '' scheme , in which the jamming signal can be decoded at the destinations @xmath0 and @xmath1 , but not at the eavesdropper @xmath2 . in this case , the sinr of the link from @xmath28 ( for @xmath30 ) to @xmath2 remains the same as @xmath93 given by ( [ gamma_ei ] ) . the sinr of the link from @xmath28 to @xmath111 ( for @xmath112 ) is modified as follows : @xmath113 the oskj scheme is taken into consideration in the numerical results section as a reference . this , however , is not the `` ideal '' jamming scheme since the artificial interference from the jammers only degrades the eavesdropper links . as we have discovered and will discuss in section slowromancap5@ , in some particular scenarios , the oskj scheme is outperformed by the osw and ssw schemes presented above , for the jamming has changed the value of @xmath114 given in ( [ t ] ) . in this section , we firstly do some quantitative analysis on the asymptotic performance of both the proposed jamming and non - jamming schemes in high transmitted power range . then , we provide a qualitative discussion of the secrecy performance of different selection schemes in some typical scenarios based on the system model in section slowromancap2@. without loss of generality , we take the os scheme for example . with high transmitted power @xmath16 , we can get @xmath115 we can see that @xmath116 grows rapidly as @xmath16 increases , while @xmath117 converges to a value that depends only on the relative distances between the sources , the eavesdropper and the relay . therefore , the ergodic secrecy capacity @xmath118 $ ] also increases rapidly with the transmitted power @xmath16 . based on ( [ gamma_i_os_approx ] ) and ( [ gamma_ei_os_approx ] ) , the slope of the curve of @xmath119 $ ] versus @xmath16 ( measured by db ) can be approximately calculated as @xmath120}}{{\partial p_s } } \nonumber\\ = & ~ \frac{{\partial \mathbb{e}\left [ { \frac{1}{2}\log _ 2 \frac{{10^{p_s /10 } |h_{s1,r } |^2 \cdot 10^{p_s /10 } |h_{s2,r } |^2 } } { { \gamma_{e1}^{os } \cdot \gamma^{os}_{e2 } } } } \right]}}{{\partial p_s } } \nonumber\\ = & ~ \frac{{\partial \mathbb{e}\left [ { \frac{1}{2}\log _ 2 10^{2p_s /10 } } \right]}}{{\partial p_s } } + \frac{{\partial \mathbb{e}\left [ { \frac{1}{2}\log _ 2 \frac{{|h_{s1,r } |^2 |h_{s2,r } |^2 } } { { \gamma^{os}_{e1 } \cdot \gamma^{os}_{e2 } } } } \right]}}{{\partial p_s } } \nonumber\\ = & ~ \frac{{\partial \left ( { \frac{{p_s } } { { 10}}\log _ 2 10 } \right)}}{{\partial p_s } } \nonumber\\ = & ~ \frac{1}{{10}}\log _ 2 10 \nonumber\\ \approx&~ 0.3322 \end{aligned}\ ] ] for the other non - jamming schemes ( i.e. cs , ss ) , we note that they share the same asymptotic performance as the os scheme with a linear increment of slope about 0.3322 as the transmitted power @xmath16 increases . we use the same method as in the previous analysis for the non - jamming selections to analyze the asymptotic performance of the proposed jamming schemes . as the transmitted power @xmath16 increases to a relatively high value , it yields @xmath121 @xmath122 it is clear that both @xmath92 and @xmath93 are independent of @xmath16 , which means that for high @xmath16 , the ergodic secrecy rate @xmath123 $ ] stops increasing and converges to a fixed value . consider the asymptotic performance of the os scheme that grows linearly with the increment of @xmath16 as described by ( [ slope_approx ] ) , it is safe to predict that there will be a crossover point , @xmath124 , between the ergodic secrecy rate v.s . transmitted power curve with jamming and the one with non - jamming . in a power range below @xmath124 , the jamming scheme outperforms the non - jamming one , while above this point , the jamming scheme loses its advantage in providing higher ergodic secrecy capacity . we note that the analysis above can apply to any scheme with continuous jamming ( i.e. , os - msisr , os - mmisr , ss - msisr , and ss - mmisr ) , which indicates that they share the same asymptotic behavior as the @xmath16 increases . in another word , the proposed selection techniques ( except for osw and ssw ) behave better than the non - jamming schemes only within a certain transmitted power range . fortunately , in a practical case , @xmath16 is always limited in a relatively low range and will not increase infinitely . this is a common configuration in which the the eavesdropper @xmath2 has similar distance with two sources @xmath0 and @xmath1 and the intermediate nodes spread randomly within the considered area . with a relatively far distance in between , the interference link between @xmath5 and @xmath7 becomes weak . as predicted in the previous subsection , within a certain transmitted power range ( less than the crossover point @xmath124 ) , the selection approaches with continuous jamming are able to provide a higher ergodic secrecy rate than the non - jamming schemes . this gain proves the introduction of jamming in selection schemes as an effective technique . outside this range , the secrecy rates of the conventional non - jamming schemes continue to grow with a slope of 0.3322 as verified by ( [ slope_approx ] ) , whereas those of the continuous jamming schemes converge to a fixed value . inside this scope , the jamming schemes lose their efficiency in providing a better secrecy performance for the system . we note that in some particular scenarios , the system s integrated secrecy performance is not measured by the sum of the total secrecy rates , but by the minimum secrecy capacity of all the source nodes in the system . in this situation , os - mmisr and ss - mmisr can optimize the overall secrecy performance of the whole system . for the hybrid schemes , the osw and ssw schemes are able to provide better secrecy performance in the whole transmitted power scope , since it overcomes the bottleneck caused by negative interference on the relay - destination links . under the condition that all the intermediate nodes are located very close to each other , we note that the continuous jamming selections will lose its efficiency in meeting the secrecy constraints . specifically , we will discuss two extreme situations in which the intermediate nodes cluster is near to one of the destination nodes @xmath28 , and to the eavesdropper @xmath2 , respectively . there are two reasons that make the proposed jamming schemes inefficient . firstly , the nodes of the relay / jammer cluster gather too close to each other , such that the selected jammer in the first phase @xmath5 has too much negative impact on the selected relay @xmath7 , which further decreases the sinrs in the second phase . secondly , the jamming code from @xmath8 in the second phase also has an overly - strong interference on the destination to the one it stays close with . aside from the first reason presented above , in this configuration , the direct link between the relay @xmath7 and the eavesdropper @xmath2 gets too strong , which will seriously sabotage the secrecy performance of selection with continuous jamming . on the other hand , the hybrid protocols ( osw and ssw ) will still be the most effective schemes in this configuration , since the system s secrecy performance considered here is measured by the ergodic secrecy rate . this is the situation in which the eavesdropper @xmath2 can get the communicating information most easily , since the direct link between @xmath2 and any one of the source nodes is strong , which makes the introduction of jamming very necessary . the jamming schemes should be efficient within quite a large power range , and the hybrid schemes should still perform as the best selection techniques within the whole power scope . simulation environment with @xmath125 , @xmath126.,width=460 ] in this section , we will provide computer simulations in order to validate the analysis in the previous section . the simulation environment takes into account two sources @xmath0 and @xmath1 , one eavesdropper @xmath2 , and a intermediate node cluster consisting of @xmath125 nodes . all the nodes are located in a 2d square topology within a @xmath127 unit square . for simplicity , the source nodes and the relay transmit with the same power , i.e. @xmath128 . the relay and jammer nodes transmit with a relay - jammer power ratio @xmath129 . as assumed in section slowromancap2@ , the power of the awgn is @xmath130 . the path - loss exponent is set to @xmath126 . in this paper , the adopted performance metric is the ergodic secrecy rate . meanwhile some results are also provided in terms of secrecy outage probability @xmath131 $ ] , where @xmath132 $ ] denotes probability , and @xmath133 is the target secrecy rate . in the first simulation , we assume a scenario where @xmath0 , @xmath1 and @xmath2 are located at @xmath134 , @xmath135 and @xmath136 , respectively . the intermediate nodes spread randomly within the square space , as shown in fig . [ fig_2 ] . for different selection techniques.,width=460 ] fig . [ fig_3 ] shows the ergodic secrecy rate versus the transmitted power @xmath128 of different selection schemes . we can observe that selection algorithms with jamming outperform their non - jamming counterparts within a certain transmitted power range ( less than @xmath137 ) , where the ergodic secrecy rate of the os - msisr scheme is approximately higher than that of the os scheme by 1 bit per channel use ( bpcu ) . outside this range ( @xmath138 ) , the secrecy rate of os - msisr converges to a power - independent value which is approximately 4.1 bpcu , whereas the ergodic secrecy rate of os continues to grow with a slope of 0.3322 , as proved by ( [ slope_approx ] ) . this validates the secrecy performance analysis in section slowromancap4@. in addition , we can see that in this relay topology , the suboptimal schemes ( ss - msisr , ss - mmisr ) which are based on average channel knowledge perform almost the same as the optimal schemes ( os - msisr , os - mmisr ) , which implies that in this configuration where the intermediate nodes are sparsely distributed , an average channel knowledge may also provide enough information in order to get optimal relay selection . for different selection techniques with @xmath139 bpcu.,width=460 ] in fig . [ fig_3 ] , a comparison between the os - msisr and os - mmisr shows that the os - msisr scheme has slightly higher ergodic secrecy capacity by about 0.25 bpcu than os - mmisr does corresponding to transmitted power @xmath16 . the same comparison result can be observed from the ss - msisr and ss - mmisr schemes , which matches our previous analysis . furthermore , it can be seen that osw performs better than any other selection techniques with or without continuous jamming . at a low power range where @xmath140 , the osw scheme performs slightly better than os - msisr , but much better than os ( by about 1.2 bpcu ) , for the reason that in this range continuous jamming is almost always needed . after @xmath16 grows much higher than @xmath124 , osw outperforms both the other two schemes by a large scale . for the suboptimal case , we can see that ssw provides almost the same performance as the osw scheme in this relay topology , which validates the practical value of this hybrid scheme . an observation of the performance of oskj scheme shows that it outperforms all the other selection techniques , providing the highest ergodic secrecy rate when the transmitted power increases due to its ability of the destinations to decode the artificial interference in this oskj scheme . .,width=460 ] within this configuration , we also compare the performance of different selection techniques measured by secrecy outage probability , which is shown in fig . [ fig_4 ] . the target secrecy rate @xmath141 is set as 0.2 bpcu . it can be seen that selection schemes with jamming provies lower secrecy outage probability within a certain transmitted power range ( @xmath140 , @xmath142 ) . outside this range , the conventional selection without jamming achieves better secrecy outage probability . regarding the hybrid protocols , the osw scheme outperforms the non - switching selection techniques . in fig . [ fig_5 ] , it deals with a configuration where the intermediate nodes cluster , which also includes @xmath125 nodes , is located closely near to one of the two users ( e.g. , node @xmath0 , without loss of generality ) . we can see the ergodic secrecy rate of the proposed selection schemes in this topology differs greatly from that in the previous configuration . we observe that continuous jamming schemes ( i.e. os - msisr , os - mmisr , ss - msisr , and ss - mmisr ) are inefficient here , which converge to less than 0.5 bpcu , validating our discussion in section slowromancap4@. .,width=460 ] on the other hand , osw and ssw still outperform all the other selection techniques by a quite large scale ( more than 4 bpcu when @xmath16 is very high , as shown in fig . [ fig_5 ] . we also note that in this topology , the osw and ssw schemes perform even better than the oskj scheme , which seems to be an interesting result . further investigation reveals that the involvement of @xmath5 node in oskj causes a different value of @xmath114 with that of osw and ssw , which results in lower secrecy rates in oskj than in osw and ssw schemes . this indicates that the proposed osw / ssw schemes may perform even better than the `` ideal '' case where the destinations can mitigate the artificial interference . all of these validate the value of the selection techniques with intelligent switching in potential practical use . in fig . [ fig_6 ] , we set the intermediate nodes cluster closely to the eavesdropper @xmath2 . here the jamming schemes also perform worse than non - jamming ones in most of the transmitted power range . it also shows the range where continuous jamming schemes perform better than non - jamming schemes in this topology is slightly larger than that of the previous one , since there is no strong @xmath143 link here . regarding to the hybrid schemes , osw and ssw still perform as the best selection techniques in providing the highest secrecy rate . is close to @xmath0.,width=460 ] finally , we place the eavesdropper @xmath2 near to one of the two sources ( taken @xmath0 for example ) to examine the results . the location of eavesdropper @xmath2 is set to @xmath144 , the intermediate nodes are spread randomly across the considered rectangle area , as shown in the inset of fig . [ fig_7 ] . we get a similar simulation result with that of the first configuration , in which the eavesdropper @xmath2 has the same distance with @xmath0 and @xmath1 . the non - jamming schemes ( cs , os and ss ) here are less effective in promoting the secrecy performance . on the contrary , the selection techniques with continuous jamming ( os - msisr , os - mmisr , ss - msisr and ss - mmisr ) provide much higher secrecy capacity in a large transmitted power range ( @xmath145 ) . within this power range , the hybrid schemes ( osw and ssw ) perform slightly better than the continuous jamming techniques because jamming is almost always needed in this configuration . outside this regime , where the non - jamming scheme performs better , the difference between the intelligent switching and continuous jamming increases and the hybrid schemes still perform as the most efficient schemes . this paper has studied the joint relay and jammer selection in two - way cooperative networks with physical layer secrecy consideration . the proposed scheme achieves an opportunistic selection of one conventional relay node and one ( or two ) jamming nodes to increase security against eavesdroppers based on both instantaneous and average knowledge of the eavesdropper channels . the selected relay node helps enhance the information transmission between the two sources via an af strategy , while the jamming nodes are used to produce intentional interference at the eavesdropper nodes in different transmission phases . we found that the proposed jamming schemes ( i.e. os - msisr , os - mmisr , ss - msisr , and ss - mmisr ) are effective within a certain transmitted power range for scenarios with sparsely distributed intermediate nodes . meanwhile the non - jamming schemes ( cs , os , and ss ) are preferred in configurations where the intermediate nodes are confined close to each other . the osw scheme which switches intelligently between jamming and non - jamming modes is very efficient in providing the highest secrecy rate in almost the whole transmitted power regime in two - way cooperative networks , but it requires an instantaneous eavesdropper channel knowledge . on the other hand , the suboptimal switching scheme , ssw , which is based on the average knowledge of the eavesdropper channel and therefore much practical , provides a comparable secrecy performance with the osw scheme . e. d. silva , a. l. d. santos , l. c. p. albini , and m. lima , `` identity - based key management in mobile ad hoc networks : techniques and applications , '' _ ieee wireless communicaiton , _ vol . 15 , no . 5 , pp . 4652 , oct . 2008 . l. dong , z. han , a. p. petropulu , and h. v. poor , `` secure wireless communications via cooperation , '' in _ proceedings of 46th annual allerton conference on communication , control , and computing _ , uiuc , illinois , usa , sep . l. dong , z. han , a. p. petropulu , and h. v. poor , `` amplify - and - forward based cooperation for secure wireless communications , '' in _ proceedings of ieee international conference on acoustics , speech , and signal processing _ , taipei , taiwan , apr . e. tekin and a. yener , `` achievable rates for the general gaussian multiple accesswire - tap channel with collective secrecy , '' in _ proceedings of the 44th annual allerton conference on communication , control , and computing _ , monticello , il , sep . e. tekin and a. yener , `` the multiple access wire - tap channel : wireless secrecy and cooperative jamming , '' in _ proceedings of the information theory and applications workshop _ , san diego , ca , jan . 2007 . x. he and a. yener , `` on the equivocation region of relay channels with orthogonal components , '' in _ proceedings of the 41st annual asilomar conference on signals , systems , and computers _ , pacific grove , ca , nov . 2007 . x. he and a. yener , `` end - to - end secure multi - hop communication with untrusted relays is possible , '' in _ proceedings of the 42nd annual asilomar conference on signals , systems , and computers_,pacific grove , ca , nov . 2008 . y. oohama , `` capacity theorems for relay channels with confidential messages , '' in _ proceedings of ieee international symposium on information theory _ , nice , france , jun . 2007 . b. rankov and a. wittneben , `` achievable rate regions for the two - way relay channel , '' in _ proceedings of ieee international symposium on information theory _ , seattle , usa , jul . 2006 . x. he and a. yener , `` on the role of feedback in two - way secure communication , '' in _ proceedings of the 42nd annual asilomar conference on signals , systems , and computers _ , pacific grove , ca , nov . 2008 . alam , k. zarifi , s. affes , and a. ghrayeb , `` design and performance analysis of distributed relay selection techniques in wireless networks , '' in _ biennial symposium on communications ( qbsc ) _ , pp . 163167 , may 2010

broad emission lines ( bels ) are a defining property of quasar spectra . nearly 25 years ago baldwin ( 1977 ) discovered an anti - correlation between the equivalent width in iv civ@xmath6 , @xmath7(iv civ ) , and the continuum luminosity , @xmath8(1450 ) , measured at 1450 in the quasar rest - frame . this result was based on a sample of 20 quasars spanning @xmath9 orders of magnitude in continuum luminosity in the redshift range @xmath10 . it has become known as the `` baldwin effect '' ( hereafter beff ) . the beff was subsequently confirmed in iv civ@xmath6 , as well as other bels such as ly@xmath11 and ovi@xmath1 ( vron - cetty , vron , & tarenghi 1983 ; baldwin , wampler , & gaskell 1989 ; kinney , rivolo , & koratkar 1990 ; osmer , porter , & green 1994 ; zheng , kriss , & davidsen 1995 ; green , forster , & kuraszkiewicz 2001 ; see osmer & shields 1999 for a recent review ) . a major impetus for studying the beff was that it might be useful for calibrating active galactic nuclei ( agns ) luminosities , e.g. based on @xmath7(iv civ ) . the agns could then be used as cosmological standard candles . but in the following years studies of bigger quasar samples revealed a large scatter in the anti - correlation of the continuum luminosity vs. emission line strength ( baldwin , wampler , & gaskell 1989 ; zamorani et al.1992 ) . the beff is nonetheless important as a diagnostic of agn structure and , perhaps , metal abundances ( korista , baldwin , & ferland 1998 ) . the relation between the continuum luminosity and the relative emission line strengths and ratios can be used to study the evolution and physics of the quasar phenomenon ( baldwin 1999 ) . in particular , this correlation can be used to test model predictions for the dependence of the shape of the continuum spectral energy distribution as a function of luminosity ( binette et al.1989 ; netzer , laor , & gondhalekar 1992 ; zheng & malkan 1993 ; wandel 1999a , b ) . the most fundamental problem , however , is that the physical cause of the beff remains unknown . it was suggested by mushotzky & ferland ( 1984 ) that the observed relation can be explained by an anti - correlation of the ionization parameter , u , and the continuum luminosity , with @xmath12 , where q(h ) is the number of hydrogen ionizing photons emitted per second by the central continuum source , r is the distance between the continuum source and the emission line region , and @xmath13 is the hydrogen density in the line - emitting cloud . assuming an additional relation of decreasing covering factor with increasing continuum strength , the observed beff could be well described for iv civ@xmath6 . however , this model does not naturally explain the beff in the full range of measured lines . it predicts , for example , a lack of a beff for ly@xmath11 even though it is clearly detected ( e.g. , kinney , rivolo , & koratkar 1990 ; osmer , porter , & green 1994 ; laor et al.1995 ; green 1996 ) . another important clue to the physical cause of the beff is that the strength of the relationship ( i.e. , the slope of the @xmath14 @xmath15 anti - correlation ) seems to depend on the ionization energy of the emission - line species ( zheng et al.1995 ; espey & andreadis 1999 ) . in particular , the equivalent widths of high - ionization lines like ovi@xmath1 decrease more dramatically with @xmath15 than lines with moderate ionization energies like iv civ@xmath6 or low ionization energies like mgii@xmath16 and balmer emission lines . our results in the present paper confirm this claim , and improve upon the overall empirical characterization of the beff correlations . the beff might result from a more fundamental correlation between the continuum luminosity , @xmath15 , and the shape of the ionizing ( euv soft x - ray ) continuum ( binette et al.1989 ; zheng & malkan 1993 ; zheng et al.1997 ; korista et al.1998 ) . netzer ( 1985 , 1987 ) and netzer et al.(1992 ) suggested accretion disc models to explain the observed continuum and emission line correlations . recently , wandel ( 1999a , b ) added to this the growth in black hole mass by accretion in the accretion disc model . his analysis predicts that the continuum luminosity increases towards higher black hole mass and that the shape of the ionizing continuum becomes softer . hence , it is suggested that the beff is driven by a softening of the ionizing continuum towards higher luminosities . this model is attractive because the uv x - ray spectral softening has been well documented by observations ( tananbaum et al.1986 ; wilkes et al.1994 ; green et al.1995 ) , and because the spectral softening provides a natural explanation for steeper beff slopes in higher ionization lines . an additional possibility is that the beff is driven , at least in part , by a trend for higher metallicities in more luminous agns . korista et al.(1998 ) presented theoretical results showing the dependence of the emission line equivalent widths on both the continuum shape and the metallicity of the gas . their proposal is based on evidence for higher metal abundances in more luminous quasars ( hamann & ferland 1993 , 1999 ; dietrich et al . 1999 ; dietrich & wilhelm - erkens 2000 ) , and on the suggestion that more luminous quasars reside in more massive host galaxies , which will naturally be more metal rich ( cen & ostriker 1999 ; pettini 1999 ; kauffmann & haehnelt 2000 ; granato et al.2001 ; see hamann & ferland 1999 for a review ) . our combination of ground - based and satellite data also provides the opportunity to study evolutionary aspects of the line strengths in detail e.g. by measuring the same rest - frame uv lines across a wide range of redshifts . one serious complication affecting previous work is that luminosity and redshift are often correlated in quasars samples , because the quasars are selected from magnitude - limited surveys . most studies have supported the original claim by baldwin ( 1977 ) , that w@xmath17 scales inversely with luminosity and that there is no significant trend with redshift @xmath18 ( e.g. , kinney , rivolo , & koratkar 1990 ; osmer et al.1994 ; francis & koratkar 1995 ; wilkes et al.1999 ) . however , some recent work based on the large bright quasar survey ( lbqs ) claims that the relationship to redshift is even stronger than with luminosity ( forster et al.2001 ; green , forster , & kuraszkiewicz 2001 ) . the present paper is the first in a series in which we examine the emission - line properties in a large sample of @xmath19 type 1 agns . the sample includes seyfert1 galaxies and both radio - loud and radio - quiet quasars spanning an unprecedented wide range in both redshift ( @xmath20 ) and intrinsic luminosity ( @xmath21 orders of magnitude ) . the database and data processing will be described in detail in a future paper . a major concern regarding beff studies is that selection effects in agn samples , e.g. pertaining to emission - line strengths , might bias some measurements of the beff . our sample has the advantage of being drawn from a variety of surveys , with objects selected based on radio properties , grism spectroscopy , or broad - band color criteria . hence , no specific selection criteria were applied for the data . another key advantage is that our sample includes a wide range of luminosities at various redshifts . we can therefore address the separate redshift and luminosity dependences in the emission line data . there are several interesting problems we wish to address . the most basic issue is to quantify the nature of empirical correlations with @xmath15 and/or @xmath18 . in an upcoming study we will further examine correlations with other agn properties such as radio - loudness ( baldwin 1977 ; sargent , steidel , & boksenberg 1989 ; steidel & sargent 1991 ; francis & koratkar 1995 ) . we will also report on trends with @xmath15 or @xmath18 among various emission - line metallicity indicators . here we present an initial analysis on the nature of the beff . our approach is to construct composite spectra for specific @xmath15 and @xmath18 intervals , thus providing high signal - to - noise spectra and allowing us to study trends in both weak and strong emission lines . the agn sample is described briefly in 2 and the calculation and analysis of the composite spectra in 3 and 4 . the main results appear in several figures in 5 . we compare our results with prior studies and discuss them in the context of suggested models in 6 . throughout this paper we use the cosmological parameters h@xmath22 kms@xmath23mpc@xmath23 , @xmath24 , and @xmath25 ( carroll , press , & turner 1992 ) . introducing @xmath26 ( netterfield et al.2002 ) instead of @xmath27 , the luminosities at the highest redshifts would be @xmath28% smaller , while at low redshifts they would reach a maximum of @xmath29% larger at redshift @xmath30 . we have compiled a large sample of rest - frame visible and ultraviolet spectra for @xmath31 type 1 agns . a majority of the spectra were obtained by several groups for different studies over the last 20 years using ground - based instruments as well as _ international ultraviolet explorer ( iue ) _ and _ hubble space telescope ( hst ) _ ( bahcall et al.1993 ; baldwin et al.1989 ; chaffee et al.1991 ; corbin & boroson 1996 ; kinney et al.1991 ; lanzetta , turnshek , & sandoval 1993 ; laor et al.1995 ; sargent et al.1988,1989 ; schneider , schmidt , & gunn 1991a , b ; steidel 1990 ; steidel & sargent 1991 ; steidel unpublished ; storrie - lombardi et al.1996 ; weymann et al.1991,1998 ; wills et al.1995 ; zheng et al.1997 ) . in total , the sample contains 351 quasar spectra measured with _ hst_. furthermore , we observed a large number of the quasars in our sample at redshifts @xmath32 ( constantin et al.2002 ; dietrich et al.1999 ; dietrich & wilhelm - erkens 2000 ; dietrich et al.2002a , b ) . for this study we exclude broad - absorption line quasars ( balqsos ) , although there are indications that their emission line properties do not differ from non - bal quasars ( weymann et al.1991 ) . we also excluded several quasar spectra with extremely poor signal - to - noise ratios . the sample we investigate consists of @xmath19 type 1 agns . because this agn sample was compiled from many independent projects , it encompasses a wide range of selection criteria . most of the quasars especially at high redshifts were discovered by color selection techniques ( e.g. , storrie - lombardi et al.1996 ) or found by objective prism surveys ( e.g. , chaffee et al.1991 ; sargent et al.1988,1989 ; schneider at al.1991a , b ) . we used the radio flux densities given in vron - cetty & vron ( 2001 ) to determine the radio loudness for the quasars . following kellermann et al.(1989 ) we calculated the parameter @xmath33 with @xmath34 as the radio flux density at @xmath35ghz and @xmath36 as the b - band flux , each in the rest frame . our classifications of radio - loud quasars are consistent with classifications available in the literature ( wills et al.1995 ; bischof & becker 1997 ; wilkes et al.1999 ; stern et al.2000 ) . in total our sample contains @xmath37 radio - loud quasars and @xmath38 radio - quiet agns . kinney et al.(1990 ) and pogge & peterson ( 1992 ) pointed out that a significant fraction of the scatter in the beff in seyfert1 galaxies is caused by variability . they found that observations obtained at multiple epochs for individual sources display a beff - like correlation ( the _ intrinsic _ beff ) , but with a significantly steeper slope than the beff observed for an ensemble of agns . the physical mechanisms which drive each effect might be different , i.e. , the ensemble beff might be due primarily to variations in black hole mass while the intrinsic beff might be caused by changes of the accretion rate @xmath39 ( green 1999 ) . for @xmath40 of the @xmath41 low luminosity agns ( log@xmath42 , for @xmath43 in ergs@xmath23 ) , the spectra are means from intense monitoring campaigns ( crenshaw et al.1996 ; collier et al.1998,2001 ; dietrich et al.1993 ; goad et al.1999 ; kaspi et al.1996 ; korista et al.1995 ; peterson et al.1994 ; reichert et al.1994 ; santos - lle et al.2001 ; stirpe et al.1994 ; wanders et al.1997 ) . at the high luminosity end of the luminosity distribution , it has been shown that quasars vary on longer timescales and with smaller amplitudes ( e.g. , kaspi et al.2000 ) . for these reasons , the scatter caused by intrinsic variability is not expected to strongly influence our analysis of the present large quasar sample , particularly when comparisons of composite spectra are employed . we transformed each quasar spectrum to the rest - frame using a redshift measured from iv civ@xmath6 . to determine the redshift we fit a gaussian profile to the upper part of the iv civ@xmath6 emission line having @xmath44% of the peak intensity . we corrected each spectrum for galactic extinction ( seaton 1979 ) , based on the neutral hydrogen column density @xmath45 expressed in units of @xmath46@xmath47 , assuming @xmath48 ( dickey & lockman 1990 ) . the redshift distribution of the quasar sample is presented in figure 1 as a function of continuum luminosity , @xmath49 . the quasars cover a redshift range of @xmath0 and nearly @xmath50 orders of magnitude in luminosity . for most of the redshift range a reasonable coverage in luminosity is achieved , spanning at least @xmath51 orders of magnitude for @xmath52 . the radio - loud and radio - quiet quasars are displayed with different symbols in figure 1 . their distributions across luminosity and redshift are very similar within out sample , although the fraction of radio - loud quasars increases for redshifts @xmath53 . a common approach to study the beff is to measure the emission line properties of individual quasars . however , the results are often limited especially for the weaker emission lines by the low signal - to - noise ratio of individual spectra . there can also be significant problems introduced by including upper limits on weak lines in individual spectra . to avoid those problems , we computed composite spectra in different intervals of redshift and luminosity . the analysis of these spectra has the advantage of more clearly representing the average properties of the sample used to calculate each composite . figure 1 indicates the ranges in luminosity and redshift used to compute the composites . except for the low and high luminosity ends of our sample , each composite spectrum is based on more than @xmath54 individual quasar spectra . the scatter introduced by different individual spectra is taken into account by calculating a root - mean - square ( standard deviation ) spectrum for each composite , and using that to estimate the uncertainties in the measured equivalent widths ( 4 ) . before combining the individual quasar spectra , we normalized the flux densities to unity in a @xmath54 wide continuum range centered at @xmath55 by multiplying each spectrum with a suitable factor . we used these normalized quasar spectra to calculate the mean and rms spectrum for each composite , and the number of spectra contributing in each wavelength bin . to avoid the influence of strong narrow absorption lines , we developed a search routine to detect strong narrow absorption features and exclude them from the calculation of the composite spectrum . to search for strong absorption lines we applied the following method ( a more detailed description will be given in a subsequent paper on the properties of the entire quasar sample , dietrich et al.2002c ) . first , a preliminary mean spectrum is calculated for the sample of spectra under study . we then divided each individual spectrum by this preliminary mean to derive a ratio spectrum . each ratio spectrum is smoothed with a running boxcar function , which provides in addition to the mean also an rms value for each wavelength box . the resulting rms spectrum indicates the scatter in flux for each wavelength element in the immediate spectral region . a narrow absorption line will cause a sharp increase in the rms spectrum . the contaminated spectral regions of individual spectra , identified by these rms spikes , are excluded from the calculation of the final composite spectrum , and no interpolation has been applied . multiple tests with different agns and different sample sizes show that this technique sucessfully removes strong narrow absorption lines from our composite spectra ( see below ) without affecting the emission lines . significant effects of narrow absorption lines remain only in the ly@xmath56 forest ( @xmath57 ) for composites that include quasars with redshift @xmath52 . the resulting composite quasar spectrum for the entire sample is shown in figure 2 . a single power law fit to the continuum at @xmath58 is also displayed ; it has a slope of @xmath59 ( @xmath60 ) . this spectral index is consistent with the slope reported for other composite spectra ( brotherton et al.2001 ; vandenberk et al.2001 ; telfer et al.2002 ) . the bottom panel of figure 2 shows the number of quasars contributing to each individual wavelength element . for every wavelength @xmath61 this composite quasar spectrum is based on at least @xmath62 quasars . to measure the emission line fluxes we analyzed each composite spectrum with spectral fitting software provided by midas . we employed a powerlaw continuum , an ultraviolet and optical fe - emission template , a balmer continuum emission template , and multiple gaussian profiles for the broad emission lines in multi - component fits using @xmath63 minimization . for the ultraviolet wavelength range we used the fe emission template which vestergaard & wilkes ( 2001 ) carefully derived from hst spectra of izw1 ( kindly provided by m.vestergaard ) . the primary aim was to measure the broad and prominent feii - emission near mgii@xmath16 . however , we also used the uv template to correct for feii and feiii emission blended with iii ciii]@xmath64 and other lines across the wavelength range @xmath65 . for the optical wavelength region around h@xmath66 and h@xmath67 , we used the empirical feii emission template obtained from izw1 ( kindly provided by t.boroson ; boroson & green 1992 ) . we adjusted the line widths in these fe emission templates to match the measured widths of iv civ@xmath6 , mgii@xmath16 , or h@xmath3 . we calculated several bound - free balmer continuum emission spectra for @xmath68k , @xmath69@xmath70 , and @xmath71 following the formalism suggested by grandi ( 1982 ) . these balmer continuum emission templates were supplemented for @xmath72 with high order balmer emission lines with @xmath73 , i.e. h@xmath74 and higher ( storey & hummer 1995 ) . the individual high - order balmer lines were represented by gaussian profiles and coadded . the multi - component fit of a balmer continuum , two fe - emission templates , and a power law continuum yields independent measurements of the ultraviolet and optical fe - emission strengths . furthermore , the spectral index , @xmath56 , of the power law continuum does not include contributions from balmer continuum or from broad fe emission . the uncertainties associated with the fit of balmer continuum emission and fe emission templates were estimated from the @xmath63 distribution for each fit ( cf . , dietrich et al.2002a ) . the emission line fluxes were measured after subtracting the power law continuum , balmer continuum , and fe emission spectra . the line fluxes were determined by multi - component gaussian fits . in general , we fit the emission lines with a `` broad '' and a `` narrow '' gaussian component . for iv civ@xmath6 , however , it was necessary to introduce in addition a very broad component ( which turned out to have typically fwhm @xmath75 kms@xmath23 ) . to determine the fluxes of ly@xmath11 and nv@xmath5 , which are severely blended , we used the components obtained from the iv civ@xmath6 profile as templates ( preserving their widths and redshifts ) . please see figure 2 in dietrich & hamann ( 2002d ) for examples of fitted profiles . note that most of the ly@xmath11 forest absorption on the blue side of the ly@xmath11 profile is corrected by the methode we applied to calculated the composite spectra . the ovi@xmath1 emission line is subject to severe corruption by ly@xmath11 forest absorption . however , in composite spectra it can be assumed that the same fraction of the ovi@xmath1 emission line flux and the adjacent continuum flux is absorbed by the ly@xmath11 forest . hence , the equivalent width is not expected to be significantly altered by ly@xmath11 forest absorption . the ovi@xmath1 emission line is also contaminated by ly@xmath66 emission in the blue wing , whose relative strength shows a wide range , ly@xmath66/ovi@xmath76 ( laor et al.1995 ) . our fits to the ovi@xmath1 emission line profiles were achieved with a broad and narrow gaussian profile , and no additional component due to ly@xmath66 emission was necessary . the iii ciii]@xmath64 emission line complex was reconstructed with four gaussian profiles , corresponding to aliii@xmath77 , siiii]@xmath78 , and broad and narrow gaussian profiles for iii ciii]@xmath64 . the correction for fe emission allows a better measurement of mgii@xmath16 , heii@xmath79 , niii]@xmath80 , and the iii ciii ] emission line complex ( aliii@xmath77 , siiii]@xmath78 , iii ciii]@xmath64 ) . the measurement of the optical feii emission with an appropriate scaled template made it possible to measure the heii@xmath81 emission line which is otherwise severely blended with feii(37,38 ) multiplets . the cumulative errors in the emission line flux measurements and @xmath14 were calculated as the square root of the sum of i ) the uncertainties in the @xmath63 fitting and ii ) the rms scatter that is introduced by both the noise and distribution of the individual spectra contributing to that particular composite spectrum , squared . to search for a dependence of emission line properties , particularly @xmath7 , on cosmic time ( redshift ) , we selected a narrow luminosity range . the location of this luminosity range was also selected to achieve a comparable number of individual quasars contributing to each redshift bin in the @xmath18 @xmath49 plane . to cover a redshift range as large as possible for an almost constant luminosity , we selected a luminosity range of @xmath82 ( i.e. , @xmath83 ) . this range contains @xmath84 quasars with @xmath85 and an average luminosity of log@xmath86 . we computed composite spectra for redshift bins with @xmath87 centered on redshift @xmath88 , respectively ( fig.1 ) . figures 3a and 3b show the composite spectra for each redshift bin . these composite spectra are normalized by the power law continuum which we derived from the multi - component fits in 4 . as a first impression the relative line strengths in the normalized composite quasar spectra stay nearly constant with redshift @xmath18 . however , some emission lines show a marginal trend for increasing equivalent width at the highest redshifts . to quantify this result , we measured the equivalent width of most emission lines in the ultraviolet wavelength range ( @xmath89 ) and of the broad ultraviolet feii - emission feature ( @xmath90 ) . the results for @xmath14 as a function of redshift are displayed in figures 4a and 4b . in general , the equivalent widths of the lines remain constant within @xmath91% . the stronger emission lines , however , like ly@xmath11 , iv civ@xmath6 , and iii ciii]@xmath64 , show marginal trends for a relation of @xmath14 and redshift . @xmath14 stays nearly constant within less than @xmath92% for redshifts less than @xmath93 , but then for @xmath94 there are marginal indications for a slight increase of @xmath14 at constant luminosity . close inspection of figures 4a and 4b indicates that in particular the equivalent widths measured for the composites with @xmath95 are larger compared to lower redshifts . most of the quasar spectra in our sample with @xmath95 were obtained by constantin et al.(2002 ) . they found evidence that in spite of strong contamination by ly@xmath11-forest absorption , ly@xmath11 is stronger for quasars with @xmath96 than in their low-@xmath18 counterparts . next we divided the quasar sample into eleven subsets with respect to the luminosity @xmath97 . except for the lowest luminosity bin , each composite spectrum covers a range of @xmath98 in luminosity starting at log@xmath99 ergs@xmath23 ( fig.1 ) . the two lowest luminosity composite spectra consist of less than 10 objects . figures 5a and 5b show the composite spectra ordered by luminosity for the ultraviolet wavelength range ( @xmath100 ) , while figure 6 shows the longer wavelengths ( @xmath101 ) . each composite spectrum is normalized by the power law continuum fit we derived from the multi - component analysis ( 3 ) . these normalized spectra show very obviously a strong relation between the equivalent width , @xmath14 , and the continuum luminosity represented by @xmath49 , i.e. the baldwin effect , for most of the emission lines . figures 7a and 7b show the measured values of @xmath14 for each luminosity interval . the beff can be seen easily in strong lines such as iv civ@xmath6 , ovi@xmath1 , ly@xmath11@xmath102 , iii ciii]@xmath64 , and mgii@xmath16 ( figures 5a and 6 ) . taking advantage of the high signal - to - noise ratio of the composite spectra , we also detected this relation for weaker emission lines such as heii@xmath79 . furthermore , emission lines of the same ion but at different wavelengths ( oiii]@xmath103 , [ oiii]@xmath2 and heii@xmath79 , heii@xmath104 ) possess nearly identical slopes ( tab.1 ) . for the first time , we report on the beff for niv]@xmath105 , oiii]@xmath103 , and niii]@xmath80 . in spite of the large scatter , the broad fe emission features in the ultraviolet show some indications of a beff , while within the errors , the optical feii emission lacks a beff . in addition to the broad emission lines , we detect the baldwin effect also for the prominent forbidden emission line [ oiii]@xmath2 ( figure 6 ) , which is regarded as a typical narrow - line region ( nlr ) emission line . in contrast to the strong beff in most of the emission lines visible in figures 5a , 5b , and 6 , the balmer lines h@xmath67 and h@xmath66 show only weak indications of a baldwin effect , which might be restricted to the line center of the line profiles . close inspection of the nv@xmath5 emission line ( figures 5a and 8) also indicates that this line does not exhibit a beff like other high ionization lines . lines having the same equivalent widths in each individual spectrum will cancel out in difference spectra ( provided that their profiles are also the same ) , while lines whose equivalent widths are a function of luminosity will still be present in difference spectra . note that nv@xmath5 has on average cancelled out in the difference spectra as shown in figure 8 . hence , the difference spectra near nv and the careful deblending of the ly@xmath11-nv@xmath5 emission line complex both reveal that nv@xmath5 remains nearly constant in @xmath14 over @xmath21 orders of magnitude in continuum luminosity . zheng , fang , & binette ( 1992 ) , espey et al.(1993 ) , and espey & andreadis ( 1999 ) have suggested that there is a correlation between the ionization energy , @xmath106 , of a specific emission line and the strength ( slope ) of the beff . in contrast to espey & andreadis ( 1999 ) , we use the ionization energy necessary to create the ion for collisionally excited lines like iv civ@xmath107 and mgii@xmath16 , and the ionization energy needed to ionize the ion for recombination lines like ly@xmath11 and heii . the beff slope of the lines under study shows a relatively close linear correlation with @xmath106 ( figures 7a , b ) . we determined the slope @xmath3 , where log@xmath108 , of the beff from a linear least square fit to log@xmath14 vs. log@xmath109 . the fits were calculated taking into account the full uncertainties in log@xmath14 . first , we calculated the slopes across the entire luminosity range , shown by the dotted line in figures 7a and 7b . however , several of the emission lines indicate that the slope of the beff is flatter at lower luminosities ( @xmath110 ) . therefore , we calculated additional linear fits for the luminosity range log@xmath111 ( dashed - dotted lines ) . it is interesting to note that the lines showing a change in the beff slope have high ionization energies with @xmath112ev . the slopes of the log@xmath14 vs. log@xmath109 relations are listed in table 1 and plotted against @xmath106 in figure 9 for the entire range ( filled symbols ) and for log@xmath111 ( open symbols ) . notice , that the slope of the beff becomes steeper for increasing @xmath113 . however , as has already been shown by earlier studies , nv@xmath5 deviates significantly from this trend ( espey & andreadis 1999 ) . nv@xmath5 stays constant or shows a slightly decreasing w@xmath17 towards higher luminosity . the slope @xmath3 for ovi@xmath1 is of particular interest for testing the hypothesis of a steeper beff towards higher ionization energies since @xmath114 is more than twice @xmath115 and nearly @xmath116 times @xmath117 . assuming a constant ratio of ly@xmath66/ovi@xmath118 does not significantly change the slope of the beff for ovi@xmath1 however , @xmath3(ovi@xmath1 ) would become steeper by nearly @xmath119%if ly@xmath66 follows the same beff as ly@xmath11 and ly@xmath66/ly@xmath11=@xmath120 ( laor et al.1995 ) . to quantify the relationship of the beff slopes @xmath3 to @xmath106 , we calculated linear least square fits to the points in figure 9 excluding the outlying data point for nv@xmath5 . the resulting slope , @xmath121 , is the same with ( @xmath122 ) or without ( @xmath123 ) the ovi@xmath1 line included . this result indicates that ovi@xmath1 does not dominate the @xmath3 versus @xmath106 relation . in general , the slopes of the baldwin effect derived here are consistent with prior investigations . however , the large range in both luminosity and redshift that distinguishes our quasar sample reveals correlations which were missed in some other studies . in particular , we can more clearly separate the trends with @xmath124 and @xmath18 , and our use of composite spectra across a wide luminosity range more accurately probes the beff slopes , e.g. , for weaker lines . in the following we provide some more detailed comparisons with earlier work . the slopes we derived for ly@xmath11@xmath102 , iv civ@xmath6 , iii ciii]@xmath64 , and mgii@xmath16 , are in good agreement with prior studies ( kinney et al.1990 ; osmer et al.1994 ; laor et al.1995 ; green 1996 ; turnshek 1997 ; zamorani et al.1992 ; zheng et al.1995 ; wang et al.1998 ; espey & andreadis 1999 ) . the steeper slopes for iv civ@xmath6 in osmer et al.(1994 ) , laor et al.(1995 ) , and green ( 1996 ) , @xmath125 compared to @xmath126 here , might be caused by the luminosity range of their quasar samples , @xmath127 ergs@xmath23 . using only the @xmath7(iv civ ) measurements for @xmath127 ergs@xmath23 in our sample , we calculate a slope of the baldwin effect of @xmath128 for iv civ@xmath6 ( tab.1 ) . a lack of the baldwin effect for iv civ@xmath6 and iii ciii]@xmath64 is reported by wilkes et al.(1999 ) . they conclude that this is caused by the narrow - line - seyfert1 galaxies ( nlsy1s ) in their sample which decrease the strength of the baldwin effect ; a significant beff is found for both iv civ and iii ciii ] if the nlsy1s galaxies in their sample are omitted . we note that these nlsy1s are also low luminosity agn . as can be seen in figures 5a and 5b , the baldwin relation is significantly flatter for low luminosities than at higher luminosities . any agn sample favoring a narrow luminosity range can dilute or bias the derived beff . similarly , the steeper beff slopes reported by laor et al.(1995 ) , for oi@xmath129 , cii@xmath130 , siiv@xmath131 , and heii@xmath79 can be attributed to their smaller luminosity range ( @xmath132 ergs@xmath23 ) . the slope for the baldwin effect for the high ionization emission line ovi@xmath1 ( @xmath133ev ) is also consistent with prior investigations . the reported slope @xmath66(ovi@xmath1 ) varies between @xmath134 ( zheng et al.1995 ; turnshek 1997 ) and @xmath135 ( laor et al.1995 ) , with most recent studies favoring a less steep anti - correlation ( @xmath136 ; espey & andreadis 1999 ) . wilkes et al.(1999 ) found only a marginal anti - correlation of @xmath137ovi@xmath138 and continuum luminosity . however , they noted that this result is based on just ten objects . green ( 1996 ) and green et al.(2001 ) did not detect a significant baldwin effect for ovi@xmath1 in their analysis of the large bright qso survey ( lbqs ) data set , taking all quasar spectra into account , i.e. including upper limit measurements as well . however , using only those spectra with detected ovi@xmath1 the baldwin effect is clearly present . a possible reason for the missing baldwin effect for ovi@xmath1 in their study might be the very narrow luminosity range ( @xmath139 ) covered by lbqs quasars at these wavelengths . a comparison of our figure 7a with figure 2 in green et al.(2001 ) shows that we measure the same range of ovi@xmath1 equivalent widths for this luminosity , @xmath140(ovi@xmath141 . hence , we conclude that the non - detection of the baldwin effect for this high - ionization line in green et al.(2001 ) is predominatly caused by their very small luminosity range . in contrast to wills et al.(1999 ) , we detect a significant baldwin effect for the prominent nlr emission line [ oiii]@xmath2 ( fig.7b ) . the slope @xmath3 of the anti - correlation is nearly identical to the one which we derived for oiii]@xmath103 ( tab.1 , fig.9 ) . again , the lack of a beff for [ oiii]@xmath2 in wills et al.(1999 ) might be caused by the covered luminosity range of less than two orders of magnitude . in agreement with prior investigations , we found no significant baldwin effect for nv@xmath5 ( osmer et al.1994 ; laor et al.1995 ; turnshek 1997 ; espey & andreadis 1999 ) . instead of a decreasing equivalent width for increasing continuum strength , @xmath142 stays nearly constant , even though it has a high value of @xmath143ev . although much observational and theoretical effort has been spent to decipher the process which causes the baldwin effect , the mechanism is still unclear . several models have been suggested which emphasize different aspects , i.e. , contributions from optically thin clouds and luminosity - dependent ionization parameter , covering factor , ionizing continuum shape , and chemical composition of the gas . the dependence of the strength of the baldwin effect on the ionization energy indicates that a decreasing covering factor towards higher luminosities is not sufficient to explain the observed beff for each emission line . to explain the observed beff for iv civ@xmath6 , mushotzky & ferland ( 1984 ) presented model calculations suggesting that the beff is caused by a decrease of the ionization parameter , @xmath144 , as well as the covering factor towards higher luminosities . although this model predicts a weaker baldwin effect for lower luminosity agn , it also predicts that ly@xmath11 will lack a beff or show only a weak beff . a luminosity dependence of the covering factor might be indicated independently by measurements of x - ray absorption ( e.g. , lawrence & elvis 1982 ) . a variant of this picture was outlined by shields , ferland , & peterson ( 1995 ) , who suggested that the luminosity dependence could be caused by a luminosity - dependent covering factor of optically thin clouds which would emit preferentially high - ionization lines . the presence of this component in the blr is suggested from several aspects of variability in seyfert galaxies , as well as other arguments . while luminosity dependence of coverage by high - ionization clouds in the blr is apparently still a viable scenario for understanding the beff , this interpretation suffers from the lack of a strong physical basis for predicting this behavior ; explanations are ad hoc or phenomenological at best . several models have been advanced that focused primarily on the spectral energy distribution ( sed ) of the ionizing continuum , and its consequences . this approach was motivated by increasing evidence that the sed of the continuum becomes softer for more luminous ( e.g. , sargent & malkan 1982 ; binette et al.1989 ; schulz 1992 ; netzer et al.1992 ; zheng et al.1992 ; zheng & malkan 1993 ; green 1996 , 1998 ; wang et al.1998 ) . empirical models have been suggested which combine a powerlaw continuum and a thermal uv bump . detailed investigations of x - ray and ultraviolet continuum observations , i.e. , the overall ionizing continuum shape , yield indications that the sed becomes softer for increasing luminosity . in particular , it has been shown that @xmath145 , the two - point power index which connects the ultraviolet at 2500 to the x - ray continuum at 2kev , increases significantly with luminosity ( tananbaum et al.1986 ; wilkes et al.1994 ; green et al.1995 ) . this softening can be understood by a shift of the thermal uv - bump towards longer wavelengths in more luminous quasars , i.e. , @xmath146 increases more than @xmath147 for increasing luminosity resulting in a greater @xmath145 ( cf . , binette et al.1989 ; zheng & malkan 1993 ) . early models of geometrically thin and optically thick accretion discs focused on the consequences of the disc inclination and the location of gas relative to the disc ( netzer 1985 , 1987 ) . although these models can not explain the beff for several orders of magnitude in luminosity they provide an explanation for scatter in the @xmath15 vs. @xmath14 relation at a given luminosity . the spectral energy distribution ( sed ) of the accretion disc continuum depends on the luminosity which is in turn related to the mass of the black hole and the accretion rate . these dependencies were suggested as the primary mechanism to cause the baldwin effect . recently , wandel ( 1999a , b ) presented an evolutionary scenario which connects the growth of the black hole mass , the accretion rate and the continuum luminosity . as the mass of the black hole increases and consequently also the luminosity , the peak temperature of the uv bump decreases . this results in a shift of the peak of the thermal uv bump to longer wavelengths , i.e. , the continuum becomes softer . netzer et al.(1992 ) calculated the log@xmath148 vs. log@xmath149 relation for a wide range of black hole masses , accretion rates , and inclination angles in geometrically thin accretion disc models . the spread of the calculated @xmath148 at a given luminosity is mainly caused by the different inclination angles of the accretion disc . assuming that our composite spectra represent an average over a variety of disc inclinations , the beff which we measured for ly@xmath11 is consistent with their model predictions . the observed ionization dependence of the baldwin effect provides strong evidence for a luminosity dependent sed of the ionizing continuum and thus favors accretion disc models . however , the detailed models discussed by wandel ( 1999b ) show that the seds of accretion discs depend in a complex way on the mass of the black hole and the accretion scenario . in particular , the cutoff energy of the thermal uv - bump is expected to depend on the black hole mass . these models also predict that luminous quasars with massive black holes should show low cutoff energies , while seyfert1 galaxies with intermediate massive black holes should display high cutoff energies . the low luminosity nlsy1 galaxies with low mass black holes should also exhibit high cutoff energies . different accretion scenarios would introduce additional scatter to this relation . in summary , accretion disc models provide a reasonable explanation of the baldwin effect , in particular for the ionization energy @xmath106 dependent strength . although these models indicate that the luminosity - dependent sed is a major mechanism which drives the beff , the large possible parameter space allows for additional effects . evidence for effects in addition to variations of the spectral energy distribution is given by the failure to explain the lack of a baldwin effect for another high ionization emission line , nv@xmath5 . baldwin et al.(1995 ) and korista et al.(1997 , 1998 ) introduced the `` locally optimally - emitting cloud '' ( loc ) model , which assumes that the blr cloud ensemble covers a wide range of internal densities and occurs over a wide range in distance from the central continuum source . under these conditions the emitted spectrum is controlled by powerful selection effects and the typical observed quasar spectrum can be naturally produced ( korista et al.1997 , 1998 ) . korista et al.(1998 ) computed for a wide range of different continuum slopes ( @xmath150 ) , densities ( @xmath151@xmath70 ) , and metallicities ( 0.2 to 10 times solar ) the resulting emission line spectra . in recent years there has been growing evidence that the chemical composition of quasar gas can reach several times solar metallicity with a trend of higher metallicities in more luminous quasars ( hamann & ferland 1993 , 1999 ; ferland et al.1996 , dietrich et al.1999 ; dietrich & wilhelm - erkens 2000 ) . the emission lines studied by korista et al.(1998 ) follow the general trend for @xmath14 diminishing with softer continuum shapes . assuming the softer continua are related to higher luminosities , this yields the baldwin effect . however , in contrast to those emission lines , nv@xmath5 lacks a baldwin effect even though its high ionization energy @xmath143ev suggests that the decrease of @xmath14(nv ) should be strong . korista et al.(1998 ) introduced metallicity as an additional parameter to provide a possible explanation for the lack of a baldwin effect in nv@xmath5 . they found in their calculations that the equivalent width of ly@xmath11 , iv civ@xmath6 , or ovi@xmath1 show only a weak dependence on the gas metallicity , but nv@xmath5 strongly depends on it . in particular , they assume that nitrogen scales like a secondary element , so that its abundance relative to the other metals increases linearly with the overall metallicity ( e.g. , n / o @xmath152 o / h , see hamann et al.2002 ) . korista et al.(1998 ) assume that the metallicity increases with increasing agn luminosity ( hamann & ferland 1999 ) . with this abundance behavior , korista et al.(1998 ) showed that the equivalent width of nv@xmath5 is not expected to display a beff . therefore , the decreasing equivalent width that should occur in nv@xmath5 as part of the normal beff is approximately compensated by the increasing relative abundance of nitrogen . the net result is that nv@xmath5 shows essentially no beff . overall , the beff slopes that we derived for the emission lines in our quasar sample are in good agreement with the predictions by korista et al . ( table 1 ) . the observed slopes of iv civ@xmath153 and heii@xmath79 are smaller than expected , while the slope in niii]@xmath80 is steeper . however , if only measurements with @xmath154 ergs@xmath23 are taken into account , the observations yield a comparable steep anti - correlation as predicted by korista et al.(1998 ) . we will discuss these comparisons further in a forthcoming paper on the trends in metallicity in this dataset ( see also 6.4 below ) . a common prediction of models that involve a softer continuum sed towards higher luminosities is a steeper slope of the baldwin effect for lines with increasing ionization energy @xmath106 ( zheng et al.1992 ; zheng & malkan 1993 ; korista et al.1998 ; wandel 1999a ) . figures 7a , b and 9 provide clear evidence for this prediction . high ionization lines like iv civ@xmath6 , heii@xmath79 , and heii@xmath104 , and especially ovi@xmath1 are particularly valuable . these emission lines respond to photoionizing energies of @xmath155ev , @xmath156ev , and @xmath157ev , respectively . hence , they can be used to probe the location of the cutoff energy of the thermal uv - bump . the strong beff in these high - ionization lines is expected if the ionizing continuum becomes softer with increasing continuum luminosity . in this case the relative number of ionizing photons with @xmath158ev decreases as the uv bump is moving towards longer wavelengths . while there is only a marginal trend for a slightly steeper baldwin effect for heii than for iv civ , the higher ionization ovi@xmath1 line displays a significantly stronger beff ( fig.9 ) . the decrease of the relative number of ionizing photons with a softer continuum , caused by the shift of the uv bump to longer wavelengths , results in weaker high - ionization lines compared to lower ionization ones . the shift of the uv bump to longer wavelengths will affect emission lines at higher ionization energy first . in addition , the relative continuum strength beneath the emission lines will be enhanced , starting at short ultraviolet wavelengths . both effects together result in smaller @xmath14 for higher luminosities . in addition to the general trend of steeper beff for higher @xmath106 , the curvature or flattening of the beff towards lower luminosities can be accommodated in the framework of some of the luminosity - dependent sed models ( e.g. , netzer et al . 1992 ; wandel 1999a ) . the flattening at low luminosities has led to suggestions that seyfert1 nuclei do not participate in the baldwin effect but have instead equivalent widths independent of luminosity ( e.g. , wampler et al.1984 ) . a close inspection of figures 7a , b provides some evidence for the tendency of a flatter slope @xmath3 of the beff at low luminosities in comparison to higher luminosities , as can be seen for several emission lines ( siiv@xmath131 , niv]@xmath105 , iv civ@xmath6 , heii@xmath79 , oiii]@xmath103 , niii]@xmath80 , heii@xmath104 ) . the turnover appears to occur near @xmath159 , in quite good agreement with predictions ( wandel 1999a ) . within this context it is interesting to note that the steepening of the baldwin effect for @xmath160 , is observed for emission lines with @xmath161ev . for several elements we have measured more than one emission line strength . the comparison of the strength of the baldwin effect may provide information about the emission line region and the mechanism which causes the observed anti - correlation of @xmath14 and continuum luminosity . for hydrogen we have measured two balmer emission lines ( h@xmath162 , h@xmath3 ) and ly@xmath11 . the slope of the baldwin effect of h@xmath162 and h@xmath3 is very similar and within the uncertainties consistent with zero , i.e. no baldwin effect for balmer lines . the missing baldwin effect for these lines provide some argument against a luminosity dependent covering factor , at least for the gas component which is the dominant contributor to these emission lines . while the equivalent width @xmath7(h@xmath66 ) remains nearly constant , the flux ratio ly@xmath11/h@xmath66 decreases by a factor @xmath163 from low to high luminosities ( @xmath164 to @xmath165 ) . the different behaviour of the balmer lines and ly@xmath11 might be also related to the complex physical processes of ly@xmath11 and h@xmath3 emission ( netzer et al.1995 ) . the prominent beff of ly@xmath11 and the lack of a beff for h@xmath66 may be caused in part by a relative stronger increase of the local continuum beneath ly@xmath56 compared to the optical wavelength range of h@xmath66 . however , ly@xmath11 shows a significant beff which fits well in the slope @xmath3 vs. @xmath106 relation ( fig.7a,9 ) . ly@xmath11 might be blended with heii@xmath102 and ov]@xmath166 ( ferland et al.1992 ; shields et al.1995 ) . for standard blr conditions the contribution of heii@xmath102 and ov]@xmath166 are negligible ( less than @xmath28% , priv.comm.k.korista ) . but for higher densities and higher ionization parameter u , particularly for optically thin conditions , these emission lines can be quite strong compared to ly@xmath11 ( shields et al.1995 ) . because ov]@xmath166 has an ionization energy of @xmath167ev , this emission line might show a strong beff . if optically thin gas is a significant fraction of the blr gas , it might be possible that the moderate beff of ly@xmath11 might be caused at least in part by ov]@xmath166 . however , if ov]@xmath166 is a significant contamination of ly@xmath11 it is expected from the results in shields et al.(1995 ) that heii@xmath79 and iv civ@xmath6 should be stronger than observed . in addition , variability behavior in some objects suggests that the optically thin gas may contribute most strongly to the line wings ( e.g. , ferland et al.1990 ; peterson et al.1993 ) , but the part of the emission lines that changes the most with luminosity has a narrow fwhm ( e.g. , osmer et al.1994 ) . although we can not exclude that ov]@xmath166 provides some contribution to ly@xmath11 , it does appear that ly@xmath11 itself shows a moderate baldwin effect in contrast to the balmer emission lines . the two emission lines of he@xmath168 which we measured , heii@xmath79 and heii@xmath104 , show very similar slopes for the baldwin effect . this behavior is expected for heii@xmath79 and heii@xmath104 as typical recombination lines . the similar slope of the beff provides also some evidence that luminosity - dependent radiative transfer effects , collisional effects or dust effects can not affect these emission lines . another line pair we measured is oiii]@xmath103 and [ oiii]@xmath169 . while the ionization energy necessary to create o@xmath170 is @xmath171ev , the excitation energies of these lines differ by a factor of @xmath172 . if the ionizing continuum becomes softer for increasing luminosity , the temperature in the emission line region should drop and hence oiii]@xmath103 should be more affected than [ oiii]@xmath169 . therefore , we might expect oiii]@xmath103 to have a steeper beff than [ oiii]@xmath169 . observationally their beffs are the same . however , the comparison of these two lines is very complicated because they differ in critical density by several orders of magnitudes ( @xmath173@xmath70 for [ oiii]@xmath169 and @xmath174@xmath70 for oiii]@xmath103 ) . in particular , [ oiii]@xmath169 forms in the narrow line region ( nlr ) , which might be heated and ionized additionally by shocks , and is , in any case , spatially distinct from the blr . the different kinematics of the oiii emitting gas is also given by the significantly different profiles we used to measure the emission line flux and equivalent width . the oiii]@xmath103 emission line was fit with a broad and narrow gaussian profile ( with an average fwhm of @xmath175kms@xmath23 and @xmath176kms@xmath23 , respectively ) . the two gaussian components we used to fit [ oiii]@xmath169 result in a profile different compared to oiii]@xmath103 , with average fwhms of @xmath177kms@xmath23 and @xmath178kms@xmath23 . the overall average fwhm of the fitted oiii]@xmath103 and [ oiii]@xmath169 lines are @xmath179 kms@xmath23 and @xmath180 kms@xmath23 , respectively . we have studied three emission lines of nitrogen . nv@xmath5 lacks within the errors a beff . in contrast , the intercombination lines niv]@xmath105 and niii]@xmath80 exhibit a baldwin effect that is consistent with the overall trend of @xmath3 vs. @xmath106 ( fig.9 ) . it might be possible that additional effects influence the emission properties of these nitrogen intercombination lines at higher luminosities . we will discuss these issues in our forthcoming paper on elemental abundances . we have investigated a large sample of @xmath19 type1 agn covering the redshift range from @xmath181 and nearly 6 orders of magnitude in continuum luminosity . to enhance the signal - to - noise ratio , minimize the influence of peculiarities of individual quasars , and investigate weak as well as strong emission lines , we computed composite spectra representing narrow intervals in redshift and luminosity . the emission line fluxes were derived using multi - component gaussian fits after removing a powerlaw continuum fit , a balmer continuum emission template , and a uv and optical iron emission template from the composite spectra . our main results are the following . * in composite spectra spanning the full redshift range at nearly constant luminosity we detect no strong trend in the line @xmath14 with redshift i.e. , with cosmic time . however , there is a marginal tendency for the highest redshift quasars ( @xmath95 ) to show slightly stronger emission lines than their counterparts at lower redshift . * in the composite spectra ranked by luminosity we find a significant baldwin effect in nearly all emission lines in the ultraviolet to optical domain . the only exceptions are nv@xmath5 , h@xmath3 , h@xmath162 , and optical feii which remain constant in @xmath14 within the uncertanties . the lack of a beff for the high - ionization feature nv@xmath5 suggests that the chemical composition of the gas is an additional parameter that can strongly influence the equivalent width of this and possibly other lines . * we detect a strong baldwin effect for the prominent nlr emission line [ oiii]@xmath169 . the strength of the beff for [ oiii]@xmath182 is very similar to the beff measured in oiii]@xmath103 . * the slope @xmath3 of the baldwin effect , where log@xmath183log@xmath184 , shows a significant correlation with the ionization energy , @xmath106 , needed to produce the lines . * the slope of the baldwin effect , @xmath3 , tends to be steeper at higher luminosities , @xmath185ergs@xmath23 , compared to the lower luminosity regime . * the baldwin effect , its steepening towards higher luminosities , and the correlation of the slope @xmath3 with @xmath113 can all be well explained in the context of a luminosity dependent spectral energy distribution of the ionizing continuum . assuming that the sed can be described as a combination of a powerlaw continuum and a thermal uv bump , the ionizing continuum becomes softer for increasing luminosity as the uv bump is shifted to longer wavelengths . this behaviour can be explained with accretion disc models as suggested by netzer et al.(1992 ) and wandel ( 1999a , b ) . we are grateful to numerous colleagues for providing many of these spectra in digital form , in particular j.a.baldwin , t.a.boroson , m.r.corbin , a.laor , w.l.w.sargent , d.p.schneider , c.c.steidel , r.j.weymann , and w.zheng . we also thank desika narayanan for reducing the spectra obtained at lick observatory . md and fh have benefited from support from nasa grant nag 5 - 3234 and nsf through ast-99 - 84040 . mv greatfully acknowledges financial support from the columbus fellowship . bahcall , j.n . , bergeron , j. , boksenberg , a. , et al . 1993 , , 87 , 1 baldwin , j.a . 1977 , , 214 , 679 baldwin , j.a . , wampler , e.j . , & gaskell , c.m . 1989 , , 338 , 630 baldwin , j.a . , ferland , g.j . , korista , k. , & verner , d. 1995 , , 455 , l119 baldwin , j.a . 1999 , in `` quasars and cosmology '' , aspconf.ser.162 , eds . g.ferland & j.a.baldwin , p.475 binette , l. , prieto , a. , szuszkiewicz , e. , & zheng , w. 1989 , , 343 , 135 bischof , o.b . & becker , r.h . 1997 , , 113 , 2000 boroson , t.a . & green , r.f . 1992 , , 80 , 109 brotherton , m.s . , et al . 2001 , , 546 , 775 carroll , s.m . , press , w.h . , & turner , e.l . 1992 , , 30 , 499 cen , r. & ostriker , j.p . 1999 , , 519 , l109 chaffee , f.h . , foltz , c.b . , hewett , p.c . , et al . 1991 , , 102 , 461 collier , s. , et al . 1998 , , 500 , 162 collier , s. , et al . 2001 , , 561 , 146 constantin , a. , shields , j.c . , hamann , f. , foltz , c.b . , & chaffee , f.h . 2002 , , 565 , 50 corbin , m.r . & boroson , t.a . 1996 , , 107 , 69 crenshaw , d.m . , 1996 , , 470 , 322 dickey , j.m . & lockman , f.j . 1990 , , 28 , 215 dietrich , m. , et al . 1993 , , 408 , 416 dietrich , m. , et al . 1999 , , 352 , l1 dietrich , m. & wilhelm - erkens , u. 2000 , , 354 , 17 dietrich , m. , appenzeller , i. , vestergaard , m. , & wagner , s. 2002a , , 564 , 581 dietrich m. , hamann f. , appenzeller i. , vestergaard m. , & wagner s.j . 2002b , in eso astrophysics symposia `` lighthouses of the universe : the most luminous celestial objects and their use for cosmology '' , eds . r.sunyaev , et al . , in press dietrich , m. & hamann , f. 2002c , in prep . dietrich , m. & hamann , f. 2002d , in prep . espey , b.r . , lanzetta , k.m . , turnshek , d.a . 1993 , baas , 25 , 1448 espey , b.r . & andreadis , s.j . 1999 , in `` quasars and cosmology '' , aspconf.ser.162 , eds . g.ferland & j.a.baldwin , p.351 ferland , g.j . , korista , k.t . , & peterson , b.m . 1990 , , 363 , l21 ferland , g.j . , peterson , b.m . , horne , k. , welsh , w.f . , & nahar , s.n . 1992 , , 387 , 95 ferland , g.j . , baldwin , j.a . , korista , k.t . , et al . 1996 , , 461 , 683 forster , k. , green , p.j . , aldroft , t. , vestergaard , m. , foltz , c.b . , & hewett , p.c . 2001 , , 134 , 35 francis , p.j . & koratkar , a.p . 1995 , , 274 , 504 goad , m. , et al . 1999 , , 524 , 707 granato , g.l . , silva , l. , monaco , p. , et al . 2001 , , 324 , 757 grandi , s.a . 1982 , , 255 , 25 green , p.j . , schartel , n. , anderson , s.f . , 1995 , , 450 , 51 green , p.j . 1996 , , 467 , 61 green , p.j . 1998 , , 498 , 170 green , p.j . 1999 , in `` quasars and cosmology '' , aspconf.ser.162 , eds . g.ferland & j.a.baldwin , p.265 green , p.j . , forster , k. , & kuraszkiewicz , j. 2001 , , 556 , 727 hamann , f. & ferland , g.j . 1993 , , 418 , 11 hamann , f. & ferland , g.j . 1999 , , 37 , 487 hamann , f. , korista , k.t . , ferland , g.j . , warner , c. , & baldwin , j.a . 2002 , , 564 , 592 kaspi , s. , et al . 1996 , , 470 , 336 kaspi , s. , smith , p.s . , netzer , h. , et al . 2000 , , 533 , 631 kauffmann , g. & haehnelt , m.g . 2000 , , 311 , 576 kellermann , k.i . , sramek , r. , schmidt , m. , shaffer , d.b . , & green , r. 1989 , , 98 , 1195 kinney , a.l . , rivolo , a.r . , & koratkar , a.p . 1990 , , 357 , 338 kinney , a.l . , bohlin , r.c . , blades , j.c . , & york , d.g . 1991 , , 75 , 645 korista , k.t . , et al . 1995 , , 97 , 285 korista , k.t . , baldwin , j.a . , & ferland , g.j . 1997 , , 108 , 401 korista , k.t . , baldwin , j.a . , & ferland , g.j . 1998 , , 507 , 24 lanzetta , k.m . , turnshek , d.,a . , & sandoval , j. 1993 , , 84 , 109 laor , a. , bahcall , j.n . , jannuzi , b.t . , schneider , d.p . , & green , r.f . 1995 , , 99 , 1 lawrence , a. & elvis , m. 1982 , , 256 , 410 mushotzky , r. & ferland , g.j . 1984 , , 278 , 558 netterfield , c.b . , 2002 , , 571 , 604 netzer , h. 1985 , , 216 , 63 netzer , h. 1987 , , 225 , 55 netzer , h. , et al . 1995 , , 448 , 27 netzer , h. , laor , a. , & gondhalekar , p.m. 1992 , , 254 , 15 osmer , p.s . , porter , a.c . , & green , r.f . 1994 , , 436 , 678 osmer , p.s . & shields , j.c . 1999 , in `` quasars and cosmology '' , aspconf.ser.162 , eds . g.ferland & j.a.baldwin , p.235 peterson , b.m . , 1993 , , 402 , 469 peterson , b.m . , et al . 1994 , , 425 , 622 pettini , m. 1999 , in proc.of eso workshop `` chemical evolution from zero to high redshift '' , ed.j.walsh & m.rosa , lnp , p.233 pogge , r.w . & peterson , b.m . 1992 , , 103 , 1084 reichert , g.a . , et al . 1994 , , 425 , 582 santos - lle , m. , et al . 2001 , , 369 , 57 sargent , w.l.w . & malkan , m.a . 1982 , , 254 , 22 sargent , w.l.w . , boksenberg , a. , & steidel , c.c . 1988 , , 68 , 539 sargent , w.l.w . , steidel , c.c . , & boksenberg , a. 1989 , , 69 , 703 schneider , d.p . , schmidt , m. , & gunn , j.e . 1991a , , 101 , 2004 schneider , d.p . , schmidt , m. , & gunn , j.e . 1991b , , 102 , 837 schulz , h. 1992 , in `` physics of active galactic nuclei '' , eds . w.j.duschl & s.j.wagner ( berlin : springer ) , p.235 seaton , m.j . 1979 , , 187 , 73 shields , j.c . , ferland , g.j . , & peterson , b.m . 1995 , , 441 507 steidel , c.c . 1990 , , 74 , 37 steidel , c.c . & sargent , w.l.w . 1991 , , 382 , 433 stern , d. , djorgovski , s.g . , perley , r.a . , de carvalho , r.r . , & wall , j.v . 2000 , , 119 , 1526 stirpe , g.m . , et al . 1994 , , 425 , 609 storey , p.j . & hummer , d.g . 1995 , , 272 , 41 storrie - lombardi , l.j . , mcmahon , r.g . , irwin , m.j . , & hazard , c. 1996 , , 468 , 121 tananbaum , h. , avni , y. , green , r.f . , schmidt , m. , & zamorani , g. 1986 , , 305 , 57 telfer , r.c . , zheng , w. , kriss , g.a . , & davidsen , a.f . 2002 , , 565 , 773 turnshek , d.a . 1997 , in asp conf.ser.128 , `` mass ejection from agn '' , eds . n.arav , i.shlosman , & r.j.weymann ( san francisco : asp ) , p.52 vanden berk , d.e . , et al . 2001 , , 122 , 549 vron - cetty , m .- p . , vron , p. , & tarenghi , m. 1983 , , 119 , 69 vron - cetty , m .- & vron , p. 2001 , `` a catalogue of quasars and active nuclei '' , 10th edition vestergaard , m. & wilkes , b.j . 2001 , , 134 , 1 wampler , e.j . , gaskell , c.m . , burke , w.l . , & baldwin , j.a . 1984 , , 276 , 403 wandel , a. 1999a , , 527 , 649 wandel , a. 1999b , , 527 , 657 wanders , i. , et al . 1997 , , 113 , 69 wang , t .- g . , lu , y .- j . , & zhou , y .- y . 1998 , , 493 , 1 weymann , r.j . , morris , s.l . , foltz , c.b . , & hewett , p.c . 1991 , , 373 , 23 weymann , r.j . , jannuzi , b.t . , lu , l. , et al . 1998 , , 506 , 1 wilkes , b.j . , tananbaum , h. , worrall , d.m . , et al . 1994 , , 92 , 53 wilkes , b.j . , kuraszkiewicz , j. , green , p.j . , mathur , s. , & mcdowell , j.c . 1999 , , 513 , 76 wills , b.j . , thompson , k.l . , han , m. , et al . 1995 , , 447 , 139 wills , b.j . , laor , a. , brotherton , m.s . , et al . 1999 , , 515 , l53 zamorani , g. , marano , b. , mignoli , m. , zitelli , v. , & boyle , b.j . 1992 , , 256 , 238 zheng , w. , fang , l .- z . , & binette , l. 1992 , , 392 , 74 zheng , w. & malkan , m.a . 1993 , , 415 , 517 zheng , w. , kriss , g.a . , & davidsen , a.f . 1995 , , 440 , 606 zheng , w. , kriss , g.a . , telfer , r.c . , 1997 , , 475 , 469 fig.1 redshift distribution of the agn sample as a function of intrinsic luminosity at @xmath55 . the open symbols represent radio - loud quasars and the filled symbols radio - quiet quasars . the dashed horizontal ( vertical ) lines indicate the luminosity ( redshift ) ranges which were used to calculate composite spectra . at the left side ( top ) of the figure is the number of the individual spectra contributing to each composite spectrum . for the luminosity range @xmath186 composite spectra were calculated for the redshift intervals marked by the vertical dashed lines . fig.2 composite spectrum of the entire agn sample ( top panel ) . the dashed line shows a power law fit to the continuum ( @xmath187 ) yielding @xmath59 @xmath188 . the bottom panel displays the number of agn contributing to each wavelength element . fig.3 ( a ) normalized composite spectra are shown for redshift bins of @xmath87 , starting at @xmath189 ( cf . , fig.1 ) and nearly constant luminosity ( @xmath186 ) . the spectra were normalized by a power law continuum fit . the horizontal dashed lines indicate the continuum level for the individual normalized composite spectra which were vertically shifted for better display . the normalized continuum strength is shown for the spectrum at the bottom of the figure . this same scale applies to all other spectra . \(b ) same as fig.3a , but with an expanded vertical scale to display the dependence of the relative strength of weaker emission lines as a function of redshift . strong emission lines with flat tops are truncated for easier display . fig.5 ( a ) normalized composite spectra are shown for the luminosity bins ( @xmath190 ) as displayed in fig.1 . the spectra were normalized with the corresponding power law continuum fit . the horizontal dashed lines indicate the continuum - level for the individual normalized composite spectra which were vertically shifted for better display . the normalized continuum strength is shown for the spectrum at the bottom of the figure and it applies for the other spectra too . \(b ) same as fig.5a , but with an expanded vertical scale to display the dependence of the relative strength of weaker emission lines as a function of luminosity . strong emission lines with flat tops are truncated for easier display . fig.7 ( a ) line equivalent widths , @xmath7 , as a function of increasing continuum luminosity @xmath109 . we calculated linear fits to @xmath14(l ) for the entire luminosity range ( dashed line ) , as well as a luminosities log@xmath191 ( dashed - dotted line ) . fig.8 difference spectra with respect to the composite spectrum log@xmath192 ergs@xmath23 are displayed to illustrate that @xmath7(nv@xmath5 ) is nearly constant , regardless how @xmath14(nv ) is measured , while the other emission lines display a prominant beff . the dotted line indicates the location of nv@xmath5 . fig.9 the slope of the baldwin effect as displayed in fig.7a and 7b as a function of the ionization energy @xmath106 needed to create the specific ions . different lines of the same ion ( heii@xmath79 , heii@xmath104 and oiii]@xmath103 , [ oiii]@xmath2 ) show nearly identical slopes . the filled symbols represent the slopes based on the entire luminosity range , corresponding to the dashed lines in figures 7a , b . the slopes of the baldwin effect for higher luminosities only ( log@xmath111 ) are plotted as open symbols . lrccc line & @xmath106&entire @xmath109-range & log@xmath193 & korista et al.(1998 ) + & [ ev]&@xmath3&@xmath66&@xmath3 + ovi@xmath1 & @xmath194&@xmath195&@xmath195&@xmath196 + nv@xmath5 & @xmath197&@xmath198&@xmath199&@xmath200 + heii@xmath79 & @xmath201&@xmath202&@xmath203&@xmath204 + heii@xmath104 & @xmath201&@xmath205&@xmath206 & + iv civ@xmath6 & @xmath207&@xmath208&@xmath209&@xmath210 + niv]@xmath105 & @xmath211&@xmath212&@xmath213&@xmath214 + oiii]@xmath103 & @xmath215&@xmath216&@xmath217&@xmath218 + [ oiii]@xmath2&@xmath215&@xmath219 & & + siiv@xmath131 & @xmath220&@xmath221&@xmath219&@xmath222 + niii]@xmath80 & @xmath223&@xmath224&@xmath225&@xmath226 + aliii@xmath77 & @xmath227&@xmath228&@xmath228&@xmath229 + iii ciii]@xmath64 & @xmath230&@xmath221&@xmath231&@xmath232 + siiii]@xmath78&@xmath233&@xmath202&@xmath234&@xmath235 + ly@xmath11@xmath102 & @xmath236&@xmath237&@xmath208&@xmath238 + h@xmath67@xmath239&@xmath236&@xmath240&@xmath240 & + h@xmath66@xmath241 & @xmath236&@xmath242&@xmath243 & + cii@xmath130 & @xmath244&@xmath245&@xmath246&@xmath247 + siii@xmath248 & @xmath249&@xmath250&@xmath246 & + feiiuv & @xmath251&@xmath252 & & + feiiopt & @xmath251&@xmath240 & & + mgii@xmath16 & @xmath253&@xmath221&@xmath221&@xmath254 + oi@xmath129 & @xmath255&@xmath256&@xmath257&@xmath200 +

cooperation and defection are the two strategies that are usually at the heart of every social dilemma @xcite . while cooperative individuals contribute to the collective welfare at a personal cost , defectors choose not to . due to the resulting lower individual fitness of cooperators the selection pressure acts in favor of the defectors , thus designating the evolution of cooperation as a dilemma standing on its own . established by @xcite , evolutionary game theory @xcite provides a competent theoretical framework to address the subtleties of cooperation among selfish and unrelated individuals . the prisoner s dilemma game in particular , is considered a paradigm for tackling the problem of cooperation @xcite . the game promises a defecting individual the highest fitness if facing a cooperator . at the same time , the exploited cooperator is worse off than a defector playing with another defector . according to the fundamental principles of darwinian selection , cooperation extinction is therefore inevitable . this unadorned scenario is actually realized in the well - mixed prisoner s dilemma game , where defectors reign supreme @xcite . relaxing the inevitability of a social downfall constituted by the well - mixed prisoner s dilemma is the snowdrift or hawk - dove game @xcite , where mutual defection is individually less favorable than a cooperation - defection pair - up . accordingly , the snowdrift game allows for stable coexistence of cooperators and defectors in well - mixed populations @xcite . completing the triplet is the stag - hunt game @xcite , which together with the prisoner s dilemma and the snowdrift game , forms the standard set of social dilemmas that is frequently explored in the current literature [ see _ e.g. _ @xcite ] . compared with the prisoner s dilemma , the stag - hunt game offers more support for cooperative individuals in that the reward for mutual cooperation is higher than the temptation to defect . still , cooperation in the stag - hunt game is compromised by the fact that mutual defection is individually more beneficial than being an exploited cooperator , as recently highlighted by @xcite . an important realization by the pursuit of cooperation in the context of social dilemmas was the fact that the outcome of evolutionary games in structured populations can be very different from the well - mixed case . in a pioneering work , @xcite showed that the introduction of spatial structure via nearest neighbor interactions enabled the cooperators to form clusters on the square lattice and so protect themselves against the exploitation by defectors . following this discovery , the impact of the spatial structure on the evolution of cooperation has been investigated in detail @xcite , and the subject has since been reviewed comprehensively on different occasions @xcite . notably , the theoretical conjecture that spatial structure may promote cooperation , or at least sustain a multitude of competing strategies has been confirmed experimentally @xcite , but there also exist evidences that spatial structure may not necessarily favor cooperation @xcite . since the impact of the spatial structure on the evolution of cooperation depends on the governing social dilemma , and due to the difficulties associated with the payoff rankings in experimental and field work @xcite , it is certainly good practice to test new mechanisms aimed at promoting cooperation on different evolutionary games . the recent shift from evolutionary games on regular grids to evolutionary games on complex networks [ for the latter see _ e.g. _ @xcite ] can be considered a step towards more realistic conditions . indeed , the shift is by no means trivial and bears fascinating results , as recently reviewed by @xcite . quite remarkably , scale - free networks @xcite turned out to sustain cooperation by all three above - described social dilemmas @xcite , owing predominantly to the heterogeneity that characterizes their degree distribution . following this seminal discovery , several studies have since elaborated on different aspects of cooperation on scale - free networks , as for example its dynamical organization @xcite , evolution under clustering @xcite , mixing patterns @xcite , memory @xcite and payoff normalization @xcite , as well as its robustness in general @xcite and under intentional attack and error @xcite . the body of literature devoted to the study of evolutionary games on complex network is extensive , aside from the scale - free architecture hosting the prisoner s dilemma @xcite and the snowdrift game @xcite , covering also small - world @xcite , social as well as other real - world networks @xcite . notably , the impact of different interaction topologies has also been studied for evolutionary games outside the realm of the above - described social dilemmas . examples include the rock - paper - scissors game @xcite , the ultimatum game @xcite or the public goods game @xcite , and indeed many more studies of the latter games on complex network are expected in the near future . besides the conditions generated by spatiality and complex interaction networks , many different mechanisms have been identified that can promote or otherwise affect the evolution of cooperation , and we mention them here briefly . aside from network reciprocity inherent to games on graphs and complex networks , other prominent rules promoting cooperative behavior are kin selection @xcite , direct reciprocity @xcite , indirect reciprocity @xcite and group selection @xcite , as recently reviewed in @xcite . moreover , voluntary participation @xcite , social diversity @xcite , asymmetric influence of links and partner selection @xcite , heterogeneous teaching activity @xcite , and the impact of long - term learning @xcite have been suggested as interesting possibilities that may emerge in real - life systems . the necessary overlap between interaction and replacement graphs @xcite has also been recognized as an important agonist in the evolution of cooperation . furthermore , the importance of time scales in evolutionary dynamics @xcite , the role of finite population size @xcite , and the impact of noise and uncertainties on evolution in general @xcite have been investigated as well . very recently , random explorations of strategies @xcite and simultaneous adoptions of different strategies depending on the opponents @xcite have also been identified as potent promoters of cooperation . some of these mechanisms will be described more accurately in the subsequent sections , but otherwise the reader is referred to the original works for details . in the focus of this mini review are evolutionary games with coevolutionary rules . initiated by @xcite and by @xcite , and in some sense motivated by then very vibrant advances in network growth and evolution @xcite , the subject has evolved into a mushrooming avenue of research that offers new ways of ensuring cooperation in situations constituting a social dilemma . coevolutionary rules constitute a natural upgrade of evolutionary games since in reality not only do the strategies evolve in time , but so does the environment , and indeed many other factors that in turn affect back the outcome of the evolution of strategies . coevolutionary rules can affect the links players make ( or brake ) @xcite ( see section [ interactions ] ) , the size of the network ( or population ) @xcite ( see section [ growth ] ) , the teaching activity ( or reproduction capability ) @xcite ( see section [ teaching ] ) and mobility of players @xcite ( see section [ mobility ] ) , their age @xcite ( see section [ aging ] ) , as well as several other factors @xcite ( see section [ related ] ) that eventually affect the outcome of the underlying evolutionary game . although the majority of coevolutionary rules studied so far affects the network architecture and size , it is important to distinguish these studies from previous , partially closely related works where networks also change or evolve in the course of time @xcite ; in particularly so , since the term ` coevolution ' has in the past been used quite frequently and for rather different processes . in the continuation of this paper we will review recent advances on evolutionary games with coevolutionary rules , affecting , as mentioned above , the interaction network , the reproduction capability of players , their reputation , mobility or age , more thoroughly . before that , however , we give in section [ games ] a more technical description of the evolutionary games and strategy adoption rules that we will encounter throughout the paper . following the main body of the review given in section [ rules ] , we conclude our work and give an outlook in section [ final ] . as noted in the first paragraph of section [ intro ] , the three main social dilemmas involving pairwise interactions are constituted by the prisoner s dilemma game , the snowdrift game and the stag - hunt game . at least one of these three games is employed in the majority of the works we will review below , and hence we give a more accurate description of them in what follows . irrespective of which game applies , players can choose either to cooperate or to defect . notably , other strategies , such as loners [ see _ e.g. _ @xcite ] or punishers [ see _ e.g. _ @xcite ] are also possible , but their inclusion to evolutionary games with coevolutionary rules has not yet been considered . in general , mutual cooperation yields the reward @xmath0 , mutual defection leads to punishment @xmath1 , and the mixed choice gives the cooperator the sucker s payoff @xmath2 and the defector the temptation @xmath3 . the standard scaled parametrization entails designating @xmath4 and @xmath5 as fixed , while the remaining two payoffs can occupy @xmath6 and @xmath7 . then , if @xmath8 we have the prisoner s dilemma game , @xmath9 yields the snowdrift game , and @xmath10 the stag - hunt game , as schematically depicted in fig . [ fig : schematic_games ] . without much loss of generality , this parametrization is often further simplified for the prisoner s dilemma game , so that @xmath11 is the only free parameter while @xmath4 and @xmath12 are left constant ( thick red line in fig . [ fig : schematic_games ] ) . however , since then the condition @xmath13 is not strictly fulfilled , this version is traditionally referred to as the weak prisoner s dilemma game @xcite . an option is also to use @xmath11 , @xmath14 , @xmath5 and @xmath15 , thus strictly adhering to the prisoner s dilemma payoff ranking @xmath8 while still having a single tunable parameter in the form of the ratio @xmath16 . for the snowdrift game one can , in a similar fashion , introduce @xmath17 $ ] such that @xmath18 and @xmath19 [ see _ e.g. _ @xcite ] , thereby again decreasing the effective dimensionality of the parameter space by one . note also that @xmath20 characterizes the cost - to - benefit ratio @xcite and in fact constitutes a diagonal in the snowdrift quadrant of the @xmath21 parameter plane , as shown in fig . [ fig : schematic_games ] by the dotted blue line . it is worth mentioning that other types of parametrization of two - strategy games are possible as well @xcite , but we focus on the one presented above since it is the most widely used , thus enabling an efficient comparison of different works . parameter plane encompassing the stag - hunt ( sh ) , the prisoner s dilemma ( pd ) and the snowdrift ( sd ) game . borders between games are denoted by dashed green lines . dotted blue diagonal depicts the @xmath20-parametrization of the snowdrift game , while the thick red line shows the span of the weak prisoner s dilemma game having @xmath11 as the only main parameter . the upper left quadrant represents the so - called harmony game ( hg ) . the latter , however , does not constitute a social dilemma because there cooperation is always the winning strategy.,width=283 ] the most frequently employed setup entails that initially each player @xmath22 is designated either as a cooperator @xmath23 or defector @xmath24 with equal probability , and is placed on one of the nodes of the network with degree @xmath25 . evolution of the two strategies is then performed in accordance with a pairwise comparison rule , during which players accumulate their payoffs @xmath26 by playing the game with their neighbors . subsequently , player @xmath22 tries to enforce its strategy @xmath27 on player @xmath28 in accordance with some probability @xmath29 to be specified below . during the simulation procedure the player @xmath22 and one of its neighbors @xmath28 are chosen randomly , whereby in accordance with the random sequential update each player is selected once on average during @xmath30 ( network size ) such elementary steps , together constituting one full monte carlo step @xcite . alternatively , players can be selected sequentially , albeit this may cause artificial effects . independently on whether synchronized or the random sequential update is used , however , the time evolution is always discrete . the probability of strategy adoption @xmath29 can be defined in several ways . if the degree @xmath25 of all players is the same and does not change in time , the fermi function @xmath31 } \label{fermi}\ ] ] is a viable option , as proposed by @xcite . in eq . [ fermi ] @xmath32 denotes the amplitude of noise @xcite , or equivalently its inverse ( @xmath33 ) the so - called intensity of selection @xcite . in the @xmath34 limit player @xmath22 always succeeds in enforcing its strategy to player @xmath28 if only @xmath35 but never otherwise . for @xmath36 , however , strategies performing worse may also be adopted based on unpredictable variations in payoffs @xcite or errors in the decision making , for example . importantly , if the degree distribution of the interaction network ( note that this is a property that may likely change due to a coevolutionary rule ) , at any instance of the game , deviates from the case where all players have the same degree , the application of the fermi function may introduce additional effects since then the impact of the same value of @xmath32 effectively varies from one player to the other . indeed , if the degree distribution characterizing the interaction network is heterogeneous , a more successful player ( _ i.e. _ having a larger payoff ) can pass its strategy with the probability @xmath37 where @xmath38 is the largest of the two degrees @xmath25 and @xmath39 , and @xmath40 for the prisoner s dilemma game , @xmath41 for the snowdrift game and @xmath42 for the stag - hunt game ( note that the ranking of payoff elements for each specific game ensures the positive sign of eq . [ pach ] . ) introduced by @xcite , it is still a popular choice surpassing the difficulties associated with the fermi function described above , albeit with the downside of being unable to adjust the level of uncertainty by strategy adoptions . finally , we mention another frequently used strategy adoption rule in coevolutionary models ; namely the so - called richest - following ( or ` learning from the best ' ) rule @xcite , where the focal player always imitates the strategy of its most successful neighbor @xcite . contrary to the preceding two strategy adoption rules , the richest - following is completely deterministic , in fact exercising the strongest selection between players . naturally , there also exist other microscopic strategy adoption rules , such as the win - stay - lose - shift rule where the focal player has restricted information on its neighbors , for which the reader is advised to consult the comprehensive review by @xcite for more details . we will use the notation introduced above throughout this work unless explicitly stated otherwise . also , any deviations with respect to the employed initial setup , simulation procedure or the definition of strategy adoption probability will be noted when applicable . while it is obvious that strategies of players engaging in evolutionary games evolve in time , the fact that other properties characterizing either their individual attributes or the environment in which the game is staged may simultaneously evolve as well gained foothold only in recent years . yet the preceding transitions from well - mixed populations to spatial grids and further to complex networks , and in particular their success in explaining the evolution of cooperation , are inviting to further extensions of the theoretical framework , and indeed , the introduction of coevolutionary rules seems like the logical next step . it should need little persuasion to acknowledge that links we make with others change in time , that all of us age , that our roles in life evolve , and that the society we are part of may itself be subject to transformations on a global scale . coevolutionary rules aim to integrate these processes into the framework of evolutionary games . perhaps the biggest challenge thereby is , how to do this without directly ( or obviously ) promoting cooperation . for example , if one introduces a rule that , in the course of time , cooperators should aim to link only with cooperators and defectors only with defectors , it should come as no surprise that such a coevolutionary rule will likely favor the evolution of cooperation . it is demanding , however , to explore and identify successful mechanisms that do not attribute special , not to say fictitious , cognitive skill to players , and do not use a discriminative set of rules for every participating strategy . thus , coming up with plausible coevolutionary rules is not straightforward , and care must be exercised in order to give both strategies equal credentials . simply because a strategy is bad for social welfare it should not be assumed that the individuals adopting it are less skilful or sly than their opponents . in fact , rather the opposite seems to apply . for example , defectors should be assumed being just as skilful by selecting appropriate partners as cooperators . in the following we will review coevolutionary rules affecting the interactions between players ( section [ interactions ] ) , population growth ( section [ growth ] ) , teaching activity ( section [ teaching ] ) , mobility ( section [ mobility ] ) and aging ( section [ aging ] ) of players , as well as related aspects ( section [ related ] ) of individual and global characteristics that may affect strategy dominance in evolutionary games . coevolutionary rules frequently affect how players link with one another and this section reviews examples thereof . as we have mentioned above , the result of a game with a partner may influence the durability of such a connection . in particular , an unsatisfied player can easily brake a link to look for a more beneficial interaction with another partner . notably , the network itself does thereby not shrink or grow in size ( for the latter see section [ growth ] ) . instead , our aim in this subsection is to explore possible rearrangements of an existing network that is driven by the success of players participating in the governing evolutionary game . since coevolutionary rules affecting the interactions between players were proposed first @xcite , the pertaining literature that has accumulated thus far is rather extensive . works can be partitioned into those that employed strategy independent rules for link adaptations @xcite and those that considered strategies or their performances as factors potentially affecting the rewiring @xcite . notably , the latter distinction is rather crude and sometimes not completely accurate since the rewiring can be performed based on a secondary player property , like reputation @xcite , attractiveness @xcite or satisfaction @xcite , which are typically related with strategy performance over time . it is indeed possible to further distinguish the proposed coevolutionary rules introducing dynamical interactions to those by which the change of the interaction network is driven by the urge to increase the payoff of the focal player directly @xcite , and those by which the rewiring serves also the increase of the payoff but on a global scale , _ i.e. _ independently of the payoff of the focal player that is affected by the link adaptation @xcite . in the latter case it is thus not necessary to calculate the players payoff prior to rewiring because solely its strategy determines the ` life ' of a link . [ ! ht ] summarizing the above , a simplified but useful classification of interaction - updating rules is presented in fig . [ fig : class_of_updates ] . as suggested in the works mentioned last in the preceding paragraph , the lifetime of a link may depend primarily on the strategies of the players that are connected with it ( type a ) . from this point of view it is straightforward to establish that defector - defector links are short - lived if compared to cooperator - cooperator links since the former are not beneficial for neither of the two involved players , while the later yield mutual gains for both . the second set of coevolutionary rules evaluates the payoffs originating from the investigated link prior to its potential deletion , while the actual removal takes place only if a new neighbor may yield higher benefits ( type b ) @xcite . and finally , the third set of coevolutionary rules considers the strategy adoption process as pivotal for deciding which links to delete and which to keep ( type c ) . an example thereof is that the invaded player looses all its links except the one with the donor of the new strategy @xcite , as depicted in fig . [ fig : class_of_updates](c ) . there are several real - life situations that can be modeled by the latter rule . from a biological viewpoint , the coevolutionary rule can be linked with an invasion of the subordinate species and the subsequent replacement by a newborn of the victor . a similar phenomenon can be observed in human societies when one changes a job . typically then the links to former coworkers fade and eventually brake , and new ties are formed primarily with the coworkers from the new working place . notably , it falls within the same logical set of rules if the player that has successfully passed its strategy is allowed to increase the number of neighbors that are directly connected to it , as was proposed by @xcite . it should not be overlooked , however , that the strategy adoption process , triggering the deletion and/or addition of links , is itself inherently routed in the payoff difference of the considered players . an important feature of coevolutionary rules molding the interactions among players is also the time scale separation between link and strategy adaptations , as reported in @xcite . as the cited works suggest , the time scale separation can drastically influence the final output of coevolutionary games . this effect will also be discussed in the present review . in what follows , we will review the coevolutionary rules presented in some of these works more accurately . in agreement with the actual time - line we start with the work of @xcite , who proposed a coevolutionary rule in which a randomly chosen player @xmath22 is connected to a new neighbor at random . if the new link increases the average payoff of the focal player the latter accepts it and disconnects from the neighbor it scores worst against . note that this coevolutionary rule indirectly favors the establishment of cooperator - cooperator links ( this pair - up yields the highest average payoff ) and at the same time facilitates the deletion of defector - defector links . the coevolutionary rule was paired up with strategy mutation @xcite , by which a mutation is accepted if it yields a higher payoff for player @xmath22 than the initial strategy [ type b ; see fig . [ fig : class_of_updates](b ) ] . starting from a random network with poissonian degree distribution , it was shown that this coevolutionary rule leads to cooperative nash equilibria in an iterative prisoner s dilemma game with the additional property that no agent can improve its payoff by changing its neighborhood . according to the authors , the later may be interpreted as a sort of ` network nash equilibrium ' @xcite . notably , this coevolutionary rule also affects the initial network structure in that the later evolves to a statistically stationary state with a broad degree distribution , suggesting scale - free behavior and giving rise to small - world properties , among others . following their preceding seminal contribution [ see @xcite ] , @xcite proposed a coevolutionary rule affecting only defector - defector pairs with the motivation that in this pair - up _ both _ players might be better off if searching for a new partner in the context of the prisoner s dilemma game [ type a ; see fig . [ fig : class_of_updates](a ) ] . it was shown that , starting from a random network with a given average degree and the richest - following strategy adoption rule , even a small probability @xmath43 of searching for a new partner from a defector - defector configuration may substantially promote cooperation . indeed , as low as @xmath44 were shown to uphold practically complete cooperator dominance across the whole span of the weak prisoner s dilemma game ( see fig . [ fig : schematic_games ] ) . with respect to the network topology , it was reported that the coevolutionary rule facilitates the formation of a hierarchical interaction structure and may also introduce small - world properties if the search for new partners is constrained to the neighbors of the neighbors . however , unlike as shown by @xcite , the occasional ( depending on @xmath43 ) break - up of defector - defector pairs has not been found leading to broad or even scale - free degree distributions . these findings were subsequently extended @xcite , where it was elaborated on the spontaneous emergence of cooperators with extremely high payoffs and the important role of this so - called ` leaders ' for the global sustenance of cooperation . as such , these works can be considered as an important prelude to the realization of the fact that scale - free networks constitute an extremely favorable environment for the evolution of cooperation irrespective of the governing social dilemma @xcite . a simple but still plausible coevolutionary rule affecting links between players has been proposed by @xcite . exemplifying type a class of interaction updating [ see fig . [ fig : class_of_updates](a ) ] , players adopting either the strategy @xmath45 ( cooperate ) or @xmath46 ( defect ) were designated a propensity to form new links denoted by @xmath47 and @xmath48 , such that @xmath49 links were formed at rates @xmath50 , where @xmath51 $ ] . moreover , each link was assigned a specific lifetime depending on the strategy of the two connected players given by @xmath52 , where @xmath53 is the corresponding link death rate . with these definitions the authors were able to specify mean field equations governing the so - called active linking dynamics of the network . this coevolutionary rule has been tested on the prisoner s dilemma and the snowdrift game subject to the fermi function ( see eq . [ fermi ] ) governing the strategy adoption [ for additional set - ups see @xcite ] . it was shown that if the time scale associated with active linking is much smaller than the one associated with strategy updating the proposed coevolutionary rule leads to an effective rescaling of the governing payoff matrix , and thus a shift in the played evolutionary game . for example , the prisoner s dilemma game transforms to the coordination game , while the snowdrift game transforms to the harmony game [ for details on the coordination game see _ e.g. _ @xcite ] . in both cases the cooperation is promoted , in turn designating the proposed coevolutionary rule as a simple and analytically tractable means of understanding how selfish and unrelated individuals may be led to adopting the cooperative strategy . on the other hand , if the ratio between the time scales associated with active linking and strategy updating is not small , the interplay between these two dynamical processes leads to a progressive crossover between the analytic results obtained for very fast active linking and the evolutionary dynamics of strategies taking place on static graphs . the latter were found to exhibit different degrees of heterogeneity depending on the parameters determining active linking , yet in general complying well with real social networks having fast decaying tails in their degree distributions . notably , compared to the earlier works reviewed above , an important observation made in the two papers by @xcite was that the impact of coevolutionary rules may depend significantly on the time scales associated with the strategy and structure ( link ) evolution . for example , @xcite too commented on the time scale separation in their model , yet the promotion of cooperation was thereby not notably affected [ both slow ( @xmath54 ) and fast ( @xmath55 ) rewiring of @xmath56 links was found to be highly effective ] . active linking dynamics has also been investigated in repeated games incorporating direct reciprocity @xcite , where additionally the productivity of every link connecting two players was evaluated prior to potential rewiring . moreover , the active linking model proposed by @xcite was recently extended by @xcite to account for the impact of different reactions to adverse ties . in particular , @xcite additionally introduced individual behavioral types of players through different values of @xmath57 , separating those that tend to break their links frequently ( @xmath57 close to @xmath58 ) from those that tend to break them rarely ( @xmath57 close to @xmath59 ) . in this way both topology and strategy dynamics become interrelated it was shown that populations in which individuals are allowed to handle their social contacts diversely are more prone to cooperative behavior than those in which such diversity is absent . similarly as in @xcite , it was shown that by an appropriate time scale separation between strategy and network dynamics the diverse behavioral preferences can also introduce a transformation of the governing social dilemma , yet so that each individual perceives the same game differently . also building on the time scale separation between rewiring and strategy updating is another paper by @xcite , where players are able to decide which links they want to maintain and which they want to change based on local information about their neighbors [ type b ; see fig . [ fig : class_of_updates](b ) ] . a link change is initiated if player @xmath22 is dissatisfied with its connection to player @xmath28 , which is the case if the strategy of player @xmath28 is to defect . however , player @xmath28 also assesses the quality of its link to @xmath22 in the same fashion . if both @xmath22 and @xmath28 are satisfied ( which practically means that both are cooperators ) the link between them remains intact . if @xmath22 wants to remove the link and @xmath28 not ( @xmath60 and @xmath61 ) , the probability @xmath62 given by the fermi function ( see eq . [ fermi ] ) is invoked . if realized , player @xmath22 is allowed to redirect to a random neighbor of @xmath28 . if not , @xmath22 stays linked with @xmath28 . if both @xmath22 and @xmath28 are defectors , and thus both want to remove the link , then rewiring takes place such that the new link keeps attached to @xmath22 with probability @xmath62 or to @xmath28 with probability @xmath63 . finally , the authors introduce a ratio defined as the time scale associated with the evolution of strategies @xmath64 ( for simplicity equal to one ) divided by the time scale associated with the rewiring of the network @xmath65 , showing that there exists a critical value for this ratio above which cooperators wipe out defectors . moreover , the emerging networks exhibit an overall heterogeneity that is maximal at the critical value and can be compared well with diversity associated with realistic social networks . the coevolutionary rule proposed by @xcite was extended by allowing individuals to adjust their social ties @xcite , with the extension that each player was assigned an individual willingness @xmath66 to rewire unwanted social interactions . accordingly , players with small @xmath67 can be considered as loyal to their partners and resilient to change , while those with @xmath68 are swift in altering their links . it was shown that the highest cooperation levels can be achieved when the propensity to change links is highly strategy - dependent . more precisely , it was found very beneficial for the evolution of cooperation if defectors changed their partners frequently while cooperators behaved oppositely , _ i.e. _ kept their partners for as long as possible . this is indeed expect since defectors are unable to establish social ties under mutual agreement with their partners . on the other hand , cooperators are typically much more prone to establishing long - term relations and loyalty . ultimately , these two facts lead to the evolution of heterogeneous interactions networks where cooperators are known to prevail over defectors [ see _ e.g. _ @xcite ] . related to the work of @xcite is the recent paper by @xcite , the difference being that in the latter only cooperators are allowed to switch their partners if they act as defectors , and moreover , the new partner is sought randomly from the whole population . it is found that under such a coevolutionary rule there exists an optimal , rather than critical [ compare with @xcite ] , time scale separation between rewiring and strategy updating for which cooperation thrives best . also , the resulting interaction topology is different in that the network typically becomes divided into isolated communities of cooperators and defectors due to the selective coevolutionary rule targeting only mixed strategy pairs . notably , a preceding study by @xcite considered partner switching also with the aid of reputation , which was defined similar to image scoring proposed a decade earlier by @xcite . it was found that coevolutionary switching of partners based on the reputation of nearest and next - nearest neighbors , _ i.e. _ preferentially targeting players with a higher reputation , is significantly more effective in promoting cooperation than seeking a new partner randomly from the whole population . thus , these results underline the importance of indirect reciprocity @xcite also when individuals can adjust their social ties . links are beneficial for all involved ) , can pass its strategy to a defector that is weakened by neighboring defectors ( panel a ) . subsequently , the invaded player looses its links to other players , except the one with the donor of the new strategy ( panel b ) . due to random link additions , the successful invasion of cooperators will repeat itself sooner or later depending on @xmath69 ( panel c ) , ultimately resulting in the disintegration of the defector cluster ( panel d ) . note that this process can not work in the opposite direction , _ i.e. _ defectors can not invade a cluster of cooperators . the necessary condition for this mechanism to work is the emergence of quasi - homogeneous groups , which occur if strategy adoptions happen frequently between new link additions , _ i.e. _ if @xmath69 is large enough.,width=283 ] in addition to the studies reviewed above , similar coevolutionary rules were used to study how scale - free networks emerge in social systems @xcite , how cooperation in the prisoner s dilemma game can be established via the interplay between dynamical interactions and game dynamics @xcite or interaction stochasticity @xcite , how social dilemmas in general can thereby be resolved @xcite , as well as other sophisticated models @xcite were considered . we refer the interested reader to the original works for further details , while here we proceed with the review of some of the studies that employed strategy independent rules for link adaptations @xcite . belonging to the third type of interaction - updating coevolutionary rules [ type c ; see fig . [ fig : class_of_updates](c ) ] is the model proposed by @xcite , where whenever player @xmath22 adopts a new strategy all its links , except from the one with the donor of the new strategy , are deleted ( see fig . [ fig : multilevel_selection ] ) , and moreover , all individuals are allowed to form a new link with a randomly chosen player with which they are not yet connected after every @xmath69 full monte carlo steps . note that the random additions of links counteract the deletions following each strategy adoption , in turn largely preserving the initially random topology and the heterogeneity of the interaction network @xcite . it was shown that at a sufficiently large time scale separation between link deletions and additions , constituted by @xmath69 , this coevolutionary rules evokes the spontaneous emergence of a powerful multilevel selection mechanism , which despite the persistent random topology of the evolving network , maintains cooperation across a substantial portion of the @xmath21 parameter plane . importantly , the promotion of cooperation is thereby not realized by some final outcome of a coevolutionary rule , as is for example the case in @xcite , but is the consequence of a dynamical processes that affects the adoption of strategies on the macroscopic level of evolutionary game dynamics . as fig . [ fig : multilevel_selection ] illustrates , the latter manifests as multilevel selection @xcite that strongly promotes cooperation in all major types of social dilemmas . [ ! ht ] grid , obtained at an optimal ( @xmath70 ; top panel ) and a too large ( @xmath71 , bottom panel ) connectivity originating from the coevolutionary rule proposed by @xcite . full black ( open green ) boxes depict the positions of influential defectors ( cooperators ) while yellow ( white ) pixels depict the players who are within ( out of ) their range of influence . if the influential players are separated by large disjunct territories of influence ( bottom panel ) the network reciprocity is not functioning well.,width=264 ] conceptually fitting to the third type of interaction - updating coevolutionary rules [ type c ; see fig . [ fig : class_of_updates](c ) ] is also the model introduced by @xcite , where each player @xmath22 that successfully passes its strategy ( _ i.e. _ reproduces in a biological scenario ) is allowed to form a new link with one randomly selected neighbor from its current neighborhood , thereby increasing its degree @xmath25 by one . thus , successful players are allowed to grow compact large neighborhoods that are centered around their initial four nearest neighbors . as it is generally assumed , the payoff of any given player is accumulated from all the links with its neighbors . hence , without the normalization by degree , the more links a player has the higher its payoff is expected to be . for the sake of an easier depiction of player distributions , we start from an interaction graph that can be represented by a square lattice . evidently , the additions of new links will drive the initial topology away from two dimensions , yet still allowing us to capture relevant details of strategy distributions via a square lattice representation , as shown in fig . [ fig : coev_range ] . notably , the coevolutionary rule is independent on whether @xmath60 or @xmath46 , and can hence be considered as strategy independent . however , since the performance of the strategies is clearly definitive for who gets to make new links , the rule has at least conceptual similarities with some of the above - reviewed works that considered strategies as more directly decisive for the outcome of dynamical interactions . since the coevolutionary rule would eventually result in a fully connected graph ( the latter constitutes well - mixed conditions ) , the parameter @xmath72 was introduced as the maximal degree a player is allowed to obtain . accordingly , the process of making new connections is stopped as soon as the degree @xmath73 of a single player within the whole population reaches @xmath72 , whereby this limit prevents the formation of a homogeneous system and indeed constitutes the main parameter affecting the impact of the coevolutionary rule . starting from a square lattice , it was shown that intermediate values of @xmath74 substantially promote cooperation in the weak prisoner s dilemma game ( see fig . [ fig : schematic_games ] ) governed by eq . [ pach ] , which was attributed to the formation of highly heterogeneous interactions networks ensuring optimal transfer of information between influential players , _ i.e. _ those that have the highest degree among any other players that can adopt the strategy from the influential player via an elementary process . the coevolutionary rule was also tested against robustness to time scale separation between neighborhood growth and strategy adoption via the introduction of a parameter @xmath75 , defining the probability of degree extension after a successful strategy pass . evidently , @xmath76 recovers the originally proposed model while decreasing values of @xmath75 result in increasingly separated time scales . although the impact of @xmath75 was found depending somewhat on the temptation to defect @xmath77 , in general values of @xmath78 yielded insignificantly different results if compared to the @xmath76 case . note that @xmath79 corresponds to the spatial model without coevolution , and hence it is natural that as @xmath80 the promotion of cooperation was found fading . the success of intermediate values of @xmath72 in promoting cooperation can be explained based on the emergence of heterogeneous interaction networks and the disassortative mixing of high - degree nodes @xcite . in particular , while intermediate values of @xmath72 result in a highly degree - diverse mixture of players , which generally promotes cooperation [ see also @xcite ] , too large values of @xmath72 yield just a few influential players with disjunct clouds of homogeneous regions surrounding them , as shown in the bottom panel of fig . [ fig : coev_range ] . in the later case , the lack of information exchange between hubs ( influential players having large degree ) defectors can easily survive , thus resulting only in moderate cooperation levels . the top panel of fig . [ fig : coev_range ] , on the other hand , features an optimal distribution of influential players ( _ i.e. _ those having large degree ) , where high - degree cooperators can make cooperation prevail practically across the whole system . it is worth mentioning that the optimal level of cooperation observed for an intermediate value of @xmath72 in @xcite is conceptually similar to the case when an intermediate strength of information exchange between influential players yields the optimal environment for cooperation @xcite , as is illustrated in fig . [ fig : restricted ] . in the later case a fraction @xmath81 of players that are characterized with a larger teaching capability are allowed to temporarily link with distant opponents of the same kind with probability @xmath43 , thus introducing shortcut connections among the distinguished . these additional temporary connections are able to sustain cooperation throughout the whole range of the temptation to defect @xmath77 ( see fig . [ fig : schematic_games ] ) . as fig . [ fig : restricted ] demonstrates , only minute values of @xmath43 , constituting a moderate intensity of information exchange between influential players , warrant the best promotion of cooperation . as a function of @xmath43 in the ` connected influential players ' model @xcite . parameter @xmath43 determines the intensity of the information exchange between influential players . the fraction of influential players is @xmath82 and the temptation to defect is @xmath83 . the line is just to guide the eye.,width=283 ] strategy - independent coevolutionary rules affecting the interactions between players have also been considered in the context of distinguished players populating a square lattice @xcite , in the context of dynamically changing random and scale - free networks @xcite , as well as in the context of random networks with different assortative mixing emerging due to links adaptations @xcite . again , the interested reader is referred to the original works for further details , while here we proceed with a new section devoted to the review of coevolutionary rules introducing network growth . first , it is worth noting that coevolutionary rules giving rise to population growth have been considered much less frequently than the above - reviewed rules affecting solely how players link with one another . indeed , only three works fit into this subsection , the latter being the arxiv contribution by @xcite and two recent papers by @xcite . closely related to the rather general and broad interest in network growth @xcite , the networks formed by the players participating in evolutionary games can be subject to growth as well , with motivations equivalent to those of the broader research field . although never officially published , the work by @xcite should be acknowledged as being pioneering in raising the question how the dynamics of an evolutionary game might affect network growth , and how in turn the latter affects back the prevalence of the competing strategies . for this purpose , the authors proposed a so - called payoff - based preferential attachment rule under the guidance of the @xmath20-parameterized snowdrift game ( see fig . [ fig : schematic_games ] ) and the fermi strategy adoption rule given by eq . [ fermi ] . indeed , the preferential attachment rule by @xcite is practically identical to the seminal growth and preferential attachment model proposed by @xcite , only that in the former the probability of linking a new player to an existing player @xmath22 is not determined by its degree @xmath25 but rather by its accumulated payoff until that time . not surprisingly then ( note that in the absence of normalization , similarly as the degree of a player , its payoff will typically also increase by one during an update ) , the coevolutionary rule was found leading to the emergence of scale - free interaction networks that are characterized by the degree distribution @xmath84 ; the coefficient @xmath57 thereby depending on the scaling of the probability of linking a new player to an existing player . in accordance with an earlier study by @xcite , the emerging scale - free topology due to the coevolutionary rule was found highly beneficial for the evolution of cooperation in the snowdrift game . notably , the authors also investigated the average path length and the assortative mixing of the emerging networks , as well as the wealth distribution of players . the former two were found to be in agreement with observations from realistic social networks , while the latter was found consistent with the pareto law . the work by @xcite also introduces an evolutionary preferential attachment rule that is based on the payoffs of existing players , albeit the weak prisoner s dilemma is employed as the governing game and the strategy adoption probability is quantified according to eq . more precisely , the network growth starts with @xmath85 fully connected players and proceeds by adding a new player with @xmath86 links to the existing ones at equally spaced time intervals @xmath87 . the probability that any player @xmath22 , ( having payoff @xmath26 ) in the network receives one of the @xmath88 new links was defined as @xmath89 } \label{attach}\ ] ] where the sum runs over all the players forming the network at time @xmath90 . moreover , the parameter @xmath91 controls the weight of the payoffs during the network growth . for @xmath92 all nodes are equiprobable , corresponding to the weak selection limit [ see _ e.g. _ @xcite for recent works related to the latter ] , while for @xmath93 the players with the highest payoffs are much more likely to attract the newcomers . the authors also specified the time interval @xmath94 for payoff evaluations and potential strategy adoptions , focusing explicitly on @xmath95 ( typically @xmath96 , although smaller and larger values were also commented on ) , so that accordingly the network growth was considered to be faster than the evolutionary dynamics . it was shown that the weak selection limit results in networks having degree distributions with exponentially decaying tails , while the strong selection limit ( @xmath93 ) yields highly heterogeneous scale - free interaction networks . in agreement with the earlier findings obtained on static graphs @xcite , it was confirmed that higher levels of cooperation are attainable on heterogeneous rather than homogeneous topologies , albeit that the distribution of strategies with respect to the degree of nodes forming the network is different . more precisely , cooperators were not found occupying the main hubs as on static graphs , but rather the nodes with an intermediate degree , thus indicating that the interplay between the local structure of the network and the hierarchical organization of cooperation is guided by the competition between the network growth and the evolutionary dynamics . notably , similar differences in the microscopic organization of the steady state composition of strategies were found on static scale - free networks when the payoffs were subjected to normalization @xcite , although the discrepancies reported by @xcite were solely the consequence of the coevolutionary growth process . indeed , in a recent study @xcite this coevolutionary rule has been studied further to confirm that the reported promotion of cooperation hinges not only on the final heterogeneity of the resulting network but also vitally on the particularities of the growth process itself . in addition , it was shown that under strongly payoff dominated growth conditions so - called super - hubs can emerge , which attract most of the links from the other nodes . although under such conditions cooperation was found thriving even for high temptations to defect , it was also noted that the robustness of these findings may be compromised , or at least not so strong as on static scale - free networks @xcite , due to the extreme heterogeneity of the star - like structures that can be brought about by the coevolutionary network growth . with the above we conclude the review of coevolutionary rules affecting the interaction network , either in terms of links players form with one another ( see section [ interactions ] above ) or the actual number of players participating in the game and the related network size . we proceed with the review of coevolutionary rules affecting individual properties of players , such as their teaching activity ( see section [ teaching ] ) , mobility ( see section [ mobility ] ) or age ( see section [ aging ] ) . note , however , that some of the above - reviewed coevolutionary rules already incorporated and/or affected personal features of players , such as for example the loyalty to their partners @xcite or influence @xcite , albeit always in conjunction with the coevolution of the interaction network . in what follows , the links and the size of the network are not affected by the coevolutionary rules unless explicitly noted otherwise . heterogeneity of players has been explored as a beneficial condition for cooperation in several forms @xcite . it can be easily accepted that players are not perfectly identical within a population . some have higher reputation or stronger influence than others . these differences can be detected via a biased direction of strategy adoptions . more precisely , players with higher reputation can spread their strategy more easily than if having an average or low reputation . in other words , their activity to teach a neighbor a new strategy is higher . it turned out that one of the individual quantities that influences the evolution of cooperation most effectively is the teaching activity @xcite . notably , teaching activity can also be referred to as the influence or reproduction rate @xcite , with the logical assumption that influential individuals are much more likely to reproduce , _ i.e. _ have a higher teaching activity , than players with low influence . teaching activity ( or the synonyms we pointed out ) can be introduced into the framework of evolutionary game theory via a modified fermi strategy adoption rule @xmath97 } \label{teachfermi}\ ] ] where @xmath98 characterizes the strength of influence ( or teaching activity ) of player @xmath22 . obviously , @xmath99 for all @xmath22 returns eq . [ fermi ] , whereby it is important to acknowledge that even if @xmath100 but the same for all @xmath22 the evolutionary outcome of strategy abundance remains the same , only the relaxation times lengthen . quenched ( non - evolving ) distributions of @xmath98 may promote cooperation even on homogenous lattice - type interaction topologies @xcite , while their application on complex networks reveals further that players with large teaching activity play a similar role as hubs in highly degree heterogenous graphs , such as scale - free networks @xcite . we refer the reader to the original works for further details on models using quenched distributions of @xmath98 , while here we proceed with the review of the two papers by @xcite that thus far considered the teaching activity as an evolving property of individual players . ) , which were found beneficial for the evolution of cooperation irrespective of the underlying interaction network.,width=283 ] in a social context the strategy adoption can be considered as learning from the more successful player . accepting this point of view , it is straightforward to consider a player who has successfully passed a strategy as the one having a higher reputation , and thus a higher teaching activity than other players . implementing this idea into a coevolutionary rule , we proposed that whenever player @xmath22 successfully passes its strategy the influence @xmath98 increases by a constant positive value @xmath101 according to @xmath102 @xcite . this coevolutionary rule is illustrated in fig . [ fig : rule_evol_teach ] . it should be noted that in this model the term ` reputation ' does not necessarily have a positive meaning , and thus may be in contradiction with the same term used elsewhere @xcite , where players who cooperated were awarded a higher reputation , which expectedly yielded higher levels of cooperation . moreover , for the sake of simplicity it was assumed that the evolution of @xmath103 stops as soon as the highest @xmath98 reaches @xmath58 @xcite . starting from a nonpreferential setup , initially assigning @xmath104 to every player irrespective of its strategy , it was found that there exists an optimal intermediate value of @xmath105 for which cooperation in the weak prisoner s dilemma as well as the @xmath20-parameterized snowdrift game ( see fig . [ fig : schematic_games ] ) is enhanced best . it is in fact understandable that only an intermediate value of the increment @xmath106 was found warranting the optimal heterogeneity of the distribution of @xmath98 . namely , if @xmath106 is small then the values of @xmath98 simply increase homogeneously for all the players , while large values of @xmath106 result in a very quick halt of the coevolutionary process , either way resulting in a rather homogeneous distribution of the teaching activity . indeed , for both considered evolutionary games it was found that using moderate @xmath106 the final distribution of @xmath103 is exponential , in turn attributing the promotion of cooperation to the spontaneously emerging highly heterogenous plethora of differently influential players , as shown in fig . [ fig : teaching_dist ] . it was also shown that the effectiveness of the coevolutionary rule increases with the increasing uncertainty by strategy adoptions @xmath32 , and that the rule is robust to variations of the updating scheme . for example , it was verified that an alternative coevolutionary rule , by which @xmath98 was allowed to grow also past @xmath58 only that then @xmath98 was normalized according to @xmath107 ( @xmath108 being the maximal out of all @xmath98 at any given time ) to ensure that the teaching activity remained bounded to the unit interval , yielded similar results as the halted version . as a result of the coevolutionary rule introduced by @xcite . results were obtained for the weak prisoner s dilemma game staged on a square lattice . parameter values were : @xmath109 , @xmath110 and @xmath111.,width=283 ] a two - fold extension of the above work was made in @xcite . first , the coevolutionary rule was no longer considered to be strategy independent . note that in the preceding work @xmath102 was executed irrespective of the strategy of player @xmath22 . conversely , in @xcite this rule was applied separately either only for @xmath60 ( cooperators ) or only for @xmath112 ( defectors ) . second , the evolution of cooperation was examined in all three major social dilemma types defined on the @xmath21 parameter plane ( see fig . [ fig : schematic_games ] ) . it was shown that both versions of the coevolutionary rule promote cooperation irrespective of the underlying game . opposite to intuitive reasoning , however , it was revealed that the exclusive coevolutionary promotion of players spreading defection is more beneficial for cooperation than the likewise direct promotion of cooperators . this was attributed to the fact that the coevolutionary promotion of defectors results in a larger fraction of players that are at least once affected by the coevolution , ultimately leading to a stronger segregation of the population into active ( those having @xmath113 ; note that the latter is the initial teaching activity assigned to all ) and virtually ( or comparably ) inactive ( those having @xmath114 ) players than the coevolutionary rule affecting cooperators . according to previous findings on the impact of static distributions of heterogeneity @xcite , the stronger expressed segregation was found directly responsible for the better promotion of cooperation when defectors rather than cooperators were subjected to coevolution . as we have already mentioned ( see _ e.g. _ section [ interactions ] ) , the time scale separation of coevolutionary processes may decisively affect the final output of such models . this was observed for the coevolution of teaching activity as well . more precisely , the time scale separation between the coevolution of teaching activity and strategy adoption can be tuned via the introduction of a parameter @xmath75 , defining the probability of increasing @xmath98 after a successful strategy pass . evidently , @xmath76 recovers the two originally proposed models while decreasing values of @xmath75 result in increasingly separated time scales . although the impact of @xmath75 was found depending somewhat on the type of the considered coevolutionary rule , in general , values of @xmath115 yielded insignificantly different results if compared to the @xmath76 case , thus indicating that the findings are robust to this type of alterations [ note that , as in @xcite , @xmath79 corresponds to the model without coevolution , and hence it is natural that as @xmath80 the promotion of cooperation was found fading ] . more precisely , however , since the fraction of cooperators was found increasing rather steadily with increasing values of @xmath75 , especially for the coevolutionary rule affecting defectors , it is optimal to keep the coevolutionary process affecting the teaching activity of players paced similarly fast as the main evolution of strategies , _ i.e. _ @xmath116 . we thus emphasize , that the above - reviewed coevolutionary models affecting the teaching activity have revealed that a simple ` successful become more successful ' principle can result in a heterogenous hierarchy of individual properties of players , such that optimal conditions for the evolution of cooperation are warranted . a similarly positive impact of heterogeneity on the spread of the cooperative strategy was also detected on heterogenous interaction networks @xcite , hence conceptually linking these two seemingly disjoint promoters of cooperation . it was acknowledged already by @xcite that players finding themselves in an unprofitable or undesirable situation frequently choose moving in order to free themselves from the negative consequences of that situation . accordingly , mobility can be considered as being a coevolutionary process in the sense of strategy and/or position alterations that ultimately determine the environment of players . although we were unable to locate coevolutionary terminology associated with mobility , we review here advances on this topic made during the last decade , and indeed consider movements of players during the evolution of strategies as being guided by rules of coevolution . the impact of diffusion on the outcome of a spatial prisoner s dilemma game via empty sites was questioned first by @xcite . therein , weak quenched disorder introduced in the form of empty sites on a square lattice was found beneficial for cooperation in the prisoner s dilemma game subject to the richest - following strategy update rule . in a follow - up paper @xcite the approach was extended to allow diffusion of players to nearest - neighbor empty sites with a certain probability . in particular , two ways of implementing the mobility were considered . first , each player was allowed to make an attempt at moving only after payoff accumulation and potential strategy adaptations were executed in parallel , or second , the moving was attempted prior to the evolution of strategies . importantly , the moving of players was considered to be brownian random walk like , _ i.e. _ diffusive , not relying on any type of explicit , genotypic or phenotypic assortment , and also being strategy - independent . due to this minimalist set - up the study provided rather general insights into possible effects of mobility . it was shown that mobility may indeed promote cooperation since it increases the ability of cooperator clusters to invade and overtake isolated defectors . on the other hand , mobility may also allow defectors to escape retaliation from a former partner and lead to stronger mixing in a population due to increasing interaction ranges of players , both of which are known to damp the evolutionary success of cooperators . thus , the impact of mobility in the form introduced by @xcite is not clear cut . as noted by the authors , further work on this is in progress . importantly , it was also emphasized that mobility may be subject to more deliberate coevolutionary rules , taking into account personal preferences of players , their strategies , as well as aims . an example of the latter was studied by @xcite , who introduced success - driven migration as a possible mediator leading towards cooperation in populations of selfish and unrelated individuals even under noisy conditions . in particular , success - driven migration [ see also @xcite ] was implemented so that , before the strategy adoption , player @xmath22 was allowed to explore potential payoffs that it would receive if occupying one of the empty sites in the migration neighborhood . the latter was typically restrained to nearest and next - nearest neighbors of player @xmath22 . if the potential payoff was found to be higher than in the current location , player @xmath22 moved to the site offering the highest payoff and , in case of several sites with the same payoff , to the closest one . on the other hand , if the current location offered the highest payoff among all the empty sites within the migration neighborhood , player @xmath22 did not move . it was found that this fairly simple and very plausible migration rule promotes cooperation in the prisoner s dilemma game on a square lattice ( with a fraction of empty sites to accommodate moving ) irrespective of the noise introduced to the system . in fact , three types of noise were considered to attest to the robustness of cooperation facilitation due to the introduced mobility of players . first was the introduction of mutation with probability @xmath75 , second was the introduction of random movements not considering the expected success ( payoff ) with probability @xmath20 , while third was the combination of the two . additionally , different update rules , adding birth and death processes , as well as introducing a small fraction of individuals defecting unconditionally were considered as well . irrespective of all these factors cooperation was found prevailing in a large region of the parameter space defining the prisoner s dilemma game if only the players were allowed to execute success - driven migration . recently , mobility is getting increasing attention as a means to promote cooperation in social dilemmas @xcite . in the model of @xcite two types of players are introduced , and a random walk of the influential individuals is possible irrespective of their strategies . the mobility of influential cooperators can have two positive impacts on the evolution of cooperation . first , they can spread the cooperative strategy among the non - influential players having a lower teaching activity , and second , when two influential players with opposite strategies meet , the cooperator can prevail and thus ensure an effective information exchange between cooperating hubs . note that the importance of the latter has as already been emphasized in section [ interactions ] ( see _ e.g. _ fig . [ fig : restricted ] and the pertaining text ) . as expected based on preceding works considering mobility as a coevolutionary process , the final outcome of the competition between mobile , and thus influential , cooperators and defectors is highly sensitive to changes in the speed of moving . when the latter is too high the influential cooperators can not benefit from their cooperative neighborhoods because they abandon them too soon . similarly , influential defectors eschew the negative feedback effect originating from defecting neighbors ( note that @xmath56 links are nonprofitable for both players ) because they leave them too fast as well . indeed , high moving speeds generate conditions mimicking the well - mixed regime which is damning for cooperators . thus , only moderate mobility of influential players has been found effectively supporting the evolution of cooperation . we refer the reader to the original works for further details , noting that mobility seems a promising avenue of research for future explorations of coevolutionary rules . as the last coevolutionary process we consider aging . indeed , aging is always present , tailoring our interactions with others and postulating a finite lifespan during which we are able to exercise them . it thus seems natural to consider aging as an integral part of every evolutionary development , and certainly evolutionary games constitute a prime example thereof . nonetheless , studies taking aging explicitly into account within this realm of research are few . @xcite recently noted that lifespan might play an important role in the evolution of cooperation , albeit their study focused on the coevolution of choosiness ( see also section [ related ] ) rather than age . moreover , concepts similar to aging were recently introduced in the voter model @xcite , showing that age and memory - dependent transition rates can have a positive effect on consensus formation . a recent work focusing on aging within evolutionary games is due to @xcite , and in the following we present a summary of the proposed coevolutionary rules for aging as well as their main implications . the newborn is not supported by its ancestor ( the @xmath56 link is detrimental for both ) , and hence a neighboring cooperator with high age can conquer the site of the newborn defector . this procedure occurs repeatedly , ultimately resulting in a practically blocked ( more precisely an oscillating ) front between @xmath45 and @xmath46 regions ( left panel ) . importantly , a similar blocking is not present around old ( and thus influential ) cooperators because their cooperator - cooperator links help newborn cooperators to achieve higher age , in turn supporting the overall maintenance of cooperative behavior ( right panel ) . in both panels the size of players is proportional to their age ( _ i.e. _ their teaching activity ) . dashed ( solid ) arrows denote attempted ( successful ) strategy adoption processes.,width=283 ] since age is often associated with knowledge and wisdom an individual is able to accumulate over the years , it was introduced through a simple tunable function that maps age to teaching activity ( see section [ teaching ] ) of the corresponding player . more precisely , @xmath98 in eq . [ teachfermi ] was related to the integer age @xmath117 in accordance with the function @xmath118 , where @xmath119 , denoting the maximal possible age of a player , serves the bounding of @xmath98 to the unit interval , and @xmath120 determines the level of heterogeneity in the @xmath121 mapping . evidently , @xmath122 corresponds to the classical ( homogeneous ) spatial model with @xmath99 characterizing all players , @xmath123 ensures that @xmath98 and @xmath124 have the same distribution , whereas values of @xmath125 impose a power law distribution of strategy transfer capability . although different age distributions of players were considered also as quenched system states , the focus was on the study of aging as a coevolutionary process , entailing death and newborns . two rules were considered separately , both starting with @xmath124 being assigned randomly from a uniform distribution within the interval @xmath126 $ ] to all players , and subsequently increasing all @xmath124 by @xmath58 after each full monte carlo step . also inherent to both was that @xmath124 was set to zero for all players whose age exceeded @xmath127 ( effectively this means that a newborn follows the dead player ) . the difference was in the way age of players that have just adopted a new strategy from one of their neighbors was handled . in the coevolutionary model a their age was left unchanged , while in the coevolutionary model b they were considered as newborns , _ i.e. _ as soon as player @xmath22 adopted a new strategy its age was set to @xmath128 . notably , rules a and b can be interpreted rather differently . from a purely biological viewpoint the more successful player replaces the neighbor with its own offspring , who therefore initially has a limited strategy transfer capability , which corresponds to rule b. on the other hand , especially in social systems , strategy adoptions may not necessarily involve death and newborns , but may indicate solely a change of heart , preference , or way of thinking , whereby this situation corresponds to rule a. nevertheless , newborns in a social context can be considered those that changed their strategy recently , and therefore have a low reputation initially . interestingly , it was found that the small difference between coevolutionary rules a and b may have significant consequences for the evolution of cooperation . foremost , it was found that rule b promotes cooperation remarkably better than rule a. however , the difference could not be explained by the resulting heterogeneity of the distributions of @xmath98 , for example via a similar reasoning as introduced by @xcite , since both rules return power law distributed values with rather similar slopes ( @xmath129 for a and @xmath130 for b ) . in fact , it was shown that the coevolutionary rule b introduces a new powerful mechanism for promotion of cooperation acting solely on a microscopic player - to - player basis , and as such is thus virtually not detectable by statistical methods assessing the heterogeneity of the system . the mechanism was found relying on a highly selective promotion of cooperator - cooperator and defector - defector pairs , which hinders influential defectors ( those having @xmath124 close to @xmath127 ) to spread their strategy effectively across the spatial grid . in particular , rule b always leads to influential players being surrounded by newborns . thereby it is important to note that whenever an old defector , with a high strategy transfer capability @xmath98 , is imitated by one of the neighbors , further spreading of defection is blocked because the newborn defector has no chance to pass strategy @xmath46 further . at that time a neighboring cooperator with high age can strike back and conquer the site of the newborn defector . as a result the whole procedure starts again , which ultimately results in a practically blocked ( more precisely an oscillating ) front between @xmath45 and @xmath46 regions . crucially , a similar blocking mechanism is not present around old ( and thus influential ) cooperators because their cooperator - cooperator links help newborn cooperators to achieve higher age , in turn supporting the overall maintenance of cooperative behavior . the main differences in the propagation of different strategy pair - ups are summarized in fig . [ fig : blocked ] , while an example of the resulting spatial distribution of players is presented in fig . [ fig : aging_dist ] . in the latter a player is considered as influential if its age exceeds that of any of its neighbors by at least @xmath131 ( qualitatively similar snapshots can be obtained by choosing different thresholds as well ) . square lattice , obtained by considering players who have adopted a new strategy as newborns [ coevolutionary rule b in @xcite ] . full black ( open green ) boxes depict influential defectors ( cooperators ) while yellow and white are all other non - influential defectors and cooperators , respectively . the snapshot demonstrates clearly that the propagation of defectors is blocked in space as a consequence of age - related teaching activity.,width=264 ] in model a the situation is significantly different since cooperative domains , created around old players with high @xmath98 , can not prevail long . namely , the central cooperator who built up the cooperative domain eventually dies , and the arriving newborn with an accordingly low strategy transfer capability simply can not maintain this domain further , thus giving defectors an opportunity to win it over . as a consequence of the dynamical origin of the observed cooperation - promoting mechanism brought about by rule b , it is expected that it will work in other cases too , for example when the interaction graph is characterized by a different topology , by other evolutionary games , or by separated time scales between aging and strategy adaptations , as was shown already by @xcite . finally , it is important to note that the observed cooperation - promoting mechanism relying on a dynamical process is robust even if non - monotonous mappings between @xmath124 and @xmath98 are considered ( the oldest individuals may not necessarily be the most influential ) . indeed , the promotion of cooperation remains intact as long as the plausible assumption that very young players should have none or very little influence is adhered to . similarly as mobility reviewed in section [ mobility ] , we note that aging as a coevolutionary process seems very liable to further studies as well , and we hope this brief summary succeeded in wetting the appetite for them . aside from thus far reviewed coevolutionary rules , there exist examples @xcite we were unable to classify into the above subsections . without going into much details as it exceeds the scope of this mini review , we briefly describe some of these related approaches , but refer the reader to the original works for further details . @xcite , for example , studied the simultaneous evolution of learning rules and strategies , whereby the former were determined endogenously based on the success of strategies observed in the neighborhood of any given player . it was shown that endogenous learning rules put more weight on the proper understanding of each player s own experience rather than on the experience of an observed neighbor . coevolving learning rules were recently considered also by @xcite , showing that imitation is frequently displaced by replication , in turn leading to a rapid decrease of cooperation in the spatial prisoner s dilemma game . on the other hand , imitation was found to be superior to global but stochastic imitation , thereby facilitating cooperative behavior . the coevolutionary selection of strategy adoption rules was consider by @xcite as well , where the uncertainty in the fermi function ( see eq . [ fermi ] ) was subject to evolution as a player - specific property . in particular , instead of a single @xmath32 value authors introduced different @xmath132 values where @xmath133 , which were then assigned randomly to the players . as we have already noted in section [ games ] , the uncertainty by strategy adoptions can originate from different sources , ranging from unpredictable variations in payoffs to errors in the decision making @xcite . the parameter @xmath32 , however , can also be considered as characterizing the willingness of a player to risk a payoff quantity during a strategy change . therefore , by using different values of @xmath32 , not only the better strategy but also the way of strategy adoption can be the subject of an imitation process . values ( @xmath134 ) demonstrate the selection of the most appropriate strategy adoption uncertainty @xmath135 ( promoting cooperation best ) , as indicated by the red solid line . this directly implies the spontaneous selection of an optimal strategy adoption rule within the scope of the fermi function ( see eq . [ fermi ] ) , as reported by @xcite . dashed blue lines depicted the extinction of the other @xmath132 values . initially , @xmath136 different @xmath132 values were distributed on the square lattice of size @xmath137 . the temptation to defect was @xmath109.,width=283 ] accordingly , aside from the fact that player @xmath28 could adopt the strategy of player @xmath22 according to eq . [ fermi ] ( @xmath138 ) , an additional independent trail with the same probability was made also for the adoption of the imitation rule ( @xmath139 ) . it was shown that , if the system is seeded by random initial conditions , the proposed coevolutionary rule drives the system towards a state where a single @xmath135 value ( only one of the initial @xmath132 values ) prevails . this final strategy adoption uncertainty is closely related with the parameter value warranting the highest cooperation level if a given value of @xmath32 is used for all players at a certain value of the temptation to defect @xmath77 ( see fig . [ fig : schematic_games ] ) . naturally , the prevailing @xmath135 value thus depends also on the topology of the interaction network . the selection process is illustrated in fig . [ fig : n20_time ] , where @xmath136 different @xmath132 values were initially assigned to the players @xmath140 . summarizing the main observation , it was shown that a darwinian selection rule affecting a model parameter can spontaneously lead to the prevalence of the value that ensures an optimal level of cooperation in the system . for further details we refer the reader to the original work of @xcite . the evolution of altruistic behavior under coevolutionary rules was studied also in what can be considered more explicitly biologically or even humanly motivated settings @xcite . for example , internal norms , being a pattern of behavior enforced in part by internal sanctions , such as shame , guilt and loss of self - esteem , were found to provide support for the evolution of altruistic norms , and moreover , via a gene - culture coevolution argument an explanation was provided as to why individually fitness - reducing internal norms are likely to be prosocial rather than socially harmful @xcite . although mentioned already in section [ aging ] , the study of @xcite in fact focuses on the coevolution of choosiness , the later relying on cooperativeness being used by other individuals as a choice criterion . in such a setting competition to be more generous than others can emerge , and in this case the evolution of cooperation between unrelated individuals can be driven by a positive feedback between increasing levels of cooperativeness and choosiness . it was shown that , in situations where individuals have the opportunity to engage in repeated pairwise interactions , the evolution of cooperation depends critically on the amount of behavioral variation that is being maintained in the population by processes such as mutation . finally , we note that @xcite investigated the evolution of cooperation in a so - called nongrowth dynamic network model with a death - birth dynamics based on tournament selection , @xcite as well as @xcite studied the coevolution of group structure rather than graph structure in the context of generalized reciprocity , while @xcite considered evolving heterogeneous games as means to sustain cooperation . interestingly , in the later study the players had individual payoff elements assigned to them for calculating their final payoffs . accordingly , within the realm of the proposed coevolutionary rule a player could adopt not only the strategy of the neighbor but also its individual payoff matrix elements . it was found that if starting with a random heterogeneous distribution of payoffs , eventually only a small number of definite payoff matrices remained while the others went ` extinct ' . on the other hand , if the initial rank of individual payoff elements agreed with those constituting social dilemma games , the prevailing one was found to be the stag - hunt game . considering the latter result in the light of findings reported within the evolving adoption rules model by @xcite , it is possible to raise the question if coevolutionary rules as a selection mechanism can spontaneously drive the system into a state where mutual cooperation ensures the maximal average payoff . indeed , further studies are necessary to clarify this issue . in sum , there are few boundaries to imagination when considering what coevolutionary rules might affect , and certainly , it seems like all facets of existence can be brought into consideration . as we hope the above mini review on coevolutionary games clearly shows , coevolution is certainly a promising concept to follow , as it constitutes the most natural upgrade of evolutionary games in the sense that not only do the strategies evolve in time , but so does the environment , and indeed many other factors that in turn affect back the outcome of the evolution of strategies . some of these coevolutionary processes are of a finite duration , and thus on their own do not necessarily affect the outcome of evolutionary games but do this only indirectly due to the environment that they produce , while others are lasting , introducing dynamical alterations that affect the evolution of cooperation on a continuous basis . in the future , it should be of interest to further elaborate on the question whether coevolution itself may promote cooperation by introducing dynamical mechanisms , or if mainly the final outcome of a coevolutionary process , if it exists , is the one vital for the sustenance of cooperation . often , however , it is the interplay of both that facilitates the promotion of cooperation , as it was already shown in some of the works . it should also be considered which coevolutionary processes end sooner or later , and which are those that at least in principle should last forever . for example , the growth of a city can be considered something that has a finite duration due to environmental constrains , while aging , on the other hand , is a natural ingredient of every living organism , and as such it should only make sense to consider it evolving permanently . although social dilemmas may emerge at different levels of human and animal interactions , their occurrence is by no means limited to these examples . the applicability of the concept of evolutionary games extends across the whole of social and natural sciences , with examples ranging from the rna virus @xcite , atp - producing pathways @xcite and biochemical systems @xcite , to traffic congestion @xcite and climate change @xcite , to name but a few . in this sense coevolutionary rules should be applied to evolutionary games in the broadest possible sense , with specially adapted motivation fitting to the research avenue of the main evolutionary process . moreover , while focusing predominantly on resolving social dilemmas , coevolutionary rules have thus far not been considered for many other game types , as for example the public goods game , the ultimatum game or the rock - scissors - paper game . these gaps should be interesting to fill as well , in particular when striving towards universal concepts underlying cooperation in the broadest possible sense . the authors acknowledge support from the slovenian research agency ( grant z1 - 2032 ) , the hungarian national research fund ( grant k-73449 ) , the bolyai research grant , and the slovene - hungarian bilateral incentive ( grant bi - hu/09 - 10 - 001 ) . chen , y. , qin , s .- , yu , l. , zhang , s. , 2008 . emergence of synchronization induced by the interplay between two prisoner s dilemma games with volunteering in small - world networks . e 77 , 032103 . helbing , d. , schnhof , m. , stark , h .- u . , holyst , j. a. , 2005 . how individuals learn to take turns : emergence of alternating cooperation in a congestion game and the prisoner s dilemma . adv . complex syst . 8 , 87116 . holme , p. , trusina , a. , kim , b. j. , minnhagen , p. , 2003 . prisoner s dilemma in real - world acquaintance networks : spikes and quasiequilibria induced by the interplay between structure and dynamics . e 68 , 030901 . kim , b. j. , trusina , a. , holme , p. , minnhagen , p. , chung , j. s. , choi , m. y. , 2002 . dynamic instabilities induced by asymmetric influence : prisoner s dilemma game in small - world networks . e 66 , 021907 . milinski , m. , sommerfeld , r. d. , krambeck , h .- j . , reed , f. a. , marotzke , j. , 2008 . the collective - risk social dilemma and the prevention of simulated danderous climate change . natl . acad . usa 105 , 22912294 . pacheco , j. m. , santos , f. c. , 2005 . network dependence of the dilemmas of cooperation . in : mendes , j. f. f. ( ed . ) , science of complex networks : from biology to the intertnet and www , aip conf . aip , melville , ny , pp . 90100 . zimmermann , m. g. , eguluz , v. m. , miguel , m. s. , 2001 . cooperation , adaption and the emergence of leadership . in : zimmermann , j. b. , kirman , a. ( eds . ) , economics and heterogeneous interacting agents . springer , berlin , pp . 7386 .

astrometric space missions , especially the esa ( european space agency ) cornerstone mission gaia , see e.g. _ perryman , et al . _ @xcite , are in preparation to attain microarcsecond ( @xmath2 ) level of accuracy in absolute positional measurements of stars and other celestial objects . this unprecedented accuracy of astrometric observations makes it necessary to account for many subtle effects which were totally negligible before . a practical model for astrometric observations with an accuracy of @xmath3 has been formulated by _ @xcite , where the influence of the gravitational fields inside the solar system were taken into account . furthermore , a number of additional effects potentially observable at this level of accuracy due to various gravitational fields generated outside of the solar system were also briefly discussed in that investigation . one of them is the gravitational light deflection of one companion of a binary system in the gravitational field of the other companion ( without loss of generality , throughout the investigation the light deflection of component b at component a is considered , hence , component a is considered to be the massive body , while component b is the light - source ) . this effect would change the apparent position of one component of a binary system . while it is clear that this effect is relatively small and even at the level of @xmath3 observable only for edge - on binary systems , the huge amount of binaries generate some hope that there are relevant systems where this light deflection effect becomes detectable . for instance , gaia will observe @xmath4 stars brighter than @xmath5 apparent magnitude . detailed numerical simulations predict the detection of about @xmath6 ( resolved , astrometric , eclipsing , spectroscopic ) binary systems by gaia mission , see _ zwitter & munari _ @xcite , which is a considerable increase compared to the @xmath7 binary systems known so far , see `` _ washington double star catalog _ '' @xcite . let us consider this argument a bit more quantitatively . for a simple estimate of the expected order of magnitude in light deflection , the classical lens equation ( in the form given by eq.(67 ) in _ fritelli , et al . _ @xcite , eq . ( 24 ) in _ bozza _ @xcite or eq . ( 23 ) in _ zschocke _ @xcite ) is applied . in terms of orbital elements of a binary system it can be written as follows : @xmath8 here , @xmath9 is the light deflectin angle , @xmath10 is the inclination , @xmath11 is the semi - major axis and @xmath12 is the distance of center - of - mass of the binary system from the observer , and the schwarzschild radius is @xmath13 , where @xmath14 is the gravitational constant and @xmath15 is the speed of light , and @xmath16 is the stellar mass of component a of the binary system . from ( [ classical_lens_1 ] ) one obtains the maximal possible light deflection for edge - on binaries , i.e. attained for the case when the inclination is exactly @xmath17 : @xmath18 here , @xmath19 is the astronomical unit , @xmath20 stands for parallax of one arcsecond , and @xmath21 is the solar mass . according to this formula , the choice of moderate values like @xmath22 and @xmath23 would result into significant light deflection effect on microarcsecond level even at large distances of about @xmath24 . a meaningful value for the density of stars in the solar neighborhood , @xmath25 , implies already @xmath7 binaries inside a sphere of @xmath26 . thus , one might conclude among them there are a few relevant edge - on binary systems . however , in order to estimate quantitatively the number of relevant systems , one needs to know the probability for the occurrence of such edge - on binaries which have a given light deflection depending on their orbital parameters like inclination , mass , distance and semi - major axis . such a relation between a given light deflection and orbital parameters is given by a so - called inclination formula . since the inclinations of binary systems are of course randomly distributed , it is meaningful to resolve such an inclination formula in terms of inclination . as it has been shown by _ klioner , et al . _ @xcite ( see appendix [ appendix_kms ] for some basic steps ) , such an inclination formula can be obtained by means of the analytical solution of light deflection in standard post - newtonian approach , and is given as follows : @xmath27 for a better illustration of the inclination formula , relation ( [ kms_d ] ) is rewritten in terms of angular degrees instead of radians : @xmath28 where also @xmath29 has been used . according to this relation , the inclination @xmath10 of a binary system with stellar mass @xmath16 and at distance @xmath12 must not deviate from the edge - on value @xmath17 too much in order to have a given light deflection @xmath9 . for example , for a hypothetical binary star with @xmath22 situated at a distance of @xmath30 the light deflection effect attains @xmath3 only if @xmath31 , which means that the probability to observe this binary at a favorable inclination is only about @xmath32 . but even this estimation is still much too optimistic . as a concrete example of todays high - precision astrometry , let us consider one important parameter about the astrometric accuracy of the gaia mission : the accuracy of one individual positional measurement in the most ideal case ( bright star , i.e. @xmath33 magnitude ) amounts to be @xmath34 , which implies @xmath35 . furthermore , inside a sphere of @xmath36 around the sun almost every star and binary system is known already by the data of _ research consortium on nearby stars ( recons ) _ @xcite . since inside that sphere there is no binary system having a light deflection on microarcsecond level , one has to take at least @xmath37 . by taking into account these both remarks , one obtains @xmath38 in relation ( [ condition_0 ] ) . therefore , even in the best case one has to conclude @xmath39 , that means ( for @xmath22 ) , the probability to observe such a binary at a favorable inclination is practically only about @xmath40 . and the binaries must be , in fact , almost edge - on in order to have a light deflection which can be detected by todays astrometry . accordingly , while relation ( [ classical_lens_2 ] ) triggers the hope about the existence of many relevant binary systems , from relation ( [ condition_0 ] ) one concludes that the number of relevant binary systems is considerably reduced . an estimation of the total amount of binaries depends on many different parameters , like mass , semi - major axis , inclination and distances of the binaries . therefore , a simple estimation is not so straightforward as one might believe . moreover , the relation ( [ classical_lens_2 ] ) has been obtained with the aid of classical lens equation , while relation ( [ condition_0 ] ) has been obtain by means of standard post - newtonian approach . these both approaches have different regions of validity . however , a rigorous treatment of the problem of light deflection in binary systems implies the need of an analytical formula which is valid for such kind of extreme astrometric configurations like the binary systems are . recently , a generalized lens equation has been derived by _ zschocke _ @xcite , which allows to determine the light deflection of binary systems on microarcsecond level . one aim of this study is , therefore , to reobtain the criteria ( [ classical_lens_2 ] ) and ( [ condition_0 ] ) as stringent conditions from one and the same approach , i.e. with the aid of generalized lens equation . this is possible , because in the corresponding limits the generalized lens equation agrees with the classical lens equation and the standard post - newtonian solution . another aim is , to derive an inclination formula like ( [ kms_d ] ) for binary systems which allows to determine the needed inclination for a given light deflection angle and as a function of the orbital parameters . for that one has to take into account the distribution of stellar masses and the distribution of semi - major axes in binary systems . finally , the aim of this study is , to determine the total number of relevant binaries having a light deflection on microarcsecond level , and to investigate the possibility to detect this effect of light deflection by todays high precision astrometry . the article is organized as follows : in section [ section1 ] some basics about orbital elements of binary systems are given . the generalized lens equation and the inclination formula are presented in section [ section2 ] . in section [ condition1 ] two stringent conditions on the orbital elements of binary systems ( astrometric , spectroscopic , eclipsing and resolved binaries ) are presented , which allow to determine whether or not the binary system will have a light deflection of a given magnitude . the total number of binaries which have a given light deflection for an infinite time of observation is estimated in section [ section3a ] , while the more practical case of a finite time of observation is considered in section [ section3b ] . the special case of resolved binaries is considered in section [ condition2 ] . for that , the specific instrumentation of gaia mission is considered in some detail as the most modern astrometric mission with todays highest possible accuracy . a summary is given in section [ summary ] . consider a binary system , component @xmath11 with mass @xmath41 at coordinate @xmath42 and component @xmath43 with mass @xmath44 at coordinate @xmath45 . in order to express the light deflection effect in terms of orbital elements , spherical coordinates are introduced , illustrated by fig . [ fig : orbital_elements ] . , semi - major axis @xmath11 , inclination @xmath10 , eccentricity @xmath46 , eccentric anomaly @xmath47 , periapsis @xmath48 , and mass ration @xmath49 . the orbit of the binary system spans the @xmath50-plane and the @xmath51-axis is perpendicular to the orbital plane . the @xmath52-axis is oriented along the semi - major axis of the orbit of the binary system , while the @xmath53-axis is perpendicular to the @xmath52-axis . the vector @xmath54 is directed from the center - of - mass ( cms ) of the binary system , see eq . ( [ inclination_3 ] ) , to the observer . the center of spherical coordinate system is located at the cms of binary system , i.e. @xmath55 . the inclination @xmath56 is the angle between @xmath54 and @xmath51-axis ; @xmath57 is called edge - on and @xmath58 corresponds to retrograde orbit . the dotted line indicates the projection of @xmath54 onto orbital @xmath50-plane , i.e. z - component of @xmath54 equals zero . the angle between this projection and @xmath52-axis is called argument of periapsis @xmath59 . the orbital elements semi - major axis @xmath11 , eccentricity @xmath60 and mass ratio @xmath49 govern uniquely the geometric shape of both ellipses . the eccentric anomaly @xmath61 ( not plotted here ) , is defined in eq . ( [ eccentric_anomaly_5 ] ) of appendix [ appendixb ] and determines the actual position of the bodies a and b on their orbit.,width=453 ] the center of coordinate system is located at the cms , i.e. @xmath62 thus , the vector @xmath54 , which points from cms to the observer , is given by @xmath63{c } \displaystyle r\,\cos \omega\;\sin i \\ \nonumber\\ \displaystyle r\,\sin \omega\;\sin i \\ \nonumber\\ \displaystyle r\,\cos i \end{array}\right)\ , , \label{inclination_5}\ ] ] where @xmath64 , the argument of periapsis is denoted by @xmath48 , and @xmath10 is the inclination , see fig . [ fig : orbital_elements ] . the solution of equation of motion yields for vectors @xmath42 and @xmath45 the expression given by eqs . ( [ appendixb_56 ] ) - ( [ appendixb_62 ] ) . the vector @xmath65 points from the mass center of massive body to the observer , and vector @xmath66 points from the mass center of massive body to the source , see also fig . [ fig : orbital_elements ] . the coordinates of these vectors can be expressed by the orbital elements of the binary star as follows : @xmath67{c } \displaystyle r\,\cos \omega\;\sin i - \frac{a\ , \left(\cos e - { \rm e } \right)}{1 + \frac{\displaystyle m_a}{\displaystyle m_b}}\\ \nonumber\\ \displaystyle r\,\sin \omega\;\sin i - \frac{a\,\sqrt{1 - { \rm e}^2}\sin e}{1 + \frac{\displaystyle m_a}{\displaystyle m_b}}\\ \nonumber\\ \displaystyle r\,\cos i \end{array}\right)\ , , \label{inclination_10}\ ] ] @xmath68{c } a \ ; \left(\cos e - { \rm e } \right ) \\ \nonumber\\ \displaystyle a\ ; \sqrt{1 - { \rm e}^2 } \sin e \\ \nonumber\\ \displaystyle 0 \end{array}\right)\ , . \label{inclination_15}\ ] ] here , @xmath11 is the semi - major axis , @xmath46 is the eccentricity , and @xmath47 is the eccentric anomaly , see appendix [ appendixb ] . the vectors ( [ inclination_10 ] ) and ( [ inclination_15 ] ) will be used to express the light deflection in terms of orbital elements of the binary system . a scheme of light propagation of a signal emitted at component b in the gravitational field of component a is shown in fig . [ fig : distance1 ] . the vector @xmath65 points from the mass center of massive body to the observer , and vector @xmath66 points from the mass center of massive body to the source , and we define @xmath69 , the absolute value @xmath70 and unit vector by @xmath71 . furthermore , the impact vector @xmath72 is defined , and its absolute value is denoted by @xmath73 . the schwarzschild radius of massive body , i.e. of component a of the binary system is denoted by @xmath74 . for determining the light deflection in weak gravitational fields there are two essential approaches : standard post - newtonian approach , e.g. _ brumberg _ @xcite , and classical lens equation e.g. eq.(67 ) in _ fritelli , et al . _ @xcite , eq . ( 24 ) in _ bozza _ @xcite , or eq . ( 23 ) in _ zschocke _ @xcite . while the first approach is restricted by the condition @xmath75 , the second approach is only valid for the case that source and observer are far from the massive body , especially for @xmath76 and @xmath77 ( for geometrical illustration of @xmath78 , @xmath79 and @xmath80 see fig . [ fig : distance1 ] ) . however , in binary systems extreme configurations are possible like @xmath81 or @xmath82 . therefore , in order to investigate the light deflection in binary systems one needs a generalized lens equation which is valid in such extreme configurations where the standard post - newtonian approach as well as the classical lens equation can not be applied . recently , _ zschocke _ @xcite has derived a generalized lens equation , which allows to determine the light deflection in such extreme astrometric configurations like the binary systems are : @xmath83 actually , the lens equation has two solutions , but here only one solution is considered , while the second solution represents just the second image of one and the same source which is not relevant in our investigation . the generalized lens equation is valid up to terms of the order @xmath84 , and the absolute value of their total sum can be shown to be smaller or equal to @xmath85 . here , @xmath86 is chandrasekhar s impact parameter , see _ chandrasekhar _ @xcite , where @xmath87 being the orbital momentum and @xmath47 is the energy of the photon in the gravitational field of massive body . basically , the light - ray of component b can not be observed if @xmath88 is smaller than the radius of massive body a. for stars , the radius is much larger than schwarzschild radius @xmath89 , hence @xmath90 . furthermore , the generalized lens equation ( [ generalized_lens_1 ] ) is finite for @xmath91 and @xmath92 , both of which are possible astrometric configurations in binary systems . furthermore , in _ zschocke _ @xcite it has been shown that the generalized lens equation ( [ generalized_lens_1 ] ) yields in the appropriate limits the correct standard post - newtonian solution and the classical lens equation , hence provides a bridge between these essential approaches . ) or ( [ generalized_lens_3 ] ) , respectively . the parameters chosen are : distance @xmath93 , semi - major axis @xmath94 inclination @xmath95 , mass @xmath96 , mass ratio @xmath97 , eccentricity @xmath98 , argument of periapsis @xmath99.,width=302 ] in the following , eq . ( [ generalized_lens_1 ] ) is applied in order to determine the light deflection in binary systems . for that , the coordinates @xmath66 and @xmath65 are used in the form as given by eqs . ( [ inclination_10 ] ) and ( [ inclination_15 ] ) , respectively . a typical light - curve of a binary system , calculated by means of generalized lens equation ( [ generalized_lens_1 ] ) is shown in fig . [ fig : lightcurve ] . now , an inclination formula is derived from the generalized lens equation ( [ generalized_lens_1 ] ) . the impact of eccentricity is neglected , circular orbits @xmath100 are considered , implying @xmath101 . thus , the coordinates @xmath66 and @xmath65 are simplified to the expressions ( [ appendixc_5 ] ) and ( [ appendixc_10 ] ) given in appendix [ appendixc ] . furthermore , the maximal value of light deflection of a binary system is of interest , i.e. the astrometric configuration @xmath102 is considered here . then , by inserting these coordinates in the generalized lens equation ( [ generalized_lens_1 ] ) one obtains ( see eq . ( [ appendixc_24_c ] ) in appendix [ appendixc ] ) up to terms of the order @xmath103 : @xmath104 the minimal value @xmath105 and maximal value @xmath106 of light deflection for the astrometric position @xmath102 follow from eq . ( [ inclination_22 ] ) : @xmath107 where in ( [ inclination_24 ] ) terms of the order @xmath108 have been neglected . the expression ( [ inclination_22 ] ) can be reconverted in terms of inclination ( see appendix [ appendixd ] ) : @xmath109 where @xmath110 the inclination formula ( [ inclination_30_a ] ) yields the upper limit for @xmath111 of a binary system in order to have a given value of light deflection @xmath9 . note , that the values of @xmath9 can not be chosen arbitrarily , but they are restricted by @xmath112 and @xmath113 given by eqs . ( [ inclination_24 ] ) and ( [ inclination_25 ] ) , respectively . the inclination formula ( [ inclination_30_a ] ) can considerably be simplified . from ( [ inclination_35 ] ) and ( [ inclination_40 ] ) one obtains by series expansion @xmath114 by means of ( [ inclination_25 ] ) , the last term in ( [ comparison_15 ] ) can be estimated to be smaller than @xmath115 . here , it should be underlined that @xmath116 even at large distances @xmath117 and small values for semi - major axis @xmath118 . thus , one obtains @xmath119 up to terms of the order @xmath120 and @xmath121 ; here the relation @xmath122 for @xmath123 has been used . it should be underlined , that the applicability of ( [ inclination_49 ] ) is restricted by the condition ( [ appendix_kms_e ] ) given in appendix [ appendix_kms ] and by @xmath124 . here , it should be noticed that @xmath125 , even in such an extreme case like @xmath93 , @xmath126 and @xmath127 one obtains a small number @xmath128 ; for an analytical proof use the exact expression for @xmath112 . due to @xmath116 , the impact of semi - major axis is of lower order and can be neglected in the inclination formula . the inclination formula ( [ inclination_49 ] ) can be expressed in terms of dimensionless quantities as follows : @xmath129 note , that expression ( [ inclination_50 ] ) agrees with an inclination formula ( [ kms_d ] ) derived at the first time by _ klioner , et al . _ @xcite ; the arguments of their work are represented in appendix [ appendix_kms ] . the simplified inclination formula ( [ inclination_50 ] ) is not only useful for straightforward estimations about the order of magnitude , but it s simple structure provides also an obvious comprehension about the interplay of the individual terms . in this section , two stringent conditions on the orbital elements are highlighted for binaries having a given light deflection . these strict conditions are valid for any binary system : astrometric , spectroscopic , eclipsing and resolved binaries . the first stringent condition follows from the maximal light deflection angle ( [ inclination_25 ] ) , given by @xmath130 it represents a strict criterion for the maximal light deflection of a binary system with given schwarzschild radius of component a , given semi - major axis @xmath11 , and given distance @xmath12 between binary system and the observer . the second stringent condition on the orbital elements follows from the inclination formula in the simplified form as given by eq . ( [ inclination_50 ] ) . for a better illustration , this condition is given in terms of angular degrees instead of radians . using @xmath29 one obtains @xmath131 according to this strict condition , the inclination @xmath10 of a binary system with mass @xmath16 ( recall @xmath16 is the stellar mass of component a ) and at distance @xmath12 must not exceed the given value in order to have a light deflection @xmath9 . both these stringent conditions eqs . ( [ condition_1 ] ) and ( [ condition_2 ] ) were already stated in the introductory section by eqs . ( [ classical_lens_2 ] ) and ( [ condition_0 ] ) , respectively . however , is should be underlined here , that both ( [ condition_1 ] ) and ( [ condition_2 ] ) were obtained with the aid of one and the same approach , namely the generalized lens equation , while ( [ classical_lens_2 ] ) and ( [ condition_0 ] ) were obtained by means of the classical lens equation and the post - newtonian solution , that means by two different approaches . the observability of light deflection effect in binaries implies the realization of both these stringent conditions ( [ condition_1 ] ) and ( [ condition_2 ] ) simultaneously for a given binary system . but even if a given binary system satisfies these both conditions , the observability of light deflection effect is not guaranteed , because the astrometric position @xmath102 has to be reached during the time of observation . nonetheless , as soon as the orbital elements @xmath12 , @xmath11 , @xmath89 and @xmath10 of the binaries are known , these both stringent conditions ( [ condition_1 ] ) and ( [ condition_2 ] ) allow to find a possible candidate for being a relevant binary system for a given light deflection @xmath9 depending on the instrumentation of the observer . however , as it will be shown in section [ section3a ] and section [ section3b ] , the existence of such systems is highly improbable . in the previous section , the conditions on orbital parameters for a binary system have been determined in order to have a given magnitude of light deflection @xmath9 . in this section , the total number of such relevant binaries is determined . in order to estimate the total number of binaries having a given light deflection @xmath9 , the following formula is applied : @xmath132 here , @xmath133 is the density of binaries , @xmath134 is the semi - major axis distribution of binary systems , @xmath135 is the mass distribution of stars where @xmath136 is the mass - ratio of massive body ( component a ) and solar mass . the probability distribution @xmath137 to find a binary system with given inclination @xmath56 , is a function of distance @xmath12 , semi - major axis @xmath11 , mass ratio @xmath138 , and the given light deflection angle @xmath9 . according to ( [ inclination_30_a ] ) , the probability distribution @xmath137 is given by ( the inclination of binary systems is of course a random distribution ) @xmath139 where @xmath140 and @xmath141 are given by eqs . ( [ inclination_35 ] ) and ( [ inclination_40 ] ) . for the minimal distance of a binary system from the sun one can safely assume @xmath142 . from ( [ inclination_25 ] ) follows that beyond a sphere with radius @xmath143 only a very few exceptional binary systems might have a light deflection of more than @xmath3 . the sun is located in the _ orion arm _ which has about @xmath144 across and approximately @xmath145 in length . for the estimate according to ( [ numerical_results_5 ] ) , it is meaningful to assume that the stars are homogeneously distributed inside the _ orion arm_. for the uniform star density @xmath146 , a value which is in line with the data of _ recons _ @xcite . furthermore , a common presumption is , that about @xmath147 percent of all stars are components of a binary or multiple system , see _ duquennoy & mayor _ @xcite and _ halbwachs , et al . then one obtains for the density of binaries @xmath148 let us now consider the distribution of semi - major axis @xmath11 in binary systems . statistical investigations show that the distribution of binary semi - major axis is flat in a logarithmic scale over the range of six orders of magnitude , that means assumed to be valid in the large range @xmath149 , see _ kouwenhoven & de grijs _ @xcite . the lower limit @xmath150 is determined by the semi - major axis at which roche lobe overflow occurs , while the upper limit @xmath151 depends on how large the averaged star density is . the logarithmic distribution is known as `` _ pik s law _ '' ( 1924 ) named after its discoverer , and is given by @xmath152 , a law which has also been confirmed by recent investigations , see _ poveda , et al . this distribution is a consequence of the process of star formation as well as of the dynamical history of the binary system , and one can take this law as a given fact for numerical studies . accordingly , see appendix [ probability_distribution ] : @xmath153 furthermore , for the mass distribution @xmath154 let us recall the initial mass function ( imf ) which is the probablity that a star is newly formed with a stellar mass @xmath16 and is frequently assumed to be a power law @xmath155 . originally , the imf has been introduced by _ salpeter _ @xcite for solar neighborhood region , who assumed the value @xmath156 and a validity region for stars with masses between @xmath157 and @xmath158 . during the past decades the imf has been refined by several investigations . especially , the numerical values of slope parameter @xmath159 and regions of validity have been proposed in subsequent investigations , e.g. _ scalo _ @xcite , _ robin , et al . _ ninkovic & trajkovska _ @xcite , _ ninkovic _ @xcite , _ kroupa _ @xcite ; for a review see _ kroupa _ @xcite . moreover , the imf does not necessarily coincide with the real mass distribution of stars , because imf describes mass distribution of a star formation , while the solar neighborhood mainly consists of evolved stars of main sequence . here , for simplicity this distribution is used as a given fact with @xmath160 , and the proposed region of validity @xmath161 , and @xmath162 is assumed . according to imf , one finds for @xmath163 ( see appendix [ probability_distribution ] ) @xmath164 in order to motivate that distribution further , one can compare ( [ mass_distribution_7 ] ) with the _ recons _ data @xcite where one finds a fair agreement . using ( [ inclination_30_b ] ) - ( [ mass_distribution_7 ] ) , the results of the estimate ( [ numerical_results_5 ] ) are shown in fig . [ fig : number_binaries_2 ] ; recall , when evaluating this integral one has to take into account the boundary conditions given by eqs . ( [ inclination_24 ] ) and ( [ inclination_25 ] ) . according to fig . [ fig : number_binaries_2 ] , in total there are about @xmath165 binaries having a light deflection of at least @xmath166 . here , the observation time is assumed to be infinite , so that it is guaranteed that the optimal configuration @xmath102 is reached . ) , having parameters such that the light deflection of component b at component a is larger than a given value for @xmath9 . note , that for one individual observation the astrometric accuracy of gaia is about @xmath34 and the end - of - mission accuracy is about @xmath167 in the most ideal case ( bright star ) . the observation time is assumed to be infinite , so that it is guaranteed that the optimal configuration @xmath102 is reached.,width=302 ] in eq . ( [ numerical_results_5 ] ) the number of binaries with a given maximal possible light deflection @xmath9 has been determined , just by taking for eccentric anomaly the value @xmath102 , that means the ideal configuration where the light deflection takes its maximal value ( note , the eccentricity @xmath168 ) . it is , however , obvious that during the most part of the orbital motion one will have @xmath169 and the light deflection will be much smaller than the maximal possible light deflection angle @xmath9 . on the other side , the orbital period @xmath170 of relevant binaries , given by eq . ( [ appendixb_55 ] ) , will easily exceed the time of observation ; for instance , the gaia mission time is about @xmath171 . therefore , it will be not very probable , that the component b will be just at the relevant position near the value @xmath102 , where the light deflection becomes observable on microarcsecond level . in order to determine that number of observable relevant binaries , one has to extend eq . ( [ numerical_results_5 ] ) as follows , @xmath172 here , @xmath173 is the probability for the binary system to be in the region @xmath47 , where the light deflection is larger than a given value for @xmath9 . ) where the binary system reaches the optimal configuration @xmath102 during an assumed observation time of @xmath174 , and having orbital parameters such that the light deflection of component b at component a is larger than a given value for @xmath9 . note , that for one individual observation the astrometric accuracy of gaia is about @xmath34 and the end - of - mission accuracy is about @xmath167 in the most ideal case ( bright star).,width=302 ] in the very same way , as applied for the derivation of the inclination formula ( [ inclination_30_a ] ) , one can reconvert ( [ appendixc_24_d ] ) in terms of eccentric anomaly @xmath47 and one finds the eccentric anomaly formula : @xmath175 where @xmath140 and @xmath141 are given by eqs . ( [ inclination_35 ] ) and ( [ inclination_40 ] ) . for a given value of light deflection @xmath9 , the formula ( [ observable_30 ] ) yields the value of eccentric anomaly @xmath47 of a binary system characterized by semi - major axis @xmath11 and schwarzschild radius ( or mass ) @xmath89 at a distance @xmath12 . however , the values of @xmath9 can not be chosen arbitrarily , instead they are restricted by @xmath176 and @xmath177 given by ( of course , only astrometric positions with @xmath178 are taken into account , because for the area @xmath179 the light deflection is negligible ) : @xmath180 where in ( [ observable_20 ] ) terms of the order @xmath108 have been neglected . these expressions resemble the corresponding expressions in eqs . ( [ inclination_24 ] ) and ( [ inclination_25 ] ) . according to eq . ( [ observable_30 ] ) , the region where the binary system has a light deflection larger or equal @xmath9 is given by @xmath181 . one has also to take into account , that during gaia mission time @xmath182 the component b moves along the orbit and could move into the region @xmath181 . therefore , the probability @xmath173 that the binary system is during gaia mission time at least ones inside the relevant astrometric position with the value @xmath47 in ( [ observable_30 ] ) , is given by @xmath183 where the operator @xmath184 is defined by @xmath185{c } \displaystyle x \quad { \rm if } \quad x < 1\,,\\ \displaystyle 1 \quad { \rm if } \quad x \ge 1\ , . \end{array } \label{observable_37}\end{aligned}\ ] ] the probability distribution ( [ observable_35 ] ) has to be implemented in eq . ( [ observable_5 ] ) in order to determine the number of binary systems having a given light deflection @xmath9 and to be observable during gaia mission time @xmath182 . the results of eq . ( [ observable_5 ] ) are shown in fig . [ fig : number_binaries_3 ] ; recall , when evaluating this integral one has to take into account the boundary conditions given by eqs . ( [ inclination_24 ] ) and ( [ inclination_25 ] ) . accordingly , there are only a very few binaries @xmath186 having a light deflection of at least @xmath166 , if the astrometric position @xmath102 is reached during an assumed observation time of @xmath174 . in this section , the special case of a resolved binary system is considered . in order to investigate the detectability of the light deflection effect in binaries , todays most modern astrometric mission , the esa cornerstone mission gaia , and its instrumentation which provides the highest possible astrometric accuracy at the moment , will be considered here as a concrete example . the core of gaia optical instrumentation consists of two identical mirror telescopes , astro-1 and astro-2 , with a rectangular pupil whose dimensions are @xmath187 , @xmath188 , and @xmath189 is the effective focal length . the intensity is given by , see _ lattanzi , et al . _ @xcite and _ lindgren _ @xcite : @xmath190 where @xmath191 , @xmath192 , @xmath193 and @xmath194 the width and length of rectangular mirror , @xmath195 is the wavelength of incident light - ray , and @xmath196 are the angle of observation , i.e. the angle between the axis of the rectangular aperture and the line between aperture center and observation point . the intensity of incident light - ray at @xmath197 is denoted by @xmath198 . the function @xmath199 in eq . ( [ rectangular_5 ] ) is the ( by @xmath200 normalized ) point spread function ( psf ) for monochromatic incident light with wavelength @xmath195 for a rectangular aperture . the optical spectrum of stars is @xmath201 . in fig . [ fig : resolving_power1 ] the psf for an incident monochromatic light - ray with @xmath202 is represented for gaia telescopes . ) . the incident monochromatic light - ray has a wavelength of @xmath203 . the parameters of the rectangular telescope are : @xmath204 , @xmath205.,width=453 ] most of the light is concentrated in the central bright rectangular shaped pattern . the length @xmath206 and width @xmath207 of this rectangle is determined by the first zero - roots of ( [ rectangular_5 ] ) at @xmath208 and @xmath209 , respectively . from that it follows , that @xmath210 , and @xmath211 . furthermore , if the diffraction pattern is shown on a screen at a distance @xmath212 , then the length and width is given by , see _ hog , et al . _ @xcite : @xmath213 where @xmath212 is the focal length of the optic , i.e. of the rectangular gaia mirror . accordingly , the given numerical values @xmath187 , @xmath188 , and @xmath214 result in @xmath215 and @xmath216 for the length and width of `` _ airy rectangle _ '' of gaia optics . note , that the `` _ airy rectangle _ '' has the same order of magnitude than the pixel size ( @xmath217 ) of the @xmath218 ccds ( charge - coupled device ) sensors of astrometric field part of the focal plane . in order to separate two pointlike sources , the distance between their centers of the rectangle has to be larger than either @xmath219 or @xmath220 . since @xmath221 , in this study the better resolution value @xmath220 is used , which corresponds to a resolution angle of @xmath222 the resolving power is the minimal angular distance between two objects to get separable by gaia instrumentation . with the parameters given above one obtains the resolving power @xmath223 of gaia optics : @xmath224 in what follows this parameter is of fundamental importance in order to determine the ability of gaia to determine the light deflection in binary systems . in this section , the question is addressed which and how many binary systems can be separated by gaia instrumentation among all those relevant binaries found in the previous section ; see fig . [ fig : number_binaries_3 ] . in average , gaia will observe each object @xmath225 times , but will not constantly observe these objects during mission time . however , for simplicity the scanning law of gaia is approximated by assuming a permanent observation of all objects during the whole mission time . furthermore , visual binaries are considered in this section , i.e. binaries which are separable by gaia telescopes . the both largest telescopes of gaia have a resolution angle @xmath223 discussed in the previous section , see eqs . ( [ rectangula_30 ] ) and ( [ minimal_alpha_10 ] ) . for binary systems , this resolution angle @xmath223 corresponds to a minimal distance between the components a and b to get separable within gaia optics . using ( [ expression_d ] ) and ( [ rectangula_30 ] ) one obtains the condition @xmath226 where @xmath12 is the distance between the remote objects and gaia observer . this condition is by far much more important than taking into account the effect of finite radius of the stars , which would imply @xmath227 , where @xmath228 being the radius of component @xmath11 . by inserting the extreme case @xmath229 into ( [ inclination_22 ] ) , one obtains @xmath230 where @xmath231 has been used , and terms of higher order @xmath232 are neglected . relation ( [ visual_5 ] ) is an expression for the maximal light deflection angle of a binary system when taking into account the resolving power of gaia . ( [ visual_5 ] ) is a much stricter restriction than the generalized lens equation ( [ inclination_22 ] ) , because ( [ visual_5 ] ) determines the light deflection angle only of those binary systems having a resolution angle @xmath223 of gaia optical instrumentation , while ( [ inclination_22 ] ) determines the light deflection angle of any possible binary system . note , that from eq . ( [ visual_5 ] ) follows the maximal possible distance of visual binaries for a given light deflection : @xmath233 where in the last expression the optimal values @xmath234 and @xmath235 has been used . the condition ( [ condition_3 ] ) can also be written by @xmath236 these both conditions ( [ condition_3 ] ) and ( [ condition_4 ] ) imply rather extreme orbital parameters on visual binaries . for instance , condition ( [ condition_3 ] ) implies a maximal distance of @xmath237 for solar - mass type binaries even with a huge semi - major axis of @xmath238 , while condition ( [ condition_4 ] ) implies a large semi - major axis for solar - mass type binaries outside a sphere of @xmath239 . it is almost certain , that such extreme parameters will not be realized in nature . in this study , the light deflection in binary systems has been considered . while there is absolutely not any doubt about the existence of this relativistic effect , is has not been observed so far . to investigate this effect of light deflection , an inclination formula ( [ inclination_30_a ] ) has been derived by means of generalized lens equation ( [ generalized_lens_1 ] ) obtained recently by _ zschocke _ @xcite , and these both equations are the theoretical basis for investigating the light deflection effect in binary systems . a simplified inclination formula has been presented by eq . ( [ inclination_50 ] ) and its validity has been discussed in some detail . this simplified inclination formula has also been obtained by _ klioner , et al . _ @xcite by an independent approach . furthermore , two stringent conditions on the orbital parameters have been given by eqs . ( [ condition_1 ] ) and ( [ condition_2 ] ) . these both stringent conditions are relations between the orbital elements of a ( resolved , astrometric , eclipsing , spectroscopic ) binary system for a given magnitude of light deflection , and allow to find a relevant binary system in a straightforward way . in section [ section3a ] , the total number of binaries with a given light deflection has been determined by means of the semi - major axis distribution according to `` _ pik s law _ '' and the mass distribution according to `` _ salpeter s mass distribution _ '' . since the inclinations are randomly distributed , the inclination formula allows to estimate the total number of relevant binaries with the aid of eq . ( [ numerical_results_5 ] ) . it turns out , that in total there exist about @xmath0 binaries having orbital parameters such that the light deflection amounts to be at least @xmath3 ; see fig . [ fig : number_binaries_2 ] . in section [ section3b ] a finite time of observation of @xmath240 years ( gaia mission time ) has been considered , which considerably reduces the total number of relevant binaries . clearly , this case is of more practical importance , since a restricted time window of observation is in more agreement with reality than the first scenario . by taking into account the probability to find the system in the ideal astrometric position @xmath102 where the light deflection becomes maximal , it has been found by evaluating the corresponding integral ( [ observable_5 ] ) that there is not any of the relevant binary systems in the ideal position @xmath102 during gaia mission time ; see fig . [ fig : number_binaries_3 ] . thus , while in principle a few binaries will have a significant light deflection , the effect could not be detected due to the restricted finite time window of observation . furthermore , the special case of resolved binaries has been considered in section [ condition2 ] . the astrometric instrumentation of the esa cornerstone mission gaia , see e.g. _ perryman , et al . _ @xcite , has been considered in some detail in order to decide whether or not this subtle effect of light deflection can be observed . two conditions for resolved binaries were presented in eqs . ( [ condition_3 ] ) and ( [ condition_4 ] ) for such special kind of binary systems . it has been shown , however , that even for the gaia mission , which is an outstanding milestone of progress in astrometry , such binary systems must have rather extreme orbital parameters in order to reach todays level of detectability , i.e. on microarcsecond level . the existence of such exotic binaries is , however , highly improbable . in summary , the main results are presented by the inclination formulae ( [ inclination_30_a ] ) and its simplified version ( [ inclination_50 ] ) , the stringent conditions ( [ condition_1 ] ) and ( [ condition_2 ] ) and by the diagrams fig . [ fig : number_binaries_2 ] and fig . [ fig : number_binaries_3 ] . accordingly , one comes to the conclusion that the detectability of light deflection in binary systems reaches the technical limit of todays high precision astrometry and might be detected only in case of a very few and highly exotic binary systems . it is , however , very unlikely that such extreme binaries might exist . it seems that the detection of the light deflection effect in binary systems needs an astrometric accuracy of better than about @xmath241 . thus , only astrometric missions of the next generation can accept the challenge to detect this subtle effect of relativity . this work was partially supported by the bmwi grants 50qg0601 and 50qg0901 awarded by the deutsche zentrum fr luft- und raumfahrt e.v . ( dlr ) . enlighting discussions with prof . sergei a. klioner , prof . michael h. soffel and prof . francois mignard are gratefully acknowledged . 99 bozza , v. , 2008 , phys . rev . d , 78 , 103005 brumberg , v. a. 1991 , _ essential relativistic celestial mechanics _ , ( bristol : adam hilder ) chandrasekhar , s. , 1983 , _ the mathematical theory of black holes _ ( oxford : clarendon ) duquennoy , a. & mayor , m. , 1991 , a&a , 248 , 485 fritelli , s. , kling , t.p . & newmann , e.t . , 2000 , phys . d , 64 , 064021 halbwachs , j.l . , mayer , m. , udry , s. & arenou , f. , 2003 , a&a , 397 , 159 hog , e. , bastian , u. & seifert , w. , 1997 , _ optical design of gaia _ , www.en.scientificcommons.org klioner , s.a . , 2003 , aj 125 , 1580 klioner , s.a . , mignard , f. & soffel , m. , 2003 , _ astrometric signature of gravitational microlensing on the components of edge - on binary systems _ , unpublished klioner , s.a . & zschocke , s. , 2010 class . quantum grav . , 27 , 075015 kouwenhoven , m.b.n . & de grijs , r. , 2008 , a&a , 480 , 103 kroupa , p. , 2001 , mnras , 322 , 231 kroupa , p. , 2002 , science , 295 , 82 . landau , l.d . & lifshitz , e.m . , 1976 , _ mechanics _ , vol . 1 , ( oxford pergamon press ) lattanzi , m. , carollo , d. & gai , m. , 1998 , _ aberrated point spread function s for the mms configuration and astrometric error _ , technical report , sag - ml-014 ; available at the gaia document archive http://www.rssd.esa.int/llink/livelink . lindgren , l. , 1998 , _ point spread functions for gaia including aberrations _ , technical report from lund observatory , sag - ll-025 ; available at the gaia document archive http://www.rssd.esa.int/llink/livelink ninkovic , s. , 1995 , bull . belgrade , 151 , 1 ninkovic , s. & trajkovska , v. , 2006 , serb . aj , 172 , 17 pik , e.j . , 1924 , tartu obs . publ . , 25 perryman , m.a.c . , de boer , k. , gilmore , g. et al . , a&a , 2001 , 369 , 339 poveda , a. , allen , c. & hernandez - alcantara , a. , 2007 , in proc . 240 ; arxiv : astro - ph/0705.2021 recons , 2012 , http://www.recons.org/ robin , a.c . , reyle , c. , derriere s. & picaud , s. , 2003 , a&a , 409 , 523 salpeter , e.e . , 1955 , aj , 121 , 161 scalo , j.m . , 1986 , fundam , 11 , 1 wds , 2011 , http://ad.usno.navy.mil/wds/wds.html zschocke , s. 2011 , _ a generalized lens equation for light deflection in weak gravitational fields _ , class . quantum grav . 28 , 125016 zwitter , t. & munari , u. , 2004 , _ the environment and evolution of double and multiple stars _ , revmexaa ( sc ) , 21 , 25 the calculations in this appendix follow mainly _ landau & lifshitz _ @xcite . consider two massive bodies , one component having a mass @xmath41 and spatial coordinate @xmath42 , and second component with a mass @xmath44 and spatial coordinate @xmath45 , respectively . they orbit around their common center of mass @xmath242 , @xmath243 the lagrangian @xmath244 of the two - body problem is given by @xmath245 where @xmath246 is the potential . with the aid of relative coordinate @xmath247 and reduced mass @xmath248 , the two - body problem can be transformed into into one - body problem , @xmath249 polar coordinates @xmath250 yield @xmath251 the orbital angular momentum @xmath87 is conserved @xmath252 by means of which one obtains for the total energy of the two - body system the expression @xmath253 from eq . ( [ appendixb_25 ] ) one deduces @xmath254 - \frac{l^2}{\overline{m}^2\;r_{ab}^2}\right)\ ; , \label{appendixb_30}\end{aligned}\ ] ] and from eq . ( [ appendixb_30 ] ) one obtains up to an integration constant @xmath255 - \frac{l^2}{\overline{m}^2\;r^2_{ab}}\right)^{-1/2}\ , , \label{appendixb_35 } \\ \nonumber\\ \phi & = & \int d r_{ab}\ ; \frac{\overline{m}}{r^2_{ab}}\ ; \left ( 2 \,\overline{m } \left [ e - u \left(r_{ab}\right ) \right ] - \frac{l^2}{r_{ab}^2}\right)^{-1/2}\ , , \label{appendixb_36}\end{aligned}\ ] ] where in the second relation eq . ( [ appendixb_20 ] ) has been used ; note that ( [ appendixb_36 ] ) is the relation between @xmath256 and @xmath257 and is called orbital equation . the eqs . ( [ appendixb_35 ] ) and ( [ appendixb_36 ] ) are the general integral solutions of a two - body problem . in order to integrate these eqs . ( [ appendixb_35 ] ) and ( [ appendixb_36 ] ) one has to specify the potential @xmath246 . in case of kepler problem one has @xmath258 ( [ appendixb_36 ] ) can be integrated and yields @xmath259 where the axes are chosen such that the mentioned integration constant vanishes . furthermore , by introducing the eccentricity @xmath260 ( possible values of eccentricity are between @xmath261 ; @xmath262 corresponds to a circular orbit ) , @xmath263 the solution ( [ appendixb_41 ] ) can be written as @xmath264 note the expressions of semi - major axis @xmath11 and semi - minor axis @xmath43 , @xmath265 to solve the integral ( [ appendixb_35 ] ) , one substitutes @xmath266 where @xmath47 is called eccentric anomaly . then , one obtains for the integral in eq . ( [ appendixb_35 ] ) the expression @xmath267 and the solution is given by @xmath268 where the integration constant vanishes , i.e. the particle at @xmath269 is in periastron . the eqs . ( [ appendixb_46 ] ) and ( [ appendixb_50 ] ) are the general solutions of two - body problem . they can be rewritten as @xmath270 in case of ellipse , @xmath271 in periastron , @xmath272 in apastron , and for a complete orbit @xmath47 runs from @xmath271 to @xmath273 . thus , one obtains for the orbital period the expression @xmath274 note , the solution @xmath54 in cartesian coordinates , @xmath275 and @xmath276 : @xmath277{l } x \\ \displaystyle y \end{array}\right ) , \label{appendixb_56 } \\ \nonumber\\ x & = & a \left ( \cos e - { \rm e } \right ) \ , , \label{appendixb_57 } \\ \nonumber\\ y & = & a \left ( 1 - { \rm e}^2 \right)^{1/2 } \;\sin e \ , . \label{appendixb_58}\end{aligned}\ ] ] the coordinates of the bodies a and b , i.e. their orbits , are given by @xmath278 accordingly , the geometry of the orbit is determined by two orbital parameters : semi - major axis a and eccentricity e. in order to know the position of one celestial body , either component a or component b , two additional orbital parameters are needed , namely orbital period @xmath170 and true anomaly @xmath279 . a geometrical representation of the coordinates of the components of a binary star is given in fig . [ fig : ellipse ] for the case of @xmath280 . and @xmath281 , respectively . the semi - major axis of the binary system is chosen @xmath282 and eccentricity is taken @xmath283 . the coordinates of mass center are @xmath55 . the massive bodies a and b are always in opposition to each other.,width=226 ] for the inclination formula the impact of eccentricity on light deflection is neglected , thus @xmath284 , implying that @xmath285 is taken . then , for the vectors from massive body to observer @xmath65 and from massive body to source @xmath66 , one has @xmath286{c } \displaystyle \sin i - \epsilon_1\,\cos e\\ \nonumber\\ \displaystyle - \epsilon_1\,\sin e \\ \nonumber\\ \displaystyle \cos i \end{array}\right ) , \label{appendixc_5}\ ] ] @xmath287{c } \displaystyle \cos e\\ \nonumber\\ \displaystyle \sin e \\ \nonumber\\ \displaystyle 0 \end{array}\right ) , \label{appendixc_10}\ ] ] where the small parameter @xmath288 has been introduced . from eqs . ( [ appendixc_5 ] ) and ( [ appendixc_10 ] ) one obtains for vector @xmath71 , where @xmath69 , the expression @xmath289{c } \displaystyle \sin i + \epsilon_2\,\cos e\\ \nonumber\\ \displaystyle \epsilon_2\,\sin e \\ \nonumber\\ \displaystyle \cos i \end{array}\right ) , \label{appendixc_15}\ ] ] where the small parameter @xmath290 has been introduced . note that ( [ appendixc_10 ] ) and ( [ appendixc_15 ] ) yields @xmath291 using ( [ appendixc_5 ] ) - ( [ epsilon_2 ] ) , the generalized lens equation ( [ generalized_lens_1 ] ) reads @xmath292 where @xmath293 and @xmath294 by series expansion one obtains up to terms of order @xmath295 : @xmath296 where @xmath297 with @xmath298 . the minimal and maximal light deflection angle are @xmath299 in this study , the maximal possible light deflection effect is of interest . accordingly , two configurations are relevant : @xmath300 up to order @xmath301 which is just eq . ( [ inclination_22 ] ) , and @xmath302 up to order @xmath295 . furthermore , it is useful to take into account only astrometric positions with @xmath178 , because otherwise the light deflection is for sure negligible . from ( [ appendixc_24_c ] ) one obtains @xmath303 from ( [ appendixd_10 ] ) one obtains @xmath304 eq . ( [ appendixd_15 ] ) represents an quadratic equation for the expression @xmath305 , which has the following both solutions for the inclination @xmath10 : @xmath306 where @xmath307 eq . ( [ appendixd_21 ] ) represents two solutions , however only the one with the plus - sign is valid . this can be shown as follows . for the value @xmath308 the light deflection has to be @xmath309 , according to ( [ inclination_25 ] ) . by inserting @xmath113 in eqs . ( [ appendixd_30 ] ) and ( [ appendixd_35 ] ) one obtains @xmath310 and @xmath311 . if one inserts @xmath308 for @xmath140 and @xmath141 into eq . ( [ appendixd_21 ] ) one obtains the relation @xmath312 obviously , relation ( [ appendixd_40 ] ) is only correct for the upper sign . a very similar proof can also be done using @xmath112 which also yields that the upper sign is the correct and only solution . therefore , the inclination formula is given by ( note , that in the region under consideration @xmath313 ) @xmath314{c } \displaystyle \arcsin \left ( - \frac{p}{2 } + \sqrt{\frac{p^2}{4 } - q } \right ) \quad { \rm for } \quad 0 \le i \le \frac{\pi}{2 } \ , , \\ \\ \displaystyle \pi - \arcsin \left ( - \frac{p}{2 } + \sqrt{\frac{p^2}{4 } - q } \right ) \quad { \rm for } \quad \frac{\pi}{2 } < i \le \pi\ , . \end{array } \label{appendixd_45}\end{aligned}\ ] ] for the complete region @xmath56 one obtains for the inclination formula the following expression : @xmath315 where @xmath140 and @xmath141 are given by eqs . ( [ appendixd_30 ] ) and ( [ appendixd_35 ] ) . in this appendix , some basic steps of the calculations of _ klioner , et al . _ @xcite are presented . a scheme of the light propagation in a binary system is presented by fig . [ fig : distance1 ] . the light signal from component b , considered to be the light source , is deflected by component a , considered to be the massive body which curves the space - time . the vector @xmath65 points from the mass center of massive body to the observer , and vector @xmath66 points from the mass center of massive body to the source ; @xmath69 and its absolute value @xmath70 , and @xmath74 is the schwarzschild radius of massive body ; the explicit label a is omitted . furthermore we define the impact vector @xmath72 is defined , and its absolute value @xmath73 , see also fig . [ fig : distance1 ] . according to _ klioner & zschocke _ @xcite , the transformation of @xmath316 to the unit tangent vector @xmath317 of light - trajectory at observer is in standard post - newtonian approach given by @xmath318 note , the ppn parameter of parameterized post - newtonian formalism , which characterizes a possible deviation of the physical reality from general theory of gravity , is set equal to @xmath1 here for simplicity . this expression is valid as long as @xmath124 , but diverges for @xmath91 . thus , it is not valid for all possible binary configurations , instead one has take care to consider only such astrometric configurations with @xmath124 . by means of ( [ appendix_kms_a ] ) one obtains for the light deflection angle @xmath9 , i.e. for the angle between @xmath317 and @xmath316 , the expression @xmath319 where @xmath320 , @xmath321 has been used , and @xmath322 is the angle between @xmath66 and @xmath65 . the expression ( [ appendix_kms_b ] ) diverges for @xmath323 , which corresponds with the mentioned divergence of ( [ appendix_kms_a ] ) for @xmath91 . obviously , @xmath324 ( from eq . ( [ observable_30 ] ) it is clear that eccentric anomaly @xmath47 of binary system should be very close to zero for the light deflection effect to be observable at the level of microarcsecond , i.e. one actually could even assume @xmath325 ) and one obtains @xmath326 where @xmath327 has been used , @xmath328 , and the asymmetry of function @xmath329 . from ( [ appendix_kms_c ] ) one obtains @xmath330 where @xmath331 has been used ( recall @xmath16 is the mass of component a and @xmath21 is the solar mass ) , and the numerical values @xmath332 , @xmath333 and @xmath334 have been inserted , so that @xmath335 . the simplified inclination formula ( [ appendix_kms_d ] ) has first been obtained by _ klioner , et al . _ @xcite ; here the fact is noticed , that due to the divergence of the post - newtonian solution ( [ appendix_kms_a ] ) for @xmath336 , which corresponds to @xmath323 , the applicability of ( [ appendix_kms_d ] ) is restricted by the condition @xmath124 . using @xmath337 one obtains the validity condition for the applicability of ( [ appendix_kms_d ] ) : @xmath338 equation ( [ appendix_kms_d ] ) provides a relation between inclination @xmath10 and light deflection @xmath9 for a binary system characerized by it s distance @xmath12 from the observer and it s stellar mass @xmath16 ; note that ( [ appendix_kms_d ] ) agrees with the simplified inclination formula given by equation ( [ inclination_50 ] ) . assume a probability distribution of a quantity @xmath52 is given by @xmath339 . the probability @xmath340 , to find a system in the interval @xmath341 is given by @xmath342 where the region of validity of probability distribution @xmath339 is given by @xmath343 and @xmath344 . in the infinitesimal limit @xmath345 , one obtains by series expansion the following explicit form for the here used probability distributions : for a power law @xmath346 with @xmath163 one finds @xmath347

one of the most important current tasks in neutrino physics is to understand the full implications of the solar @xcite , atmospheric @xcite , and reactor @xcite neutrino data that became available in the last five years . while much of the recent effort has been focused on establishing the values of the neutrino masses and mixings , there has also been a growing realization that the same data may contain valuable information about other neutrino properties , such as its interactions with matter @xcite or with electromagnetic fields @xcite . any measured deviation of these properties from their standard model ( sm ) values would have profound theoretical implications . as part of this program , we wish to reexamine here the sensitivity of the recent solar neutrino measurements to the neutrino electromagnetic properties . the idea that the neutrino may possess a relatively large magnetic ( transition ) moment has been the subject of intense theoretical speculations for many years . it was originally put forth in the early 1970 s , as a possible explanation to the solar neutrino deficit @xcite : the solar neutrinos , it was argued , were `` disappearing '' as a result of the neutrino spin precession in the solar magnetic field . remarkably , this explanation much improved with time @xcite remained viable for the next three decades . only in 2002 did the kamland measurement of the reactor antineutrino disappearance @xcite conclusively rule it out as the _ dominant _ mechanism of the solar neutrino conversion . the possibility of the spin precession happening at a _ subdominant _ level , however , remains of great interest to this day , as a probe of the neutrino electromagnetic properties and , at the same time , of the magnetic fields in the solar interior . for definiteness , here we will focus on the case of majorana neutrinos . the experimental sensitivity to magnetic spin precession in this scenario is especially high , as has been long recognized @xcite : upon the spin flip , a majorana neutrino turns into an antineutrino and thanks to flavor oscillations can then be detected with some probability as an electron antineutrino , @xmath0 . the @xmath0 flux from the sun , in turn , is strongly constrained by super - kamiokande @xcite , sno @xcite , and especially kamland @xcite which can detect @xmath1 conversion of the solar @xmath2b neutrinos at the level of a few hundredths of a percent . to understand if the kamland bound places a nontrivial constraint on the neutrino magnetic moment , we need to compare it to the maximum possible @xmath0 flux from the sun . this must be done taking into account ( i ) the recently measured values of the oscillation parameters , ( ii ) the bounds on the solar magnetic field strength and ( iii ) the existing direct bounds on the neutrino magnetic moment . we also need to consider that magnetic spin flip could occur either in the convective zone ( cz ) of the sun , which comprises the outer 30% by radius , or deeper in the radiative zone ( rz ) . in the first case , the neutrino traverses a region of turbulent magnetic field ; in the second case , the field , if present , is smooth and frozen into stationary plasma . the physics of the neutrino evolution is quite different in the two cases and requires separate treatments . both the cz @xcite and the rz @xcite cases have been investigated in the literature and in both cases it has been concluded that it was _ a priori _ possible to have a signal greater than the current kamland sensitivity reach . in the case of the cz , even a bound on the neutrino magnetic moment was claimed . in our analysis , we reach different conclusions . in the case of the rz , we find that the usual treatment , based on the @xmath3 resonance , works only for small flavor mixing . the large measured value of the solar neutrino mixing angle leads to a qualitatively new effect : the resonant level crossing disappears and no measurable magnetic spin flip can take place . a new analytical result , extending the classical `` magnetic resonance '' condition @xcite to large flavor mixing , is derived here for the first time . in the case of the cz , we consider two qualitatively different models of the magnetic field . one of these models , that of the uniform kolmogorov turbulence , follows @xcite and our basic treatment of the neutrino evolution in this case agrees with that work . we , however , take what we regard as more physically justifiable magnetic field parameters . for the second model , we take a scenario of isolated flux ropes as suggested by the recent solar physics research . this scenario , to the best of our knowledge , is considered in the neutrino literature for the first time . the net result is that in both cases we estimate the maximum @xmath0 flux to be _ below _ the present kamland bound . we begin by giving the necessary background information , including the form of the interaction ( sect . [ sect : interaction ] ) , the known bounds on the neutrino magnetic moment ( sect . [ sect : directbounds ] ) , the general properties of the solar magnetic fields ( sect . [ sect : magnfields ] ) , and the form of the neutrino oscillation hamiltonian ( sect . [ sect : oschamiltonian ] ) . the case of the rz magnetic field is treated in sect . [ sect : rz ] : after reviewing the conventional approach ( sect . [ sect : rz_conventional ] ) , we argue that resonant antineutrino production is suppressed for large flavor mixing ( sect . [ sect : rz_nores ] ) , derive a corresponding general analytical condition ( sect . [ sect : rz_analyt ] ) , and finally perform a direct numerical scan of the parameter space to confirm our analytical conclusions ( sect . [ sect : rz_scan ] ) . this is followed by the analysis of the cz fields in sect . [ sect : cz ] , where the two models of the magnetic field mentioned above are considered ( in sects . [ sect : cz_kolmogorov ] and [ sect : cz_tubes ] ) . the results are summarized in sect . [ sect : conclusions ] . some additional technical details about the cz neutrino evolution can be found in the appendix . the electromagnetic moment interaction of the majorana neutrinos is given by the following dimension five operator @xmath4 where @xmath5 and @xmath6 are majorana spinors of flavors @xmath7 and @xmath8 in the 4-component notation , and @xmath9/2 $ ] , with @xmath10 and @xmath11 being the lorentz indices . it proves useful to rewrite this lagrangian in the 2-component weyl spinor notation . this notation , besides being simply more natural for ultrarelativistic fermions , brings considerable physical insight . writing the 4-component majorana spinor @xmath12 in the weyl basis as @xmath13 and introducing @xmath14 matrices @xmath15 , with @xmath16 , @xmath17 ( here @xmath18 are the usual pauli matrices ) , we rewrite eq . ( [ eq : lagr4 ] ) as @xmath19 this form makes the following important physical properties of the interaction manifest : * _ the interaction couples the neutrino and antineutrino states . _ notice in eq . ( [ eq : lagr ] ) the spinors are either both `` undotted '' or both `` dotted '' . each `` undotted '' spinor destroys a neutrino state or creates an antineutrino state . * _ no flavor - diagonal terms . _ since spinors anticommute , the @xmath20 terms in eq . ( [ eq : lagr ] ) vanish identically , and hence majorana neutrinos can not have flavor - diagonal magnetic moments @xcite . this can be easily seen , for example , using the definition of the @xmath21 matrix given above and the identity @xmath22 the _ transition _ ( @xmath23 ) moments , however which lead to a simultaneous spin and flavor change , _ i.e. _ , @xmath24 are perfectly allowed . * _ the interaction with the @xmath25 ( longitudinal ) component of the magnetic field vanishes identically , while the interaction with the transverse components @xmath26 is nonzero . _ one of the two spinors in eq . ( [ eq : lagr ] ) has a raised index . denoting it by @xmath27 , we write eq . ( [ eq : lagr ] ) in the three - dimensional notation as @xmath28 plugging in @xmath29 which neglects terms of order @xmath30 we find that the interaction picks the @xmath31 component of the matrix @xmath32 , i.e. , @xmath33 for @xmath34 , @xmath35 for @xmath36 , and @xmath37 for @xmath38 . + the same conclusion can be reached by boosting the magnetic fields to the neutrino rest frame . * the real part of @xmath39 corresponds to the magnetic dipole moment and the imaginary part to the electric dipole moment . an ultrarelativistic neutrino precesses in an external magnetic field if _ either _ the magnetic or electric dipole moments are non - zero @xcite . following the convention , we will henceforth use the term magnetic moment " to refer to both possibilities . all these properties are well - known and are derived here for completeness . the laboratory bounds on the neutrino magnetic ( transition ) moment come from measuring the cross sections of @xmath40 or @xmath41 scattering in nearly forward direction . the recent bound for the interaction involving the electron antineutrino is @xmath42 at the 90% confidence level @xcite , where @xmath43 is the bohr magneton ( @xmath44 is the electron mass , @xmath45 is its charge ) . stronger bounds , @xmath46 , exist from astrophysical considerations , particularly from the study of red giant populations in globular clusters @xcite . larger values of the transition moment would provide an additional cooling mechanism and change the red giant core mass at helium flash beyond what is observationally allowed . even allowing for new inputs ( such as updated distances to globular clusters ) or refinements in the stellar models , it appears extremely unlikely that the red giant bound could be weakened by more than a factor of a few . correspondingly , as a very conservative bound , we will take @xmath47 in the subsequent analysis . from the theoretical point of view , the neutrino does possess a nonzero transition moment , already in the sm . it is induced by loop effects : as the neutrino briefly dissociates into virtual charged particles ( @xmath48 and a charged lepton @xmath49 ) , those constituent particles couple to the photon . because the operator in eq . ( [ eq : lagr ] ) requires a helicity flip , and because in the sm the @xmath48 boson couples only to left - handed fields , the mass insertion has to be put on the neutrino line . the resulting moment is proportional to the neutrino mass and hence is highly suppressed , @xmath50 . in possible extensions of the sm , however , this need not be the case , and the effect can be proportional to the mass of the charged lepton running in the loop . for example , in the left - right symmetric model one obtains @xmath51 @xcite , where @xmath52 is the left - right mixing parameter in the model . the cz of the sun is known to host significant magnetic fields , which manifest themselves in a variety of effects on the solar surface ( sunspots , flares , prominences , etc ) . these fields are created and destroyed during each solar cycle by a combination of convective motions and differential rotation . detailed understanding of the solar cycle is still an active subject of research . nevertheless , the following general features of the cz fields are well established : * the magnetic field observed in sunspots is several kg , exceeding the average surface field strength by three orders of magnitude . sunspots usually come in pairs of opposite polarity and are thought to be manifestations of large - scale magnetic structures residing inside the cz . * the total toroidal flux in the cz at sunspot maximum can be estimated from the total flux that emerges on the surface during the solar cycle , which is around @xmath53 mx @xcite . since the same flux tube may be emerging more than once , @xmath54 mx has been argued to be the likely value @xcite . averaged over the cz , this gives a field of several kg . * the turbulent equipartition value for the magnetic field at the base of the cz can be expressed in terms of the density @xmath55 , the solar luminosity @xmath56 , and the distance to the center of the sun @xmath57 : @xmath58 kg ( e.g. , @xcite ) . * modern models of the cz fields argue that fields on the order of 100 kg may exist in the cz . for the present discussion , it is important to realize that such strong fields , if they exist , can occupy only a small fraction of the cz volume , such as the _ thin _ shear layer near the base of the cz ( ` tachocline ' ) , or be localized in isolated flux tubes @xcite . to have even half of the cz filled with such strong fields would be contrary to the flux arguments mentioned above , as well as basic energetics considerations . for an in - depth review of the cz fields and further references , the reader is urged to consult an excellent review @xcite . for the status of the present - day simulation efforts see , e.g. , @xcite . unlike the cz , the rz has no internal motions to provide an active mechanism to generate magnetic fields . on the other hand , the conductivity of the rz plasma is very high , which opens up the possibility that the rz could support large - scale magnetic fields left over from the early stages of the solar evolution . let us consider both of these points in turn . the evidence for the lack of any large - scale mixing motions in the rz comes from several sources . first , helioseismological studies show that , unlike the cz , the rz rotates as a solid body . ( this fact by itself , incidentally , provides a hint that at least a weak poloidal field is present in the rz @xcite ) . second , solar model calculations show that the stratification ( entropy gradient ) in the rz interior is quite strong , precluding large radial motions . finally , a very impressive direct confirmation of the lack of radial mixing is provided by the observed beryllium abundance at the solar surface @xcite . be is destroyed at @xmath59 k , which is the temperature at @xmath60 , @xmath61 being the solar radius , @xmath62 km . since beryllium observed at the solar surface is not noticeably depleted relative to its primordial abundance , the material at @xmath63 ( on the bottom of the cz ) and the material at @xmath64 have not been mixed during the lifetime of the sun ( except for a brief period during the pre - main sequence convection stage ) . while a large - scale magnetic field in the rz is not being presently generated , it is also not destroyed by ohmic decay over the solar lifetime , 4.7 billion years . the ohmic decay time for large - scale field configurations is in the billions of years @xcite , in fact for the lowest toroidal mode it is about 24 billion years @xcite . over time , a primordial rz field has been invoked by many authors to explain a variety of effects : from the solar neutrino problem @xcite to the so - called princeton solar oblateness measurements @xcite . what is important for the present analysis is that the field in the rz , if it exists , must be ( i ) smooth and ( ii ) bounded in strength @xcite . smoothness follows from the fact that , because of ohmic decay , any small - scale magnetic field ( @xmath65 ) features would have decayed away over the lifetime of the sun . this ensures the adiabatic character of the neutrino evolution throughout the bulk of the cz . the bounds on the field strength come from several independent arguments , such as the measurements of the solar oblateness and the stability analysis of field configurations . it follows that the amplitude of the large scale toroidal field can not exceed @xmath66 mg @xcite . combined with the bounds on the size of the neutrino transition moment , this limits the value of the magnetic term ( see below ) in the neutrino oscillation hamiltonian . since flavor oscillations change the neutrino ( and antineutrino ) flavor and the magnetic interactions convert neutrinos into antineutrinos , the evolution of the neutrino state is , in general , governed by a @xmath67 hamiltonian matrix , with the neutrinos and antineutrinos of all three flavors . for the purpose of studying the spin flip in the rz , the problem can be reduced to a @xmath68 case . sector , solar neutrinos actually oscillate between @xmath69 and a certain superposition state of @xmath70 and @xmath71 . to stay with the tradition , we denote the latter state by @xmath70 ( and the corresponding superposition antineutrino state by @xmath72 ) . ] in the basis @xmath73 , the hamiltonian is @xcite @xmath74 where @xmath75 are the transverse components of the magnetic field and the @xmath14 submatrices are given by @xmath76 here @xmath77 , @xmath78 and @xmath79 . @xmath80 is the neutrino mass - squared splitting , @xmath81 is its energy , and @xmath82 and @xmath83 are the electron and neutron number densities . the physical reason for the @xmath84 reduction is that the large atmospheric neutrino splitting effectively decouples two of the six states in the case of the smooth magnetic field . it should be noted , however , that the `` noisy '' field in the cz enables transitions between states separated by the large splitting . consequently , in the cz one may need to consider the full six - neutrino problem @xcite . historically , in discussions of spin - flavor precession ( sfp ) in the sun , special attention has been given to the sfp resonance " region @xcite , defined as the region where two of the _ diagonal _ terms in the hamiltonian ( [ eq : h]-[eq : h3 ] ) , namely those corresponding to the @xmath85 and @xmath86 entries , become equal . the corresponding resonance condition is @xmath87 compared to the traditional flavor resonance " condition , defined as the equality of the @xmath85 and @xmath88 diagonal entries in ( [ eq : h]-[eq : h3 ] ) ) , @xmath89 condition ( [ eq : res_diag ] ) is satisfied at a somewhat higher density in the sun . when the resonant sfp mechanism was introduced back in the late 1980 s , the values of the oscillation parameters were essentially unknown . in contrast , nowadays the solar and kamland experiments have told us that @xmath90 ev@xmath91 and @xmath92 @xcite . two implications of this have been recognized : ( i ) if the process of sfp operates in the sun , it must be combined with the ordinary " flavor oscillations @xcite , and ( ii ) the sfp resonance condition , as given in eq . ( [ eq : res_diag ] ) , is satisfied for @xmath2b neutrinos in the rz . -[eq : h3 ] ) for @xmath93 ev@xmath91 , @xmath94 mev . the solid / dashed lines correspond to the neutrino / antineutrino entries.,scaledwidth=47.0% ] the second point is illustrated in fig . [ fig : levels_diag ] , where the four diagonal elements of the hamiltonian are plotted as a function of the distance from the solar center . here and later in this paper , the bp04 solar model @xcite is used . in most of the recent studies of the rz spin - flip @xcite , the two processes , flavor oscillations and spin - flavor precession , are treated as occurring in spatially separated regions . the resulting probability of @xmath1 conversion is then given as a product @xmath95 a notable exception is ref . @xcite , where it is noted that the evolution in the case of the rz does not decouple into two separate problems . the numerical conclusions presented in fig . 2 of that paper , however , are quite different from ours , as discussed later . the above treatment relies on the notion of resonance , as given by eq . ( [ eq : res_diag ] ) . fundamentally , this notion comes from the analysis of a two - level system , and even there it applies only in the limit of small mixing @xcite . we wish to examine whether it can be applied to the four - level system at hand . let us first briefly review the physics behind the standard small - angle msw effect @xcite . suppose we are given a two - level neutrino system with a hamiltonian @xmath96 if @xmath97 , flavor oscillations are suppressed , as observed already in @xcite . indeed , the mixing angle is given by @xmath98 and the oscillation amplitude is proportional to @xmath99 . the physical idea behind the resonance is that this suppression is lifted if the diagonal splitting becomes small in some region along the neutrino trajectory , as a result of the crossing of the levels . suppose @xmath100 over a distance @xmath101 . then , inside this interval , @xmath102 becomes _ the dominant part of the hamiltonian _ and hence drives oscillations . if @xmath103 is sufficiently large , it may be possible to obtain large flavor conversion . fits inside the interval @xmath103 . this condition gives nothing but the usual adiabaticity parameter of the resonance @xcite . ] with this in mind , let us return to the hamiltonian in eqs . ( [ eq : h]-[eq : h3 ] ) . now , the electron neutrino state , in addition to being coupled to the @xmath72 state via the magnetic moment interaction , is also coupled to @xmath70 state via the flavor oscillation term . a simple , but crucial observation is the following : for large values of the mixing angle , even if the diagonal @xmath85 and @xmath86 terms are equal , the hamiltonian is _ still not dominated _ by the off - diagonal magnetic terms @xmath104 , but by the off - diagonal flavor oscillation terms @xmath105 . thus , there is no reason to expect that the condition in eq . ( [ eq : res_diag ] ) should automatically lead to large @xmath3 conversion . this complication is , however , easily removed by going to a basis in which the large off - diagonal flavor oscillation terms are absent . consequently , we diagonalize @xmath106 and @xmath107 given in eqs . ( [ eq : h2 ] ) and ( [ eq : h3 ] ) by performing separate flavor rotations on neutrinos and antineutrinos . we find @xmath108 in eq . ( [ eq : hmass ] ) , @xmath109 , @xmath110 , @xmath111 and @xmath112 are the eigenvalues of the oscillation hamiltonian ( [ eq : h]-[eq : h3 ] ) for vanishing magnetic field , @xmath113^{\frac{1}{2}},\\ \label{eq : deltamibar } \bar\delta_{mi } & = & -\frac{a_e + a_n}{2 } \pm \left[\left(\delta\cos 2\theta+\frac{a_e - a_n}{2}\right)^2+\delta^2\sin^2 2\theta\right]^{\frac{1}{2}},\end{aligned}\ ] ] while @xmath114 and @xmath115 are the rotation angles , @xmath116,\\ \tan 2\bar\theta & = & \delta\sin 2\theta/[\delta\cos 2\theta+(a_e - a_n)/2 ] . \label{eq : mixingangle}\end{aligned}\ ] ] the barred quantities refer to antineutrinos . let us first discuss the general properties of these equations . when the matter potentials are negligible , @xmath109 and @xmath111 both approach @xmath117 , while @xmath110 and @xmath112 both approach @xmath118 . likewise , in the same limit the rotation angles @xmath114 and @xmath115 both approach the vacuum value @xmath119 . notice that in vacuum the structure of the off - diagonal magnetic interactions in the mass basis is the same as in the flavor basis . this structure is that of an antisymmetric tensor , eq . ( [ eq : h1 ] ) , which is invariant under an @xmath120 rotation applied to both indices . more generally , this is a manifestation of the fact that the operator in ( [ eq : lagr ] ) can not convert a majorana neutrino state in vacuum into its own antiparticle , whether this state is a mass eigenstate or a flavor eigenstate . in sufficiently dense matter , however , @xmath114 and @xmath115 are , in general , different , because the matter terms for neutrinos and antineutrinos have opposite signs . thus , in matter , direct transitions between all four states become allowed . and @xmath121 have opposite signs and hence tend to cancel . the cancellation is incomplete if one of the channels is enhanced with respect to the other one by matter effects . ] for the energies at which kamland is sensitive to solar antineutrinos , @xmath122 mev , the magnitude of @xmath123 is of order @xmath124 ev . by comparison , the value of the off - diagonal magnetic terms is at most of order @xmath125 mg @xmath126 ev , or at least an order of magnitude smaller . this makes the analysis in the mass basis quite straightforward . clearly , transitions between any two states with diagonal splitting of order @xmath127 ev will be suppressed . conversions between a pair of states with smaller diagonal splitting may take place , provided that @xmath114 and @xmath115 are sufficiently different , _ i.e. _ , the matter potential is sufficiently large . ) ( top panel ) and the mixing angles @xmath114 and @xmath115 ( bottom panel ) for @xmath93 ev@xmath91 , @xmath128 , @xmath94 mev . the solid / dashed lines correspond to the neutrino / antineutrino levels.,scaledwidth=47.0% ] in the top panel of fig . [ fig : eigen_mixing ] we show the diagonal elements of the hamiltonian eq . ( [ eq : hmass ] ) for the best - fit values of the oscillation parameters . obviously , their behavior is quite different from the diagonal elements in the flavor basis , plotted in fig . [ fig : levels_diag ] . in particular , in fig . [ fig : eigen_mixing ] there is no level crossing at @xmath129 . _ the mass eigenvalues do not intersect anywhere inside the sun . _ notice that the top two levels in fig . [ fig : eigen_mixing ] get close to each other for @xmath130 . the difference @xmath131 there is small , but non - zero , and _ a priori _ could drive a transition between these two levels . the possibility of this transition requires a more detailed study . the answer comes from considering the _ adiabaticity _ of the neutrino evolution . consider an idealized problem : a neutrino mass eigenstate moving through a magnetic field which is adiabatically turned on and then turned off . the neutrino will be partially converted into the corresponding antineutrino state _ while inside the field region _ , but upon exiting the region will revert back to being fully a neutrino . this idealized description applies in our case : ( i ) at the production point in the core , for the best - fit kamland mass splitting @xmath132 ev@xmath91 , the diagonal splitting dominates , suppressing the effects of the magnetic field . the neutrino is produced predominately in the heavy eigenstate ( the top line in the upper panel of fig . [ fig : eigen_mixing ] ) . ( ii ) in the region @xmath133 , the magnetic term drives partial @xmath134 conversion ( transitions between the top two states in the top panel of fig . [ fig : eigen_mixing ] ) . ( iii ) as the effects of the magnetic term are decreased ( mainly due to the alignment of @xmath114 and @xmath115 at smaller densities ) , the @xmath134 conversion is largely undone . the adiabaticity through the bulk of the rz zone is ensured by the physics of the rz fields : as mentioned in sec . [ sect : fields ] , any structures smaller than @xmath135 must have decayed away over the lifetime of the sun . state as a function of the distance from the solar center ( in units of the solar radius ) , for @xmath93 ev@xmath91 , @xmath136 mev , @xmath137 , @xmath138 mg.,scaledwidth=47.0% ] this behavior is illustrated in fig . [ fig : evolution ] , which shows that the electron antineutrino fraction @xmath139 indeed first grows , but then decreases to unobservable levels , as the effect of the magnetic term decreases in the outer part of the rz . the residual nonzero value of @xmath139 is due to the violation of adiabaticity near the top of the rz . in summary , except in the limit of small mixing angle , eq . ( [ eq : res_diag ] ) is not the right condition for describing resonant neutrino - antineutrino conversion . a correct physical criterion should be based on comparing the diagonal elements _ in the mass basis _ , not in the flavor basis . this `` mass basis level - crossing '' would be indeed realized in the sun , had the value of the mixing angle been small . for the large measured value of the mixing angle , however , the level - crossing is simply absent and the efficiency of the neutrino - antineutrino conversion is very low . let us now derive the general analytical condition for having resonant neutrino - antineutrino conversion . referring to fig . [ fig : eigen_mixing ] , we see that , _ a priori _ , either the two light levels or the two heavy levels could intersect . the corresponding analytical conditions for these two possibilities are @xmath140 and @xmath141 where the indices `` 1 '' and 2 " refer to the light and heavy states respectively ( the - " and + " signs in eqs . ( [ eq : deltami],[eq : deltamibar ] ) ) . written out explicitly , the above two conditions read @xmath142^{\frac{1}{2 } } -\left[\left(\sqrt{2}g_f n_e/2-\delta\cos 2\theta\right)^2+\delta^2\sin^2 2\theta\right]^{\frac{1}{2}},\end{aligned}\ ] ] where @xmath7 is @xmath143 for the condition in eq . ( [ eq : cond ] ) and @xmath144 for the condition in eq . ( [ eq : condm1 ] ) . since in the sun @xmath145 always , the above condition can be satisfied only for @xmath146 for @xmath147 and only for @xmath148 for @xmath149 . in other words , in the `` dark side '' @xcite only the two light levels can intersect , while in the `` light side '' , as preferred by the data , only the two heavy levels can . with this in mind , let us solve eq . ( [ eq : cond_a ] ) . we find , in addition to the trivial vacuum solution , @xmath150 , a solution @xmath151 this is the desired general condition for having resonant antineutrino production in the solar magnetic field . once again , for @xmath149 , it solves eq . ( [ eq : cond ] ) , while in the opposite case it solves eq . ( [ eq : condm1 ] ) . ( it is also worth reminding here that ( [ eq : cond_solution ] ) has been derived assuming the neutrino is a majorana particle . ) we indeed see that eq . ( [ eq : cond_solution ] ) and the conventionally used eq . ( [ eq : res_diag ] ) agree only for @xmath152 and are substantially different for large @xmath119 . in particular , as @xmath119 is increased beyond @xmath153 the level crossing as given by eq . ( [ eq : cond_solution ] ) disappears . it can be said that there is a `` topological '' difference between the cases of small and large mixing and eq . ( [ eq : analyt_theta ] ) gives the angle at which the topology changes . the ratio @xmath154 varies with the distance to the center of the sun . correspondingly , the numerical value of @xmath155 for which the disappearance takes place varies with @xmath118 ( which controls the location in the sun where the level crossing could have happened ) . in the central region , @xmath156 , while for the outer regions @xmath157 . the corresponding variation in @xmath155 is @xmath158 let us finally discuss when each of the level - crossings , ( [ eq : cond ] ) and ( [ eq : condm1 ] ) , can affect the neutrino evolution . the answer depends on the probability of finding the initial electron neutrino in each of the two neutrino mass eigenstates . this probability , in turn , depends on the flavor mixing angle at the production point . if the matter potential @xmath159 dominates over the vacuum term @xmath160 , the mixing angle at production approaches @xmath161 and the neutrino is produced almost entirely in the heavy neutrino mass eigenstate . in the opposite limit , the neutrino is produced in the superposition of both mass eigenstates , with the probabilities @xmath162 and @xmath163 . the rough boundary between these two regimes for the @xmath16410 mev @xmath2b neutrinos lies near @xmath165 ev@xmath91 : for smaller @xmath80 including the best - fit region of @xmath166 ev@xmath91 only the condition in eq . ( [ eq : cond ] ) matters , while in the opposite case both eq . ( [ eq : cond ] ) and eq . ( [ eq : condm1 ] ) do . for @xmath167 ev@xmath91 we can thus expect sizable antineutrino production both in the `` light '' and `` dark '' sides , while for @xmath168 ev@xmath91 the effect can happen only in the light side . to confirm this analysis , we have carried out a numerical scan of the oscillation parameter space , in the range @xmath169 ev@xmath91 . the calculation was carried out for the realistic numerical solar density and chemical composition profiles @xcite . for the magnetic field profile , we have taken the lowest toroidal ohmic eigenmode , as described in @xcite and in detail in @xcite . while higher eigenmodes could also be present , the choice of the lowest mode is expected to capture the essential physics : the field is axisymmetric , smooth , and vanishes at the rz / cz boundary . the only free parameter describing the field in this case is the overall mode normalization . we have performed scans for three different ( highly optimistic ) normalization values of the magnetic field , @xmath170 mg , also taking a highly optimistic value of the transition moment , @xmath171 . the results of the scan are shown in fig . [ fig : scan ] . the three colored regions on the left side ( small mixing ) indicate the regions where the @xmath0 flux for the chosen normalization values of the magnetic field would exceed the kamland bound . the corresponding regions in the dark " side , @xmath147 , are also shown . all the regions agree well with the analytical analysis of the previous subsection . the figure also shows the allowed regions from the kamland analysis of their reactor antineutrino spectrum @xcite , as well as the region that gives a fit to the solar neutrino data ( unfilled contours ) . the best - fit ( `` lma - i '' ) region lies at the intersection of the solar and kamland regions and , as can be clearly seen , has no overlap with the regions where the @xmath0 flux would be measurable . we note that a similar conclusion was reached in @xcite , using , however , an analytical criterion that differs from eq . ( [ eq : analyt_theta ] ) . the relatively large @xmath0 flux cited in that work was obtained by assuming the field strength of 50 mg at the neutrino production point , which was only later shown to be not allowed @xcite . compared to the case of the rz , the case of the cz requires a different treatment . the difference has to do with the turbulent nature of the cz magnetic fields . below , for completeness , we briefly review the relevant physics . the main idea is that random variations @xcite of the magnetic field on scales equal to , or smaller than , the neutrino oscillation length lessen the suppression of the oscillations by a large diagonal mass splitting . while the neutrino follows slow variations with scales longer than the oscillation length @xmath172 adiabatically , for shorter variations the adiabaticity condition is violated . the net result of the small - scale noise " is that the neutrino state vector performs a random walk " in the oscillation space , gradually receding from its original position . to be specific , consider a simple model in which the magnetic field changes on scales of @xmath173 . the probability of spin flip upon traversing the first correlation cell , @xmath174 , is @xmath175 ( simple vacuum oscillation on scales smaller than the oscillation length ) . since the fields in different @xmath174 intervals are uncorrelated , as the neutrino traverses many domains the probabilities , not amplitudes , add up . for solar neutrinos , the experimental bound on @xmath0 flux implies @xmath176 . so long as this is satisfied , the probability of spin flip after traveling the distance @xmath177 can be written as @xmath178 . for slow variations with @xmath179 , as already mentioned , the evolution is adiabatic and the conversion probability is exponentially suppressed . one can also imagine a case when the field domains are longer than @xmath180 , but have sharp boundaries . in such a case , the conversion probability is @xmath181 we see that in any case the conversion is most efficient for @xmath182 , namely , @xmath183 . let us apply this to solar neutrinos . once again , we are interested in the high - energy @xmath2b neutrinos , as these are the neutrinos the kamland bound applies to . these neutrinos are produced almost exclusively in the heavy mass eigenstate @xmath184 and then propagate adiabatically through the rz . in the cz , these neutrinos encounter turbulent magnetic fields and , if the transition moment is large , a small fraction of them gets converted into the light antineutrino mass eigenstate @xmath185 . the net result is the flux of neutrinos in the heavy eigenstate with possibly a small admixture of the antineutrinos in the light mass eigenstate . the light mass eigenstate , in turn , is detected as an electron antineutrino with probability @xmath186 . the relevant field fluctuations are those on scales of @xmath187 mev@xmath188 ev@xmath189 km . we will denote them by @xmath190 . overall , we get @xmath191 the oscillation parameters are known , @xmath192 , @xmath193 . the extent of the region of the strong random field can be estimated to be some fraction of the cz , say @xmath194 . finally , the main issue is to estimate the size of the amplitude of the small - scale noise @xmath190 . unfortunately , the theory and simulation efforts are not yet at the level to give a robust prediction of @xmath190 ( see @xcite for the state of the art on the simulation front ) . we will next consider two models for the magnetic field . a physically sensible estimate can be obtained by assuming that ( i ) the field on the large scales is of the order of its turbulent equipartition value ; and ( ii ) that the small - scale noise follows some kind of a scaling law @xmath195 typical of turbulent systems . such approach was adopted in @xcite and , in our opinion , is a clear improvement over naive models of delta - correlated noise . the largest scale of the turbulence , at which energy pumping takes place , is estimated to be of order the typical pressure scale height in the cz . in the inner part of the cz , the pressure scale height is of order @xmath196 km @xcite . this number can be estimated directly from the solar model @xcite , and is also consistent with the value of the mixing length @xmath197 km at the base of the cz given in @xcite . although the corresponding large - scale eddies reside in the interior of the cz , they manifest themselves on the solar surface as _ supergranules _ ( @xmath198 km ) . these should not be confused with surface granules ( @xmath199 km ) whose size is dictated by the scale height near the surface . the turbulent equipartition value for the large scale turbulent fields near the bottom of the cz is estimated as @xmath200 kg @xcite . putting the numbers together , we find @xmath201 this means that , for this field model , even with the most optimistic values of the neutrino magnetic moment kamland will not have a chance to see antineutrinos from the cz spin flip unless its sensitivity is improved by as least one order of magnitude . the estimate just given uses a simple model for the turbulent magnetic field . it is possible that the field in the cz instead has a `` fibril '' nature @xcite , i.e. , is expelled by the turbulence and combines in isolated flux tubes . it has been argued that the total energy of the cz ( thermal + gravitational + magnetic ) is reduced by the fibril state of the magnetic field by avoiding the magnetic inhibition of convection @xcite . let us then consider a model in which the neutrinos travel through spatially separated flux bundles . the conversion probability in this case depends on the transverse size of a bundle . given that a sunspot typically contains @xmath202 mx @xcite and assuming superequipartition field strength inside the bundle of @xmath203 kg , one finds the transverse size of the bundle that is actually close to optimal , @xmath204 km . the thickness of the flux rope comes out to be close to the oscillation length for the @xmath2b neutrinos , a remarkable coincidence indeed ! since the total toroidal flux in the cz is estimated to be of the order of @xmath54 mx @xcite , a given neutrino encounters only several of such flux bundles . the resulting conversion probability in the most optimistic case is then close to the kamland sensitivity bound , as can be easily checked . we have shown that the large measured value of the flavor mixing angle _ qualitatively _ changes the treatment of the neutrino interaction with the magnetic field in the radiative zone of the sun . the physics of the effect is most transparent in the basis of the mass eigenstates , rather than in the flavor basis . we have shown that in the mass basis , for the allowed values of the oscillation parameters , there is no neutrino - antineutrino level crossing and , as a result , no measurable antineutrino production in the rz . a correct level crossing condition , eq . ( [ eq : cond_solution ] ) , was derived , and a detailed numerical scan carried out to confirm this conclusion . we emphasize that the recent determination of the neutrino oscillation parameters is crucial for reaching this conclusion : for small flavor mixing angle , it was possible to have large @xmath3 conversion even with magnetic fields an order of magnitude below the bound , @xmath205 mg @xcite . moreover , as seen in fig . [ fig : scan ] , even a relatively minor shift of the best - fit point , from @xmath206 ev@xmath91 and @xmath207 to , _ e.g. _ , @xmath208 ev@xmath91 and @xmath209 , would have made the resonant antineutrino production possible . the field of solar neutrinos has truly entered the precision measurement stage ! one may wonder if our analysis is completely self - consistent . after all , the oscillation parameters we use are found assuming only the flavor oscillations and neglecting possible magnetic @xmath134 transitions . this , however , is entirely justified , because the kamland bound on the @xmath0 flux ensures that the magnetic transitions can be present only at very small levels and thus can not affect the fit to the oscillation parameters . in the convective zone , the solar neutrino propagate through turbulent magnetic fields . our estimate of the probability of the spin - flavor flip in this case gives a rate that , at present , is likewise not observable . this conclusion disagrees with earlier analyses . the source of the disagreement is not in the treatment of the neutrino evolution , but in the treatment of the solar magnetic fields , which we believe were significantly overestimated in the earlier studies . overall , our conclusion is that , despite achieving a clearly impressive level of sensitivity to @xmath0 from the sun , kamland has not yet reached the level where it would be placing a meaningful constraint on the neutrino transition moment and/or solar magnetic fields . on the other hand , kamland has reached the point where it may begin probing the `` optimistic '' scenario , @xmath210 . clearly , further sensitivity improvements through accumulation of additional statistics is very desirable . i owe special thanks to andrei gruzinov for countless always very clear and helpful discussions of the solar magnetic fields and plasma physics in general , to matthias rempel for a very helpful discussion and for pointing me to an excellent set of references , and to cecilia lunardini for many useful suggestions on the draft . i also acknowledge stirling colgate for a useful discussion of the physics of the convective zone fields . additionally , i benefited greatly from stimulating correspondence with evgeny akhmedov and timur rashba that followed the initial posting of the preprint of this paper . this work was supported by the department of energy , under contract w-7405-eng-36 . as described in sect . [ sect : cz ] , simple physical arguments give the following estimates for the probability of spin - flavor flip in random magnetic fields : @xmath211 more generally , the probability of conversion is proportional to a fourier component of the field correlation function @xcite @xmath212 if @xmath213 is a slowly decaying function ( compared to @xmath214 ) with a vanishing slope for small @xmath215 , indicating no small - scale noise , for example @xmath216 $ ] , the resulting @xmath139 is exponentially small , as expected for adiabatic evolution . the quantity of interest in the isotropic turbulence model is @xmath217 . @xcite performed an estimate based on a simple kolmogorov turbulence . within this model , the effects of the magnetic field on the character of the turbulence are neglected and the simple kolmogorov scaling law for velocities is applied , @xmath218 , where @xmath219 is the energy dissipation rate per unit mass . the field was assumed to scale with @xmath220 the same way as velocities , _ i.e. _ , @xmath221 . more generally , if the magnetic energy becomes comparable to the kinetic energy of the fluid , the character of the turbulence changes ( see , e.g. , @xcite , pp . 138 - 142 ) . the rate of energy transfer to small scales is decided by the interaction of alfvn waves in plasma . recent investigations @xcite argue that in the presence of magnetic fields the turbulence becomes anisotropic in @xmath222 space ( see @xcite , appendix a , for a clear summary ) . while a detailed discussion of the latest models of magnetohydrodynamics turbulence is clearly beyond the scope of this work , the preceding example show that the exponent in the scaling law ( [ eq : bscaling ] ) is not known precisely . the correlation function can then be parameterized as @xmath223 , yielding upon integration @xmath224 as expected on physical grounds . s. fukuda _ et al . _ [ super - kamiokande collaboration ] , phys . rev . lett . * 86 * , 5651 ( 2001 ) [ arxiv : hep - ex/0103032 ] ; s. fukuda _ et al . _ [ super - kamiokande collaboration ] , phys . lett . * 86 * , 5656 ( 2001 ) [ arxiv : hep - ex/0103033 ] . q. r. ahmad _ et al . _ [ sno collaboration ] , phys . rev . lett . * 87 * , 071301 ( 2001 ) [ arxiv : nucl - ex/0106015 ] . q. r. ahmad _ et al . _ [ sno collaboration ] , phys . rev . lett . * 89 * , 011301 ( 2002 ) [ arxiv : nucl - ex/0204008 ] . q. r. ahmad _ et al . _ [ sno collaboration ] , phys . rev . lett . * 89 * , 011302 ( 2002 ) [ arxiv : nucl - ex/0204009 ] . m. b. smy [ super - kamiokande collaboration ] , nucl . suppl . * 118 * , 25 ( 2003 ) [ arxiv : hep - ex/0208004 ] . y. gando _ et al . _ [ super - kamiokande collaboration ] , phys . rev . lett . * 90 * , 171302 ( 2003 ) [ arxiv : hep - ex/0212067 ] . s. n. ahmed _ et al . _ [ sno collaboration ] , phys . lett . * 92 * , 181301 ( 2004 ) [ arxiv : nucl - ex/0309004 ] . k. eguchi _ et al . _ [ kamland collaboration ] , phys . * 92 * , 071301 ( 2004 ) [ arxiv : hep - ex/0310047 ] . b. aharmim _ et al . _ [ sno collaboration ] , phys . d * 70 * , 093014 ( 2004 ) [ arxiv : hep - ex/0407029 ] . b. aharmim _ et al . _ [ sno collaboration ] , arxiv : nucl - ex/0502021 . t. toshito [ super - kamiokande collaboration ] , arxiv : hep - ex/0105023 . y. ashie _ et al . _ [ super - kamiokande collaboration ] , phys . lett . * 93 * , 101801 ( 2004 ) [ arxiv : hep - ex/0404034 ] . k. eguchi _ et al . _ [ kamland collaboration ] , phys . lett . * 90 * , 021802 ( 2003 ) [ arxiv : hep - ex/0212021 ] . t. araki _ et al . _ [ kamland collaboration ] , arxiv : hep - ex/0406035 . n. fornengo , m. maltoni , r. t. bayo and j. w. f. valle , phys . d * 65 * , 013010 ( 2002 ) [ arxiv : hep - ph/0108043 ] . m. guzzo , p. c. de holanda , m. maltoni , h. nunokawa , m. a. tortola and j. w. f. valle , nucl . b * 629 * , 479 ( 2002 ) [ arxiv : hep - ph/0112310 ] . a. friedland , c. lunardini and c. pena - garay , phys . b * 594 * , 347 ( 2004 ) [ arxiv : hep - ph/0402266 ] . m. m. guzzo , p. c. de holanda and o. l. g. peres , phys . b * 591 * , 1 ( 2004 ) [ arxiv : hep - ph/0403134 ] . m. c. gonzalez - garcia and m. maltoni , phys . d * 70 * , 033010 ( 2004 ) [ arxiv : hep - ph/0404085 ] . o. g. miranda , m. a. tortola and j. w. f. valle , arxiv : hep - ph/0406280 . a. friedland , c. lunardini and m. maltoni , phys . d * 70 * , 111301 ( 2004 ) [ arxiv : hep - ph/0408264 ] . a. friedland and c. lunardini , arxiv : hep - ph/0506143 , to appear in phys . d. j. f. beacom and p. vogel , phys . * 83 * , 5222 ( 1999 ) [ arxiv : hep - ph/9907383 ] . e. k. akhmedov and j. pulido , phys . b * 553 * , 7 ( 2003 ) [ arxiv : hep - ph/0209192 ] . a. friedland and a. gruzinov , astropart . * 19 * , 575 ( 2003 ) [ arxiv : hep - ph/0202095 ] . b. c. chauhan , j. pulido and e. torrente - lujan , phys . d * 68 * , 033015 ( 2003 ) [ arxiv : hep - ph/0304297 ] . o. g. miranda , t. i. rashba , a. i. rez and j. w. f. valle , phys . * 93 * , 051304 ( 2004 ) [ arxiv : hep - ph/0311014 ] . o. g. miranda , t. i. rashba , a. i. rez and j. w. f. valle , phys . d * 70 * , 113002 ( 2004 ) [ arxiv : hep - ph/0406066 ] . a. b. balantekin and c. volpe , arxiv : hep - ph/0411148 . a. cisneros , astrophys . space sci . * 10 * , 87 ( 1971 ) . m. b. voloshin and m. i. vysotsky , yadernaya fizika * 44 * , 845 ( 1986 ) [ sov . j. nucl . phys . * 44 * , 544 ( 1986 ) ] . l. b. okun , yadernaya fizika * 44 * , 847 ( 1986 ) [ sov . j. nucl . phys . * 44 * , 546 ( 1986 ) ] . m. b. voloshin and m. i. vysotsky , l. b. okun , zh . fiz . , 446 ( 1986 ) [ sov . jetp*64 * , 446 ( 1986 ) ] . e. k. akhmedov , phys . b * 213 * , 64 ( 1988 ) . c. lim and w. marciano , phys . d * 37 * , 1368 ( 1988 ) . r. s. raghavan , a. b. balantekin , f. loreti , a. j. baltz , s. pakvasa and j. pantaleone , phys . d * 44 * , 3786 ( 1991 ) . a. b. balantekin and f. loreti , phys . d * 45 * , 1059 ( 1992 ) ; p. vogel and j. f. beacom , phys . d * 60 * , 053003 ( 1999 ) [ arxiv : hep - ph/9903554 ] ; e. torrente - lujan , phys . b * 494 * , 255 ( 2000 ) [ arxiv : hep - ph/9911458 ] . j. schechter and j. w. valle , phys . d * 24 * , 1883 ( 1981 ) [ erratum - ibid . d * 25 * , 283 ( 1981 ) ] . z. daraktchieva _ et al . _ [ munu collaboration ] , phys . b * 564 * , 190 ( 2003 ) [ arxiv : hep - ex/0304011 ] . g. g. raffelt , phys . lett . * 64 * , 2856 ( 1990 ) ; g. g. raffelt , phys . * 320 * , 319 ( 1999 ) . j. e. kim , phys . d * 14 * , 3000 ( 1976 ) . a. friedland , in the proceedings of _ frontiers in contemporary physics - ii _ , march 2001 , nashville , tn ; arxiv : hep - ph/0106042 . l. wolfenstein , phys . d * 17 * , 2369 ( 1978 ) . s. p. mikheev and a. y. smirnov , sov . j. nucl . phys . * 42 * , 913 ( 1985 ) [ yad . fiz . * 42 * , 1441 ( 1985 ) ] . e. k. akhmedov , zhurnal eksperimentalnoy i teoreticheskoy fiziki * 95 * , 1195 ( 1989 ) [ sov . phys . jetp * 68 * , 690 ( 1989 ) ] a. de gouvea , a. friedland and h. murayama , phys . b * 490 * , 125 ( 2000 ) [ arxiv : hep - ph/0002064 ] . a. friedland , phys . lett . * 85 * , 936 ( 2000 ) [ arxiv : hep - ph/0002063 ] . f. n. loreti and a. b. balantekin , phys . d * 50 * , 4762 ( 1994 ) [ arxiv : nucl - th/9406003 ] .

there are several excellent review articles on trapped bose - condensed atomic gases written at a relatively introductory level @xcite , with emphasis on the dynamics of the condensate at @xmath0 . in the present lectures , i will concentrate on the interplay between the condensate and non - condensate components . this topic requires a more sophisticated analysis based on the concepts and methods of many body theory . my lectures will attempt to give a very basic introduction to this kind of approach . the primary audience i have in mind are graduate students and postdocs coming from atomic and laser physics , rather than condensed matter theory . however , i hope the experts will also learn something from these lectures . in his lectures , fetter @xcite has given a detailed review of excitations in a trapped dilute bose gas at @xmath0 . he shows that a convenient way of discussing these excitations is to start from the time - dependent gross - pitaevskii ( gp ) equation for the macroscopic wavefunction @xmath1 . linearizing around the equilibrium value @xmath2 , one finds of the macroscopic wavefunction are given by the well - known bogoliubov coupled equations . for a uniform bose - condensed gas , the excitation frequencies are those first discussed by bogoliubov in 1947 @xcite . at @xmath0 , one can assume that all the atoms are in the bose condensate described by @xmath1 . in contrast , in these lectures , my main topic will be to review our understanding of what excitations are in a bose - condensed gas at _ finite _ temperatures , when there is a large number of atoms in the non - condensate ( in trapped gases , the non - condensate is often referred to as the `` thermal cloud '' ) . i will make some contact with the ideas discussed in the lectures by burnett @xcite . the first half of these lectures ( sections [ sec : elementary excitations and density fluctuations in normal systems ] - [ sec : why are excitations in a bose fluid so interesting ] ) deals with excitations whose very existence depends on self - consistent mean fields ( of various kinds ! ) , rather than on the collisions between atoms . in the standard language developed in condensed matter physics in the 1960 s , this means the excitations are in the `` collisionless region '' . the second half of the lectures ( sections [ sec : hydrodynamic oscillations in a trapped bose gas ] - [ sec : first and second sound in a uniform bose - condensed gas ] ) deals with excitations in the collision - dominated hydrodynamic region . i review the two - fluid hydrodynamic equations such as given by landau @xcite , generalized to include a trap . i give an explicit microscopic derivation of such two - fluid equations in a trapped bose gas . this extends recent work of zaremba , griffin and nikuni ( zgn ) @xcite to include the case when the condensate is not yet in local equilibrium with the non - condensate atoms . the type of questions i want to address in these lectures include : 1 . what is the difference between an elementary excitation and a collective mode ? 2 . can we _ isolate _ the dynamical role of the condensate on the nature of the excitations ? 3 . at @xmath0 , with a pure condensate , excitations in a gas must be oscillations of the condensate . in contrast , above @xmath3 , the excitations are not related to a condensate . what happens to an excitation as we go from @xmath0 to @xmath4 . how does the excitation get rid of its `` condensate '' dressing ? what is the essential physics behind the different mean - field theories of excitations which have been discussed in the recent literature : gross - pitaevskii , hartree - fock - popov , hartree - fock - bogoluibov ? what is the physics behind the dreaded beliaev second - order approximation ? this is the first approximation which includes damping of the elementary excitations even at @xmath0 @xcite . why are the excitations and collective modes in a bose - condensed system uniquely interesting , compared to all other many body systems ? the key reason is , of course , that the condensate couples and hybridizes single - particle excitations with density fluctuations . above @xmath3 , these two excitation branches are uncoupled . once this is understood , one sees why more detailed and systematic experimental studies of excitations in trapped atomic gases at finite temperature are of great importance . to give a careful discussion of all these questions would need 10 lectures . in these 3 lectures , i will give a speeded - up version . i will often use a _ uniform _ weakly interacting bose gas to illustrate the structure of the theory . while i will only sketch the math , i will still try to give a flavour of what is involved . the approach i will use to discuss these questions is based on the field - theoretic formulation of a bose - condensed system of particles . as i review elsewhere in this volume @xcite , this powerful formalism was introduced by beliaev @xcite in 1957 and extensively developed in the golden period : 1958 - 1965 . i will introduce this formalism in a very schematic manner - but even experimentalists will find it useful to know some of the language " used in this approach . there are other methods to deal with collective modes in bose gases but they are not as useful at _ isolating _ the dynamical role of the condensate , or dealing with finite temperatures . while i will always have trapped atomic bose gases in mind , much of the general theory @xcite is valid for any bose - condensed fluid ( gas or liquid ) . thus i will often make references to superfluid @xmath5he , pointing out similarities with bose gases . we work with quantum field operators : @xmath6 these operators satisfy the usual bose commutation relations , such as @xmath7=\delta ( { \bf r } - { \bf r'})$ ] . of course , if we have several hyperfine atomic states , then we have different field operators @xmath8 , where @xmath9 is the hyperfine state label ( the analogue of a spin label ) . all observables can be written in terms of these quantum field operators . two important examples are the local density operator : @xmath10 and the hamiltonian @xmath11 { \hat\psi}({\bf r } ) + { 1 \over 2 } g \int d{\bf r } \hat{\psi}^+({\bf r } ) \hat{\psi}^+({\bf r } ) \hat{\psi}({\bf r } ) \hat{\psi}({\bf r } ) . \label{eq : elementary excitations3}\ ] ] in these lectures , the two - particle interaction will always be approximated by a @xmath12-wave scattering length , appropriate to a dilute bose gas at very low temperatures @xcite . thus @xmath13 , with @xmath14 ( we set @xmath15 throughout this article ) . what we want to calculate are various kinds of correlation functions involving different local operators : @xmath16 in these tutorial lectures , i will not bother to distinguish between the various kinds of correlation functions ( time - ordered , retarded , etc ) or even the difference between correlation functions and response functions . these are discussed in all the standard textbooks on many body theory @xcite but are only important when one is doing systematic calculations . a convenient summary of the formalism is given in a recent review article on homogeneous weakly interacting bose gases @xcite . the density response function @xmath17 involves four quantum field operators and is an example of a two - particle green s function @xmath18 . one can measure @xmath19 by coupling a weak external field to the local density in the system @xmath20 linear response theory gives @xmath21 for the density response ( see chapters 5 and 9 of ref.@xcite and chapter 2 of ref.@xcite ) . for a uniform system with @xmath22 , we have @xmath23 fourier transforming @xmath24 and @xmath25 , the linear response equation in ( [ eq : elementary excitations7 ] ) reduces to @xmath26 if @xmath27 has a pole at @xmath28 , ( for further details , see section [ sec : density fluctuation spectrum in the mean - field approximation ] ) then when @xmath29 and @xmath30 of the external potential satisfy @xmath28 , @xmath31 as given by ( [ eq : elementary excitations8 ] ) can be very large even though @xmath32 is small . we note that @xmath33 and thus clearly the pole at @xmath28 is the signature of an oscillating density fluctuation . the basic correlation function in the field - theoretic approach is given by @xmath34 this single - particle green s function @xmath35 involves two quantum field operators . it describes creating an atom at @xmath36 , let it propagate through the system to @xmath37 and then destroying the atom . all other higher - order correlation functions such as @xmath38 can be constructed out of combinations of @xmath39 . for example , the _ lowest order _ contributions to the density response function in ( [ eq : elementary excitations5 ] ) are @xmath40 as we discuss in section [ sec : green s function formulation ] , the terms in the last line of ( [ eq : elementary excitations11 ] ) vanish in a normal bose system but are finite for @xmath41 . the poles of @xmath35 correspond to what are called elementary excitations ( or quasiparticles ) . in a uniform system , we have @xmath42 one can show @xcite that these single - particle excitations determine the thermodynamic properties of interacting systems . however , it is very difficult to directly measure the spectrum of @xmath39 since one needs an external field which couples to @xmath43 , ie , an atom reservoir . later we will see that what makes a bose - condensed system unique is that we can easily access @xmath39 as a result of the effects of the bose condensate . finally we introduce the key idea of a single - particle self - energy through dyson s equation : @xmath44 where @xmath35 is the interacting single - particle green s function , @xmath45 is the non - interacting single - particle green s function and all effects of the two - particle interactions are contained in the self - energy function @xmath46 . in a uniform system , we can fourier transform this dyson equation to give @xmath47 where @xmath48 . this is easily solved to give @xmath49}.\label{eq : elementary excitations15}\ ] ] thus we see that @xmath35 may have a single - particle pole at the quasiparticle energy @xmath50 in general , @xmath51 , where @xmath52 describes the damping of the single - particle excitations . field - theoretic calculations @xcite involve a systemmatic ( diagrammatic ) procedure to calculate @xmath53 and from this to obtain @xmath54 . we note that the self - energy @xmath55 is by definition highly non - perturbative . as an illustration , let us consider the self - energy to first order in @xmath56 . this hartree - fock approximation is shown in fig.1 . in our @xmath12-wave approximation , the total self - energy is simply @xmath57 and therefore ( [ eq : elementary excitations15 ] ) gives @xmath58}. \label{eq : elementary excitations18}\ ] ] hence the normal hf excitation energy has the dispersion relation @xmath59 the simplest approximation for the density response function introduced in section [ sec : elementary excitations and density fluctuations in normal systems ] is @xcite @xmath60 in a uniform bose system , the fourier transform of this gives @xmath61\over \omega_1 - \omega_2 - \omega } a({\bf k } , \omega_1 ) a({\bf k } - { \bf q } , \omega_2 ) . \label{eq : density fluc2}\ ] ] here @xmath62 is the bose distribution function and @xmath63 single - particle spectral density @xmath64 . if we use @xmath65 as given in ( [ eq : elementary excitations18 ] ) , we find ( [ eq : density fluc2 ] ) reduces to @xmath66 we note that @xmath67 has a continuum of poles given by @xmath68 . it is easy to understand the physics which gives rise to this `` ideal - gas '' spectrum . the fourier transform of the local density operator in ( [ eq : elementary excitations2 ] ) is given by ( for a uniform system ) @xmath69 clearly @xmath70 creates a `` particle - hole '' density fluctuation with the following features : @xmath71 the spectrum ( [ eq : density fluc3 ] ) describes a broad incoherent superposition of particle - hole states @xcite and is not a true collective mode , such as we discuss next . the mean field approximation ( mfa ) for the density response is also called other names : scf ( self - consistent field ) , rpa ( randon phase approximation ) , but all involve the same physics . the mfa was introduced by bohm and pines in the period 1951 - 1953 , work which had a pivotal effect in our understanding collective effects in all many - particle systems . we recall the linear response expression in ( [ eq : elementary excitations7 ] ) , where @xmath72 is the _ full _ density response function for interacting bose gas . if we introduce the self - consistent hartree mean - field : @xmath73 then we can approximate the linear response equation in ( [ eq : elementary excitations7 ] ) by @xmath74 . \label{eq : density fluc7}\ ] ] the system is assumed to respond as if the atoms propagate independently ( as described by @xmath75 but are moving in an effective field @xmath76 . in a uniform system , the fourier transform of ( [ eq : density fluc7 ] ) is @xmath77,\label{eq : density fluc8}\ ] ] which gives the well - known mfa expression for @xmath72 @xcite : @xmath78 this result for @xmath79 may have new poles given by zeros of the denominator , @xmath80 assuming that it is distinguisable from the incoherent ideal gas density fluctuation spectrum given by ( [ eq : density fluc3 ] ) , this pole at @xmath81 is called a zero sound mode ( a plasmon in charged systems ) . this language was first introduced in 1957 by landau in fermi systems @xcite but the concept is generally applicable in any interacting many body system . physically , it is clear that zero sound is a `` collisionless '' density oscillation arising from dynamic self - consistent mean fields . we make a few comments about such `` zero sound '' collective modes : 1 . @xmath82 and @xmath83 ( the poles of @xmath54 ) are _ both _ states of an interacting many - particle system . however the collective mode disappears if there are no interactions , while the single - particle excitations still exist in a non - interacting gas . because of the low density , dynamic mean fields in normal systems are too weak to allow the existence of a well - defined ( weakly - damped ) zero sound mode in the bose gases of current interest ( for @xmath84 ) . if @xmath85 , we can then approximate @xmath86 . the situation is quite different for @xmath87 , when a coherent mean - field due to the condensate is present . one _ also _ expects a collective pole to appear in @xmath88 in the collision - dominated hydrodynamic region . however this sound wave pole is induced by rapid collisions producing local equilibrium , as we discuss in sections [ sec : hydrodynamic oscillations in a trapped bose gas ] - [ sec : first and second sound in a uniform bose - condensed gas ] . ordinary sound is not described by the mfa discussed above , ie , it is not the result of dynamic mean fields . all the formalism we have been discussing for @xmath89 can be extended in a natural way to bose - condensed systems @xcite making use of the fundamental decomposition which separates out the condensate and non - condensate parts of the quantum fields : @xmath90 here the average is over a restricted ensemble @xcite consistent with @xmath91 . the most profound way of doing this is to add a symmetry - breaking field @xmath92 \label{eq : green function2}\ ] ] and work with @xmath93 , taking the limit @xmath94 at the end . this gives the system a `` hunting license '' @xcite to have finite value of @xmath95 and one finds @xmath96 a key point is that if @xmath97 , then a direct consequence is the existence of `` anomalous '' or `` off - diagonal '' propagators @xcite @xmath98 these describe new condensate - induced correlations between non - condensate atoms at different space - time points @xmath99 and @xmath100 . in a sense , these anomalous correlation functions are as important as the macroscopic wavefunction @xmath101 . we also note that @xmath102 is the `` pair '' function that burnett @xcite discusses in detail , using a different formalism . > from now on , we leave the symmetry - breaking label on the averages implicit . in addition , if @xmath103 we find that correlation functions involving three non - condensate field operators can be finite @xmath104 in particular , @xmath105 . this describes the condensate - induced coupling of non - condensate density fluctuations @xmath106 and the single - particle field fluctuations ( see section [ sec : why are excitations in a bose fluid so interesting ] ) . clearly one has to work with a @xmath107 matrix single - particle propagator @xmath35 when @xmath91 , namely @xmath108 where we have introduced a @xmath107 matrix green s function for the non - condensate atoms @xmath109 beliaev @xcite effectively showed ( in modern matrix notation ) @xmath110 where we use the standard convention that repeated indices @xmath111 are summed over . this is the famous dyson - beliaev equation , involving a 2 x 2 matrix self - energy @xmath112 . clearly all components @xmath113 will share the same single - particle excitation spectrum . in a uniform system , we have @xmath114 which is a set of linear algebraic equations which are easy to solve for @xmath115 and @xmath116 . we will now use this beliaev formalism to discuss various simple approximations for the self - energies @xmath117 . the interaction energy in ( [ eq : elementary excitations3 ] ) , namely @xmath118 splits into various distinct contributions when we use ( [ eq : green function1 ] ) to separate out the condensate parts ( see fig . 2 ) . at @xmath119 we can ignore the @xmath120 and @xmath121 contributions because so few atoms are in non - condensate . this is the famous bogoliubov approximation @xcite , and it is equivalent to the linearized gp theory . in beliaev language , the bogoliubov self - energies are shown in fig . 3 . for a uniform gas , these give @xcite @xmath122 } \nonumber\\ { \tilde g}_{12}(p , \omega ) & = & { -n_cg\over \omega^2 - \left[\epsilon^2_p + 2\epsilon_p n_cg\right]}\ .\label{eq : green function11}\end{aligned}\ ] ] these bogoliubov single - particle green s functions contain the same physics as discussed in fetter s lectures @xcite . they clearly have poles at the frequencies @xmath123 , where @xmath124 at low @xmath125 , the single - particle excitation is phonon - like @xmath126 , with the bogoliubov phonon velocity @xmath127 it turns out that this simple @xmath0 bogoliubov theory already exhibits most of the structure which will always arise in bose - condensed systems . this is why it plays the role of the `` h - atom '' in discussions of bose - condensed fluids @xcite . in this approximation , as we have stressed , the starting point assumes that all atoms are in the condensate and hence @xmath128 . however , we can still use @xmath129 in ( [ eq : green function11 ] ) to estimate the depletion . this makes use of the relation @xmath130 which is an exact formula for @xmath131 ( see ch.4 of ref . @xcite ) in terms of the so - called single - particle _ spectral density _ [ see ( [ eq : density fluc2 ] ) ] . the spectral density given by the bogoliubov approximation above is @xmath132 showing the characteristic negative energy pole at @xmath133 this last feature emphasizes that creating an atom in a bose - condensed system involves a coherent weighted combination of creating an excitation _ and _ destroying an excitation . calculating the depletion , one finds @xmath134 \nonumber\\ & = & \int d{\bf q}\left\{n^0(e_q)u^2_q -n^0(-e_q)v^2_q\right\}\nonumber\\ & = & \int d{\bf q}\left\{v^2_q + n^0(e_q)(u^2_q + v^2_q)\right\}\nonumber\\ & = & { 8\over 3 } n_c \left(n_c a^3\over\pi\right)^{1/2 } , \label{eq : green function16}\end{aligned}\ ] ] a result first obtained by bogoliubov @xcite and reviewed in detail by fetter @xcite . we have made use of the key identity which the bose distribution satisfies , @xmath135 . \label{eq : green function17}\ ] ] we see from ( [ eq : green function16 ] ) that the @xmath0 depletion of the condensate formally arises from the negative energy pole in @xmath136 . it is easy to evaluate the density response function @xmath137 in ( [ eq : density fluc1 ] ) using the bogoliubov spectral density in ( [ eq : green function15 ] ) . the negative energy poles are seen to give rise to new poles at @xmath138 which have finite weight even at @xmath0 . these are in addition to the `` normal '' particle - hole poles of @xmath67 at @xmath139 , as given by ( [ eq : density fluc3 ] ) . at finite @xmath140 , a lot of atoms are thermally excited out of the condensate in a dilute bose gas . these `` normal '' self - energies must be included . in the beliaev self - energy formalism , this was first done by popov in 1965 @xcite . we have to add the _ ordinary _ hartree - fock self - energies , as shown in fig.4 . this gives the first - order popov approximation ( the excited atoms are treated as as ideal bose gas , and thus the approximation is not self - consistent ) @xmath141 these self - energies lead to [ compare with ( [ eq : green function11 ] ) ] @xmath142 with @xmath143[eq : beyond3 ] where @xmath144 the `` chemical potential '' @xmath145 as defined in ( [ eq : beyond4 ] ) here was introduced by hugenholtz and pines @xcite at @xmath146 ( and generalized to @xmath147 later @xcite ) . hp showed that if @xmath148 had a gapless excitation spectrum in the @xmath149 limit , then the true chemical potential @xmath150 must satisfy @xmath151 . in the popov approximation , calculation gives @xcite @xmath152 and thus @xmath153 , where @xmath154 is the ideal gas condensate density at temperature @xmath140 . the generalized gp equation determines the chemical potential , which we find convenient to denote by @xmath155 . in the popov approximation , this equation is given by @xmath156 and hence in a uniform gas , we see that @xmath157 . in the popov approximation , the single - particle green s functions @xmath158 in ( [ eq : beyond2 ] ) have poles given by @xmath159^{1/2}.\label{eq : beyond7}\ ] ] this is formally the same as the @xmath0 bogoliubov excitation frequency , except that now @xmath154 is temperature - dependent . for a more detailed discussion , see chapter 3 of the recent review by shi and griffin @xcite . the _ full _ hartree - fock bogoliubov ( hfb ) approximation involves calculating @xmath117 self - consistently using the complete matrix @xmath107 beliaev propagator , as shown in fig . we note that above @xmath3 , the hfb reduces to the usual self - consistent hartree - fock approximation . moreover , the hfb is the best _ single - particle _ approximation for a bose - condensed system , in a variational sense . within this limitation , the hfb will give the best single - particle approximation for the thermodynamic properties . however , there is one bad " aspect of the full hfb which was noticed in the 1960 s and which has been the subject of a several recent theoretical papers on bose - condensed gases . namely , the hfb does not obey the hugenholtz - pines ( hp ) theorem . in a uniform gas , this means that the single - particle excitations will have an energy gap in the long wavelength limit _ _ @xmath160 0 . it is easy to check this by calculating the two chemical potentials introduced earlier : @xmath161 the static equation of motion for @xmath162 using the hfb approximation gives @xmath163{\sqrt n_{c}}= \mu_{c}\sqrt n_{c}\rightarrow\mu_{c } = g(n+\tilde{n}+\tilde{m}).\label{eq : beyond9}\ ] ] one `` solution '' of this problem is to leave out @xmath164 ( which is the source of the problem ) but keep @xmath165 . this corresponds to the self - consistent hfp(opov ) approximation @xcite which , while approximate , has several nice features . it gives the correct excitation spectrum for both @xmath166 0 and @xmath84 . moreover , the spectrum is gapless at all temperatures . the origin of this problem " with the hfb is clear . the hfb keeps all self - energies which are first order in the interaction @xmath56 . however , by computing these self - energy diagrams using self - consistent @xmath113 propagators , one is clearly bringing in terms to all orders in @xmath56 . this brings one into dangerous territory ! moreover , one easily can check that @xmath164 must be at least of order @xmath56 . thus the @xmath167 contribution to @xmath168 is at least of @xmath169 ) . this suggests that to `` fix - up '' the hfb as a theory of excitations , we have to include all self - energy contributions to at least second order in @xmath56 . moreover , it is not obvious that it is consistent to calculate @xmath170 self - consistently @xcite and ignore @xmath164 , since the lowest order interaction contributions to both quantities are found to be of the same order . indeed , one can show for a uniform gas that keeping the lowest order asymptolic correction to the ideal gas result gives @xmath171 , where @xmath172 is the critical density of an ideal gas at temperature @xmath140 @xcite . however , we note that an expansion around the ideal gas results is not necessarily a good approximation in trapped gases ( see also the discussion of the results in ( [ eq : two - fluid7 ] ) . at @xmath146 , the popov approximation is a poor guide for corrections to the bogoliubov approximation since a calculation analogous to ( [ eq : green function16 ] ) gives @xmath173 @xcite . so , finally , we come to the work of beliaev @xcite who , in 1957 , evaluated the @xmath117 self - energies keeping all contributions up to order @xmath174 . this was originally done at @xmath146 . for finite temperatures @xmath175 , beliaev s work has been recently extended by several authors @xcite . when one includes the second - order diagrams for @xmath117 , we note that one must be careful to also treat the hfb first - order diagrams correctly to order @xmath174 since there are many terms which cancel . as expected , the beliaev - type second order calculation cures all the problems of the first - order hfb approximation . we make a few comments on the results for a uniform bose gas . the self - energies are found to be @xmath176 with the renormalized interaction defined by @xmath177\label{eq : beyond11}\ ] ] and @xmath178.\label{eq : beyond12}\ ] ] here the anomalous and normal densities , @xmath179 and @xmath180 , are computed using the hfp approximation for @xmath181 . @xmath182 is the second - order approximation to the @xmath183-matrix , as shown in fig . 6 . using these results to calculate the chemical potential , we find ( for @xmath184 @xmath185 this last result gives precisely the hfb result for @xmath186 , and hence the hp relation is satisfied . in the last line of ( [ eq : beyond13 ] ) , we have used the key result that for @xmath187 , @xmath188 , as proven at all temperatures by talbot and griffin @xcite . we note that one can rewrite the expression in ( [ eq : beyond13 ] ) in the alternative form @xmath189,\label{eq : beyond14}\ ] ] where we have introduced a `` renormalized '' anomalous density @xmath190.\label{eq : beyond15}\end{aligned}\ ] ] in contrast to @xmath191 in ( [ eq : beyond12 ] ) , @xmath192 has no ultraviolet divergence from large @xmath193 contributions . proukakis , burnett and coworkers @xcite have pointed out that the self - consistent ladder diagram approximation for the @xmath194matrix ( see ref.@xcite and ch.4 of ref.@xcite ) can be expressed in terms of @xmath195 . in a uniform gas , this relation is simply @xmath196 . this suggests that the effect of @xmath197 can be partially included by working in terms of the self - consistent @xmath194matrix @xcite . we have seen in section [ sec : beyond the bogoliubov approximation ] that beliaev s second - order calculation satisfies the hp theorem . it has another feature whose significance was only realized in the early 1960 s . using the above @xmath0 beliaev expression ( [ eq : beyond14 ] ) for the chemical potential @xmath150 , one finds at @xmath0 ( a result first found by lee and yang using a different approach ) @xmath198 , \label{eq : approximation1}\ ] ] where we have used [ see ( [ eq : green function16 ] ) ] @xmath199 one also finds that the phonon pole of @xmath200 has a velocity given by @xmath201 , \label{eq : approximation3}\ ] ] again expressed in terms of the total density @xmath202 . one may easily check that @xmath203 that is , the beliaev phonon excitation at @xmath0 has a velocity precisely equal to the _ compressional _ sound velocity . this last result raises the basic question : why does the elementary excitation ( pole of @xmath204 have a velocity corresponding to that of a density fluctuation ( pole of @xmath205 ? later this strange identity was proven to be correct at @xmath0 to _ all _ orders of perturbation theory by gavoret and nozi@xmath206res @xcite . to be precise , at @xmath207 in any bose - condensed fluid ( liquid or gas ) , gn showed ( for @xmath184 @xmath208 where @xmath209 is given by the compressibility as in ( [ eq : approximation4 ] ) . we discuss this equivalence in section [ sec : why are excitations in a bose fluid so interesting ] . let us consider the full time - dependent hfb equation of motion for @xmath210 , given by @xmath211 \phi({\bf r},t ) . \label{eq : approximation6}\ ] ] there are various approximations to this key equation which have been used in recent discussions of trapped bose gases : * _ gp : _ ignore @xmath165 and @xmath164 completely . this is valid at @xmath146 @xcite . * _ static hfp : _ keep @xmath212 but set @xmath213 and @xmath214 @xcite . * _ static hfb : _ keep @xmath215 but set @xmath216 and @xmath217 @xcite . as we have noted , this produces an energy gap in the excitation spectrum ( since @xmath218 . * _ dynamic hfb : _ treat all dynamic mean - fields due to @xmath219 and @xmath220 on an equal basis in a generalized mean - field calculation of the density response function @xmath221 . one finds @xcite that the poles of @xmath72 given by this kind of calculation @xmath222 are _ identical _ to the poles of @xmath181 given by the second - order beliaev approximation @xmath223 ) . recent work on excitations at @xmath2240 in trapped gases is easily understood in terms of such a generalized linear response calculation of @xmath72 . for example , minguzzi and tosi @xcite only keep fluctuations in @xmath225 and @xmath226 but @xmath227 is completely ignored ( dynamic hfpopov ) ; in contrast , giorgini @xcite keeps fluctuations in both @xmath226 and @xmath227 but only those induced by the condensate fluctuations @xmath225 . let us briefly discuss the excitation frequencies in trapped gases at finite temperatures @xcite . for @xmath228 @xmath3 , gp theory @xmath229 predictions @xcite are in excellent agreement with the jila @xcite and mit @xcite experiments . for @xmath230 @xmath3 , however , there are significant differences between most current theoretical results and the measured excitation energies @xcite . the problem clearly is that mean - field theories must include the collective dynamics of _ both _ the condensate and the non - condensate . the beliaev treatment does this , as do the papers @xcite mentioned above , as well as work based on a collisionless kinetic equation @xcite . these improved treatments can correctly obtain the in - phase rigid oscillation of the condensate and non - condensate equilibrium profiles ( the so - called kohn mode ) , as discussed in section [ sec : in - phase oscillation of the condensate and non - condensate ] . we have seen in section [ sec : physics of the beliaev approximation ] that at @xmath146 , the elementary excitations have the same spectrum as density fluctuations . in fact , one can prove this equivalence to be a direct consequence of the bose broken symmetry , @xmath231 . thus , this equivalence is valid at all @xmath232 . of course , as soon as @xmath233 , the phonon velocity is no longer simply given by the compressional sound velocity ( [ eq : approximation4 ] ) as found at @xmath146 . this fundamental equivalence was only proven and understood in the early 1960 s ( gavoret and nozi@xmath206res @xcite , hohenberg and martin @xcite , bogoliubov @xcite ) . it was implicitly assumed by landau @xcite in his 1941 paper and later by feynman @xcite in his 195354 papers on excitations . of course , the clear distinction between elementary excitations and density fluctuations in many - body systems was only clarified in the period starting from 1957 . only then were bose fluids realized to be very special @xcite . today , we understand this key hidden assumption of the work of landau and of feynman has its microscopic basis in @xmath234 . let us try to understand the origin of this equivalence of the field fluctuation and density fluctuation spectrum . the density operator in a bose - condensed system can be decomposed using ( [ eq : green function1 ] ) as follows : @xmath235 where the first three terms correspond to @xmath236 . clearly density fluctuations in a bose - condensed system ( ie , @xmath237 ) have a direct coupling to the single - particle field fluctuations , due to the possibility of atoms coming in or out of condensate reservoir . in momentum space , we recall that ( [ eq : excitations1 ] ) is equivalent to @xmath238 making the same decomposition as in ( [ eq : excitations1 ] ) for the density response function ( [ eq : elementary excitations5 ] ) gives @xmath239 we note that in the bogoliubov approximation , we have @xmath240 and the equivalence of @xmath241 and @xmath181 spectrum is then trivial ( see also @xcite ) @xmath242 to illustrate how the hybridization of single - particle and density fluctuation arises , we discuss the time - dependent hfp approximation on which ref . @xcite is based . for simplicity , we consider the uniform gas case . in the mfa of section [ sec : density fluctuation spectrum in the mean - field approximation ] , we have @xmath243 , \label{eq : excitations5}\ ] ] and hence for @xmath244 , @xmath245 using this result in the hfp equation of motion ( [ eq : approximation6 ] ) , ie , with @xmath246 , we find @xmath247 in this approximation , which only keeps the @xmath248 fluctuations induced by the condensate , the equation of motion for @xmath249 reduces to the gp equation , but with @xmath250 . thus the characteristic poles will be determined by the zeroes of @xmath251 \propto [ \omega^2 - ( \epsilon^2_p + 2 g n_c\epsilon_p ) ] [ 1 - 2g\chi^0 ] - 4g^2 2n_c \epsilon_p \chi^0 . \label{eq : excitations8}\ ] ] this is the same denominator ( up to a common factor ) exhibited by the response functions obtained by minguzzi and tosi @xcite . this calculation illustrates how the bogoliubov single - particle mode and the zero sound density fluctuation are coupled and hybridized . more generally , one can prove that ( at all @xmath252 : @xmath253 where @xmath254 is a bose broken - symmetry vertex function which vanishes if @xmath255 . using the dielectric formalism " developed in the early 1970 s ( for a review , see ch.5 of ref . @xcite ) , one can prove that the self - energies @xmath117 are such that @xmath72 and @xmath181 have the same poles and that there are none specifically associated with @xmath256 . the advantage of this diagrammatic formalism @xcite is that one manifestly sees that the poles of @xmath181 and @xmath72 are identical , within a given approximation for the building blocks " of the dielectric formalism . we now see why we can study elementary excitations from the density fluctuation spectrum when @xmath237 . to directly measure @xmath257 , we need an external perturbation which can change the number of particles in the system . bec indirectly provides us with such a probe ! this equivalence of the single - particle excitations with density fluctuations lies at the heart of the phenomenon of superfluidity . it essentially _ restricts _ the possible excited states to density fluctuations and thus ensures the stability of superfluid motion . however , as nozires says in a very lucid review article @xcite , `` that the @xmath146 superfluid equations merge with those of ordinary hydrodynamics does not alter the fact that real understanding must be based on a microscopic description based on long - range order induced by a condensate '' . in the previous sections , we have been discussing collective oscillations of the condensate and non - condensate in the mean - field or _ collisionless _ region . we now turn to the collision - dominated _ hydrodynamic _ region @xmath258 , where @xmath259 is the wavelength of the collective mode and @xmath260 is the collisonal mean - free - path of the elementary excitations @xcite . experiments in trapped bose gases are just now starting to probe the hydrodynamic two - fluid region but this region is very rich in new physics and deserves careful study . it is unfortunate that the collisionless region at @xmath0 described by the gross - pitaevskii equation is also commonly referred to as `` hydrodynamic '' theory @xcite . i guess we are stuck with this terminology , although the terms `` quantum hydrodynamics '' or `` superfluid hydrodynamics '' would be useful compromises . we will limit our analysis to _ finite _ temperatures @xmath261 , where the number of atoms in the non - condensate is comparable to the condensate @xmath262 . we will concentrate on the _ derivation _ of two - fluid hydrodynamic equations starting from a simple microscopic model and try to clarify the physics . these two - fluid hydrodynamic equations are expressed in terms of fluctuations of the two components : @xmath263 and @xmath264 . we can use them to `` derive '' the two - fluid equations in the conventional landau form , which are written in terms of the fluctuations of local thermodynamic variables ( temperature , entropy , pressure , etc ) . we mainly consider the linearized version of the zgn - type two - fluid equations , which give the hydrodynamic normal modes of the coupled system of two components . as a concrete example , we discuss the _ in - phase _ rigid motion of the equilibrium density profiles of the condensate and non - condensate . we need to find equations of motion for both the condensate and the non - condensate . as a first approximation @xcite , we can use the time - dependent hf popov equation of motion ( see section [ sec : physics of the beliaev approximation ] ) , for the order parameter @xmath265 @xmath266\phi({\bf r } , t ) , \label{eq : hf popov equation motion1}\ ] ] where @xmath267 and @xmath268 is the local non - condensate density . it is clear that this equation of motion for @xmath269 is not closed since it involves @xmath270 if we treat the non - condensate in ( [ eq : hf popov equation motion1 ] ) _ statically _ , this means that the condensate moves in the _ static _ hartree - fock mean field of non - condensate given by @xmath271 @xcite more generally , we need to have an equation of motion for the fluctuations in @xmath272 of the non - condensate ( ie , the density of the excited atoms ) , which will be discussed in section [ sec : dynamics of the non - condensate atoms ] . we can rewrite equation of motion in ( [ eq : hf popov equation motion1 ] ) in the _ quantum hydrodynamic _ variables for phase and amplitude , defined by @xmath273 where the superfluid velocity is given by @xmath274 using ( [ eq : hf popov equation motion3 ] ) , ( [ eq : hf popov equation motion1 ] ) can be shown to be equivalent to ( all quantities depend on @xmath275 and @xmath183 ) @xmath276 & = & -\mbox{\boldmath$\nabla$}\mu_c,\label{eq : hf popov equation motion5}\end{aligned}\ ] ] where the _ superfluid chemical potential _ is given by ( in the dynamic hfp approximation ) @xmath277 the first term in ( [ eq : hf popov equation motion6 ] ) is the `` quantum pressure '' term , which is _ ignored _ in the so - called thomas - fermi ( tf ) approximation to the condensate equation of motion @xcite . if we omit the non - condensate term , equations ( [ eq : hf popov equation motion5 ] ) and ( [ eq : hf popov equation motion6 ] ) describe the well - known gross - pitaevskii approximation @xcite . following ref.@xcite , we base our discussion on the simplest possible _ kinetic equation _ for the distribution function @xmath278 valid in the semi - classical limit : @xmath279 this distribution function is determined by the quantum boltzmann equation @xmath280 f({\bf r , p } , t ) = \left.{\partial f({\bf r , p } , t)\over\partial t}\right|_{\mbox{collision } } , \label{eq : semi - classical limit2}\ ] ] where the effective potential @xmath281.\label{eq : semi - classical limit3}\ ] ] includes the dynamic hf field . once we solve ( [ eq : semi - classical limit2 ] ) for @xmath282 we can find the _ non - condensate _ density from @xmath283 in our simple model , the excited atoms are still particle - like , ie , @xmath284 this last fact leads to many simplifications but restricts the results to finite temperatures close to @xmath3 . we now give the explicit form for the collision term in ( [ eq : semi - classical limit2 ] ) : @xmath285 + c_{12}[f ] . \label{eq : semi - classical limit6}\ ] ] the term @xmath286 $ ] describes collisions between excited atoms and is given by @xcite , @xmath287 & = & 2g^2 \int{d{\bf p}_1\over(2\pi)^3}{d{\bf p}_2\over(2\pi)^3}{d{\bf p}_3\over(2\pi)^3}(2\pi)^3\delta({\bf p}+{\bf p}_1 - { \bf p}_2 - { \bf p}_3)\nonumber\\ & \times & 2\pi\delta(e+e_1-e_2-e_3 ) \left[(1+f)(1+f_1)f_2f_3-ff_1(1+f_2)(1+f_3)\right ] . \label{eq : semi - classical limit7}\end{aligned}\ ] ] in contrast , @xmath288 $ ] describes collisions involving one condensate atom . this kind of collision brings about equilibrium between the _ excited _ atoms and _ condensate _ atoms and it is given by ( these collision terms were ignored in ref.@xcite ) @xmath289&= & 2g^2 \int{d{\bf p}_1\over(2\pi)^3}{d{\bf p}_2\over(2\pi)^3}{d{\bf p}_3\over(2\pi)^3}(2\pi)^3\delta({\bf p}_1 + { \bf p}_s - { \bf p}_2 - { \bf p}_3)\nonumber\\ & & \qquad\hbox{}\times 2\pi\delta(e+\epsilon_c - e_2-e_3)(2\pi)^3[\delta({\bf p}-{\bf p}_1 ) -\delta({\bf p}-{\bf p}_2 ) - \delta({\bf p}-{\bf p}_3)]\nonumber\\ & & \qquad\hbox{}\times \left[n_c(1+f_1)f_2f_3-n_cf_1(1+f_2)(1+f_3)\right ] , \label{eq : semi - classical limit8}\end{aligned}\ ] ] with @xmath290 and @xmath291 . these two collision terms were first derived by kirkpatrick and dorfman @xcite , who considered a uniform gas and worked in a frame in which @xmath292 . our present calculation @xcite is for trapped bose gas and works in the lab frame . as a result , ( [ eq : semi - classical limit8 ] ) takes into account that a condensate atom has energy @xmath293 and momentum @xmath294 , where @xmath155 is defined in ( [ eq : hf popov equation motion6 ] ) . we note that : * @xmath286 $ ] conserves number , momentum and energy of the colliding excited atoms and therefore @xmath295 = 0 \nonumber\\ & \int & d{\bf p}{\bf p } c_{22}[f ] = 0 \nonumber\\ & \int & d{\bf p } e_p c_{22 } [ f ] = 0 \label{eq : semi - classical limit9}\end{aligned}\ ] ] * @xmath288 $ ] conserves momentum and energy of the colliding atoms , with the condensate atoms having energy @xmath296 and momentum @xmath297 . therefore , we have @xmath298 = 0 \nonumber\\ & \int & d{\bf p } ( e_p - \epsilon_c ) c_{12 } [ f ] = 0 , \label{eq : semi - classical limit10}\end{aligned}\ ] ] but @xmath299 \equiv \gamma_{12}[f ] \neq 0 . \label{eq : semi - classical limit11}\ ] ] the fact that @xmath300 is finite follows since @xmath301 does _ not _ conserve the number of _ excited _ atoms ; @xmath300 will be referred to as a source term . as usual in dealing with the collision - dominated hydrodynamic region , we will assume that the collisions @xmath286 $ ] produce _ local thermal equilibrium _ among the excited atoms . this is described by the _ bose distribution : @xmath302}-1 } , \label{eq : semi - classical limit12}\ ] ] where now @xmath303 and @xmath304 all depend on @xmath305 . one can easily checks that @xmath306 $ ] vanishes exactly . this uses the key identity for bose distribution given by ( [ eq : green function17 ] ) . in contrast , one finds that substituting @xmath307 in ( [ eq : semi - classical limit8 ] ) gives @xmath308 & = & 2g^2 \int{d{\bf p}_1\over(2\pi)^3}{d{\bf p}_2\over(2\pi)^3}{d{\bf p}_3\over(2\pi)^3}(2\pi)^3\delta({\bf p}_1+{\bf p}_s - { \bf p}_2 - { \bf p}_3)\nonumber\\ & \times & 2\pi\delta(e_1+\epsilon_c - e_2-e_3)(2\pi)^3[\delta({\bf p}-{\bf p}_1 ) -\delta({\bf p}-{\bf p}_2 ) - \delta({\bf p}-{\bf p}_3)]\nonumber\\ & \times & \left[n_c - n_c e^{-\beta({\tilde\mu } ' - \mu_c)}\right](1 + f_{01 } ) f_{02 } f_{03 } , \label{eq : semi - classical limit13}\end{aligned}\ ] ] where @xmath309 in a _ linearized _ theory . we note that @xmath310 $ ] in ( [ eq : semi - classical limit13 ] ) vanishes when @xmath311 ie , only if the condensate atoms are in _ diffusive equilibrium _ with the non - condensate atoms . the same kind of factor arises in the theory of the growth dynamics of a condensate in trapped bose gases developed by gardiner , zoller and coworkers @xcite . we next derive _ hydrodynamic equations _ for the non - condensate atoms by taking moments of the kinetic equation : @xmath312 f({\bf r , p } , t ) = c_{22}[f ] + c_{12}[f ] . \label{eq : semi - classical limit15}\ ] ] we obtain @xcite @xmath313 & \rightarrow & { \partial{\tilde n}\over\partial t}+\mbox{\boldmath$\nabla$}\cdot{\tilde n}{\bf v}_n = \gamma_{12}\nonumber\\ \int d{\bf pp } [ \mbox{kinetic eq . } ] & \rightarrow & m{\tilde n}\left({\partial\over\partial t}+{\bf v}_n\cdot\mbox{\boldmath$\nabla$}\right){\bf v}_n = -\mbox{\boldmath$\nabla$}{\tilde p}-{\tilde n}\mbox{\boldmath$\nabla$}u - m({\bf v}_n - { \bf v}_s)\gamma_{12}\nonumber\\ \int d{\bf p}p^2 [ \mbox{kinetic eq . } ] & \rightarrow & { \partial { \tilde p}\over\partial t}+\mbox{\boldmath$\nabla$}\cdot({\tilde p}{\bf v}_n ) = -{2\over 3}{\tilde p}\mbox{\boldmath$\nabla$}\cdot{\bf v}_n\nonumber\\ & \quad & \quad + { 2\over 3}[\mu_c + { 1\over 2}m({\bf v}_n - { \bf v}_s)^2 - u]\gamma_{12}. \label{eq : semi - classical limit16}\end{aligned}\ ] ] here @xmath300 is defined in ( [ eq : semi - classical limit11 ] ) and @xmath314 with the local fugacity @xmath315 . in sections [ sec : hydrodynamic oscillations in a trapped bose gas ] and [ sec : dynamics of the non - condensate atoms ] , we derived a set of 7 equations for non - condensate and the 4 equations for the condensate . this gives a coupled set of equations for the variables : @xmath316.\label{eq : two - fluid1}\ ] ] > from ( [ eq : semi - classical limit13 ] ) , one sees @xmath300 is function of @xmath317 , which in turn depends on @xmath318 and @xmath304 . thus our system of hydrodynamic equations is closed and can be solved since it involves 11 local variables and 11 equations . for reasons discussed earlier , the explicit effect of the @xmath319 collisions has disappeared in the hydrodynamic equations ( [ eq : semi - classical limit16 ] ) . however , the @xmath319 collisions justify and enforce the local equilibrium form ( [ eq : semi - classical limit12 ] ) used for @xmath320 which describes the excited atoms . in contrast , @xmath301 is still explicitly present in ( [ eq : semi - classical limit16 ] ) through the @xmath300 source term . this means that the condensate is not in _ diffusive _ equilibrium with non - condensate . we recall from ( [ eq : semi - classical limit13 ] ) that @xmath321 , \label{eq : two - fluid2}\ ] ] and thus @xmath322 $ ] vanishes only when @xmath323 . one might expect that this condition would arise in the case of strong @xmath301 collisions . a detailed study @xcite shows that even in this case , one must allow for fluctuations in the variable @xmath324 and thus we can not simply set @xmath325 since @xmath326 in ( [ eq : semi - classical limit16 ] ) , we have to generalize the hf popov description to ensure that [ compare with the result in ( [ eq : hf popov equation motion5 ] ) ] @xmath327 this is needed for the continuity equation to be correct @xmath328 where @xmath329 as discussed in detail in ref.@xcite , a simple microscopic model based on the neglect of anomalous pair correlation functions @xmath330 leads to both ( [ eq : two - fluid3 ] ) and ( [ eq : semi - classical limit16 ] ) . we can linearize our coupled two - fluid hydrodynamic equations given by ( [ eq : hf popov equation motion5 ] ) and ( [ eq : semi - classical limit16 ] ) around the static equilibrium solution , where @xmath331 . the fluctuations @xmath332 are precisely the quantities which can be directly measured in trapped gases . the only static equilibrium functions we need are @xmath333 and @xmath334 , which are given by ( [ eq : semi - classical limit17 ] ) , @xmath335 where @xmath336 and @xmath337 is given by the _ static _ hfp equation ( [ eq : beyond6 ] ) for @xmath338 . we note that these static thermodynamic properties for a bose - condensed gas are quite different from the expressions first given by lee and yang ( 1958 ) for a uniform gas @xcite . they calculated only the first - order interaction corrections to the thermodynamic properties of an ideal bose gas , using @xmath339 here @xmath340 is the critical density at the temperature @xmath140 for an ideal uniform bose gas . the non - perturbative approximation we use for thermodynamic properties [ as illustrated by ( [ eq : two - fluid6 ] ) ] is consistent with the lee - yang results when we expand to first order in @xmath56 @xcite . we can use our generalized zgn equations to derive two - fluid equations in the traditional form first given by landau ( 1941 ) . in linearized form , these are @xcite @xmath341 where we have included an external potential , @xmath342 is the local entropy density and @xmath343 of course , the landau equations describe the case when the superfluid and normal fluid are in dynamic local equilibrium with each other . the landau form of two - fluid equations is very natural for superfluid @xmath5he , for which they were developed . they involve the usual thermodynamic variables : pressure , entropy and temperature fluctuations . however , in a trapped gas , the more natural variables are those used in a zgn - type formulation @xcite , namely @xmath344 and @xmath345 . in the generalized zgn equations discussed in ref . @xcite , the fluctuations around static equilibrium involve fluctuations in @xmath346 , namely [ see ( [ eq : two - fluid2 ] ) ] @xmath347 this leads to an additional equation of motion for @xmath348 , with a characteristic relaxation time @xmath349 . this gives rise to a new relaxational mode of the condensate . the zgn limit @xcite corresponds to @xmath350 while the landau two - fluid limit corresponds to @xmath351 while the relaxational mode is not included , the zgn equations appear to give a reasonable first approximation for the other hydrodynamic modes , as shown in the explicit results given in ref.@xcite . we note that landau two - fluid equations given by ( [ eq : two - fluid8 ] ) have been recently used by shenoy and ho @xcite to work out the hydrodynamic normal modes of a trapped superfluid gas . the landau two - fluid equations , which assume total local equilibrium , are valid at all temperatures . our present analysis is restricted to finite @xmath352 , but could be generalized to the case of very low temperatures . at @xmath0 , of course , the distinction between the collisionless and hydrodynamic region is not a useful one ( see , moreover , the comment at the end of section [ sec : why are excitations in a bose fluid so interesting ] ) . however , at @xmath0 , we have @xmath353 and @xmath354 and then the landau two - fluid equations formally reduce to hydrodynamic equations for _ one _ component : @xmath355 these equations are the basis of recent work by pitaevskii and stringari @xcite . in conjuction with the local density approximation @xcite @xmath356 + u_{\rm ext}({\bf r } ) , \label{eq : two - fluid12}\ ] ] where @xmath357 is the chemical potential of a _ uniform _ interacting bose - condensed gas at @xmath0 , such as given by ( [ eq : approximation1 ] ) , one may use ( [ eq : two - fluid11 ] ) to find collective mode frequencies past the mfa ( or gp approxiamation ) . two interesting collective modes exhibited by the zgn equations involve _ rigid _ centre - of - mass oscillations of the equilibrium density profiles : @xmath358 one finds @xcite an in - phase solution ( kohn mode ) with @xmath359 , with a frequency given by the trap frequency @xmath360 ; and an out - of - phase solution , with @xmath361 and @xmath362 in opposite directions . the latter is the _ analogue _ of the second sound mode in liquid @xmath5he . first let us work out the static equilibrium solution @xmath363 of ( [ eq : semi - classical limit15 ] ) , which must satisfy @xmath364 = 0 . \label{eq : oscillation of the cond2}\ ] ] the solution is easily verified to be @xmath365 where ( within the thomas - fermi approximation for @xmath366 ) @xmath367 . \label{eq : oscillation of the cond4}\end{aligned}\ ] ] using ( [ eq : oscillation of the cond3 ] ) , we find @xmath368 is given by ( [ eq : two - fluid6 ] ) , with @xmath369 we next consider the in - phase solution described by ( [ eq : oscillation of the cond1 ] ) , with @xmath370 ( not dependent on the position @xmath275 ) . this is a solution of zgn equations with @xmath371 , where @xmath360 is the parabolic trap frequency in @xmath372 direction @xmath373 . the proof proceeds as follows ( in the linearized theory ) . given ( [ eq : oscillation of the cond1 ] ) , we have @xmath374 and hence @xmath375 \nonumber \\ & = & \mbox{\boldmath$\nabla$ } \left({\mbox{\boldmath$\eta$ } } \cdot [ 2 g \mbox{\boldmath$\nabla$}{\tilde n}_0 + g \mbox{\boldmath$\nabla$}n_{c0 } ] \right).\label{eq : oscillation of the cond7}\end{aligned}\ ] ] this result is equivalent to [ using ( [ eq : oscillation of the cond4 ] ) ] @xmath376 with @xmath377 the solution of ( [ eq : oscillation of the cond8 ] ) is clearly @xmath378 in a similar way , one can show that the hydrodynamic equations for the non - condensate lead to the same equation of motion ( [ eq : oscillation of the cond8 ] ) . the rigid motion of the non - condensate density profile corresponds to a distribution function given by @xmath379 this can be verified to solve the kinetic equation ( [ eq : semi - classical limit15 ] ) , using the fact that @xmath380 = 0 $ ] , and leads to the expression in ( [ eq : oscillation of the cond1 ] ) , namely @xmath381 this in - phase dipole mode was first exhibited by zgn in the hydrodynamic region @xcite . it is an important test of the proper treatment of the non - condensate dynamics . this kind of mode is generic , existing in the collisionless region as well @xcite . the simplicity of the zgn two - fluid equations allow a very transparent discussion of first and second sound in a uniform gas @xcite . we find two sound - wave solutions : @xmath382 , @xmath383 . first sound has a velocity given by ( to lowest order in @xmath56 ) @xmath384 and mainly involves the non - condensate , with @xmath385 ( but in - phase ) . second sound has a velocity given by ( to lowest order in @xmath56 ) @xmath386 and mainly involves the oscillation of the condensate , with @xmath387 ( and out - of - phase ) . it is clear that second sound is the collision - dominated hydrodynamic mode which is the _ analogue _ of the collisionless bogoliubov phonon ( [ eq : green function13 ] ) which we discussed in section [ sec : green s function formulation ] . the frequency vanishes ( becomes `` soft '' ) above @xmath3 and it is the expected goldstone mode of the bose broken - symmetry @xcite . in contrast to liquid @xmath5he , both second and first sound in a bose gas involve density fluctuations ( ie , second sound is not a temperature wave in a gas ! ) . can we study first and second sound pulse propagation in cigar - shaped traps ? in this volume , ketterlee @xcite reviews the beautiful pulse propagation experiments carried out at mit at low temperatures in the collisionless region @xcite . if we could get into the hydrodynamic region , one should be able to see _ two _ pulses propagate roughly with the speeds @xmath388 and @xmath389 given above . this would be very dramatic and a direct confirmation of the superfluid dynamics described by the two - fluid equations . inside the broad , relatively uniform condensate distribution along the axis of the cigar - shaped trap , one could excite both first and second sound pulses , with the relative intensity of each propagating pulse given by @xcite @xmath390 the velocity of second sound would give [ using ( [ eq : sound velocity2 ] ) ] a direct measurement of the condensate density @xmath391 . perhaps of more interest , the velocity of the first sound pulse would give a relatively _ direct _ measurement [ using ( [ eq : sound velocity1 ] ) ] of the non - condensate density @xmath392 `` underneath '' the condensate . in sections [ sec : dynamics of the non - condensate atoms ] and [ sec : the two - fluid hydrodynamic equations ] , we discussed the linearized form of the generalized two - fluid equations @xcite which describe how the condensate relaxes to equilibrium with the excited atoms . these hydrodynamic equations with @xmath393 ( i.e. @xmath394 ) should also be useful in the study of how a condensate grows and comes into equilibrium with the thermal cloud @xcite . while based on a different formalism , our theory has points of contact with the work by gardiner , zoller and co - workers on the kinetics of condensate growth @xcite . however , our analysis , while assuming the existence of a condensate , takes the dynamical evolution of the non - condensate fully into account . the two - fluid dynamics of trapped bose - condensed gases promises to be a very rich subject , as discussed in ref . the region where @xmath393 is not easily studied in superfluid @xmath5he , largely because local thermal equilibrium is almost too easy to reach in a liquid . however , some years ago pitaevskii @xcite treated the regime very close to @xmath395 , when the superfluid density @xmath396 was not necessarily in equilibrium with the normal fluid density @xmath397 . one of the interesting aspects of trapped bose gases is that even if the centre of the trap is at large enough density so that one is in the hydrodynamic region described by dynamic local thermal equilibrium , this description will always break down in the low density non - condensate thermal cloud far enough from the trap centre . recent theoretical studies @xcite have shown that the cross - over region between the hydrodynamic and collisionless regions in this low density region plays the dominant role as the source of damping of certain kinds of hydrodynamic oscillations . more generally , i would like to emphasize that the study of collective modes in trapped bose gases finally provides us with the opportunity of making a quantitative test of the complex dynamics of a system with a bose - broken symmetry . as i discuss elsewhere in this volume @xcite , the complexity of dealing with a liquid such as superfluid @xmath5he never allowed one to make a thorough confrontation of the microscopic theory of bose - condensed systems with experiment @xcite . bose gases finally allow us to do this , in a much cleaner fashion @xcite . i would hope that future studies on trapped bose gases will also stimulate new interest in the dynamics of superfluid @xmath5he as a bose - condensed liquid . the work presented here on the two - fluid hydrodynamic equations has been done in close collaboration with eugene zaremba and tetsuro nikuni . in addition , i have benefitted from stimulating discussions with many of my colleagues , and especially would like to thank k.burnett , m.l . chiofalo , e. cornell , w. ketterle , l.p . pitaevskii , g.v . shlyapnikov , d.m . stamper - kurn , s.stringari and p. zoller . 1 . hartree - fock self - energies in a normal bose gas . various interaction terms involving condensate ( wiggly line ) and non - condensate ( solid line ) atoms . self - energy diagrams in the bogoliubov approximation @xmath229 . self - energy diagrams in the first - order popov approximation . the propagators are for an ideal bose gas . 5 . self - consistent hartree - fock - bogoliubov ( hfb ) approximation for self - energies . ladder diagram approximation for the @xmath183-matrix .

it is extraordinary but true that the galaxy of the local group whose global stellar populations are least well observed is our own milky way . deeply immersed within the optically opaque , dusty lanes of our galaxy s spiral arms , astronomers have been frustrated for centuries in their attempts to map the stellar populations of the milky way . a complete census of our galaxy for every member of even one class of star would have seemed like an impossible goal even a decade ago . major advances in instrumentation are transforming this daunting task from near - impossibility to increasingly likely . wide - field , high - resolution , sensitive surveys , particularly in the near - infrared and x - ray parts of the electromagnetic spectrum ( where the galaxy is relatively transparent ) , are key to locating and characterizing all members of one or more classes of stellar object . the goal of the project described in this paper is to detect and spectrographically characterize at least 90% of the wolf - rayet stars in the milky way within ten years . a set of well - defined tests of stellar evolution theory will follow from detections of complete samples of wolf - rayet and other related stars . for example , the radial abundance gradient across our galaxy and the increase of the wr / o number - ratio with increasing z suggests that many more wr stars will be found in the inner parts of the milky way than in the outer regions . in addition , our previous hst survey of the hii regions in the sciii galaxy ngc 2403 @xcite suggested that the distribution of wolf - rayet and red supergiant stars ( rsg ) is a sensitive diagnostic of the recent star - forming history of these large complexes : young cores of o and wr stars are surrounded by older halos containing rsg . theory predicts that the number - ratio wr / o increases with increasing metallicity ; thus , relatively fewer wr stars form at lower z. we will also be able to determine if superclusters , dominated by wolf - rayet stars , are common in the milky way . finally , we note that wr stars are predicted to end their lives as supernovae , and in rare cases as gamma ray bursts . the wolf - rayet stars in the milky way may be abundant enough for one to erupt as a type ib or ic supernova within a few generations . this comes from the assumption that the mw contains @xmath06000 wr stars , each lasting @xmath1 yrs . the clear identification of a wr star as the progenitor of one of these eruptions would be a dramatic confirmation of a key prediction of stellar evolution theory . wolf - rayet ( wr ) stars are massive ( with initial masses greater than @xmath2 at @xmath3 ) stars with strong winds ( @xmath4 ) displaying the heavier elements created by what are normally internal nuclear processes . distinctive spectra with strong , broad emission lines of helium , and either nitrogen ( wn ) or carbon ( wc ) are the defining observational characteristics of wr stars . as they have relatively short lifetimes ( about @xmath5 years ) , wr stars are excellent tracers of star formation , and they are also believed to be type ib or ic supernova progenitors , because they have removed their outer h - rich layers ( wn ) or even he - rich layers ( wc / o ) . about 300 wr stars have been previously identified in the milky way @xcite , with distribution models predicting @xmath01000 - 6500 total expected ( @xcite ; @xcite ) . optical narrow band surveys have been severely limited by interstellar extinction @xcite , and so the natural solution is to turn to the near infrared . emission line magnitudes of 40 known wr stars are presented in appendix a. a model of the milky way , predicting the numbers and distributions of wr stars visible in the k band is presented in appendix b. in an initial attempt at a narrowband near - infrared survey @xcite had limited success , while @xcite were somewhat more successful with their color - based selection of objects from the 2mass @xcite and glimpse @xcite surveys . utilizing a new narrow - band survey , described in @xmath6 [ observations ] along with the candidate selection methods , we have found 41 new wr stars : 15 wn and 26 wc . spectrographic follow up and data reduction is described in @xmath6 [ specobsred ] . our resulting k - band spectra , line measurements , and subtype classification are presented in @xmath6 [ results ] . distances to the new wr stars are calculated and their distribution within the galaxy considered in @xmath6 [ distribution ] . our conclusions are summarized in @xmath6 [ conclusion ] . the most reliable optical technique to detect individual wolf - rayet stars in crowded fields consists of subtracting a normalized continuum image from an image obtained with a narrowband filter centered on the heii 4686 line . this works well for individual wr stars with equivalent widths of heii(4686 ) @xmath7 or larger , and even for dense , unresolved clusters that include a very small fraction of wr stars ( @xcite ) . unfortunately , dust extinction makes this technique infeasible for the large majority of galactic wr stars . only in the near - infrared can we hope to detect the wr stars farther than about 5 kpc . a near - infrared survey of the plane of the galaxy was carried out under the umbrella of the smarts consortium ( see http://www.astro.yale.edu/smarts/ ) . the imaging data were taken over approximately 200 nights in 2005 - 2006 on the cerro tololo inter - american observatory ( ctio ) 1.5 meter telescope , using the universit de montral s cpapir camera . images are 35@xmath8 on a side , with a plate scale of 1.03@xmath9 per pixel , and cover 1@xmath10 above and below the galactic plane from galactic longitude _ _ l__=-90@xmath10 to _ _ l__=60@xmath10 . each of the 1200 fields was imaged in the j and k bands , and through a selection of narrow bands designed for identifying wr stars . motivated by the wr infrared spectra in figer et al . ( 1997 ) , we purchased a filter set ( table [ filters ] ) which targeted the emission line features he i 2.062 @xmath11 m , c iv 2.081 @xmath11 m , h i br@xmath12 2.169 @xmath11 m , and he ii 2.192 @xmath11 m . in addition , two narrow - band continuum filters were used which were selected to be relatively devoid of emission lines , one blueward ( 2.033 @xmath11 m ) and one redward ( 2.255 @xmath11 m ) of the emission line filters . these were then used to linearly interpolate a continuum magnitude at each of the emission line bands , so we could calculate the difference in measured and interpolated magnitudes , @xmath13 , indicative of emission or absorption in that band . in this paper , we are working from a catalogue of the calculated @xmath13 values for all of the stars contained in the survey area which had been processed by the time of our spectroscopic follow - up . this consisted of about 75% of the survey area , and in general we excluded those areas most crowded with stars , in particular the approximately eight degrees in longitude closest to the galactic center . the more than @xmath14 science and dome flat images of the survey require a customized , streamlined pipeline to reduce this large amount of data . the pipeline was constructed ( by jg and dz ) in idl and uses the 2mass catalogue extensively as a reference for both astrometry and photometry . each of the @xmath15 fields has been imaged in each of the @xmath16 filters at @xmath17 dither positions with separations of @xmath18 . dome flats were used to flatten each of the images for sensitivity and chip defects . these @xmath17 dither positions were median combined together without shifting to remove most of the stellar sources and to create a sky image . the sky image was subtracted from each image . sources for each field / filter were matched to 2mass to determine the world coordinate system ( wcs ) . once a wcs was fit , another iterative process was performed to minimize the residuals to 2mass and determine the best geometric distortion solution . the idl procedure warp_tri applied the geometric distortion solution . the idl procedure hastom aligned all the images taken in a dither series . these images were combined to create a final deep exposure . sources were identified as wr candidates through emission in the narrowband filters . this made it necessary to have not only the magnitude of a source in the narrowband filter , but also the magnitude of the continuum at that wavelength . as a result , each narrowband filter had to be examined concurrently with the cont1 and cont2 filters . sources on the final deep exposures were detected using the idl procedure find . aperture photometry was then carried out using the idl procedure aper with a 2-pixel aperture and a sky annulus from 10 to 20 pixels . objects matched to the 2mass catalog ( at least 100 in each field ) were used to determine a zero point for each filter in each deep image . an object with a flat spectrum through our filters has the same magnitude in all 6 filters . sources were considered matched across filters if their positions were consistent within 0.5 pixel . these matched sources with 2mass - calibrated magnitudes were then used to construct emission - magnitude diagrams ( emds ) . an emd was constructed for each narrowband filter with the wavelength - interpolated continuum magnitude vs continuum - subtracted narrowband magnitude . the stars scatter around a continuum - subtracted narrowband magnitude of zero . in each magnitude bin we calculate the standard deviation sigma of the continuum - subtracted narrowband magnitude . objects that are 5@xmath19 or more in the negative direction from the locus of stars are considered candidates . we determined the offset to convert instrumental magnitudes to apparent magnitudes using the 2mass ks band catalogue . the images were divided into an 8x8 grid , with each of the 64 areas having an individually determined offset , to compensate for an observed color dependence ( probably due to variable reddening ) across the field . the idl procedure aper determined the magnitude for the source at the coordinates given by 2mass . once the offsets to convert to apparent magnitude were calculated , find , an idl procedure , identified sources in an image that were a given deviation above the background . aper found the instrumental magnitude for the sources , which were then converted to apparent magnitudes . the cont1 filter was taken as a reference image and hastrom aligned all filters from a field . the sources from cont1 were then matched to the sources from the cont2 image and sources within 0.5 pixels were kept as matches . the matched sources were then compared to the sources of a narrowband filter . the result was a list of sources found within 0.5 pixels of each other in both of the continuum filters and the narrowband filter . a linear fit was found between the cont1 and cont2 filter magnitudes of each source . then , using the central wavelength for the narrowband filter , an interpolated continuum value at the narrowband wavelength was determined . the magnitude of the source in the narrowband filter was then subtracted from the interpolated continuum value , giving the negative emission magnitude for the source in that narrowband filter . the emission magnitude for a source was also estimated by subtracting the cont1 and the cont2 magnitude from the narrowband magnitude , resulting in a total of three estimates for the emission magnitude of a source . an emd of continuum magnitude vs. emission magnitude was created for each narrowband filter and the standard deviation was determined for the sources in bins of 1 magnitude . sources that had emission magnitudes of @xmath20 or greater from the center of the emd were marked as wr candidates . 1 arcmin @xmath21 1 arcmin finder charts were produced for each wr candidate , showing the candidate in all filters of the survey and also in xdss red and xdss infrared images . these are presented in appendix c. candidates with emission magnitudes that were similar to those of the known wr stars covered in the survey were selected for spectrographic follow up . this refined list of candidates was blinked by eye to remove any candidates that did not resemble stars . in this initial , exploratory phase of the survey we used two techniques when selecting targets for spectroscopic follow - up . we began by selecting targets with such powerful emission lines that the star appeared brighter in the narrow - band images when compared to the continuum images even when examined by eye . this corresponds to a minimum 0.5 to 1 magnitude difference in brightness between the narrow- and broad - band images . we initially selected candidates displaying a brightening of at least 0.5 magnitudes in at least one narrow band image relative to the continuum . this resulted in the detection of 34 new planetary nebulae ( which will be reported elsewhere ) whose very strong , sharp emission lines , he i 2.058 @xmath11 m and br@xmath12 2.166 @xmath11 m , fell within our he i and br@xmath12 narrow - band imaged fields . a few of the planetary nebulae were slightly resolved on the br@xmath12 narrow - band images . no new wr stars have yet been found this way . our second , much more successful , technique relied on using known wr stars to calibrate our selection of targets . forty known wr stars were selected within the survey area , and patterns were found in their narrow band @xmath13 values which distinguished broad subtypes of wr stars . ( see appendix [ earlywork ] ) . wc stars generally showed strong ( -0.8 magnitudes or less ) emission excess in the c iv filter , and slightly weaker emission ( between -0.4 and -0.8 magnitudes ) in the he i filter , due to the blue side of the c iv line extending into the range of the he i narrow - band filter . early wn stars generally showed moderate emission in the br@xmath12 filter and the he ii filter , but slight absorption ( 0.1 magnitudes ) in both he i and c iv . using these criteria , 173 candidate targets were selected which appeared at least @xmath20 brighter in their narrow - band filter than did other stars in the field and which also fit the criteria suggested by the known wrs . though we are reliably detecting stars to magnitude 14 - 14.5 in all of the filters ( by judicious choice of exposure times ) , during this exploratory stage candidates were selected to have emission - band magnitudes brighter than k = 11.5 . this is because the initial spectrographic follow - up is being done with a 1.5 m telescope ( see below ) . ( thousands of uncrowded @xmath20 candidates as faint as faint as k = 14.5 will be the subjects of future papers ) . there are strong selection effects for those wr subtypes which were used to determine the selection criteria , and as a result , no wc9 or late wn type ( @xmath22wn6 ) stars have yet been found . improvements now underway in survey image reduction will permit discovery of the less strongly distinguished subtypes . the spectrographic follow - up data were taken between 28 april and 6 june 2007 , with the near - infrared ( 0.8 - 2.5 @xmath11 m ) simon spectrograph @xcite of the universit de montral mounted on the 1.5 meter telescope at ctio . simon has a scale of 0.46 " /pixel on the ctio 1.5 meter telescope . targets were observed in the k band with a resolving power of r @xmath0 1500 . each target was observed 5 times , with a nod to move the target along the slit between each observation . total integration times ranged from 10 to 30 minutes per candidate . all data were reduced using iraf routines . images were dark subtracted and flat fielded to remove any instrument signature ; spectra were then extracted using the apall task . this task also provided sky subtraction by fitting the background on either side of the object along the slit . the five exposures were scaled and median - combined . standard stars , which were observed periodically throughout the night , were similarly reduced , and corrections made for br@xmath12 absorption in the telluric standard . object spectra were then divided by the temporally closest standard - star spectra to remove , as best as possible , the atmospheric absorption features , particularly those at 2.008 @xmath11 m and 2.059 @xmath11 m , and a fainter feature at 2.199 @xmath11 m . wavelength calibrations were done using the atmospheric oh emission observed off - target by the spectrograph during each object observation . lines were first identified using the identify task and were compared with the coordinate list ohlines.dat included therein ( @xcite ) . the wavelength solution was then refitted to each observed target using the reidentify task . in general , the root - mean - square error of the residuals of wavelength fits was less than 1 , and in most cases between 0.1 and 0.3 . these dispersion solutions were each applied to the corresponding target object , and then corrected for the intrinsic heliocentric motion using the rvcorrect task to identify the heliocentric velocity and the dopcorrect task to doppler shift the wavelength scale . from our target list of 173 candidates we have discovered 41 new wolf - rayet stars : 15 of type wn and 26 of type wc . right ascension and declination , as well as j , h and k@xmath23 magnitudes , were obtained from 2mass , and are listed in table [ tbl - pos ] . all 2mass objects were then referenced in nomad @xcite to obtain b , v , and r magnitudes , when available , which are also included in table [ tbl - pos ] . spectra are grouped by assigned spectral type , and are presented in figures [ wn4 ] through [ wc82 ] . the iraf task splot was used to to fit the continuum and then optimize and deblend gaussian functions to fit the observed emission lines in all spectra . this gave measurements of line centers , equivalent widths ( ews ) and the full - widths - at - half - maximum ( fwhms ) . a number of errors contribute to reduce the accuracy of these measurements . blending of many lines creates the largest emission features , and introduces errors into line center measurements up to 10 . it also skews the shape of the feature to be a poor fit to a gaussian . in general , the rms error of the residuals for the fits was about 10% of the continuum value . additionally , the continuum was quite difficult to determine precisely due to the abundance of emission lines throughout the spectrum , especially in wc stars , resulting in errors in the equivalent width measurements . a number of our wn detections were rather faint , with peak fluxes only twice the continuum level , resulting in lower signal - to - noise ratios for these spectra . finally , ground - based observatories must peer through the murky atmosphere , so that the strong c iv 2.08 @xmath11 m line in wc stars falls on the edge of the equally strong atmospheric absorption feature at 2.06 @xmath11 m . division by a standard star removes most of this feature ; however , changes in atmospheric conditions between object and standard - star observations result in residual features on the blue side of the emission line . the strong emission lines visible in the spectra , while easily distinguished as belonging to either type wc or wn , are the result of overlapping blends of emission lines of various elements . in the optical , wr stars are categorized by looking at ratios of equivalent widths of various nitrogen species for wn , carbon and oxygen species for wc , and he ii and he i for both . however , the heavy blending of lines present in the k band makes it more difficult to find either isolated spectral lines or distinguishing ratios of blended lines @xcite . spectral subtypes were assigned following @xcite . these are presented in tables [ tbl - eqws - wn ] and [ tbl - eqws - wc ] , along with the ratio of equivalent widths used for categorization . the ratio @xmath24 was used to categorize the wn stars , and the ratio @xmath25 was used to categorize the wc stars . also following @xcite , wn stars with fwhm(he ii 2.189 ) @xmath26 @xmath27 were classified as broad / strong , and the letter b appended to the subtype designation . subtypes thus assigned are expected to be accurate to within one subtype . the 40 known wr stars were similarly assigned subtypes , which agreed within one subtype of their published spectral classifications . those wc stars with especially broad , heavily blended c iv and c iii lines , as presented in figure [ wce ] and which match our observed spectra of wr19 ( wc4 ) and spectra observed in @xcite of wr146(wc4 ) , wr143(wc5 ) and wr150(wc5 ) , are classified more generally as wce . blended lines complicate the calculations of accurate radial velocities , as there is no longer any fixed line - center with which to compare theoretical and actual wavelengths . the purest line in our k band spectra , he ii at 21891 , allows measurement of radial velocity with respect to the motion of the sun in principal ; however , this line is extremely weak or not discernable in most of the wc stars . line - center measurements are further complicated by the difficulty of accurately fitting gaussian curves to wr emission lines , as described in section [ spectra ] , resulting in errors of 5 - 10 in determination of peak wavelengths . this corresponds to errors on the order of 50 - 150 km / s . in general , the direction of motion with respect to the sun follows the clockwise rotation of the galaxy ; however , as noted , the error on these measurements is large . measured radial velocities ( based on the measured line centers , which may be shifted , depending on details of the line formation mechanism ) for wns and , when possible , wcs , are included in tables [ tbl - eqws - wn ] and [ tbl - eqws - wc ] . distances to all new wr stars were estimated from the 2mass j , h and ks@xmath23 color excesses , using the method of spectroscopic parallax described in @xcite . intrinsic @xmath28 and @xmath29 colors specific to wr subtype were taken from @xcite and used to calculate e@xmath30 and e@xmath31 . extinction ratios taken from @xcite then allow two calculations for @xmath32 : @xmath33 and @xmath34 the average of these values , @xmath35 , was used to calculate the distance modulus , taking the apparent k@xmath23 magnitude from 2mass and the subtype - specific absolute magnitude , m@xmath36 , from @xcite . derived k@xmath23 band extinctions and galactocentric distances , r@xmath37 , are listed in table [ tbl - dist ] . calculations for r@xmath37 assume the iau standard solar galactocentric distance r@xmath38=8.0 kpc . , and the known galactic wr stars are on the same scale in figure 2 . because these calculations are based on the inherent absolute magnitude and @xmath28 and @xmath29 colors for each subtype , the errors are highly dependent on the accuracy of these measurements . the measured scatter about the adopted color values is approximately 0.02 magnitudes ( table a1 in @xcite ) , which is negligible in a@xmath36 compared to the average scatter of 0.4 magnitudes from the adopted m@xmath36 values . the redundant calculations of @xmath32 also provide some indication of the reliability of our measurement . the two values are generally in agreement to within 0.2 magnitudes , especially for the wns , however they can differ by as much as 0.66 magnitudes for some of the wcs , indicating that more accurate subtype specific colors and absolute magnitudes are needed . these uncertainties give typical errors on the order of 25% in our distance measurements , though they may range as high as 40% in some cases . the distances to new stars were constrained by the limiting observable magnitude of @xmath3911.5 . using the overall average extinction @xmath40 , we can calculate the typical measurable distances by subtype . as all discovered wn stars were early types , we can distinguish the faintest observed , strong and weak - lined wn as having average distances of 8.0 and 9.4 kpc , respectively , while observations of the faintest wc stars yield typical distances of 8.9 kpc . in figure [ galaxyplot ] our new wr stars ( in bold ) have been over - plotted with the previously - known wrs onto the plane of the galaxy , with distances to the known stars taken from the 7th catalogue of galactic wolf - rayet stars @xcite . the galactic center is labeled , and circles of radius 4 , 8 , and 12 kpc are plotted . the new wr stars largely follow the distribution pattern established by the known stars , though we can see that we are beginning to push out to larger heliocentric distances . we also find new stars without optical counterparts within a few kpc of our sun , reinforcing the necessity of wr surveys in the near infrared . @xcite , along with the more recent reanalysis in @xcite , maintain that wr stars trace the spiral structure of the galaxy . one arm may be seen along roughly the 8 kpc radius , and an inner arm can perhaps begin to be traced along the inner 4kpc radius . however , the distance error bars are not trivial , so that firm conclusions about the utility of wr stars as spiral tracers should not yet be drawn . we have discovered 41 new galactic wr stars , 15 of type wn and 26 of type wc , using a new , near - infrared narrow - band survey of the galactic plane . the reduced extinction from dust and gas in the near infrared makes this the optimal method for future discovery of the thousands of undetected galactic wr stars . of the 254 total candidates observed spectrographically , 75 proved to be emission line objects . all of the emission line objects that were not wr stars ( 34 objects ) were planetary nebulae ( pn ) . as the key goal of this survey is the detection of new galactic wr stars , we amended our selection criteria to eliminate likely pn . our modified selection criteria yielded 173 wr star candidates which were observed spectrographically : 41 proved to be new wr stars . with such a 23% detection rate , we have barely scratched the surface of the wealth of new wr stars expected to be discovered within our survey area with the available data . an initially fairly simple sky - subtraction methodology resulted in relatively scattered color - magnitude diagrams , raising our cut for emission objects to 5@xmath19 . it also meant that most of our non - detections were erroneously selected objects with featureless spectra . improved sky subtraction ( using entire nights of data , median - filtered in each filter as skyflats ) will allow us to lower this limit to 3@xmath19 and will improve the detection rate of emission - line objects . we expect this survey to yield thousands of additional discoveries in the coming years . our survey limits will be pushed fainter by the use of a larger infrared telescope for spectroscopic follow - up . as we increase the number of known stars , we will also increase the statistical significance of distribution plots , and subtype abundances , allowing us to learn more about our galaxy s structure and composition . the galactic center is expected to prove an especially rich area for discovery , but it is still largely terra incognita as the crowding of stars there is very high . the vast majority of galactic wolf - rayet stars remain to be discovered , but we now have a proven technique to continue the search . benjamin , r.a . , churchwell , e. , babler , b.l . , bania , t. m. , clemens , d. p. , cohen , m. , dickey , j. m. , indebetouw , r. , jackson , j. m. , kobulnicky , h. a. , lazarian , a. , marston , a. p. , mathis , j. s. , meade , m. r. , seager , s. , stolovy , s. r. , watson , c. , whitney , b. a. , wolff , m. j. , & wolfire , m. g. 2003 , , 115 , 953 buckalew , b.a . , kobulnicky , h.a . , dufour , r.j . 2005 , , 157 , 30 clark , j.s . & negueruela , i. 2002 , , 396 , l25 crowther , p.a . , hadfield , l.j . , clark , j.s . , negueruela , i. & vacca , w.d . 2006 , , 372 , 1407 conti , p. s. , & vacca , w. d. 1990 , , 100 , 431 doyon , r. , nadeau , d. & vallee , ph . 2000 , asp conf . 195 , 548 drissen , l. , moffat , a.f.j . & shara , m. m. 1993 , , 105 , 1400 drissen , l. , roy , j - r , moffat , a.f.j . & shara , m. m. 1999 , , 117 , 1249 drimmel , r. , & spergel , d. n. 2001 , , 556 , 181 figer , d.f . , mclean , i.s . & najarro , f. 1997 , , 486 , 420 hadfield , l.j . , crowther , p.a . , schild , h. & schmutz , w. 2005 , , 439 , 265 hadfield , l.j . , van dyk , s.d . , morris , p.w . , smith , j.d . , marston , a.p . , peterson , d.e . 2007 , , 376 , 248 homeier , n.l . , blum , r.d . , pasquali , a. , conti , p.s . & damineli , a. 2003 , , 408 , 153 indebetouw , r. , et al . 2005 , , 619 , 931 massey , p. , conti , p.s . , moffat , a.f.j . & shara , m.m . 1987 , , 99 , 816 meynet , g. , & maeder , a. 2005 , , 429 , 581 moffat , a.f.j . , drissen , l. shara , m.m . 1994 , , 436 , 183 moffat , a.f.j . & shara , m.m . 1987 , , 320 , 266 shara , m.m . , moffat , a.f.j . , smith , l.f . , niemela , v.s . , potter , m. & lamontagne , r. 1999 , , 118 , 390 shorlin , s.l . , turner , d.g . , & pedreros , m.h . 2004 , , 116 , 170 skrutskie , m. f. , cutri , r. m. , stiening , r. , weinberg , m. d. , schneider , s. , carpenter , j. m. , beichman , c. , capps , r. , chester , t. , elias , j. , huchra , j. , liebert , j. , lonsdale , c. , monet , d. g. , price , s. , seitzer , p. , jarrett , t. , kirkpatrick , j. d. , gizis , j. e. , howard , e. , evans , t. , fowler , j. , fullmer , l. , hurt , r. , light , r. , kopan , e. l. , marsh , k. a. , mccallon , h. l. , tam , r. , van dyk , s. , & wheelock , s. 2006 , , 131 , 1163 smartt , s.j . , & rolleston , w.r.j . 1997 , , 481 , 47 steed , a.j . & baker , d.j . 1979 , , 18 , 3386 tutukov , a.v . , & cherepashchuk , a.m. 2003 , astronomy reports , 47 , 386 vanbeveren , d. , van bever , j. & belkus , h. 2007 , , 662 , 107 van der hucht , k.a . 2001 , new astronomy review , 45 , 135 van der hucht , k.a . 2006 , , 458 , 453 zacharias , n. , monet , d.g . , levine , s.e . , urban , s.e . , gaume , r. , & wycoff , g.l . 2005 , vizier catalog i/297 mms , jg and dz acknowledge with gratitude hilary lipsitz , whose ongoing support has been essential to the success of this program . afjm , rd , ld are grateful to nserc ( canada ) and fqrnt ( quebec ) for financial aid . we also thank the american museum of natural history for essential funding , and a careful referee for excellent suggestions . most of the infrared imaging was expertly carried out by claudio aguilera , alberto miranda and alberto pasten . * appendix a * in tables 6 and 7 we give a summary of the emission line @xmath13 values measured for the known wrs found in the survey area . all survey stars were compared with finder charts for the known wr stars to ensure that photometric measurements were made for the correct star . all stars were found by eye to be present in the survey images , though some had not been identified photometrically due to nebulosity and occasional out - of - focus images . the @xmath13 magnitude for each narrow - band filter gives the difference in magnitude from the interpolated continuum magnitude . negative values of @xmath13 indicate emission , positive values indicate absorption for lines in the narrow - band emission - line filter , assuming there are no other problems . several fields were imaged multiple times , and on more than one night . this allowed us to estimate the accuracy of the emission line magnitudes in tables 6 and 7 as @xmath410.03 magnitudes . * appendix b * following shara et al . ( 1999 ) , we assume that the star distribution expressed in cylindrical galactic coordinates ( i.e. @xmath42 ) follows that of the interstellar dust and adopt an axisymmetric exponential disk formulation with an outwards flair taken from drimmel & spergel ( 2001 ) : @xmath43 with , + @xmath44 a linear flair is added by imposing the following functional form to @xmath45 : @xmath46 h_o & r(r , l ) \leq r_f \end{array } \right.\ ] ] in these equations , @xmath47 is the local stellar density , @xmath48 pc and @xmath49 pc are respectively the solar galactocentric distance and distance from the galactic plane . the scale length @xmath50 and the constants @xmath51 , @xmath52 and @xmath53 which define the scale height @xmath45 are taken from the fit to the fir emission of the interstellar dust of drimmel & spergel ( 2001 ) and are respectively @xmath54 pc , @xmath55 pc , @xmath56 and @xmath57 pc . although the assumption that the star distribution follows the dust is justified for o stars , it must be modified for wr stars to account for the observed metallicity dependence of their galactic distribution . using the results of maeder & meynet ( 1994 ) for the metallicity dependence of the wr / o number ratio and the radial metallicity distribution of the galactic disk from smartt & rolleston ( 1997 ) , shara et al . ( 1999 ) found that the wr star density can be expressed as : @xmath58 where , + @xmath59 knowing the local surface density of wr stars from van der hucht ( 2001 ) : @xmath60 and applying this constraint to @xmath61 we obtain : @xmath62 finally , the conversion of the wr star density from a spatial ( @xmath61 ) to a magnitude ( @xmath63 ) dependence is done in two steps . first , we use the conservation of stars in both systems to write : @xmath64 then , the radial distance @xmath65 is converted to the magnitude @xmath66 using the inverse square - law of light attenuation accounting for interstellar extinction : @xmath67 where the extinction @xmath68 allowing for a spherical hole in the interstellar dust at the center of the galaxy is ( see drimmel & spergel 20001 ) : @xmath69 a_k(0.5r_o , l , z)e^{- \left ( \frac{r(r , l ) - 0.5r_o}{2500}\right)^2 } & r(r , l ) < 0.5r_o \end{array } \right.\ ] ] with @xmath70 mag / pc ( mathis 1990 ) and the intrinsic magnitude of wr stars in the @xmath71-band @xmath72 mag ( van der hucht 2001 ) . these equations have been solved numerically to determine the star density @xmath73 to an accuracy better than @xmath74 using a monte - carlo method with a sobol quasi - random number generator in @xmath75 . by integrating over the whole galaxy the model predicts @xmath76 wr stars which is very close to the @xmath77 predicted by van de hucht ( 2001 ) . further validation can be carried out by applying this model to the @xmath78-band where most wr stars have been detected so far . it is then possible to compare the observed to the predicted number of wr stars . to do so , we adopt @xmath79 mag ( van der hucht 2001 ) and @xmath80 mag / pc . according to the model , the number of wr stars observable up to a magnitude of v = 15 , 12 and 10 are respectively 153 , 79 and 42 . these numbers are very close to the respective observed numbers reported by van der hucht ( 2001 ) of 159 , 80 and 36 . it is again apparent how important it is to continue the search for new wr stars in the infrared . figure [ con95 ] shows the contour plot representing the necessary target k magnitude to detect @xmath81 of all wr stars along a line of sight . figure [ numwr ] shows the number of wr stars expected per cpapir field and figure [ numwr2 ] presents the same information for @xmath82 . figure 6 shows the cumulative number of expected galactic wr stars as a function of k magnitude . at k= 11 , 12 , 13 and 14 , the expected numbers of galactic wr stars are 1200 , 2500 , 4200 and 5400 . * appendix c * we present in figures 7a through 7 g the finder charts for the 41 new wolf - rayet stars discovered in our survey . l r r r r c c c r r r 668_4 & 10 16 26.22 & -57 28 05.7 & 283.26 & -0.64 & - & - & 17.46 & 11.73 & 10.38 & 9.53 + 740_21 & 11 16 03.53 & -61 26 58.3 & 291.79 & -0.66 & 16.76 & 16.82 & 15.97 & 13.16 & 11.57 & 10.47 + 740_16 & 11 19 42.96 & -61 27 12.4 & 292.20 & -0.51 & 13.67 & 12.35 & 13.03 & 11.50 & 10.99 & 10.33 + 768_6 & 11 46 06.66 & -62 47 12.7 & 295.54 & -0.86 & 15.56 & 17.17 & 14.11 & 12.15 & 11.36 & 10.78 + 772_17 & 11 50 04.24 & -62 52 15.4 & 296.00 & -0.83 & 18.23 & - & 16.47 & 12.57 & 11.58 & 10.94 + 776_3 & 11 55 52.11 & -62 45 02.2 & 296.62 & -0.56 & 18.38 & - & 14.94 & 11.65 & 10.76 & 10.14 + 791_12c & 12 13 28.29 & -62 41 42.9 & 298.59 & -0.14 & - & - & - & 14.05 & 12.67 & 11.53 + 808_14 & 12 28 41.91 & -63 25 46.1 & 300.39 & -0.67 & 20.28 & - & 17.82 & 12.78 & 11.83 & 10.94 + 808_23 & 12 28 50.99 & -63 17 00.2 & 300.40 & -0.52 & - & - & 18.61 & 13.56 & 12.55 & 11.84 + 807_16 & 12 30 03.86 & -62 50 17.1 & 300.50 & -0.07 & - & - & 19.20 & 13.16 & 12.05 & 11.09 + 816_10 & 12 38 18.78 & -63 24 19.7 & 301.46 & -0.57 & - & - & 18.43 & 12.93 & 11.76 & 10.96 + 832_25 & 12 55 44.26 & -63 35 50.0 & 303.41 & -0.73 & - & - & - & 12.48 & 11.12 & 10.22 + 856_13c & 13 03 11.08 & -63 42 16.2 & 304.23 & -0.86 & - & - & 18.72 & 13.13 & 11.92 & 11.05 + 839_12 & 13 04 50.08 & -63 04 40.2 & 304.45 & -0.25 & - & - & - & 15.69 & 14.00 & 12.54 + 845_34 & 13 12 21.30 & -62 40 12.5 & 305.33 & 0.10 & - & - & 16.24 & 10.75 & 9.57 & 8.77 + 845_35 & 13 12 27.67 & -62 44 22.0 & 305.34 & 0.03 & - & - & 17.95 & 13.16 & 11.82 & 10.71 + 847_8 & 13 12 45.35 & -63 05 52.0 & 305.34 & -0.33 & - & - & - & 13.06 & 11.34 & 10.26 + 853_9 & 13 22 16.08 & -62 30 57.4 & 306.48 & 0.14 & - & - & - & 14.77 & 12.69 & 11.55 + 858_26 & 13 28 15.87 & -62 06 23.5 & 307.22 & 0.46 & - & - & 18.70 & 12.53 & 11.40 & 10.68 + 883_18 & 13 52 02.36 & -62 26 46.0 & 309.88 & -0.39 & - & - & - & 14.59 & 12.41 & 10.99 + 885_11 & 13 54 13.45 & -61 50 01.8 & 310.27 & 0.14 & - & - & - & 13.60 & 12.07 & 10.86 + 897_5 & 14 10 10.01 & -61 15 25.5 & 312.25 & 0.18 & - & - & - & 15.30 & 12.43 & 10.60 + 903_15c & 14 12 36.54 & -61 45 32.7 & 312.38 & -0.38 & - & - & - & 13.58 & 11.33 & 9.71 + 907_18 & 14 16 27.37 & -61 17 56.2 & 312.96 & -0.09 & - & - & - & 13.60 & 11.39 & 10.01 + 956_25 & 15 01 30.11 & -59 16 12.0 & 318.88 & -0.49 & - & - & - & 13.84 & 11.83 & 10.64 + 979_11 & 15 20 35.91 & -57 27 11.9 & 321.95 & -0.20 & - & - & - & 13.88 & 12.15 & 11.06 + 1011_24 & 15 43 04.68 & -55 11 12.3 & 325.81 & -0.13 & 19.93 & 17.31 & 16.06 & 10.96 & 9.94 & 9.06 + 1053_27 & 16 11 43.70 & -51 10 16.6 & 331.67 & 0.17 & - & - & 18.53 & 10.26 & 8.97 & 8.10 + 1059_34 & 16 14 37.23 & -51 26 26.3 & 331.81 & -0.34 & - & - & - & 15.05 & 12.79 & 11.54 + 1081_21 & 16 24 58.86 & -48 56 52.4 & 334.75 & 0.27 & - & - & - & 13.28 & 11.76 & 10.73 + 1093_34 & 16 31 29.23 & -47 56 16.4 & 336.22 & 0.19 & - & - & - & 15.56 & 12.86 & 11.32 + 1093_33 & 16 31 49.06 & -47 56 04.4 & 336.26 & 0.15 & - & - & - & 15.11 & 12.86 & 11.47 + 1093_53 & 16 32 12.98 & -47 50 35.8 & 336.37 & 0.17 & - & - & - & 15.00 & 12.70 & 11.34 + 1096_22 & 16 35 23.31 & -48 09 18.0 & 336.51 & -0.43 & - & - & - & 14.90 & 12.77 & 11.43 + 1222_15 & 17 22 40.74 & -35 04 52.9 & 352.20 & 0.74 & - & - & - & 15.11 & 12.05 & 10.33 + 1385_24 & 18 13 42.47 & -17 28 12.2 & 13.15 & 0.13 & - & - & 20.02 & 11.21 & 9.70 & 8.57 + 1425_47 & 18 23 03.42 & -13 10 00.4 & 18.01 & 0.18 & 15.23 & 14.49 & 14.41 & 10.34 & 9.28 & 8.27 + 1462_54 & 18 29 33.84 & -08 39 02.1 & 22.75 & 0.87 & - & - & - & 14.07 & 12.75 & 11.93 + 1509_29 & 18 41 48.45 & -04 00 12.9 & 28.27 & 0.31 & - & - & - & 15.62 & 13.32 & 11.99 + 1613_21 & 19 06 36.53 & + 07 29 52.4 & 41.33 & 0.07 & - & - & - & 14.24 & 12.65 & 11.61 + 1671_5 & 19 20 40.38 & + 13 50 35.2 & 48.55 & -0.05 & - & - & - & 13.57 & 11.80 & 10.76 + l r r r r r r r r r r l 668_4 & -8 & 141 & -11 & 121 & -31 & 147 & -91 & 153 & 2.9 & 55 & wn5b + 740_21 & -14 & 208 & -20 & 200 & -39 & 165 & -203 & 242 & 5.2 & -82 & wn4b + 768_6 & -6 & 38 & -19 & 126 & -42 & 120 & -101 & 106 & 2.4 & 41 & wn5 + 772_17 & -3 & 38 & -12 & 113 & -53 & 126 & -95 & 121 & 1.8 & - & wn6 + 776_3 & - & - & -19 & 109 & -60 & 93 & -72 & 94 & 1.2 & - & wn6 + 808_23 & - & - & - & - & -46 & 101 & -141 & 126 & 3.1 & 68 & wn5 + 816_10 & - & - & -29 & 227 & -48 & 140 & -137 & 136 & 2.9 & -14 & wn5b + 847_8 & -8 & 70 & -47 & 142 & -65 & 119 & -114 & 120 & 1.8 & -96 & wn6 + 853_9 & - & - & -29 & 95 & -36 & 66 & -92 & 97 & 2.6 & 27 & wn6 + 858_26 & - & - & -26 & 96 & -38 & 74 & -56 & 78 & 1.5 & 55 & wn6 + 907_18 & -16 & 171 & -19 & 140 & -70 & 175 & -150 & 153 & 2.1 & -68 & wn5b + 956_25 & -11 & 202 & -35 & 222 & -31 & 163 & -180 & 366 & 5.8 & -315 & wn4b + 979_11 & - & - & -57 & 260 & -62 & 207 & -212 & 220 & 3.4 & -14 & wn4b + 1093_53 & - & - & -52 & 219 & -60 & 185 & -145 & 182 & 2.4 & -68 & wn5b + 1462_54 & - & - & -59 & 202 & -72 & 164 & -166 & 126 & 2.3 & 68 & wn5 + l r r r r r r r r r r l 740_16 & -987 & 255 & -226 & 330 & - & - & -50 & 241 & 4.4 & - & wce + 791_12c & -1590 & 282 & -498 & 443 & - & - & -75 & 228 & 3.2 & - & wce + 808_14 & -948 & 306 & -392 & 477 & - & - & -140 & 623 & 2.4 & - & wce + 807_16 & -985 & 178 & -206 & 147 & -31 & 103 & -74 & 116 & 4.8 & -55 & wc7 + 832_25 & -454 & 236 & -73 & 236 & -37 & 351 & -15 & 121 & 6.2 & - & wc5 - 6 + 839_12 & -1658 & 230 & -244 & 195 & - & - & -71 & 157 & 6.8 & - & wc5 - 6 + 845_34 & -374 & 271 & -131 & 284 & - & - & -35 & 384 & 2.9 & - & wc8 + 845_35 & -1789 & 249 & -412 & 258 & -42 & 323 & -70 & 195 & 4.3 & 0 & wc7 + 856_13c & -712 & 191 & -138 & 180 & -14 & 86 & -61 & 157 & 5.2 & -164 & wc5 - 6 + 883_18 & -1093 & 301 & -373 & 453 & -10 & 103 & -77 & 320 & 2.9 & - & wce + 885_11 & -1489 & 210 & -279 & 180 & - & - & -60 & 174 & 5.3 & - & wc5 - 6 + 897_5 & -525 & 205 & -188 & 165 & -56 & 354 & -35 & 126 & 2.8 & - & wc8 + 903_15c & -165 & 186 & -66 & 123 & -13 & 125 & -14 & 125 & 2.5 & - & wc8 + 1011_24 & -884 & 244 & -275 & 226 & - & - & -83 & 360 & 3.2 & - & wc8 + 1053_27 & -303 & 222 & -81 & 213 & -59 & 223 & -84 & 202 & 3.7 & - & wc8 + 1059_34 & -729 & 210 & -242 & 181 & -28 & 366 & -44 & 199 & 3.0 & - & wc8 + 1081_21 & -419 & 219 & -214 & 174 & -75 & 185 & -60 & 143 & 2.0 & -137 & wc8 + 1093_33 & -369 & 199 & -155 & 161 & -69 & 161 & -69 & 132 & 2.4 & -219 & wc8 + 1093_34 & -475 & 226 & -178 & 222 & -62 & 212 & -84 & 187 & 2.7 & -260 & wc8 + 1096_22 & -571 & 202 & -230 & 156 & -71 & 169 & -76 & 141 & 2.5 & -82 & wc8 + 1222_15 & -551 & 197 & -171 & 154 & - & - & -45 & 230 & 3.2 & - & wc8 + 1385_24 & -537 & 220 & -245 & 196 & -63 & 361 & -33 & 138 & 2.2 & - & wc8 + 1425_47 & -1593 & 238 & -300 & 296 & -48 & 250 & -75 & 179 & 5.3 & - & wc5 - 6 + 1509_29 & -1578 & 240 & -355 & 224 & - & - & -110 & 273 & 4.4 & - & wc7 + 1613_21 & -659 & 365 & -245 & 518 & - & - & - & - & 2.7 & - & wce + 1671_5 & -335 & 215 & -94 & 210 & -49 & 337 & -20 & 151 & 3.6 & - & wc8 + l l c c r r r r r r 668_4 & wn5b & 1.23 & 1.06 & 1.14 & 9.53 & -4.77 & 13.16 & 4.28 & 8.60 + 740_21 & wn4b & 1.55 & 1.51 & 1.53 & 10.47 & -4.77 & 13.71 & 5.51 & 8.24 + 740_16 & wce & 0.37 & 0.15 & 0.26 & 10.33 & -4.59 & 14.66 & 8.56 & 9.52 + 768_6 & wn5 & 0.80 & 0.76 & 0.78 & 10.78 & -4.41 & 14.41 & 7.62 & 8.63 + 772_17 & wn6 & 0.97 & 0.87 & 0.92 & 10.94 & -4.41 & 14.43 & 7.68 & 8.60 + 776_3 & wn6 & 0.89 & 0.84 & 0.86 & 10.14 & -4.41 & 13.69 & 5.46 & 7.78 + 791_12c & wce & 1.27 & 1.02 & 1.15 & 11.53 & -4.59 & 14.97 & 9.88 & 9.46 + 808_14 & wce & 0.82 & 0.56 & 0.69 & 10.94 & -4.59 & 14.84 & 9.29 & 8.87 + 808_23 & wn5 & 1.03 & 1.00 & 1.02 & 11.84 & -4.41 & 15.23 & 11.14 & 10.02 + 807_16 & wc7 & 0.97 & 0.69 & 0.83 & 11.09 & -4.59 & 14.85 & 9.33 & 8.88 + 816_10 & wn5b & 1.07 & 0.96 & 1.02 & 10.96 & -4.77 & 14.71 & 8.76 & 8.44 + 832_25 & wc5 - 6 & 1.10 & 0.58 & 0.84 & 10.22 & -4.59 & 13.97 & 6.22 & 7.26 + 856_13c & wc5 - 6 & 0.98 & 0.53 & 0.75 & 11.05 & -4.59 & 14.89 & 9.49 & 8.46 + 839_12 & wc5 - 6 & 1.70 & 1.60 & 1.65 & 12.54 & -4.59 & 15.48 & 12.48 & 10.39 + 845_34 & wc8 & 1.04 & 0.76 & 0.90 & 8.77 & -4.65 & 12.52 & 3.19 & 7.15 + 845_35 & wc7 & 1.23 & 0.96 & 1.10 & 10.71 & -4.59 & 14.20 & 6.93 & 7.22 + 847_8 & wn6 & 1.76 & 1.67 & 1.71 & 10.26 & -4.41 & 12.96 & 3.90 & 7.01 + 853_9 & wn6 & 2.04 & 1.78 & 1.91 & 11.55 & -4.41 & 14.05 & 6.46 & 6.98 + 858_26 & wn6 & 1.12 & 1.02 & 1.07 & 10.68 & -4.41 & 14.02 & 6.37 & 6.88 + 883_18 & wce & 2.00 & 1.53 & 1.76 & 10.99 & -4.59 & 13.82 & 5.80 & 6.53 + 885_11 & wc5 - 6 & 1.42 & 1.15 & 1.28 & 10.86 & -4.59 & 14.17 & 6.81 & 6.62 + 897_5 & wc8 & 2.86 & 2.64 & 2.75 & 10.60 & -4.65 & 12.50 & 3.16 & 6.79 + 903_15c & wc8 & 2.30 & 2.26 & 2.28 & 9.71 & -4.65 & 12.08 & 2.61 & 7.01 + 907_18 & wn5b & 2.16 & 2.02 & 2.09 & 10.01 & -4.77 & 12.69 & 3.45 & 6.65 + 956_25 & wn4b & 1.90 & 1.67 & 1.79 & 10.64 & -4.77 & 13.62 & 5.31 & 5.70 + 979_11 & wn4b & 1.64 & 1.49 & 1.57 & 11.06 & -4.77 & 14.26 & 7.12 & 5.26 + 1011_24 & wc8 & 0.98 & 0.91 & 0.95 & 9.06 & -4.65 & 12.76 & 3.57 & 5.90 + 1053_27 & wc8 & 1.16 & 0.89 & 1.03 & 8.10 & -4.65 & 11.72 & 2.21 & 6.64 + 1059_34 & wc8 & 2.06 & 1.58 & 1.82 & 11.54 & -4.65 & 14.37 & 7.47 & 4.01 + 1081_21 & wc8 & 1.42 & 1.18 & 1.30 & 10.73 & -4.65 & 14.08 & 6.54 & 3.80 + 1093_34 & wc8 & 2.55 & 2.11 & 2.33 & 11.32 & -4.65 & 13.64 & 5.34 & 4.21 + 1093_33 & wc8 & 2.15 & 1.84 & 1.99 & 11.47 & -4.65 & 14.13 & 6.69 & 3.59 + 1093_53 & wn5b & 2.20 & 1.98 & 2.09 & 11.34 & -4.77 & 14.02 & 6.36 & 3.70 + 1096_22 & wc8 & 2.04 & 1.75 & 1.89 & 11.43 & -4.65 & 14.19 & 6.88 & 3.51 + 1222_15 & wc8 & 2.91 & 2.44 & 2.68 & 10.33 & -4.65 & 12.30 & 2.89 & 5.65 + 1385_24 & wc8 & 1.48 & 1.37 & 1.42 & 8.57 & -4.65 & 11.80 & 2.29 & 6.29 + 1425_47 & wc5 - 6 & 0.97 & 0.78 & 0.88 & 8.27 & -4.59 & 11.98 & 2.49 & 6.18 + 1462_54 & wn5 & 1.31 & 1.20 & 1.26 & 11.93 & -4.41 & 15.08 & 10.39 & 4.16 + 1509_29 & wc7 & 2.02 & 1.36 & 1.69 & 11.99 & -4.59 & 14.89 & 9.50 & 4.50 + 1613_21 & wce & 1.35 & 0.84 & 1.09 & 11.61 & -4.59 & 15.11 & 10.51 & 6.97 + 1671_5 & wc8 & 1.59 & 1.20 & 1.40 & 10.76 & -4.65 & 14.01 & 6.34 & 6.41 + l l r r r r 48c & wn3h+wc4 & 0.01 & 0.02 & -0.37 & -0.20 + 18 & wn4 & -0.07 & 0.06 & -0.13 & -0.32 + 35b & wn4 & 0.10 & 0.12 & -0.18 & -0.48 + 44a & wn4 & 0.19 & -0.22 & -0.42 & -0.19 + 45b & wn4 & 0.11 & - & -0.19 & 0.11 + 62a & wn4 & -0.03 & 0.01 & -0.08 & -0.13 + 31 & wn4+o8v & -0.01 & 0.03 & -0.12 & -0.21 + 51 & wn4+ob ? & -0.08 & -0.01 & -0.15 & -0.38 + 38a & wn5 & 0.04 & 0.02 & -0.19 & -0.25 + 42c & wn5 & 0.09 & 0.12 & -0.16 & -0.35 + 42d & wn5 & 0.03 & 0.06 & -0.09 & -0.29 + 45a & wn5 & 0.08 & 0.05 & -0.17 & -0.41 + 36 & wn5 - 6+ob ? & 0.00 & 0.08 & -0.25 & -0.44 + 111c & wn6 & 0.08 & 0.12 & -0.21 & -0.38 + 35a & wn6h & 0.00 & 0.01 & -0.15 & -0.13 + 21a & wn6+o / a & 0.09 & - & -0.05 & 0.02 + 28 & wn6(h)+ob ? & 0.06 & 0.06 & -0.21 & -0.22 + 47 & wn6+o5v & -0.10 & 0.01 & -0.18 & -0.19 + 55 & wn7 & -0.08 & 0.03 & -0.23 & -0.17 + 74 & wn7 & -0.06 & 0.00 & -0.17 & -0.16 + 84 & wn7 & -0.07 & 0.03 & -0.16 & -0.22 + 111d & wn7 ? & 0.03 & 0.09 & -0.04 & -0.09 + 19a & wn7:(h ) & -0.06 & 0.00 & -0.20 & -0.10 + 26 & wn7/wce & 0.03 & -0.59 & -0.16 & -0.50 + 107 & wn8 & -0.23 & -0.07 & -0.22 & -0.02 + 47b & wn9h & 0.08 & 0.03 & -0.26 & -0.06 + 108 & wn9h+ob & -0.11 & -0.02 & -0.01 & 0.12 + l l r r r r 38 & wc4 & -1.06 & -1.56 & - & -0.20 + 47c & wc5 & -0.16 & -0.42 & -0.05 & -0.08 + 32 & wc5+ob ? & -0.41 & -1.77 & -0.13 & -0.37 + 41 & wc5+ob ? & -1.06 & -2.52 & -0.40 & -79.86 + 23 & wc6 & -0.46 & - & -0.12 & -0.28 + 27 & wc6+a & -0.71 & -1.36 & -0.13 & -0.31 + 31c & wc6+ob & -1.04 & -1.51 & -0.19 & -0.28 + 38b & wc7+ob & -0.77 & -1.14 & -0.08 & -0.11 + 39 & wc7+ob ? & -0.31 & -0.76 & -0.13 & -0.17 + 42 & wc7+o7v & -0.36 & - & -0.11 & -0.12 + 50 & wc7+ob & -0.76 & -1.42 & -0.02 & -0.15 + 111b & wc9d & 0.12 & 0.18 & -0.11 & -0.19 + 48b & wc9d & -0.03 & -0.06 & - & -0.24 +